summaryrefslogtreecommitdiffstats
path: root/third_party/sqlite/src/ext
diff options
context:
space:
mode:
authorjhawkins@chromium.org <jhawkins@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98>2010-12-07 19:38:40 +0000
committerjhawkins@chromium.org <jhawkins@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98>2010-12-07 19:38:40 +0000
commitb104b771c129e5f5eb68bfa06ca6a2fc243b00d3 (patch)
tree00f2bc6ef916a2d4a2147654b16feb97384f72af /third_party/sqlite/src/ext
parent03d21b85e07e9d35d92867606c42e0f4bf74cf30 (diff)
downloadchromium_src-b104b771c129e5f5eb68bfa06ca6a2fc243b00d3.zip
chromium_src-b104b771c129e5f5eb68bfa06ca6a2fc243b00d3.tar.gz
chromium_src-b104b771c129e5f5eb68bfa06ca6a2fc243b00d3.tar.bz2
Update sqlite to 3.7.3.
BUG=none TEST=none Review URL: http://codereview.chromium.org/5626002 git-svn-id: svn://svn.chromium.org/chrome/trunk/src@68506 0039d316-1c4b-4281-b951-d872f2087c98
Diffstat (limited to 'third_party/sqlite/src/ext')
-rw-r--r--third_party/sqlite/src/ext/async/sqlite3async.c34
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/fts2/README.tokenizers0
-rw-r--r--third_party/sqlite/src/ext/fts2/fts2.c8
-rw-r--r--third_party/sqlite/src/ext/fts2/fts2_porter.c1
-rw-r--r--third_party/sqlite/src/ext/fts2/fts2_tokenizer1.c6
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/fts2/mkfts2amal.tcl0
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/fts3/README.tokenizers0
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3.c8537
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3Int.h326
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_expr.c58
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_expr.h96
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_hash.c84
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_hash.h36
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_porter.c33
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_snippet.c1205
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_tokenizer.c161
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_tokenizer.h4
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_tokenizer1.c18
-rw-r--r--third_party/sqlite/src/ext/fts3/fts3_write.c2622
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/fts3/mkfts3amal.tcl0
-rw-r--r--third_party/sqlite/src/ext/icu/README.txt3
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/README0
-rw-r--r--third_party/sqlite/src/ext/rtree/rtree.c914
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree1.test21
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree2.test6
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree3.test225
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree4.test24
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree5.test4
-rw-r--r--third_party/sqlite/src/ext/rtree/rtree6.test14
-rw-r--r--third_party/sqlite/src/ext/rtree/rtree7.test58
-rw-r--r--third_party/sqlite/src/ext/rtree/rtree8.test171
-rw-r--r--third_party/sqlite/src/ext/rtree/rtree9.test125
-rw-r--r--third_party/sqlite/src/ext/rtree/rtreeA.test220
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree_perf.tcl2
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/rtree_util.tcl3
-rw-r--r--third_party/sqlite/src/ext/rtree/sqlite3rtree.h56
-rw-r--r--third_party/sqlite/src/ext/rtree/tkt3363.test6
-rw-r--r--[-rwxr-xr-x]third_party/sqlite/src/ext/rtree/viewrtree.tcl1
38 files changed, 7938 insertions, 7144 deletions
diff --git a/third_party/sqlite/src/ext/async/sqlite3async.c b/third_party/sqlite/src/ext/async/sqlite3async.c
index 11a4761..a351eaa 100644
--- a/third_party/sqlite/src/ext/async/sqlite3async.c
+++ b/third_party/sqlite/src/ext/async/sqlite3async.c
@@ -667,9 +667,9 @@ static int asyncRead(
){
AsyncFileData *p = ((AsyncFile *)pFile)->pData;
int rc = SQLITE_OK;
- sqlite3_int64 filesize;
- int nRead;
+ sqlite3_int64 filesize = 0;
sqlite3_file *pBase = p->pBaseRead;
+ sqlite3_int64 iAmt64 = (sqlite3_int64)iAmt;
/* Grab the write queue mutex for the duration of the call */
async_mutex_enter(ASYNC_MUTEX_QUEUE);
@@ -683,13 +683,14 @@ static int asyncRead(
}
if( pBase->pMethods ){
+ sqlite3_int64 nRead;
rc = pBase->pMethods->xFileSize(pBase, &filesize);
if( rc!=SQLITE_OK ){
goto asyncread_out;
}
- nRead = (int)MIN(filesize - iOffset, iAmt);
+ nRead = MIN(filesize - iOffset, iAmt64);
if( nRead>0 ){
- rc = pBase->pMethods->xRead(pBase, zOut, nRead, iOffset);
+ rc = pBase->pMethods->xRead(pBase, zOut, (int)nRead, iOffset);
ASYNC_TRACE(("READ %s %d bytes at %d\n", p->zName, nRead, iOffset));
}
}
@@ -703,16 +704,24 @@ static int asyncRead(
(pWrite->pFileData==p) ||
(zName && pWrite->pFileData->zName==zName)
)){
+ sqlite3_int64 nCopy;
+ sqlite3_int64 nByte64 = (sqlite3_int64)pWrite->nByte;
+
+ /* Set variable iBeginIn to the offset in buffer pWrite->zBuf[] from
+ ** which data should be copied. Set iBeginOut to the offset within
+ ** the output buffer to which data should be copied. If either of
+ ** these offsets is a negative number, set them to 0.
+ */
sqlite3_int64 iBeginOut = (pWrite->iOffset-iOffset);
sqlite3_int64 iBeginIn = -iBeginOut;
- int nCopy;
-
if( iBeginIn<0 ) iBeginIn = 0;
if( iBeginOut<0 ) iBeginOut = 0;
- nCopy = (int)MIN(pWrite->nByte-iBeginIn, iAmt-iBeginOut);
+ filesize = MAX(filesize, pWrite->iOffset+nByte64);
+
+ nCopy = MIN(nByte64-iBeginIn, iAmt64-iBeginOut);
if( nCopy>0 ){
- memcpy(&((char *)zOut)[iBeginOut], &pWrite->zBuf[iBeginIn], nCopy);
+ memcpy(&((char *)zOut)[iBeginOut], &pWrite->zBuf[iBeginIn], (size_t)nCopy);
ASYNC_TRACE(("OVERREAD %d bytes at %d\n", nCopy, iBeginOut+iOffset));
}
}
@@ -721,6 +730,9 @@ static int asyncRead(
asyncread_out:
async_mutex_leave(ASYNC_MUTEX_QUEUE);
+ if( rc==SQLITE_OK && filesize<(iOffset+iAmt) ){
+ rc = SQLITE_IOERR_SHORT_READ;
+ }
return rc;
}
@@ -1131,7 +1143,6 @@ static int asyncOpen(
async_mutex_leave(ASYNC_MUTEX_LOCK);
if( rc==SQLITE_OK ){
- incrOpenFileCount();
pData->pLock = pLock;
}
@@ -1148,7 +1159,10 @@ static int asyncOpen(
}
if( rc!=SQLITE_OK ){
p->pMethod = 0;
+ }else{
+ incrOpenFileCount();
}
+
return rc;
}
@@ -1224,8 +1238,8 @@ static int asyncFullPathname(
*/
if( rc==SQLITE_OK ){
int i, j;
- int n = nPathOut;
char *z = zPathOut;
+ int n = (int)strlen(z);
while( n>1 && z[n-1]=='/' ){ n--; }
for(i=j=0; i<n; i++){
if( z[i]=='/' ){
diff --git a/third_party/sqlite/src/ext/fts2/README.tokenizers b/third_party/sqlite/src/ext/fts2/README.tokenizers
index 98d2021..98d2021 100755..100644
--- a/third_party/sqlite/src/ext/fts2/README.tokenizers
+++ b/third_party/sqlite/src/ext/fts2/README.tokenizers
diff --git a/third_party/sqlite/src/ext/fts2/fts2.c b/third_party/sqlite/src/ext/fts2/fts2.c
index ec4c108..5ec1265 100644
--- a/third_party/sqlite/src/ext/fts2/fts2.c
+++ b/third_party/sqlite/src/ext/fts2/fts2.c
@@ -320,8 +320,6 @@
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
-#include <ctype.h>
-
#include "fts2.h"
#include "fts2_hash.h"
#include "fts2_tokenizer.h"
@@ -369,13 +367,13 @@ static int fts2Corrupt(void){
*/
/* TODO(shess) Is __isascii() a portable version of (c&0x80)==0? */
static int safe_isspace(char c){
- return (c&0x80)==0 ? isspace(c) : 0;
+ return c==' ' || c=='\t' || c=='\n' || c=='\r' || c=='\v' || c=='\f';
}
static int safe_tolower(char c){
- return (c>='A' && c<='Z') ? (c-'A'+'a') : c;
+ return (c>='A' && c<='Z') ? (c - 'A' + 'a') : c;
}
static int safe_isalnum(char c){
- return (c&0x80)==0 ? isalnum(c) : 0;
+ return (c>='0' && c<='9') || (c>='A' && c<='Z') || (c>='a' && c<='z');
}
typedef enum DocListType {
diff --git a/third_party/sqlite/src/ext/fts2/fts2_porter.c b/third_party/sqlite/src/ext/fts2/fts2_porter.c
index 97a95c8..16620b9 100644
--- a/third_party/sqlite/src/ext/fts2/fts2_porter.c
+++ b/third_party/sqlite/src/ext/fts2/fts2_porter.c
@@ -29,7 +29,6 @@
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
-#include <ctype.h>
#include "fts2_tokenizer.h"
diff --git a/third_party/sqlite/src/ext/fts2/fts2_tokenizer1.c b/third_party/sqlite/src/ext/fts2/fts2_tokenizer1.c
index 1f03cc4..7e13366 100644
--- a/third_party/sqlite/src/ext/fts2/fts2_tokenizer1.c
+++ b/third_party/sqlite/src/ext/fts2/fts2_tokenizer1.c
@@ -29,7 +29,6 @@
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
-#include <ctype.h>
#include "fts2_tokenizer.h"
@@ -89,7 +88,8 @@ static int simpleCreate(
/* Mark non-alphanumeric ASCII characters as delimiters */
int i;
for(i=1; i<0x80; i++){
- t->delim[i] = !isalnum(i);
+ t->delim[i] = !((i>='0' && i<='9') || (i>='A' && i<='Z') ||
+ (i>='a' && i<='z'));
}
}
@@ -191,7 +191,7 @@ static int simpleNext(
** case-insensitivity.
*/
unsigned char ch = p[iStartOffset+i];
- c->pToken[i] = (ch>='A' && ch<='Z') ? (ch-'A'+'a') : ch;
+ c->pToken[i] = (ch>='A' && ch<='Z') ? (ch - 'A' + 'a') : ch;
}
*ppToken = c->pToken;
*pnBytes = n;
diff --git a/third_party/sqlite/src/ext/fts2/mkfts2amal.tcl b/third_party/sqlite/src/ext/fts2/mkfts2amal.tcl
index 5c8d1e9..5c8d1e9 100755..100644
--- a/third_party/sqlite/src/ext/fts2/mkfts2amal.tcl
+++ b/third_party/sqlite/src/ext/fts2/mkfts2amal.tcl
diff --git a/third_party/sqlite/src/ext/fts3/README.tokenizers b/third_party/sqlite/src/ext/fts3/README.tokenizers
index e06803a..e06803a 100755..100644
--- a/third_party/sqlite/src/ext/fts3/README.tokenizers
+++ b/third_party/sqlite/src/ext/fts3/README.tokenizers
diff --git a/third_party/sqlite/src/ext/fts3/fts3.c b/third_party/sqlite/src/ext/fts3/fts3.c
index 3e87a40..91f047e 100644
--- a/third_party/sqlite/src/ext/fts3/fts3.c
+++ b/third_party/sqlite/src/ext/fts3/fts3.c
@@ -23,9 +23,6 @@
** SQLite (in which case SQLITE_ENABLE_FTS3 is defined).
*/
-/* TODO(shess) Consider exporting this comment to an HTML file or the
-** wiki.
-*/
/* The full-text index is stored in a series of b+tree (-like)
** structures called segments which map terms to doclists. The
** structures are like b+trees in layout, but are constructed from the
@@ -48,30 +45,40 @@
** 21 bits - BBA
** and so on.
**
-** This is identical to how sqlite encodes varints (see util.c).
+** This is similar in concept to how sqlite encodes "varints" but
+** the encoding is not the same. SQLite varints are big-endian
+** are are limited to 9 bytes in length whereas FTS3 varints are
+** little-endian and can be up to 10 bytes in length (in theory).
+**
+** Example encodings:
+**
+** 1: 0x01
+** 127: 0x7f
+** 128: 0x81 0x00
**
**
**** Document lists ****
** A doclist (document list) holds a docid-sorted list of hits for a
-** given term. Doclists hold docids, and can optionally associate
-** token positions and offsets with docids.
+** given term. Doclists hold docids and associated token positions.
+** A docid is the unique integer identifier for a single document.
+** A position is the index of a word within the document. The first
+** word of the document has a position of 0.
+**
+** FTS3 used to optionally store character offsets using a compile-time
+** option. But that functionality is no longer supported.
**
-** A DL_POSITIONS_OFFSETS doclist is stored like this:
+** A doclist is stored like this:
**
** array {
** varint docid;
** array { (position list for column 0)
-** varint position; (delta from previous position plus POS_BASE)
-** varint startOffset; (delta from previous startOffset)
-** varint endOffset; (delta from startOffset)
+** varint position; (2 more than the delta from previous position)
** }
** array {
** varint POS_COLUMN; (marks start of position list for new column)
** varint column; (index of new column)
** array {
-** varint position; (delta from previous position plus POS_BASE)
-** varint startOffset;(delta from previous startOffset)
-** varint endOffset; (delta from startOffset)
+** varint position; (2 more than the delta from previous position)
** }
** }
** varint POS_END; (marks end of positions for this document.
@@ -79,19 +86,32 @@
**
** Here, array { X } means zero or more occurrences of X, adjacent in
** memory. A "position" is an index of a token in the token stream
-** generated by the tokenizer, while an "offset" is a byte offset,
-** both based at 0. Note that POS_END and POS_COLUMN occur in the
-** same logical place as the position element, and act as sentinals
-** ending a position list array.
-**
-** A DL_POSITIONS doclist omits the startOffset and endOffset
-** information. A DL_DOCIDS doclist omits both the position and
-** offset information, becoming an array of varint-encoded docids.
-**
-** On-disk data is stored as type DL_DEFAULT, so we don't serialize
-** the type. Due to how deletion is implemented in the segmentation
-** system, on-disk doclists MUST store at least positions.
-**
+** generated by the tokenizer. Note that POS_END and POS_COLUMN occur
+** in the same logical place as the position element, and act as sentinals
+** ending a position list array. POS_END is 0. POS_COLUMN is 1.
+** The positions numbers are not stored literally but rather as two more
+** than the difference from the prior position, or the just the position plus
+** 2 for the first position. Example:
+**
+** label: A B C D E F G H I J K
+** value: 123 5 9 1 1 14 35 0 234 72 0
+**
+** The 123 value is the first docid. For column zero in this document
+** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1
+** at D signals the start of a new column; the 1 at E indicates that the
+** new column is column number 1. There are two positions at 12 and 45
+** (14-2 and 35-2+12). The 0 at H indicate the end-of-document. The
+** 234 at I is the next docid. It has one position 72 (72-2) and then
+** terminates with the 0 at K.
+**
+** A "position-list" is the list of positions for multiple columns for
+** a single docid. A "column-list" is the set of positions for a single
+** column. Hence, a position-list consists of one or more column-lists,
+** a document record consists of a docid followed by a position-list and
+** a doclist consists of one or more document records.
+**
+** A bare doclist omits the position information, becoming an
+** array of varint-encoded docids.
**
**** Segment leaf nodes ****
** Segment leaf nodes store terms and doclists, ordered by term. Leaf
@@ -271,7 +291,6 @@
** deletions and duplications. This would basically be a forced merge
** into a single segment.
*/
-#define CHROMIUM_FTS3_CHANGES 1
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
@@ -279,41 +298,21 @@
# define SQLITE_CORE 1
#endif
+#include "fts3Int.h"
+
#include <assert.h>
#include <stdlib.h>
+#include <stddef.h>
#include <stdio.h>
#include <string.h>
-#include <ctype.h>
+#include <stdarg.h>
#include "fts3.h"
-#include "fts3_expr.h"
-#include "fts3_hash.h"
-#include "fts3_tokenizer.h"
#ifndef SQLITE_CORE
# include "sqlite3ext.h"
SQLITE_EXTENSION_INIT1
#endif
-
-/* TODO(shess) MAN, this thing needs some refactoring. At minimum, it
-** would be nice to order the file better, perhaps something along the
-** lines of:
-**
-** - utility functions
-** - table setup functions
-** - table update functions
-** - table query functions
-**
-** Put the query functions last because they're likely to reference
-** typedefs or functions from the table update section.
-*/
-
-#if 0
-# define FTSTRACE(A) printf A; fflush(stdout)
-#else
-# define FTSTRACE(A)
-#endif
-
#if 0
/* Useful to set breakpoints. See main.c sqlite3Corrupt(). */
static int fts3Corrupt(void){
@@ -324,80 +323,12 @@ static int fts3Corrupt(void){
# define SQLITE_CORRUPT_BKPT SQLITE_CORRUPT
#endif
-/* It is not safe to call isspace(), tolower(), or isalnum() on
-** hi-bit-set characters. This is the same solution used in the
-** tokenizer.
-*/
-/* TODO(shess) The snippet-generation code should be using the
-** tokenizer-generated tokens rather than doing its own local
-** tokenization.
-*/
-/* TODO(shess) Is __isascii() a portable version of (c&0x80)==0? */
-static int safe_isspace(char c){
- return (c&0x80)==0 ? isspace(c) : 0;
-}
-static int safe_tolower(char c){
- return (c>='A' && c<='Z') ? (c-'A'+'a') : c;
-}
-static int safe_isalnum(char c){
- return (c&0x80)==0 ? isalnum(c) : 0;
-}
-
-typedef enum DocListType {
- DL_DOCIDS, /* docids only */
- DL_POSITIONS, /* docids + positions */
- DL_POSITIONS_OFFSETS /* docids + positions + offsets */
-} DocListType;
-
-/*
-** By default, only positions and not offsets are stored in the doclists.
-** To change this so that offsets are stored too, compile with
-**
-** -DDL_DEFAULT=DL_POSITIONS_OFFSETS
-**
-** If DL_DEFAULT is set to DL_DOCIDS, your table can only be inserted
-** into (no deletes or updates).
-*/
-#ifndef DL_DEFAULT
-# define DL_DEFAULT DL_POSITIONS
-#endif
-
-enum {
- POS_END = 0, /* end of this position list */
- POS_COLUMN, /* followed by new column number */
- POS_BASE
-};
-
-/* MERGE_COUNT controls how often we merge segments (see comment at
-** top of file).
-*/
-#define MERGE_COUNT 16
-
-/* utility functions */
-
-/* CLEAR() and SCRAMBLE() abstract memset() on a pointer to a single
-** record to prevent errors of the form:
-**
-** my_function(SomeType *b){
-** memset(b, '\0', sizeof(b)); // sizeof(b)!=sizeof(*b)
-** }
+/*
+** Write a 64-bit variable-length integer to memory starting at p[0].
+** The length of data written will be between 1 and FTS3_VARINT_MAX bytes.
+** The number of bytes written is returned.
*/
-/* TODO(shess) Obvious candidates for a header file. */
-#define CLEAR(b) memset(b, '\0', sizeof(*(b)))
-
-#ifndef NDEBUG
-# define SCRAMBLE(b) memset(b, 0x55, sizeof(*(b)))
-#else
-# define SCRAMBLE(b)
-#endif
-
-/* We may need up to VARINT_MAX bytes to store an encoded 64-bit integer. */
-#define VARINT_MAX 10
-
-/* Write a 64-bit variable-length integer to memory starting at p[0].
- * The length of data written will be between 1 and VARINT_MAX bytes.
- * The number of bytes written is returned. */
-static int fts3PutVarint(char *p, sqlite_int64 v){
+int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){
unsigned char *q = (unsigned char *) p;
sqlite_uint64 vu = v;
do{
@@ -405,6908 +336,2328 @@ static int fts3PutVarint(char *p, sqlite_int64 v){
vu >>= 7;
}while( vu!=0 );
q[-1] &= 0x7f; /* turn off high bit in final byte */
- assert( q - (unsigned char *)p <= VARINT_MAX );
+ assert( q - (unsigned char *)p <= FTS3_VARINT_MAX );
return (int) (q - (unsigned char *)p);
}
-/* Read a 64-bit variable-length integer from memory starting at p[0].
- * Return the number of bytes read, or 0 on error.
- * The value is stored in *v. */
-static int fts3GetVarintSafe(const char *p, sqlite_int64 *v, int max){
+/*
+** Read a 64-bit variable-length integer from memory starting at p[0].
+** Return the number of bytes read, or 0 on error.
+** The value is stored in *v.
+*/
+int sqlite3Fts3GetVarint(const char *p, sqlite_int64 *v){
const unsigned char *q = (const unsigned char *) p;
sqlite_uint64 x = 0, y = 1;
- if( max>VARINT_MAX ) max = VARINT_MAX;
- while( max && (*q & 0x80) == 0x80 ){
- max--;
+ while( (*q&0x80)==0x80 && q-(unsigned char *)p<FTS3_VARINT_MAX ){
x += y * (*q++ & 0x7f);
y <<= 7;
}
- if( !max ){
- assert( 0 );
- return 0; /* tried to read too much; bad data */
- }
x += y * (*q++);
*v = (sqlite_int64) x;
return (int) (q - (unsigned char *)p);
}
-static int fts3GetVarint(const char *p, sqlite_int64 *v){
- return fts3GetVarintSafe(p, v, VARINT_MAX);
-}
-
-static int fts3GetVarint32Safe(const char *p, int *pi, int max){
+/*
+** Similar to sqlite3Fts3GetVarint(), except that the output is truncated to a
+** 32-bit integer before it is returned.
+*/
+int sqlite3Fts3GetVarint32(const char *p, int *pi){
sqlite_int64 i;
- int ret = fts3GetVarintSafe(p, &i, max);
- if( !ret ) return ret;
+ int ret = sqlite3Fts3GetVarint(p, &i);
*pi = (int) i;
- assert( *pi==i );
return ret;
}
-static int fts3GetVarint32(const char* p, int *pi){
- return fts3GetVarint32Safe(p, pi, VARINT_MAX);
-}
-
-/*******************************************************************/
-/* DataBuffer is used to collect data into a buffer in piecemeal
-** fashion. It implements the usual distinction between amount of
-** data currently stored (nData) and buffer capacity (nCapacity).
-**
-** dataBufferInit - create a buffer with given initial capacity.
-** dataBufferReset - forget buffer's data, retaining capacity.
-** dataBufferDestroy - free buffer's data.
-** dataBufferSwap - swap contents of two buffers.
-** dataBufferExpand - expand capacity without adding data.
-** dataBufferAppend - append data.
-** dataBufferAppend2 - append two pieces of data at once.
-** dataBufferReplace - replace buffer's data.
-*/
-typedef struct DataBuffer {
- char *pData; /* Pointer to malloc'ed buffer. */
- int nCapacity; /* Size of pData buffer. */
- int nData; /* End of data loaded into pData. */
-} DataBuffer;
-
-static void dataBufferInit(DataBuffer *pBuffer, int nCapacity){
- assert( nCapacity>=0 );
- pBuffer->nData = 0;
- pBuffer->nCapacity = nCapacity;
- pBuffer->pData = nCapacity==0 ? NULL : sqlite3_malloc(nCapacity);
-}
-static void dataBufferReset(DataBuffer *pBuffer){
- pBuffer->nData = 0;
-}
-static void dataBufferDestroy(DataBuffer *pBuffer){
- if( pBuffer->pData!=NULL ) sqlite3_free(pBuffer->pData);
- SCRAMBLE(pBuffer);
-}
-static void dataBufferSwap(DataBuffer *pBuffer1, DataBuffer *pBuffer2){
- DataBuffer tmp = *pBuffer1;
- *pBuffer1 = *pBuffer2;
- *pBuffer2 = tmp;
-}
-static void dataBufferExpand(DataBuffer *pBuffer, int nAddCapacity){
- assert( nAddCapacity>0 );
- /* TODO(shess) Consider expanding more aggressively. Note that the
- ** underlying malloc implementation may take care of such things for
- ** us already.
- */
- if( pBuffer->nData+nAddCapacity>pBuffer->nCapacity ){
- pBuffer->nCapacity = pBuffer->nData+nAddCapacity;
- pBuffer->pData = sqlite3_realloc(pBuffer->pData, pBuffer->nCapacity);
- }
-}
-static void dataBufferAppend(DataBuffer *pBuffer,
- const char *pSource, int nSource){
- assert( nSource>0 && pSource!=NULL );
- dataBufferExpand(pBuffer, nSource);
- memcpy(pBuffer->pData+pBuffer->nData, pSource, nSource);
- pBuffer->nData += nSource;
-}
-static void dataBufferAppend2(DataBuffer *pBuffer,
- const char *pSource1, int nSource1,
- const char *pSource2, int nSource2){
- assert( nSource1>0 && pSource1!=NULL );
- assert( nSource2>0 && pSource2!=NULL );
- dataBufferExpand(pBuffer, nSource1+nSource2);
- memcpy(pBuffer->pData+pBuffer->nData, pSource1, nSource1);
- memcpy(pBuffer->pData+pBuffer->nData+nSource1, pSource2, nSource2);
- pBuffer->nData += nSource1+nSource2;
-}
-static void dataBufferReplace(DataBuffer *pBuffer,
- const char *pSource, int nSource){
- dataBufferReset(pBuffer);
- dataBufferAppend(pBuffer, pSource, nSource);
-}
-
-/* StringBuffer is a null-terminated version of DataBuffer. */
-typedef struct StringBuffer {
- DataBuffer b; /* Includes null terminator. */
-} StringBuffer;
-
-static void initStringBuffer(StringBuffer *sb){
- dataBufferInit(&sb->b, 100);
- dataBufferReplace(&sb->b, "", 1);
-}
-static int stringBufferLength(StringBuffer *sb){
- return sb->b.nData-1;
-}
-static char *stringBufferData(StringBuffer *sb){
- return sb->b.pData;
-}
-static void stringBufferDestroy(StringBuffer *sb){
- dataBufferDestroy(&sb->b);
-}
-
-static void nappend(StringBuffer *sb, const char *zFrom, int nFrom){
- assert( sb->b.nData>0 );
- if( nFrom>0 ){
- sb->b.nData--;
- dataBufferAppend2(&sb->b, zFrom, nFrom, "", 1);
- }
-}
-static void append(StringBuffer *sb, const char *zFrom){
- nappend(sb, zFrom, strlen(zFrom));
-}
-
-/* Append a list of strings separated by commas. */
-static void appendList(StringBuffer *sb, int nString, char **azString){
- int i;
- for(i=0; i<nString; ++i){
- if( i>0 ) append(sb, ", ");
- append(sb, azString[i]);
- }
-}
-
-static int endsInWhiteSpace(StringBuffer *p){
- return stringBufferLength(p)>0 &&
- safe_isspace(stringBufferData(p)[stringBufferLength(p)-1]);
-}
-
-/* If the StringBuffer ends in something other than white space, add a
-** single space character to the end.
+/*
+** Return the number of bytes required to encode v as a varint
*/
-static void appendWhiteSpace(StringBuffer *p){
- if( stringBufferLength(p)==0 ) return;
- if( !endsInWhiteSpace(p) ) append(p, " ");
-}
-
-/* Remove white space from the end of the StringBuffer */
-static void trimWhiteSpace(StringBuffer *p){
- while( endsInWhiteSpace(p) ){
- p->b.pData[--p->b.nData-1] = '\0';
- }
+int sqlite3Fts3VarintLen(sqlite3_uint64 v){
+ int i = 0;
+ do{
+ i++;
+ v >>= 7;
+ }while( v!=0 );
+ return i;
}
-/*******************************************************************/
-/* DLReader is used to read document elements from a doclist. The
-** current docid is cached, so dlrDocid() is fast. DLReader does not
-** own the doclist buffer.
+/*
+** Convert an SQL-style quoted string into a normal string by removing
+** the quote characters. The conversion is done in-place. If the
+** input does not begin with a quote character, then this routine
+** is a no-op.
**
-** dlrAtEnd - true if there's no more data to read.
-** dlrDocid - docid of current document.
-** dlrDocData - doclist data for current document (including docid).
-** dlrDocDataBytes - length of same.
-** dlrAllDataBytes - length of all remaining data.
-** dlrPosData - position data for current document.
-** dlrPosDataLen - length of pos data for current document (incl POS_END).
-** dlrStep - step to current document.
-** dlrInit - initial for doclist of given type against given data.
-** dlrDestroy - clean up.
+** Examples:
**
-** Expected usage is something like:
+** "abc" becomes abc
+** 'xyz' becomes xyz
+** [pqr] becomes pqr
+** `mno` becomes mno
**
-** DLReader reader;
-** dlrInit(&reader, pData, nData);
-** while( !dlrAtEnd(&reader) ){
-** // calls to dlrDocid() and kin.
-** dlrStep(&reader);
-** }
-** dlrDestroy(&reader);
-*/
-typedef struct DLReader {
- DocListType iType;
- const char *pData;
- int nData;
-
- sqlite_int64 iDocid;
- int nElement;
-} DLReader;
-
-static int dlrAtEnd(DLReader *pReader){
- assert( pReader->nData>=0 );
- return pReader->nData<=0;
-}
-static sqlite_int64 dlrDocid(DLReader *pReader){
- assert( !dlrAtEnd(pReader) );
- return pReader->iDocid;
-}
-static const char *dlrDocData(DLReader *pReader){
- assert( !dlrAtEnd(pReader) );
- return pReader->pData;
-}
-static int dlrDocDataBytes(DLReader *pReader){
- assert( !dlrAtEnd(pReader) );
- return pReader->nElement;
-}
-static int dlrAllDataBytes(DLReader *pReader){
- assert( !dlrAtEnd(pReader) );
- return pReader->nData;
-}
-/* TODO(shess) Consider adding a field to track iDocid varint length
-** to make these two functions faster. This might matter (a tiny bit)
-** for queries.
*/
-static const char *dlrPosData(DLReader *pReader){
- sqlite_int64 iDummy;
- int n = fts3GetVarintSafe(pReader->pData, &iDummy, pReader->nElement);
- if( !n ) return NULL;
- assert( !dlrAtEnd(pReader) );
- return pReader->pData+n;
-}
-static int dlrPosDataLen(DLReader *pReader){
- sqlite_int64 iDummy;
- int n = fts3GetVarint(pReader->pData, &iDummy);
- assert( !dlrAtEnd(pReader) );
- return pReader->nElement-n;
-}
-static int dlrStep(DLReader *pReader){
- assert( !dlrAtEnd(pReader) );
-
- /* Skip past current doclist element. */
- assert( pReader->nElement<=pReader->nData );
- pReader->pData += pReader->nElement;
- pReader->nData -= pReader->nElement;
-
- /* If there is more data, read the next doclist element. */
- if( pReader->nData>0 ){
- sqlite_int64 iDocidDelta;
- int nTotal = 0;
- int iDummy, n = fts3GetVarintSafe(pReader->pData, &iDocidDelta, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- pReader->iDocid += iDocidDelta;
- if( pReader->iType>=DL_POSITIONS ){
- while( 1 ){
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- if( iDummy==POS_END ) break;
- if( iDummy==POS_COLUMN ){
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- }else if( pReader->iType==DL_POSITIONS_OFFSETS ){
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- }
- }
- }
- pReader->nElement = nTotal;
- assert( pReader->nElement<=pReader->nData );
- }
- return SQLITE_OK;
-}
-static void dlrDestroy(DLReader *pReader){
- SCRAMBLE(pReader);
-}
-static int dlrInit(DLReader *pReader, DocListType iType,
- const char *pData, int nData){
- int rc;
- assert( pData!=NULL && nData!=0 );
- pReader->iType = iType;
- pReader->pData = pData;
- pReader->nData = nData;
- pReader->nElement = 0;
- pReader->iDocid = 0;
-
- /* Load the first element's data. There must be a first element. */
- rc = dlrStep(pReader);
- if( rc!=SQLITE_OK ) dlrDestroy(pReader);
- return rc;
-}
+void sqlite3Fts3Dequote(char *z){
+ char quote; /* Quote character (if any ) */
-#ifndef NDEBUG
-/* Verify that the doclist can be validly decoded. Also returns the
-** last docid found because it is convenient in other assertions for
-** DLWriter.
-*/
-static void docListValidate(DocListType iType, const char *pData, int nData,
- sqlite_int64 *pLastDocid){
- sqlite_int64 iPrevDocid = 0;
- assert( nData>0 );
- assert( pData!=0 );
- assert( pData+nData>pData );
- while( nData!=0 ){
- sqlite_int64 iDocidDelta;
- int n = fts3GetVarint(pData, &iDocidDelta);
- iPrevDocid += iDocidDelta;
- if( iType>DL_DOCIDS ){
- int iDummy;
- while( 1 ){
- n += fts3GetVarint32(pData+n, &iDummy);
- if( iDummy==POS_END ) break;
- if( iDummy==POS_COLUMN ){
- n += fts3GetVarint32(pData+n, &iDummy);
- }else if( iType>DL_POSITIONS ){
- n += fts3GetVarint32(pData+n, &iDummy);
- n += fts3GetVarint32(pData+n, &iDummy);
- }
- assert( n<=nData );
+ quote = z[0];
+ if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){
+ int iIn = 1; /* Index of next byte to read from input */
+ int iOut = 0; /* Index of next byte to write to output */
+
+ /* If the first byte was a '[', then the close-quote character is a ']' */
+ if( quote=='[' ) quote = ']';
+
+ while( ALWAYS(z[iIn]) ){
+ if( z[iIn]==quote ){
+ if( z[iIn+1]!=quote ) break;
+ z[iOut++] = quote;
+ iIn += 2;
+ }else{
+ z[iOut++] = z[iIn++];
}
}
- assert( n<=nData );
- pData += n;
- nData -= n;
+ z[iOut] = '\0';
}
- if( pLastDocid ) *pLastDocid = iPrevDocid;
}
-#define ASSERT_VALID_DOCLIST(i, p, n, o) docListValidate(i, p, n, o)
-#else
-#define ASSERT_VALID_DOCLIST(i, p, n, o) assert( 1 )
-#endif
-/*******************************************************************/
-/* DLWriter is used to write doclist data to a DataBuffer. DLWriter
-** always appends to the buffer and does not own it.
-**
-** dlwInit - initialize to write a given type doclistto a buffer.
-** dlwDestroy - clear the writer's memory. Does not free buffer.
-** dlwAppend - append raw doclist data to buffer.
-** dlwCopy - copy next doclist from reader to writer.
-** dlwAdd - construct doclist element and append to buffer.
-** Only apply dlwAdd() to DL_DOCIDS doclists (else use PLWriter).
+/*
+** Read a single varint from the doclist at *pp and advance *pp to point
+** to the first byte past the end of the varint. Add the value of the varint
+** to *pVal.
*/
-typedef struct DLWriter {
- DocListType iType;
- DataBuffer *b;
- sqlite_int64 iPrevDocid;
-#ifndef NDEBUG
- int has_iPrevDocid;
-#endif
-} DLWriter;
-
-static void dlwInit(DLWriter *pWriter, DocListType iType, DataBuffer *b){
- pWriter->b = b;
- pWriter->iType = iType;
- pWriter->iPrevDocid = 0;
-#ifndef NDEBUG
- pWriter->has_iPrevDocid = 0;
-#endif
-}
-static void dlwDestroy(DLWriter *pWriter){
- SCRAMBLE(pWriter);
+static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){
+ sqlite3_int64 iVal;
+ *pp += sqlite3Fts3GetVarint(*pp, &iVal);
+ *pVal += iVal;
}
-/* iFirstDocid is the first docid in the doclist in pData. It is
-** needed because pData may point within a larger doclist, in which
-** case the first item would be delta-encoded.
-**
-** iLastDocid is the final docid in the doclist in pData. It is
-** needed to create the new iPrevDocid for future delta-encoding. The
-** code could decode the passed doclist to recreate iLastDocid, but
-** the only current user (docListMerge) already has decoded this
-** information.
-*/
-/* TODO(shess) This has become just a helper for docListMerge.
-** Consider a refactor to make this cleaner.
-*/
-static int dlwAppend(DLWriter *pWriter,
- const char *pData, int nData,
- sqlite_int64 iFirstDocid, sqlite_int64 iLastDocid){
- sqlite_int64 iDocid = 0;
- char c[VARINT_MAX];
- int nFirstOld, nFirstNew; /* Old and new varint len of first docid. */
-#ifndef NDEBUG
- sqlite_int64 iLastDocidDelta;
-#endif
-
- /* Recode the initial docid as delta from iPrevDocid. */
- nFirstOld = fts3GetVarintSafe(pData, &iDocid, nData);
- if( !nFirstOld ) return SQLITE_CORRUPT_BKPT;
- assert( nFirstOld<nData || (nFirstOld==nData && pWriter->iType==DL_DOCIDS) );
- nFirstNew = fts3PutVarint(c, iFirstDocid-pWriter->iPrevDocid);
-
- /* Verify that the incoming doclist is valid AND that it ends with
- ** the expected docid. This is essential because we'll trust this
- ** docid in future delta-encoding.
- */
- ASSERT_VALID_DOCLIST(pWriter->iType, pData, nData, &iLastDocidDelta);
- assert( iLastDocid==iFirstDocid-iDocid+iLastDocidDelta );
- /* Append recoded initial docid and everything else. Rest of docids
- ** should have been delta-encoded from previous initial docid.
- */
- if( nFirstOld<nData ){
- dataBufferAppend2(pWriter->b, c, nFirstNew,
- pData+nFirstOld, nData-nFirstOld);
+/*
+** As long as *pp has not reached its end (pEnd), then do the same
+** as fts3GetDeltaVarint(): read a single varint and add it to *pVal.
+** But if we have reached the end of the varint, just set *pp=0 and
+** leave *pVal unchanged.
+*/
+static void fts3GetDeltaVarint2(char **pp, char *pEnd, sqlite3_int64 *pVal){
+ if( *pp>=pEnd ){
+ *pp = 0;
}else{
- dataBufferAppend(pWriter->b, c, nFirstNew);
+ fts3GetDeltaVarint(pp, pVal);
}
- pWriter->iPrevDocid = iLastDocid;
- return SQLITE_OK;
-}
-static int dlwCopy(DLWriter *pWriter, DLReader *pReader){
- return dlwAppend(pWriter, dlrDocData(pReader), dlrDocDataBytes(pReader),
- dlrDocid(pReader), dlrDocid(pReader));
-}
-static void dlwAdd(DLWriter *pWriter, sqlite_int64 iDocid){
- char c[VARINT_MAX];
- int n = fts3PutVarint(c, iDocid-pWriter->iPrevDocid);
-
- /* Docids must ascend. */
- assert( !pWriter->has_iPrevDocid || iDocid>pWriter->iPrevDocid );
- assert( pWriter->iType==DL_DOCIDS );
-
- dataBufferAppend(pWriter->b, c, n);
- pWriter->iPrevDocid = iDocid;
-#ifndef NDEBUG
- pWriter->has_iPrevDocid = 1;
-#endif
}
-/*******************************************************************/
-/* PLReader is used to read data from a document's position list. As
-** the caller steps through the list, data is cached so that varints
-** only need to be decoded once.
-**
-** plrInit, plrDestroy - create/destroy a reader.
-** plrColumn, plrPosition, plrStartOffset, plrEndOffset - accessors
-** plrAtEnd - at end of stream, only call plrDestroy once true.
-** plrStep - step to the next element.
+/*
+** The xDisconnect() virtual table method.
*/
-typedef struct PLReader {
- /* These refer to the next position's data. nData will reach 0 when
- ** reading the last position, so plrStep() signals EOF by setting
- ** pData to NULL.
- */
- const char *pData;
- int nData;
-
- DocListType iType;
- int iColumn; /* the last column read */
- int iPosition; /* the last position read */
- int iStartOffset; /* the last start offset read */
- int iEndOffset; /* the last end offset read */
-} PLReader;
-
-static int plrAtEnd(PLReader *pReader){
- return pReader->pData==NULL;
-}
-static int plrColumn(PLReader *pReader){
- assert( !plrAtEnd(pReader) );
- return pReader->iColumn;
-}
-static int plrPosition(PLReader *pReader){
- assert( !plrAtEnd(pReader) );
- return pReader->iPosition;
-}
-static int plrStartOffset(PLReader *pReader){
- assert( !plrAtEnd(pReader) );
- return pReader->iStartOffset;
-}
-static int plrEndOffset(PLReader *pReader){
- assert( !plrAtEnd(pReader) );
- return pReader->iEndOffset;
-}
-static int plrStep(PLReader *pReader){
- int i, n, nTotal = 0;
+static int fts3DisconnectMethod(sqlite3_vtab *pVtab){
+ Fts3Table *p = (Fts3Table *)pVtab;
+ int i;
- assert( !plrAtEnd(pReader) );
+ assert( p->nPendingData==0 );
- if( pReader->nData<=0 ){
- pReader->pData = NULL;
- return SQLITE_OK;
+ /* Free any prepared statements held */
+ for(i=0; i<SizeofArray(p->aStmt); i++){
+ sqlite3_finalize(p->aStmt[i]);
}
-
- n = fts3GetVarint32Safe(pReader->pData, &i, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- if( i==POS_COLUMN ){
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &pReader->iColumn, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- pReader->iPosition = 0;
- pReader->iStartOffset = 0;
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
+ for(i=0; i<p->nLeavesStmt; i++){
+ sqlite3_finalize(p->aLeavesStmt[i]);
}
- /* Should never see adjacent column changes. */
- assert( i!=POS_COLUMN );
+ sqlite3_free(p->zSelectLeaves);
+ sqlite3_free(p->aLeavesStmt);
- if( i==POS_END ){
- assert( nTotal<=pReader->nData );
- pReader->nData = 0;
- pReader->pData = NULL;
- return SQLITE_OK;
- }
+ /* Invoke the tokenizer destructor to free the tokenizer. */
+ p->pTokenizer->pModule->xDestroy(p->pTokenizer);
- pReader->iPosition += i-POS_BASE;
- if( pReader->iType==DL_POSITIONS_OFFSETS ){
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- pReader->iStartOffset += i;
- n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- nTotal += n;
- pReader->iEndOffset = pReader->iStartOffset+i;
- }
- assert( nTotal<=pReader->nData );
- pReader->pData += nTotal;
- pReader->nData -= nTotal;
+ sqlite3_free(p);
return SQLITE_OK;
}
-static void plrDestroy(PLReader *pReader){
- SCRAMBLE(pReader);
-}
-static int plrInit(PLReader *pReader, DLReader *pDLReader){
- int rc;
- pReader->pData = dlrPosData(pDLReader);
- pReader->nData = dlrPosDataLen(pDLReader);
- pReader->iType = pDLReader->iType;
- pReader->iColumn = 0;
- pReader->iPosition = 0;
- pReader->iStartOffset = 0;
- pReader->iEndOffset = 0;
- rc = plrStep(pReader);
- if( rc!=SQLITE_OK ) plrDestroy(pReader);
- return rc;
-}
-
-/*******************************************************************/
-/* PLWriter is used in constructing a document's position list. As a
-** convenience, if iType is DL_DOCIDS, PLWriter becomes a no-op.
-** PLWriter writes to the associated DLWriter's buffer.
-**
-** plwInit - init for writing a document's poslist.
-** plwDestroy - clear a writer.
-** plwAdd - append position and offset information.
-** plwCopy - copy next position's data from reader to writer.
-** plwTerminate - add any necessary doclist terminator.
-**
-** Calling plwAdd() after plwTerminate() may result in a corrupt
-** doclist.
-*/
-/* TODO(shess) Until we've written the second item, we can cache the
-** first item's information. Then we'd have three states:
-**
-** - initialized with docid, no positions.
-** - docid and one position.
-** - docid and multiple positions.
-**
-** Only the last state needs to actually write to dlw->b, which would
-** be an improvement in the DLCollector case.
-*/
-typedef struct PLWriter {
- DLWriter *dlw;
-
- int iColumn; /* the last column written */
- int iPos; /* the last position written */
- int iOffset; /* the last start offset written */
-} PLWriter;
-
-/* TODO(shess) In the case where the parent is reading these values
-** from a PLReader, we could optimize to a copy if that PLReader has
-** the same type as pWriter.
-*/
-static void plwAdd(PLWriter *pWriter, int iColumn, int iPos,
- int iStartOffset, int iEndOffset){
- /* Worst-case space for POS_COLUMN, iColumn, iPosDelta,
- ** iStartOffsetDelta, and iEndOffsetDelta.
- */
- char c[5*VARINT_MAX];
- int n = 0;
-
- /* Ban plwAdd() after plwTerminate(). */
- assert( pWriter->iPos!=-1 );
-
- if( pWriter->dlw->iType==DL_DOCIDS ) return;
-
- if( iColumn!=pWriter->iColumn ){
- n += fts3PutVarint(c+n, POS_COLUMN);
- n += fts3PutVarint(c+n, iColumn);
- pWriter->iColumn = iColumn;
- pWriter->iPos = 0;
- pWriter->iOffset = 0;
- }
- assert( iPos>=pWriter->iPos );
- n += fts3PutVarint(c+n, POS_BASE+(iPos-pWriter->iPos));
- pWriter->iPos = iPos;
- if( pWriter->dlw->iType==DL_POSITIONS_OFFSETS ){
- assert( iStartOffset>=pWriter->iOffset );
- n += fts3PutVarint(c+n, iStartOffset-pWriter->iOffset);
- pWriter->iOffset = iStartOffset;
- assert( iEndOffset>=iStartOffset );
- n += fts3PutVarint(c+n, iEndOffset-iStartOffset);
- }
- dataBufferAppend(pWriter->dlw->b, c, n);
-}
-static void plwCopy(PLWriter *pWriter, PLReader *pReader){
- plwAdd(pWriter, plrColumn(pReader), plrPosition(pReader),
- plrStartOffset(pReader), plrEndOffset(pReader));
-}
-static void plwInit(PLWriter *pWriter, DLWriter *dlw, sqlite_int64 iDocid){
- char c[VARINT_MAX];
- int n;
-
- pWriter->dlw = dlw;
-
- /* Docids must ascend. */
- assert( !pWriter->dlw->has_iPrevDocid || iDocid>pWriter->dlw->iPrevDocid );
- n = fts3PutVarint(c, iDocid-pWriter->dlw->iPrevDocid);
- dataBufferAppend(pWriter->dlw->b, c, n);
- pWriter->dlw->iPrevDocid = iDocid;
-#ifndef NDEBUG
- pWriter->dlw->has_iPrevDocid = 1;
-#endif
-
- pWriter->iColumn = 0;
- pWriter->iPos = 0;
- pWriter->iOffset = 0;
-}
-/* TODO(shess) Should plwDestroy() also terminate the doclist? But
-** then plwDestroy() would no longer be just a destructor, it would
-** also be doing work, which isn't consistent with the overall idiom.
-** Another option would be for plwAdd() to always append any necessary
-** terminator, so that the output is always correct. But that would
-** add incremental work to the common case with the only benefit being
-** API elegance. Punt for now.
-*/
-static void plwTerminate(PLWriter *pWriter){
- if( pWriter->dlw->iType>DL_DOCIDS ){
- char c[VARINT_MAX];
- int n = fts3PutVarint(c, POS_END);
- dataBufferAppend(pWriter->dlw->b, c, n);
- }
-#ifndef NDEBUG
- /* Mark as terminated for assert in plwAdd(). */
- pWriter->iPos = -1;
-#endif
-}
-static void plwDestroy(PLWriter *pWriter){
- SCRAMBLE(pWriter);
-}
-
-/*******************************************************************/
-/* DLCollector wraps PLWriter and DLWriter to provide a
-** dynamically-allocated doclist area to use during tokenization.
+/*
+** Construct one or more SQL statements from the format string given
+** and then evaluate those statements. The success code is writting
+** into *pRc.
**
-** dlcNew - malloc up and initialize a collector.
-** dlcDelete - destroy a collector and all contained items.
-** dlcAddPos - append position and offset information.
-** dlcAddDoclist - add the collected doclist to the given buffer.
-** dlcNext - terminate the current document and open another.
-*/
-typedef struct DLCollector {
- DataBuffer b;
- DLWriter dlw;
- PLWriter plw;
-} DLCollector;
-
-/* TODO(shess) This could also be done by calling plwTerminate() and
-** dataBufferAppend(). I tried that, expecting nominal performance
-** differences, but it seemed to pretty reliably be worth 1% to code
-** it this way. I suspect it is the incremental malloc overhead (some
-** percentage of the plwTerminate() calls will cause a realloc), so
-** this might be worth revisiting if the DataBuffer implementation
-** changes.
+** If *pRc is initially non-zero then this routine is a no-op.
*/
-static void dlcAddDoclist(DLCollector *pCollector, DataBuffer *b){
- if( pCollector->dlw.iType>DL_DOCIDS ){
- char c[VARINT_MAX];
- int n = fts3PutVarint(c, POS_END);
- dataBufferAppend2(b, pCollector->b.pData, pCollector->b.nData, c, n);
+static void fts3DbExec(
+ int *pRc, /* Success code */
+ sqlite3 *db, /* Database in which to run SQL */
+ const char *zFormat, /* Format string for SQL */
+ ... /* Arguments to the format string */
+){
+ va_list ap;
+ char *zSql;
+ if( *pRc ) return;
+ va_start(ap, zFormat);
+ zSql = sqlite3_vmprintf(zFormat, ap);
+ va_end(ap);
+ if( zSql==0 ){
+ *pRc = SQLITE_NOMEM;
}else{
- dataBufferAppend(b, pCollector->b.pData, pCollector->b.nData);
- }
-}
-static void dlcNext(DLCollector *pCollector, sqlite_int64 iDocid){
- plwTerminate(&pCollector->plw);
- plwDestroy(&pCollector->plw);
- plwInit(&pCollector->plw, &pCollector->dlw, iDocid);
-}
-static void dlcAddPos(DLCollector *pCollector, int iColumn, int iPos,
- int iStartOffset, int iEndOffset){
- plwAdd(&pCollector->plw, iColumn, iPos, iStartOffset, iEndOffset);
-}
-
-static DLCollector *dlcNew(sqlite_int64 iDocid, DocListType iType){
- DLCollector *pCollector = sqlite3_malloc(sizeof(DLCollector));
- dataBufferInit(&pCollector->b, 0);
- dlwInit(&pCollector->dlw, iType, &pCollector->b);
- plwInit(&pCollector->plw, &pCollector->dlw, iDocid);
- return pCollector;
-}
-static void dlcDelete(DLCollector *pCollector){
- plwDestroy(&pCollector->plw);
- dlwDestroy(&pCollector->dlw);
- dataBufferDestroy(&pCollector->b);
- SCRAMBLE(pCollector);
- sqlite3_free(pCollector);
-}
-
-
-/* Copy the doclist data of iType in pData/nData into *out, trimming
-** unnecessary data as we go. Only columns matching iColumn are
-** copied, all columns copied if iColumn is -1. Elements with no
-** matching columns are dropped. The output is an iOutType doclist.
-*/
-/* NOTE(shess) This code is only valid after all doclists are merged.
-** If this is run before merges, then doclist items which represent
-** deletion will be trimmed, and will thus not effect a deletion
-** during the merge.
-*/
-static int docListTrim(DocListType iType, const char *pData, int nData,
- int iColumn, DocListType iOutType, DataBuffer *out){
- DLReader dlReader;
- DLWriter dlWriter;
- int rc;
-
- assert( iOutType<=iType );
-
- rc = dlrInit(&dlReader, iType, pData, nData);
- if( rc!=SQLITE_OK ) return rc;
- dlwInit(&dlWriter, iOutType, out);
-
- while( !dlrAtEnd(&dlReader) ){
- PLReader plReader;
- PLWriter plWriter;
- int match = 0;
-
- rc = plrInit(&plReader, &dlReader);
- if( rc!=SQLITE_OK ) break;
-
- while( !plrAtEnd(&plReader) ){
- if( iColumn==-1 || plrColumn(&plReader)==iColumn ){
- if( !match ){
- plwInit(&plWriter, &dlWriter, dlrDocid(&dlReader));
- match = 1;
- }
- plwAdd(&plWriter, plrColumn(&plReader), plrPosition(&plReader),
- plrStartOffset(&plReader), plrEndOffset(&plReader));
- }
- rc = plrStep(&plReader);
- if( rc!=SQLITE_OK ){
- plrDestroy(&plReader);
- goto err;
- }
- }
- if( match ){
- plwTerminate(&plWriter);
- plwDestroy(&plWriter);
- }
-
- plrDestroy(&plReader);
- rc = dlrStep(&dlReader);
- if( rc!=SQLITE_OK ) break;
+ *pRc = sqlite3_exec(db, zSql, 0, 0, 0);
+ sqlite3_free(zSql);
}
-err:
- dlwDestroy(&dlWriter);
- dlrDestroy(&dlReader);
- return rc;
}
-/* Used by docListMerge() to keep doclists in the ascending order by
-** docid, then ascending order by age (so the newest comes first).
-*/
-typedef struct OrderedDLReader {
- DLReader *pReader;
-
- /* TODO(shess) If we assume that docListMerge pReaders is ordered by
- ** age (which we do), then we could use pReader comparisons to break
- ** ties.
+/*
+** The xDestroy() virtual table method.
+*/
+static int fts3DestroyMethod(sqlite3_vtab *pVtab){
+ int rc = SQLITE_OK; /* Return code */
+ Fts3Table *p = (Fts3Table *)pVtab;
+ sqlite3 *db = p->db;
+
+ /* Drop the shadow tables */
+ fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_content'", p->zDb, p->zName);
+ fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segments'", p->zDb,p->zName);
+ fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segdir'", p->zDb, p->zName);
+ fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_docsize'", p->zDb, p->zName);
+ fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_stat'", p->zDb, p->zName);
+
+ /* If everything has worked, invoke fts3DisconnectMethod() to free the
+ ** memory associated with the Fts3Table structure and return SQLITE_OK.
+ ** Otherwise, return an SQLite error code.
*/
- int idx;
-} OrderedDLReader;
-
-/* Order eof to end, then by docid asc, idx desc. */
-static int orderedDLReaderCmp(OrderedDLReader *r1, OrderedDLReader *r2){
- if( dlrAtEnd(r1->pReader) ){
- if( dlrAtEnd(r2->pReader) ) return 0; /* Both atEnd(). */
- return 1; /* Only r1 atEnd(). */
- }
- if( dlrAtEnd(r2->pReader) ) return -1; /* Only r2 atEnd(). */
-
- if( dlrDocid(r1->pReader)<dlrDocid(r2->pReader) ) return -1;
- if( dlrDocid(r1->pReader)>dlrDocid(r2->pReader) ) return 1;
-
- /* Descending on idx. */
- return r2->idx-r1->idx;
+ return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc);
}
-/* Bubble p[0] to appropriate place in p[1..n-1]. Assumes that
-** p[1..n-1] is already sorted.
-*/
-/* TODO(shess) Is this frequent enough to warrant a binary search?
-** Before implementing that, instrument the code to check. In most
-** current usage, I expect that p[0] will be less than p[1] a very
-** high proportion of the time.
-*/
-static void orderedDLReaderReorder(OrderedDLReader *p, int n){
- while( n>1 && orderedDLReaderCmp(p, p+1)>0 ){
- OrderedDLReader tmp = p[0];
- p[0] = p[1];
- p[1] = tmp;
- n--;
- p++;
- }
-}
-/* Given an array of doclist readers, merge their doclist elements
-** into out in sorted order (by docid), dropping elements from older
-** readers when there is a duplicate docid. pReaders is assumed to be
-** ordered by age, oldest first.
-*/
-/* TODO(shess) nReaders must be <= MERGE_COUNT. This should probably
-** be fixed.
+/*
+** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table
+** passed as the first argument. This is done as part of the xConnect()
+** and xCreate() methods.
*/
-static int docListMerge(DataBuffer *out,
- DLReader *pReaders, int nReaders){
- OrderedDLReader readers[MERGE_COUNT];
- DLWriter writer;
- int i, n;
- const char *pStart = 0;
- int nStart = 0;
- sqlite_int64 iFirstDocid = 0, iLastDocid = 0;
- int rc = SQLITE_OK;
+static int fts3DeclareVtab(Fts3Table *p){
+ int i; /* Iterator variable */
+ int rc; /* Return code */
+ char *zSql; /* SQL statement passed to declare_vtab() */
+ char *zCols; /* List of user defined columns */
- assert( nReaders>0 );
- if( nReaders==1 ){
- dataBufferAppend(out, dlrDocData(pReaders), dlrAllDataBytes(pReaders));
- return SQLITE_OK;
+ /* Create a list of user columns for the virtual table */
+ zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]);
+ for(i=1; zCols && i<p->nColumn; i++){
+ zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]);
}
- assert( nReaders<=MERGE_COUNT );
- n = 0;
- for(i=0; i<nReaders; i++){
- assert( pReaders[i].iType==pReaders[0].iType );
- readers[i].pReader = pReaders+i;
- readers[i].idx = i;
- n += dlrAllDataBytes(&pReaders[i]);
- }
- /* Conservatively size output to sum of inputs. Output should end
- ** up strictly smaller than input.
- */
- dataBufferExpand(out, n);
-
- /* Get the readers into sorted order. */
- while( i-->0 ){
- orderedDLReaderReorder(readers+i, nReaders-i);
- }
-
- dlwInit(&writer, pReaders[0].iType, out);
- while( !dlrAtEnd(readers[0].pReader) ){
- sqlite_int64 iDocid = dlrDocid(readers[0].pReader);
-
- /* If this is a continuation of the current buffer to copy, extend
- ** that buffer. memcpy() seems to be more efficient if it has a
- ** lots of data to copy.
- */
- if( dlrDocData(readers[0].pReader)==pStart+nStart ){
- nStart += dlrDocDataBytes(readers[0].pReader);
- }else{
- if( pStart!=0 ){
- rc = dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid);
- if( rc!=SQLITE_OK ) goto err;
- }
- pStart = dlrDocData(readers[0].pReader);
- nStart = dlrDocDataBytes(readers[0].pReader);
- iFirstDocid = iDocid;
- }
- iLastDocid = iDocid;
- rc = dlrStep(readers[0].pReader);
- if( rc!= SQLITE_OK ) goto err;
-
- /* Drop all of the older elements with the same docid. */
- for(i=1; i<nReaders &&
- !dlrAtEnd(readers[i].pReader) &&
- dlrDocid(readers[i].pReader)==iDocid; i++){
- rc = dlrStep(readers[i].pReader);
- if( rc!=SQLITE_OK ) goto err;
- }
+ /* Create the whole "CREATE TABLE" statement to pass to SQLite */
+ zSql = sqlite3_mprintf(
+ "CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN)", zCols, p->zName
+ );
- /* Get the readers back into order. */
- while( i-->0 ){
- orderedDLReaderReorder(readers+i, nReaders-i);
- }
+ if( !zCols || !zSql ){
+ rc = SQLITE_NOMEM;
+ }else{
+ rc = sqlite3_declare_vtab(p->db, zSql);
}
- /* Copy over any remaining elements. */
- if( nStart>0 ) rc = dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid);
-err:
- dlwDestroy(&writer);
+ sqlite3_free(zSql);
+ sqlite3_free(zCols);
return rc;
}
-/* Helper function for posListUnion(). Compares the current position
-** between left and right, returning as standard C idiom of <0 if
-** left<right, >0 if left>right, and 0 if left==right. "End" always
-** compares greater.
-*/
-static int posListCmp(PLReader *pLeft, PLReader *pRight){
- assert( pLeft->iType==pRight->iType );
- if( pLeft->iType==DL_DOCIDS ) return 0;
-
- if( plrAtEnd(pLeft) ) return plrAtEnd(pRight) ? 0 : 1;
- if( plrAtEnd(pRight) ) return -1;
-
- if( plrColumn(pLeft)<plrColumn(pRight) ) return -1;
- if( plrColumn(pLeft)>plrColumn(pRight) ) return 1;
-
- if( plrPosition(pLeft)<plrPosition(pRight) ) return -1;
- if( plrPosition(pLeft)>plrPosition(pRight) ) return 1;
- if( pLeft->iType==DL_POSITIONS ) return 0;
-
- if( plrStartOffset(pLeft)<plrStartOffset(pRight) ) return -1;
- if( plrStartOffset(pLeft)>plrStartOffset(pRight) ) return 1;
-
- if( plrEndOffset(pLeft)<plrEndOffset(pRight) ) return -1;
- if( plrEndOffset(pLeft)>plrEndOffset(pRight) ) return 1;
-
- return 0;
-}
-
-/* Write the union of position lists in pLeft and pRight to pOut.
-** "Union" in this case meaning "All unique position tuples". Should
-** work with any doclist type, though both inputs and the output
-** should be the same type.
-*/
-static int posListUnion(DLReader *pLeft, DLReader *pRight, DLWriter *pOut){
- PLReader left, right;
- PLWriter writer;
- int rc;
-
- assert( dlrDocid(pLeft)==dlrDocid(pRight) );
- assert( pLeft->iType==pRight->iType );
- assert( pLeft->iType==pOut->iType );
-
- rc = plrInit(&left, pLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = plrInit(&right, pRight);
- if( rc!=SQLITE_OK ){
- plrDestroy(&left);
- return rc;
+/*
+** Create the backing store tables (%_content, %_segments and %_segdir)
+** required by the FTS3 table passed as the only argument. This is done
+** as part of the vtab xCreate() method.
+**
+** If the p->bHasDocsize boolean is true (indicating that this is an
+** FTS4 table, not an FTS3 table) then also create the %_docsize and
+** %_stat tables required by FTS4.
+*/
+static int fts3CreateTables(Fts3Table *p){
+ int rc = SQLITE_OK; /* Return code */
+ int i; /* Iterator variable */
+ char *zContentCols; /* Columns of %_content table */
+ sqlite3 *db = p->db; /* The database connection */
+
+ /* Create a list of user columns for the content table */
+ if( p->bHasContent ){
+ zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY");
+ for(i=0; zContentCols && i<p->nColumn; i++){
+ char *z = p->azColumn[i];
+ zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z);
+ }
+ if( zContentCols==0 ) rc = SQLITE_NOMEM;
+
+ /* Create the content table */
+ fts3DbExec(&rc, db,
+ "CREATE TABLE %Q.'%q_content'(%s)",
+ p->zDb, p->zName, zContentCols
+ );
+ sqlite3_free(zContentCols);
}
- plwInit(&writer, pOut, dlrDocid(pLeft));
-
- while( !plrAtEnd(&left) || !plrAtEnd(&right) ){
- int c = posListCmp(&left, &right);
- if( c<0 ){
- plwCopy(&writer, &left);
- rc = plrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( c>0 ){
- plwCopy(&writer, &right);
- rc = plrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- plwCopy(&writer, &left);
- rc = plrStep(&left);
- if( rc!=SQLITE_OK ) break;
- rc = plrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }
+ /* Create other tables */
+ fts3DbExec(&rc, db,
+ "CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);",
+ p->zDb, p->zName
+ );
+ fts3DbExec(&rc, db,
+ "CREATE TABLE %Q.'%q_segdir'("
+ "level INTEGER,"
+ "idx INTEGER,"
+ "start_block INTEGER,"
+ "leaves_end_block INTEGER,"
+ "end_block INTEGER,"
+ "root BLOB,"
+ "PRIMARY KEY(level, idx)"
+ ");",
+ p->zDb, p->zName
+ );
+ if( p->bHasDocsize ){
+ fts3DbExec(&rc, db,
+ "CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);",
+ p->zDb, p->zName
+ );
+ fts3DbExec(&rc, db,
+ "CREATE TABLE %Q.'%q_stat'(id INTEGER PRIMARY KEY, value BLOB);",
+ p->zDb, p->zName
+ );
}
-
- plwTerminate(&writer);
- plwDestroy(&writer);
- plrDestroy(&left);
- plrDestroy(&right);
return rc;
}
-/* Write the union of doclists in pLeft and pRight to pOut. For
-** docids in common between the inputs, the union of the position
-** lists is written. Inputs and outputs are always type DL_DEFAULT.
+/*
+** An sqlite3_exec() callback for fts3TableExists.
*/
-static int docListUnion(
- const char *pLeft, int nLeft,
- const char *pRight, int nRight,
- DataBuffer *pOut /* Write the combined doclist here */
-){
- DLReader left, right;
- DLWriter writer;
- int rc;
-
- if( nLeft==0 ){
- if( nRight!=0) dataBufferAppend(pOut, pRight, nRight);
- return SQLITE_OK;
- }
- if( nRight==0 ){
- dataBufferAppend(pOut, pLeft, nLeft);
- return SQLITE_OK;
- }
-
- rc = dlrInit(&left, DL_DEFAULT, pLeft, nLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrInit(&right, DL_DEFAULT, pRight, nRight);
- if( rc!=SQLITE_OK){
- dlrDestroy(&left);
- return rc;
- }
- dlwInit(&writer, DL_DEFAULT, pOut);
-
- while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){
- if( dlrAtEnd(&right) ){
- rc = dlwCopy(&writer, &left);
- if( rc!=SQLITE_OK) break;
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK) break;
- }else if( dlrAtEnd(&left) ){
- rc = dlwCopy(&writer, &right);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrDocid(&left)<dlrDocid(&right) ){
- rc = dlwCopy(&writer, &left);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrDocid(&left)>dlrDocid(&right) ){
- rc = dlwCopy(&writer, &right);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- rc = posListUnion(&left, &right, &writer);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }
- }
-
- dlrDestroy(&left);
- dlrDestroy(&right);
- dlwDestroy(&writer);
- return rc;
+static int fts3TableExistsCallback(void *pArg, int n, char **pp1, char **pp2){
+ UNUSED_PARAMETER(n);
+ UNUSED_PARAMETER(pp1);
+ UNUSED_PARAMETER(pp2);
+ *(int*)pArg = 1;
+ return 1;
}
-/*
-** This function is used as part of the implementation of phrase and
-** NEAR matching.
-**
-** pLeft and pRight are DLReaders positioned to the same docid in
-** lists of type DL_POSITION. This function writes an entry to the
-** DLWriter pOut for each position in pRight that is less than
-** (nNear+1) greater (but not equal to or smaller) than a position
-** in pLeft. For example, if nNear is 0, and the positions contained
-** by pLeft and pRight are:
-**
-** pLeft: 5 10 15 20
-** pRight: 6 9 17 21
-**
-** then the docid is added to pOut. If pOut is of type DL_POSITIONS,
-** then a positionids "6" and "21" are also added to pOut.
-**
-** If boolean argument isSaveLeft is true, then positionids are copied
-** from pLeft instead of pRight. In the example above, the positions "5"
-** and "20" would be added instead of "6" and "21".
-*/
-static int posListPhraseMerge(
- DLReader *pLeft,
- DLReader *pRight,
- int nNear,
- int isSaveLeft,
- DLWriter *pOut
+/*
+** Determine if a table currently exists in the database.
+*/
+static void fts3TableExists(
+ int *pRc, /* Success code */
+ sqlite3 *db, /* The database connection to test */
+ const char *zDb, /* ATTACHed database within the connection */
+ const char *zName, /* Name of the FTS3 table */
+ const char *zSuffix, /* Shadow table extension */
+ u8 *pResult /* Write results here */
){
- PLReader left, right;
- PLWriter writer;
- int match = 0;
- int rc;
-
- assert( dlrDocid(pLeft)==dlrDocid(pRight) );
- assert( pOut->iType!=DL_POSITIONS_OFFSETS );
-
- rc = plrInit(&left, pLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = plrInit(&right, pRight);
- if( rc!=SQLITE_OK ){
- plrDestroy(&left);
- return rc;
- }
-
- while( !plrAtEnd(&left) && !plrAtEnd(&right) ){
- if( plrColumn(&left)<plrColumn(&right) ){
- rc = plrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( plrColumn(&left)>plrColumn(&right) ){
- rc = plrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else if( plrPosition(&left)>=plrPosition(&right) ){
- rc = plrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- if( (plrPosition(&right)-plrPosition(&left))<=(nNear+1) ){
- if( !match ){
- plwInit(&writer, pOut, dlrDocid(pLeft));
- match = 1;
- }
- if( !isSaveLeft ){
- plwAdd(&writer, plrColumn(&right), plrPosition(&right), 0, 0);
- }else{
- plwAdd(&writer, plrColumn(&left), plrPosition(&left), 0, 0);
- }
- rc = plrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- rc = plrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }
- }
- }
-
- if( match ){
- plwTerminate(&writer);
- plwDestroy(&writer);
- }
-
- plrDestroy(&left);
- plrDestroy(&right);
- return rc;
+ int rc = SQLITE_OK;
+ int res = 0;
+ char *zSql;
+ if( *pRc ) return;
+ zSql = sqlite3_mprintf(
+ "SELECT 1 FROM %Q.sqlite_master WHERE name='%q%s'",
+ zDb, zName, zSuffix
+ );
+ rc = sqlite3_exec(db, zSql, fts3TableExistsCallback, &res, 0);
+ sqlite3_free(zSql);
+ *pResult = (u8)(res & 0xff);
+ if( rc!=SQLITE_ABORT ) *pRc = rc;
}
/*
-** Compare the values pointed to by the PLReaders passed as arguments.
-** Return -1 if the value pointed to by pLeft is considered less than
-** the value pointed to by pRight, +1 if it is considered greater
-** than it, or 0 if it is equal. i.e.
-**
-** (*pLeft - *pRight)
-**
-** A PLReader that is in the EOF condition is considered greater than
-** any other. If neither argument is in EOF state, the return value of
-** plrColumn() is used. If the plrColumn() values are equal, the
-** comparison is on the basis of plrPosition().
-*/
-static int plrCompare(PLReader *pLeft, PLReader *pRight){
- assert(!plrAtEnd(pLeft) || !plrAtEnd(pRight));
-
- if( plrAtEnd(pRight) || plrAtEnd(pLeft) ){
- return (plrAtEnd(pRight) ? -1 : 1);
- }
- if( plrColumn(pLeft)!=plrColumn(pRight) ){
- return ((plrColumn(pLeft)<plrColumn(pRight)) ? -1 : 1);
- }
- if( plrPosition(pLeft)!=plrPosition(pRight) ){
- return ((plrPosition(pLeft)<plrPosition(pRight)) ? -1 : 1);
- }
- return 0;
-}
-
-/* We have two doclists with positions: pLeft and pRight. Depending
-** on the value of the nNear parameter, perform either a phrase
-** intersection (if nNear==0) or a NEAR intersection (if nNear>0)
-** and write the results into pOut.
-**
-** A phrase intersection means that two documents only match
-** if pLeft.iPos+1==pRight.iPos.
-**
-** A NEAR intersection means that two documents only match if
-** (abs(pLeft.iPos-pRight.iPos)<nNear).
-**
-** If a NEAR intersection is requested, then the nPhrase argument should
-** be passed the number of tokens in the two operands to the NEAR operator
-** combined. For example:
-**
-** Query syntax nPhrase
-** ------------------------------------
-** "A B C" NEAR "D E" 5
-** A NEAR B 2
-**
-** iType controls the type of data written to pOut. If iType is
-** DL_POSITIONS, the positions are those from pRight.
-*/
-static int docListPhraseMerge(
- const char *pLeft, int nLeft,
- const char *pRight, int nRight,
- int nNear, /* 0 for a phrase merge, non-zero for a NEAR merge */
- int nPhrase, /* Number of tokens in left+right operands to NEAR */
- DocListType iType, /* Type of doclist to write to pOut */
- DataBuffer *pOut /* Write the combined doclist here */
+** This function is the implementation of both the xConnect and xCreate
+** methods of the FTS3 virtual table.
+**
+** The argv[] array contains the following:
+**
+** argv[0] -> module name ("fts3" or "fts4")
+** argv[1] -> database name
+** argv[2] -> table name
+** argv[...] -> "column name" and other module argument fields.
+*/
+static int fts3InitVtab(
+ int isCreate, /* True for xCreate, false for xConnect */
+ sqlite3 *db, /* The SQLite database connection */
+ void *pAux, /* Hash table containing tokenizers */
+ int argc, /* Number of elements in argv array */
+ const char * const *argv, /* xCreate/xConnect argument array */
+ sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */
+ char **pzErr /* Write any error message here */
){
- DLReader left, right;
- DLWriter writer;
- int rc;
-
- /* These two buffers are used in the 'while', but are declared here
- ** to simplify error-handling.
- */
- DataBuffer one = {0, 0, 0};
- DataBuffer two = {0, 0, 0};
-
- if( nLeft==0 || nRight==0 ) return SQLITE_OK;
-
- assert( iType!=DL_POSITIONS_OFFSETS );
-
- rc = dlrInit(&left, DL_POSITIONS, pLeft, nLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrInit(&right, DL_POSITIONS, pRight, nRight);
- if( rc!=SQLITE_OK ){
- dlrDestroy(&left);
- return rc;
- }
- dlwInit(&writer, iType, pOut);
-
- while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){
- if( dlrDocid(&left)<dlrDocid(&right) ){
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) goto err;
- }else if( dlrDocid(&right)<dlrDocid(&left) ){
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) goto err;
- }else{
- if( nNear==0 ){
- rc = posListPhraseMerge(&left, &right, 0, 0, &writer);
- if( rc!=SQLITE_OK ) goto err;
- }else{
- /* This case occurs when two terms (simple terms or phrases) are
- * connected by a NEAR operator, span (nNear+1). i.e.
- *
- * '"terrible company" NEAR widget'
- */
- DLWriter dlwriter2;
- DLReader dr1 = {0, 0, 0, 0, 0};
- DLReader dr2 = {0, 0, 0, 0, 0};
-
- dlwInit(&dlwriter2, iType, &one);
- rc = posListPhraseMerge(&right, &left, nNear-3+nPhrase, 1, &dlwriter2);
- if( rc!=SQLITE_OK ) goto err;
- dlwInit(&dlwriter2, iType, &two);
- rc = posListPhraseMerge(&left, &right, nNear-1, 0, &dlwriter2);
- if( rc!=SQLITE_OK ) goto err;
-
- if( one.nData){
- rc = dlrInit(&dr1, iType, one.pData, one.nData);
- if( rc!=SQLITE_OK ) goto err;
- }
- if( two.nData){
- rc = dlrInit(&dr2, iType, two.pData, two.nData);
- if( rc!=SQLITE_OK ) goto err;
- }
-
- if( !dlrAtEnd(&dr1) || !dlrAtEnd(&dr2) ){
- PLReader pr1 = {0};
- PLReader pr2 = {0};
-
- PLWriter plwriter;
- plwInit(&plwriter, &writer, dlrDocid(dlrAtEnd(&dr1)?&dr2:&dr1));
-
- if( one.nData ){
- rc = plrInit(&pr1, &dr1);
- if( rc!=SQLITE_OK ) goto err;
- }
- if( two.nData ){
- rc = plrInit(&pr2, &dr2);
- if( rc!=SQLITE_OK ) goto err;
- }
- while( !plrAtEnd(&pr1) || !plrAtEnd(&pr2) ){
- int iCompare = plrCompare(&pr1, &pr2);
- switch( iCompare ){
- case -1:
- plwCopy(&plwriter, &pr1);
- rc = plrStep(&pr1);
- if( rc!=SQLITE_OK ) goto err;
- break;
- case 1:
- plwCopy(&plwriter, &pr2);
- rc = plrStep(&pr2);
- if( rc!=SQLITE_OK ) goto err;
- break;
- case 0:
- plwCopy(&plwriter, &pr1);
- rc = plrStep(&pr1);
- if( rc!=SQLITE_OK ) goto err;
- rc = plrStep(&pr2);
- if( rc!=SQLITE_OK ) goto err;
- break;
- }
- }
- plwTerminate(&plwriter);
- }
- dataBufferReset(&one);
- dataBufferReset(&two);
- }
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) goto err;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) goto err;
+ Fts3Hash *pHash = (Fts3Hash *)pAux;
+ Fts3Table *p; /* Pointer to allocated vtab */
+ int rc; /* Return code */
+ int i; /* Iterator variable */
+ int nByte; /* Size of allocation used for *p */
+ int iCol; /* Column index */
+ int nString = 0; /* Bytes required to hold all column names */
+ int nCol = 0; /* Number of columns in the FTS table */
+ char *zCsr; /* Space for holding column names */
+ int nDb; /* Bytes required to hold database name */
+ int nName; /* Bytes required to hold table name */
+
+ const char *zTokenizer = 0; /* Name of tokenizer to use */
+ sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */
+
+ nDb = (int)strlen(argv[1]) + 1;
+ nName = (int)strlen(argv[2]) + 1;
+ for(i=3; i<argc; i++){
+ char const *z = argv[i];
+ rc = sqlite3Fts3InitTokenizer(pHash, z, &pTokenizer, &zTokenizer, pzErr);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ if( z!=zTokenizer ){
+ nString += (int)(strlen(z) + 1);
}
}
-
-err:
- dataBufferDestroy(&one);
- dataBufferDestroy(&two);
- dlrDestroy(&left);
- dlrDestroy(&right);
- dlwDestroy(&writer);
- return rc;
-}
-
-/* We have two DL_DOCIDS doclists: pLeft and pRight.
-** Write the intersection of these two doclists into pOut as a
-** DL_DOCIDS doclist.
-*/
-static int docListAndMerge(
- const char *pLeft, int nLeft,
- const char *pRight, int nRight,
- DataBuffer *pOut /* Write the combined doclist here */
-){
- DLReader left, right;
- DLWriter writer;
- int rc;
-
- if( nLeft==0 || nRight==0 ) return SQLITE_OK;
-
- rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrInit(&right, DL_DOCIDS, pRight, nRight);
- if( rc!=SQLITE_OK ){
- dlrDestroy(&left);
- return rc;
- }
- dlwInit(&writer, DL_DOCIDS, pOut);
-
- while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){
- if( dlrDocid(&left)<dlrDocid(&right) ){
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrDocid(&right)<dlrDocid(&left) ){
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- dlwAdd(&writer, dlrDocid(&left));
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
+ nCol = argc - 3 - (zTokenizer!=0);
+ if( zTokenizer==0 ){
+ rc = sqlite3Fts3InitTokenizer(pHash, 0, &pTokenizer, 0, pzErr);
+ if( rc!=SQLITE_OK ){
+ return rc;
}
+ assert( pTokenizer );
}
- dlrDestroy(&left);
- dlrDestroy(&right);
- dlwDestroy(&writer);
- return rc;
-}
-
-/* We have two DL_DOCIDS doclists: pLeft and pRight.
-** Write the union of these two doclists into pOut as a
-** DL_DOCIDS doclist.
-*/
-static int docListOrMerge(
- const char *pLeft, int nLeft,
- const char *pRight, int nRight,
- DataBuffer *pOut /* Write the combined doclist here */
-){
- DLReader left, right;
- DLWriter writer;
- int rc;
-
- if( nLeft==0 ){
- if( nRight!=0 ) dataBufferAppend(pOut, pRight, nRight);
- return SQLITE_OK;
- }
- if( nRight==0 ){
- dataBufferAppend(pOut, pLeft, nLeft);
- return SQLITE_OK;
+ if( nCol==0 ){
+ nCol = 1;
}
- rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrInit(&right, DL_DOCIDS, pRight, nRight);
- if( rc!=SQLITE_OK ){
- dlrDestroy(&left);
- return rc;
- }
- dlwInit(&writer, DL_DOCIDS, pOut);
-
- while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){
- if( dlrAtEnd(&right) ){
- dlwAdd(&writer, dlrDocid(&left));
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrAtEnd(&left) ){
- dlwAdd(&writer, dlrDocid(&right));
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrDocid(&left)<dlrDocid(&right) ){
- dlwAdd(&writer, dlrDocid(&left));
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- }else if( dlrDocid(&right)<dlrDocid(&left) ){
- dlwAdd(&writer, dlrDocid(&right));
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }else{
- dlwAdd(&writer, dlrDocid(&left));
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) break;
- }
+ /* Allocate and populate the Fts3Table structure. */
+ nByte = sizeof(Fts3Table) + /* Fts3Table */
+ nCol * sizeof(char *) + /* azColumn */
+ nName + /* zName */
+ nDb + /* zDb */
+ nString; /* Space for azColumn strings */
+ p = (Fts3Table*)sqlite3_malloc(nByte);
+ if( p==0 ){
+ rc = SQLITE_NOMEM;
+ goto fts3_init_out;
+ }
+ memset(p, 0, nByte);
+
+ p->db = db;
+ p->nColumn = nCol;
+ p->nPendingData = 0;
+ p->azColumn = (char **)&p[1];
+ p->pTokenizer = pTokenizer;
+ p->nNodeSize = 1000;
+ p->nMaxPendingData = FTS3_MAX_PENDING_DATA;
+ zCsr = (char *)&p->azColumn[nCol];
+
+ fts3HashInit(&p->pendingTerms, FTS3_HASH_STRING, 1);
+
+ /* Fill in the zName and zDb fields of the vtab structure. */
+ p->zName = zCsr;
+ memcpy(zCsr, argv[2], nName);
+ zCsr += nName;
+ p->zDb = zCsr;
+ memcpy(zCsr, argv[1], nDb);
+ zCsr += nDb;
+
+ /* Fill in the azColumn array */
+ iCol = 0;
+ for(i=3; i<argc; i++){
+ if( argv[i]!=zTokenizer ){
+ char *z;
+ int n;
+ z = (char *)sqlite3Fts3NextToken(argv[i], &n);
+ memcpy(zCsr, z, n);
+ zCsr[n] = '\0';
+ sqlite3Fts3Dequote(zCsr);
+ p->azColumn[iCol++] = zCsr;
+ zCsr += n+1;
+ assert( zCsr <= &((char *)p)[nByte] );
+ }
+ }
+ if( iCol==0 ){
+ assert( nCol==1 );
+ p->azColumn[0] = "content";
+ }
+
+ /* If this is an xCreate call, create the underlying tables in the
+ ** database. TODO: For xConnect(), it could verify that said tables exist.
+ */
+ if( isCreate ){
+ p->bHasContent = 1;
+ p->bHasDocsize = argv[0][3]=='4';
+ rc = fts3CreateTables(p);
+ }else{
+ rc = SQLITE_OK;
+ fts3TableExists(&rc, db, argv[1], argv[2], "_content", &p->bHasContent);
+ fts3TableExists(&rc, db, argv[1], argv[2], "_docsize", &p->bHasDocsize);
}
+ if( rc!=SQLITE_OK ) goto fts3_init_out;
- dlrDestroy(&left);
- dlrDestroy(&right);
- dlwDestroy(&writer);
- return rc;
-}
-
-/* We have two DL_DOCIDS doclists: pLeft and pRight.
-** Write into pOut as DL_DOCIDS doclist containing all documents that
-** occur in pLeft but not in pRight.
-*/
-static int docListExceptMerge(
- const char *pLeft, int nLeft,
- const char *pRight, int nRight,
- DataBuffer *pOut /* Write the combined doclist here */
-){
- DLReader left, right;
- DLWriter writer;
- int rc;
+ rc = fts3DeclareVtab(p);
+ if( rc!=SQLITE_OK ) goto fts3_init_out;
- if( nLeft==0 ) return SQLITE_OK;
- if( nRight==0 ){
- dataBufferAppend(pOut, pLeft, nLeft);
- return SQLITE_OK;
- }
+ *ppVTab = &p->base;
- rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft);
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrInit(&right, DL_DOCIDS, pRight, nRight);
+fts3_init_out:
+ assert( p || (pTokenizer && rc!=SQLITE_OK) );
if( rc!=SQLITE_OK ){
- dlrDestroy(&left);
- return rc;
- }
- dlwInit(&writer, DL_DOCIDS, pOut);
-
- while( !dlrAtEnd(&left) ){
- while( !dlrAtEnd(&right) && dlrDocid(&right)<dlrDocid(&left) ){
- rc = dlrStep(&right);
- if( rc!=SQLITE_OK ) goto err;
- }
- if( dlrAtEnd(&right) || dlrDocid(&left)<dlrDocid(&right) ){
- dlwAdd(&writer, dlrDocid(&left));
+ if( p ){
+ fts3DisconnectMethod((sqlite3_vtab *)p);
+ }else{
+ pTokenizer->pModule->xDestroy(pTokenizer);
}
- rc = dlrStep(&left);
- if( rc!=SQLITE_OK ) break;
}
-
-err:
- dlrDestroy(&left);
- dlrDestroy(&right);
- dlwDestroy(&writer);
return rc;
}
-static char *string_dup_n(const char *s, int n){
- char *str = sqlite3_malloc(n + 1);
- memcpy(str, s, n);
- str[n] = '\0';
- return str;
+/*
+** The xConnect() and xCreate() methods for the virtual table. All the
+** work is done in function fts3InitVtab().
+*/
+static int fts3ConnectMethod(
+ sqlite3 *db, /* Database connection */
+ void *pAux, /* Pointer to tokenizer hash table */
+ int argc, /* Number of elements in argv array */
+ const char * const *argv, /* xCreate/xConnect argument array */
+ sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
+ char **pzErr /* OUT: sqlite3_malloc'd error message */
+){
+ return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr);
+}
+static int fts3CreateMethod(
+ sqlite3 *db, /* Database connection */
+ void *pAux, /* Pointer to tokenizer hash table */
+ int argc, /* Number of elements in argv array */
+ const char * const *argv, /* xCreate/xConnect argument array */
+ sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
+ char **pzErr /* OUT: sqlite3_malloc'd error message */
+){
+ return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr);
}
-/* Duplicate a string; the caller must free() the returned string.
- * (We don't use strdup() since it is not part of the standard C library and
- * may not be available everywhere.) */
-static char *string_dup(const char *s){
- return string_dup_n(s, strlen(s));
-}
+/*
+** Implementation of the xBestIndex method for FTS3 tables. There
+** are three possible strategies, in order of preference:
+**
+** 1. Direct lookup by rowid or docid.
+** 2. Full-text search using a MATCH operator on a non-docid column.
+** 3. Linear scan of %_content table.
+*/
+static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
+ Fts3Table *p = (Fts3Table *)pVTab;
+ int i; /* Iterator variable */
+ int iCons = -1; /* Index of constraint to use */
-/* Format a string, replacing each occurrence of the % character with
- * zDb.zName. This may be more convenient than sqlite_mprintf()
- * when one string is used repeatedly in a format string.
- * The caller must free() the returned string. */
-static char *string_format(const char *zFormat,
- const char *zDb, const char *zName){
- const char *p;
- size_t len = 0;
- size_t nDb = strlen(zDb);
- size_t nName = strlen(zName);
- size_t nFullTableName = nDb+1+nName;
- char *result;
- char *r;
-
- /* first compute length needed */
- for(p = zFormat ; *p ; ++p){
- len += (*p=='%' ? nFullTableName : 1);
- }
- len += 1; /* for null terminator */
-
- r = result = sqlite3_malloc(len);
- for(p = zFormat; *p; ++p){
- if( *p=='%' ){
- memcpy(r, zDb, nDb);
- r += nDb;
- *r++ = '.';
- memcpy(r, zName, nName);
- r += nName;
- } else {
- *r++ = *p;
+ /* By default use a full table scan. This is an expensive option,
+ ** so search through the constraints to see if a more efficient
+ ** strategy is possible.
+ */
+ pInfo->idxNum = FTS3_FULLSCAN_SEARCH;
+ pInfo->estimatedCost = 500000;
+ for(i=0; i<pInfo->nConstraint; i++){
+ struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i];
+ if( pCons->usable==0 ) continue;
+
+ /* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */
+ if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ
+ && (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1 )
+ ){
+ pInfo->idxNum = FTS3_DOCID_SEARCH;
+ pInfo->estimatedCost = 1.0;
+ iCons = i;
+ }
+
+ /* A MATCH constraint. Use a full-text search.
+ **
+ ** If there is more than one MATCH constraint available, use the first
+ ** one encountered. If there is both a MATCH constraint and a direct
+ ** rowid/docid lookup, prefer the MATCH strategy. This is done even
+ ** though the rowid/docid lookup is faster than a MATCH query, selecting
+ ** it would lead to an "unable to use function MATCH in the requested
+ ** context" error.
+ */
+ if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH
+ && pCons->iColumn>=0 && pCons->iColumn<=p->nColumn
+ ){
+ pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn;
+ pInfo->estimatedCost = 2.0;
+ iCons = i;
+ break;
}
}
- *r++ = '\0';
- assert( r == result + len );
- return result;
-}
-static int sql_exec(sqlite3 *db, const char *zDb, const char *zName,
- const char *zFormat){
- char *zCommand = string_format(zFormat, zDb, zName);
- int rc;
- FTSTRACE(("FTS3 sql: %s\n", zCommand));
- rc = sqlite3_exec(db, zCommand, NULL, 0, NULL);
- sqlite3_free(zCommand);
- return rc;
-}
-
-static int sql_prepare(sqlite3 *db, const char *zDb, const char *zName,
- sqlite3_stmt **ppStmt, const char *zFormat){
- char *zCommand = string_format(zFormat, zDb, zName);
- int rc;
- FTSTRACE(("FTS3 prepare: %s\n", zCommand));
- rc = sqlite3_prepare_v2(db, zCommand, -1, ppStmt, NULL);
- sqlite3_free(zCommand);
- return rc;
+ if( iCons>=0 ){
+ pInfo->aConstraintUsage[iCons].argvIndex = 1;
+ pInfo->aConstraintUsage[iCons].omit = 1;
+ }
+ return SQLITE_OK;
}
-/* end utility functions */
-
-/* Forward reference */
-typedef struct fulltext_vtab fulltext_vtab;
-
/*
-** An instance of the following structure keeps track of generated
-** matching-word offset information and snippets.
-*/
-typedef struct Snippet {
- int nMatch; /* Total number of matches */
- int nAlloc; /* Space allocated for aMatch[] */
- struct snippetMatch { /* One entry for each matching term */
- char snStatus; /* Status flag for use while constructing snippets */
- short int iCol; /* The column that contains the match */
- short int iTerm; /* The index in Query.pTerms[] of the matching term */
- int iToken; /* The index of the matching document token */
- short int nByte; /* Number of bytes in the term */
- int iStart; /* The offset to the first character of the term */
- } *aMatch; /* Points to space obtained from malloc */
- char *zOffset; /* Text rendering of aMatch[] */
- int nOffset; /* strlen(zOffset) */
- char *zSnippet; /* Snippet text */
- int nSnippet; /* strlen(zSnippet) */
-} Snippet;
-
-
-typedef enum QueryType {
- QUERY_GENERIC, /* table scan */
- QUERY_DOCID, /* lookup by docid */
- QUERY_FULLTEXT /* QUERY_FULLTEXT + [i] is a full-text search for column i*/
-} QueryType;
-
-typedef enum fulltext_statement {
- CONTENT_INSERT_STMT,
- CONTENT_SELECT_STMT,
- CONTENT_UPDATE_STMT,
- CONTENT_DELETE_STMT,
- CONTENT_EXISTS_STMT,
-
- BLOCK_INSERT_STMT,
- BLOCK_SELECT_STMT,
- BLOCK_DELETE_STMT,
- BLOCK_DELETE_ALL_STMT,
-
- SEGDIR_MAX_INDEX_STMT,
- SEGDIR_SET_STMT,
- SEGDIR_SELECT_LEVEL_STMT,
- SEGDIR_SPAN_STMT,
- SEGDIR_DELETE_STMT,
- SEGDIR_SELECT_SEGMENT_STMT,
- SEGDIR_SELECT_ALL_STMT,
- SEGDIR_DELETE_ALL_STMT,
- SEGDIR_COUNT_STMT,
-
- MAX_STMT /* Always at end! */
-} fulltext_statement;
-
-/* These must exactly match the enum above. */
-/* TODO(shess): Is there some risk that a statement will be used in two
-** cursors at once, e.g. if a query joins a virtual table to itself?
-** If so perhaps we should move some of these to the cursor object.
+** Implementation of xOpen method.
*/
-static const char *const fulltext_zStatement[MAX_STMT] = {
- /* CONTENT_INSERT */ NULL, /* generated in contentInsertStatement() */
- /* CONTENT_SELECT */ NULL, /* generated in contentSelectStatement() */
- /* CONTENT_UPDATE */ NULL, /* generated in contentUpdateStatement() */
- /* CONTENT_DELETE */ "delete from %_content where docid = ?",
- /* CONTENT_EXISTS */ "select docid from %_content limit 1",
-
- /* BLOCK_INSERT */
- "insert into %_segments (blockid, block) values (null, ?)",
- /* BLOCK_SELECT */ "select block from %_segments where blockid = ?",
- /* BLOCK_DELETE */ "delete from %_segments where blockid between ? and ?",
- /* BLOCK_DELETE_ALL */ "delete from %_segments",
-
- /* SEGDIR_MAX_INDEX */ "select max(idx) from %_segdir where level = ?",
- /* SEGDIR_SET */ "insert into %_segdir values (?, ?, ?, ?, ?, ?)",
- /* SEGDIR_SELECT_LEVEL */
- "select start_block, leaves_end_block, root, idx from %_segdir "
- " where level = ? order by idx",
- /* SEGDIR_SPAN */
- "select min(start_block), max(end_block) from %_segdir "
- " where level = ? and start_block <> 0",
- /* SEGDIR_DELETE */ "delete from %_segdir where level = ?",
-
- /* NOTE(shess): The first three results of the following two
- ** statements must match.
- */
- /* SEGDIR_SELECT_SEGMENT */
- "select start_block, leaves_end_block, root from %_segdir "
- " where level = ? and idx = ?",
- /* SEGDIR_SELECT_ALL */
- "select start_block, leaves_end_block, root from %_segdir "
- " order by level desc, idx asc",
- /* SEGDIR_DELETE_ALL */ "delete from %_segdir",
- /* SEGDIR_COUNT */ "select count(*), ifnull(max(level),0) from %_segdir",
-};
+static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){
+ sqlite3_vtab_cursor *pCsr; /* Allocated cursor */
-/*
-** A connection to a fulltext index is an instance of the following
-** structure. The xCreate and xConnect methods create an instance
-** of this structure and xDestroy and xDisconnect free that instance.
-** All other methods receive a pointer to the structure as one of their
-** arguments.
-*/
-struct fulltext_vtab {
- sqlite3_vtab base; /* Base class used by SQLite core */
- sqlite3 *db; /* The database connection */
- const char *zDb; /* logical database name */
- const char *zName; /* virtual table name */
- int nColumn; /* number of columns in virtual table */
- char **azColumn; /* column names. malloced */
- char **azContentColumn; /* column names in content table; malloced */
- sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */
-
- /* Precompiled statements which we keep as long as the table is
- ** open.
- */
- sqlite3_stmt *pFulltextStatements[MAX_STMT];
+ UNUSED_PARAMETER(pVTab);
- /* Precompiled statements used for segment merges. We run a
- ** separate select across the leaf level of each tree being merged.
- */
- sqlite3_stmt *pLeafSelectStmts[MERGE_COUNT];
- /* The statement used to prepare pLeafSelectStmts. */
-#define LEAF_SELECT \
- "select block from %_segments where blockid between ? and ? order by blockid"
-
- /* These buffer pending index updates during transactions.
- ** nPendingData estimates the memory size of the pending data. It
- ** doesn't include the hash-bucket overhead, nor any malloc
- ** overhead. When nPendingData exceeds kPendingThreshold, the
- ** buffer is flushed even before the transaction closes.
- ** pendingTerms stores the data, and is only valid when nPendingData
- ** is >=0 (nPendingData<0 means pendingTerms has not been
- ** initialized). iPrevDocid is the last docid written, used to make
- ** certain we're inserting in sorted order.
+ /* Allocate a buffer large enough for an Fts3Cursor structure. If the
+ ** allocation succeeds, zero it and return SQLITE_OK. Otherwise,
+ ** if the allocation fails, return SQLITE_NOMEM.
*/
- int nPendingData;
-#define kPendingThreshold (1*1024*1024)
- sqlite_int64 iPrevDocid;
- fts3Hash pendingTerms;
-};
-
-/*
-** When the core wants to do a query, it create a cursor using a
-** call to xOpen. This structure is an instance of a cursor. It
-** is destroyed by xClose.
-*/
-typedef struct fulltext_cursor {
- sqlite3_vtab_cursor base; /* Base class used by SQLite core */
- QueryType iCursorType; /* Copy of sqlite3_index_info.idxNum */
- sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */
- int eof; /* True if at End Of Results */
- Fts3Expr *pExpr; /* Parsed MATCH query string */
- Snippet snippet; /* Cached snippet for the current row */
- int iColumn; /* Column being searched */
- DataBuffer result; /* Doclist results from fulltextQuery */
- DLReader reader; /* Result reader if result not empty */
-} fulltext_cursor;
-
-static fulltext_vtab *cursor_vtab(fulltext_cursor *c){
- return (fulltext_vtab *) c->base.pVtab;
-}
-
-static const sqlite3_module fts3Module; /* forward declaration */
-
-/* Return a dynamically generated statement of the form
- * insert into %_content (docid, ...) values (?, ...)
- */
-static const char *contentInsertStatement(fulltext_vtab *v){
- StringBuffer sb;
- int i;
-
- initStringBuffer(&sb);
- append(&sb, "insert into %_content (docid, ");
- appendList(&sb, v->nColumn, v->azContentColumn);
- append(&sb, ") values (?");
- for(i=0; i<v->nColumn; ++i)
- append(&sb, ", ?");
- append(&sb, ")");
- return stringBufferData(&sb);
-}
-
-/* Return a dynamically generated statement of the form
- * select <content columns> from %_content where docid = ?
- */
-static const char *contentSelectStatement(fulltext_vtab *v){
- StringBuffer sb;
- initStringBuffer(&sb);
- append(&sb, "SELECT ");
- appendList(&sb, v->nColumn, v->azContentColumn);
- append(&sb, " FROM %_content WHERE docid = ?");
- return stringBufferData(&sb);
-}
-
-/* Return a dynamically generated statement of the form
- * update %_content set [col_0] = ?, [col_1] = ?, ...
- * where docid = ?
- */
-static const char *contentUpdateStatement(fulltext_vtab *v){
- StringBuffer sb;
- int i;
-
- initStringBuffer(&sb);
- append(&sb, "update %_content set ");
- for(i=0; i<v->nColumn; ++i) {
- if( i>0 ){
- append(&sb, ", ");
- }
- append(&sb, v->azContentColumn[i]);
- append(&sb, " = ?");
- }
- append(&sb, " where docid = ?");
- return stringBufferData(&sb);
-}
-
-/* Puts a freshly-prepared statement determined by iStmt in *ppStmt.
-** If the indicated statement has never been prepared, it is prepared
-** and cached, otherwise the cached version is reset.
-*/
-static int sql_get_statement(fulltext_vtab *v, fulltext_statement iStmt,
- sqlite3_stmt **ppStmt){
- assert( iStmt<MAX_STMT );
- if( v->pFulltextStatements[iStmt]==NULL ){
- const char *zStmt;
- int rc;
- switch( iStmt ){
- case CONTENT_INSERT_STMT:
- zStmt = contentInsertStatement(v); break;
- case CONTENT_SELECT_STMT:
- zStmt = contentSelectStatement(v); break;
- case CONTENT_UPDATE_STMT:
- zStmt = contentUpdateStatement(v); break;
- default:
- zStmt = fulltext_zStatement[iStmt];
- }
- rc = sql_prepare(v->db, v->zDb, v->zName, &v->pFulltextStatements[iStmt],
- zStmt);
- if( zStmt != fulltext_zStatement[iStmt]) sqlite3_free((void *) zStmt);
- if( rc!=SQLITE_OK ) return rc;
- } else {
- int rc = sqlite3_reset(v->pFulltextStatements[iStmt]);
- if( rc!=SQLITE_OK ) return rc;
+ *ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor));
+ if( !pCsr ){
+ return SQLITE_NOMEM;
}
-
- *ppStmt = v->pFulltextStatements[iStmt];
+ memset(pCsr, 0, sizeof(Fts3Cursor));
return SQLITE_OK;
}
-/* Like sqlite3_step(), but convert SQLITE_DONE to SQLITE_OK and
-** SQLITE_ROW to SQLITE_ERROR. Useful for statements like UPDATE,
-** where we expect no results.
-*/
-static int sql_single_step(sqlite3_stmt *s){
- int rc = sqlite3_step(s);
- return (rc==SQLITE_DONE) ? SQLITE_OK : rc;
-}
-
-/* Like sql_get_statement(), but for special replicated LEAF_SELECT
-** statements. idx -1 is a special case for an uncached version of
-** the statement (used in the optimize implementation).
-*/
-/* TODO(shess) Write version for generic statements and then share
-** that between the cached-statement functions.
+/*
+** Close the cursor. For additional information see the documentation
+** on the xClose method of the virtual table interface.
*/
-static int sql_get_leaf_statement(fulltext_vtab *v, int idx,
- sqlite3_stmt **ppStmt){
- assert( idx>=-1 && idx<MERGE_COUNT );
- if( idx==-1 ){
- return sql_prepare(v->db, v->zDb, v->zName, ppStmt, LEAF_SELECT);
- }else if( v->pLeafSelectStmts[idx]==NULL ){
- int rc = sql_prepare(v->db, v->zDb, v->zName, &v->pLeafSelectStmts[idx],
- LEAF_SELECT);
- if( rc!=SQLITE_OK ) return rc;
- }else{
- int rc = sqlite3_reset(v->pLeafSelectStmts[idx]);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- *ppStmt = v->pLeafSelectStmts[idx];
+static int fulltextClose(sqlite3_vtab_cursor *pCursor){
+ Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
+ sqlite3_finalize(pCsr->pStmt);
+ sqlite3Fts3ExprFree(pCsr->pExpr);
+ sqlite3_free(pCsr->aDoclist);
+ sqlite3_free(pCsr->aMatchinfo);
+ sqlite3_free(pCsr);
return SQLITE_OK;
}
-/* insert into %_content (docid, ...) values ([docid], [pValues])
-** If the docid contains SQL NULL, then a unique docid will be
-** generated.
-*/
-static int content_insert(fulltext_vtab *v, sqlite3_value *docid,
- sqlite3_value **pValues){
- sqlite3_stmt *s;
- int i;
- int rc = sql_get_statement(v, CONTENT_INSERT_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_value(s, 1, docid);
- if( rc!=SQLITE_OK ) return rc;
-
- for(i=0; i<v->nColumn; ++i){
- rc = sqlite3_bind_value(s, 2+i, pValues[i]);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- return sql_single_step(s);
-}
-
-/* update %_content set col0 = pValues[0], col1 = pValues[1], ...
- * where docid = [iDocid] */
-static int content_update(fulltext_vtab *v, sqlite3_value **pValues,
- sqlite_int64 iDocid){
- sqlite3_stmt *s;
- int i;
- int rc = sql_get_statement(v, CONTENT_UPDATE_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- for(i=0; i<v->nColumn; ++i){
- rc = sqlite3_bind_value(s, 1+i, pValues[i]);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- rc = sqlite3_bind_int64(s, 1+v->nColumn, iDocid);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
-
-static void freeStringArray(int nString, const char **pString){
- int i;
-
- for (i=0 ; i < nString ; ++i) {
- if( pString[i]!=NULL ) sqlite3_free((void *) pString[i]);
- }
- sqlite3_free((void *) pString);
-}
-
-/* select * from %_content where docid = [iDocid]
- * The caller must delete the returned array and all strings in it.
- * null fields will be NULL in the returned array.
- *
- * TODO: Perhaps we should return pointer/length strings here for consistency
- * with other code which uses pointer/length. */
-static int content_select(fulltext_vtab *v, sqlite_int64 iDocid,
- const char ***pValues){
- sqlite3_stmt *s;
- const char **values;
- int i;
- int rc;
-
- *pValues = NULL;
-
- rc = sql_get_statement(v, CONTENT_SELECT_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iDocid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- if( rc!=SQLITE_ROW ) return rc;
-
- values = (const char **) sqlite3_malloc(v->nColumn * sizeof(const char *));
- for(i=0; i<v->nColumn; ++i){
- if( sqlite3_column_type(s, i)==SQLITE_NULL ){
- values[i] = NULL;
+/*
+** Position the pCsr->pStmt statement so that it is on the row
+** of the %_content table that contains the last match. Return
+** SQLITE_OK on success.
+*/
+static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){
+ if( pCsr->isRequireSeek ){
+ pCsr->isRequireSeek = 0;
+ sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId);
+ if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){
+ return SQLITE_OK;
}else{
- values[i] = string_dup((char*)sqlite3_column_text(s, i));
+ int rc = sqlite3_reset(pCsr->pStmt);
+ if( rc==SQLITE_OK ){
+ /* If no row was found and no error has occured, then the %_content
+ ** table is missing a row that is present in the full-text index.
+ ** The data structures are corrupt.
+ */
+ rc = SQLITE_CORRUPT;
+ }
+ pCsr->isEof = 1;
+ if( pContext ){
+ sqlite3_result_error_code(pContext, rc);
+ }
+ return rc;
}
- }
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_DONE ){
- *pValues = values;
+ }else{
return SQLITE_OK;
}
-
- freeStringArray(v->nColumn, values);
- return rc;
}
-/* delete from %_content where docid = [iDocid ] */
-static int content_delete(fulltext_vtab *v, sqlite_int64 iDocid){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, CONTENT_DELETE_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iDocid);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
-
-/* Returns SQLITE_ROW if any rows exist in %_content, SQLITE_DONE if
-** no rows exist, and any error in case of failure.
-*/
-static int content_exists(fulltext_vtab *v){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, CONTENT_EXISTS_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- if( rc!=SQLITE_ROW ) return rc;
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_DONE ) return SQLITE_ROW;
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- return rc;
-}
-
-/* insert into %_segments values ([pData])
-** returns assigned blockid in *piBlockid
-*/
-static int block_insert(fulltext_vtab *v, const char *pData, int nData,
- sqlite_int64 *piBlockid){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, BLOCK_INSERT_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_blob(s, 1, pData, nData, SQLITE_STATIC);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- if( rc!=SQLITE_DONE ) return rc;
-
- /* blockid column is an alias for rowid. */
- *piBlockid = sqlite3_last_insert_rowid(v->db);
- return SQLITE_OK;
-}
-
-/* delete from %_segments
-** where blockid between [iStartBlockid] and [iEndBlockid]
-**
-** Deletes the range of blocks, inclusive, used to delete the blocks
-** which form a segment.
-*/
-static int block_delete(fulltext_vtab *v,
- sqlite_int64 iStartBlockid, sqlite_int64 iEndBlockid){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, BLOCK_DELETE_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iStartBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 2, iEndBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
-
-/* Returns SQLITE_ROW with *pidx set to the maximum segment idx found
-** at iLevel. Returns SQLITE_DONE if there are no segments at
-** iLevel. Otherwise returns an error.
-*/
-static int segdir_max_index(fulltext_vtab *v, int iLevel, int *pidx){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_MAX_INDEX_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int(s, 1, iLevel);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- /* Should always get at least one row due to how max() works. */
- if( rc==SQLITE_DONE ) return SQLITE_DONE;
- if( rc!=SQLITE_ROW ) return rc;
-
- /* NULL means that there were no inputs to max(). */
- if( SQLITE_NULL==sqlite3_column_type(s, 0) ){
- rc = sqlite3_step(s);
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- return rc;
- }
-
- *pidx = sqlite3_column_int(s, 0);
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- if( rc!=SQLITE_DONE ) return rc;
- return SQLITE_ROW;
-}
-
-/* insert into %_segdir values (
-** [iLevel], [idx],
-** [iStartBlockid], [iLeavesEndBlockid], [iEndBlockid],
-** [pRootData]
-** )
-*/
-static int segdir_set(fulltext_vtab *v, int iLevel, int idx,
- sqlite_int64 iStartBlockid,
- sqlite_int64 iLeavesEndBlockid,
- sqlite_int64 iEndBlockid,
- const char *pRootData, int nRootData){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_SET_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int(s, 1, iLevel);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int(s, 2, idx);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 3, iStartBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 4, iLeavesEndBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 5, iEndBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_blob(s, 6, pRootData, nRootData, SQLITE_STATIC);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
-
-/* Queries %_segdir for the block span of the segments in level
-** iLevel. Returns SQLITE_DONE if there are no blocks for iLevel,
-** SQLITE_ROW if there are blocks, else an error.
-*/
-static int segdir_span(fulltext_vtab *v, int iLevel,
- sqlite_int64 *piStartBlockid,
- sqlite_int64 *piEndBlockid){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_SPAN_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int(s, 1, iLevel);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- if( rc==SQLITE_DONE ) return SQLITE_DONE; /* Should never happen */
- if( rc!=SQLITE_ROW ) return rc;
-
- /* This happens if all segments at this level are entirely inline. */
- if( SQLITE_NULL==sqlite3_column_type(s, 0) ){
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- int rc2 = sqlite3_step(s);
- if( rc2==SQLITE_ROW ) return SQLITE_ERROR;
- return rc2;
+/*
+** Advance the cursor to the next row in the %_content table that
+** matches the search criteria. For a MATCH search, this will be
+** the next row that matches. For a full-table scan, this will be
+** simply the next row in the %_content table. For a docid lookup,
+** this routine simply sets the EOF flag.
+**
+** Return SQLITE_OK if nothing goes wrong. SQLITE_OK is returned
+** even if we reach end-of-file. The fts3EofMethod() will be called
+** subsequently to determine whether or not an EOF was hit.
+*/
+static int fts3NextMethod(sqlite3_vtab_cursor *pCursor){
+ int rc = SQLITE_OK; /* Return code */
+ Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
+
+ if( pCsr->aDoclist==0 ){
+ if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){
+ pCsr->isEof = 1;
+ rc = sqlite3_reset(pCsr->pStmt);
+ }
+ }else if( pCsr->pNextId>=&pCsr->aDoclist[pCsr->nDoclist] ){
+ pCsr->isEof = 1;
+ }else{
+ sqlite3_reset(pCsr->pStmt);
+ fts3GetDeltaVarint(&pCsr->pNextId, &pCsr->iPrevId);
+ pCsr->isRequireSeek = 1;
+ pCsr->isMatchinfoNeeded = 1;
}
-
- *piStartBlockid = sqlite3_column_int64(s, 0);
- *piEndBlockid = sqlite3_column_int64(s, 1);
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- if( rc!=SQLITE_DONE ) return rc;
- return SQLITE_ROW;
+ return rc;
}
-/* Delete the segment blocks and segment directory records for all
-** segments at iLevel.
-*/
-static int segdir_delete(fulltext_vtab *v, int iLevel){
- sqlite3_stmt *s;
- sqlite_int64 iStartBlockid, iEndBlockid;
- int rc = segdir_span(v, iLevel, &iStartBlockid, &iEndBlockid);
- if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ) return rc;
-
- if( rc==SQLITE_ROW ){
- rc = block_delete(v, iStartBlockid, iEndBlockid);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- /* Delete the segment directory itself. */
- rc = sql_get_statement(v, SEGDIR_DELETE_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iLevel);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
-/* Delete entire fts index, SQLITE_OK on success, relevant error on
-** failure.
-*/
-static int segdir_delete_all(fulltext_vtab *v){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_DELETE_ALL_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sql_single_step(s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sql_get_statement(v, BLOCK_DELETE_ALL_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- return sql_single_step(s);
-}
+/*
+** The buffer pointed to by argument zNode (size nNode bytes) contains the
+** root node of a b-tree segment. The segment is guaranteed to be at least
+** one level high (i.e. the root node is not also a leaf). If successful,
+** this function locates the leaf node of the segment that may contain the
+** term specified by arguments zTerm and nTerm and writes its block number
+** to *piLeaf.
+**
+** It is possible that the returned leaf node does not contain the specified
+** term. However, if the segment does contain said term, it is stored on
+** the identified leaf node. Because this function only inspects interior
+** segment nodes (and never loads leaf nodes into memory), it is not possible
+** to be sure.
+**
+** If an error occurs, an error code other than SQLITE_OK is returned.
+*/
+static int fts3SelectLeaf(
+ Fts3Table *p, /* Virtual table handle */
+ const char *zTerm, /* Term to select leaves for */
+ int nTerm, /* Size of term zTerm in bytes */
+ const char *zNode, /* Buffer containing segment interior node */
+ int nNode, /* Size of buffer at zNode */
+ sqlite3_int64 *piLeaf /* Selected leaf node */
+){
+ int rc = SQLITE_OK; /* Return code */
+ const char *zCsr = zNode; /* Cursor to iterate through node */
+ const char *zEnd = &zCsr[nNode];/* End of interior node buffer */
+ char *zBuffer = 0; /* Buffer to load terms into */
+ int nAlloc = 0; /* Size of allocated buffer */
+
+ while( 1 ){
+ int isFirstTerm = 1; /* True when processing first term on page */
+ int iHeight; /* Height of this node in tree */
+ sqlite3_int64 iChild; /* Block id of child node to descend to */
+ int nBlock; /* Size of child node in bytes */
+
+ zCsr += sqlite3Fts3GetVarint32(zCsr, &iHeight);
+ zCsr += sqlite3Fts3GetVarint(zCsr, &iChild);
+
+ while( zCsr<zEnd ){
+ int cmp; /* memcmp() result */
+ int nSuffix; /* Size of term suffix */
+ int nPrefix = 0; /* Size of term prefix */
+ int nBuffer; /* Total term size */
+
+ /* Load the next term on the node into zBuffer */
+ if( !isFirstTerm ){
+ zCsr += sqlite3Fts3GetVarint32(zCsr, &nPrefix);
+ }
+ isFirstTerm = 0;
+ zCsr += sqlite3Fts3GetVarint32(zCsr, &nSuffix);
+ if( nPrefix+nSuffix>nAlloc ){
+ char *zNew;
+ nAlloc = (nPrefix+nSuffix) * 2;
+ zNew = (char *)sqlite3_realloc(zBuffer, nAlloc);
+ if( !zNew ){
+ sqlite3_free(zBuffer);
+ return SQLITE_NOMEM;
+ }
+ zBuffer = zNew;
+ }
+ memcpy(&zBuffer[nPrefix], zCsr, nSuffix);
+ nBuffer = nPrefix + nSuffix;
+ zCsr += nSuffix;
+
+ /* Compare the term we are searching for with the term just loaded from
+ ** the interior node. If the specified term is greater than or equal
+ ** to the term from the interior node, then all terms on the sub-tree
+ ** headed by node iChild are smaller than zTerm. No need to search
+ ** iChild.
+ **
+ ** If the interior node term is larger than the specified term, then
+ ** the tree headed by iChild may contain the specified term.
+ */
+ cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer));
+ if( cmp<0 || (cmp==0 && nBuffer>nTerm) ) break;
+ iChild++;
+ };
-/* Returns SQLITE_OK with *pnSegments set to the number of entries in
-** %_segdir and *piMaxLevel set to the highest level which has a
-** segment. Otherwise returns the SQLite error which caused failure.
-*/
-static int segdir_count(fulltext_vtab *v, int *pnSegments, int *piMaxLevel){
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_COUNT_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
+ /* If (iHeight==1), the children of this interior node are leaves. The
+ ** specified term may be present on leaf node iChild.
+ */
+ if( iHeight==1 ){
+ *piLeaf = iChild;
+ break;
+ }
- rc = sqlite3_step(s);
- /* TODO(shess): This case should not be possible? Should stronger
- ** measures be taken if it happens?
- */
- if( rc==SQLITE_DONE ){
- *pnSegments = 0;
- *piMaxLevel = 0;
- return SQLITE_OK;
+ /* Descend to interior node iChild. */
+ rc = sqlite3Fts3ReadBlock(p, iChild, &zCsr, &nBlock);
+ if( rc!=SQLITE_OK ) break;
+ zEnd = &zCsr[nBlock];
}
- if( rc!=SQLITE_ROW ) return rc;
-
- *pnSegments = sqlite3_column_int(s, 0);
- *piMaxLevel = sqlite3_column_int(s, 1);
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_DONE ) return SQLITE_OK;
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
+ sqlite3_free(zBuffer);
return rc;
}
-/* TODO(shess) clearPendingTerms() is far down the file because
-** writeZeroSegment() is far down the file because LeafWriter is far
-** down the file. Consider refactoring the code to move the non-vtab
-** code above the vtab code so that we don't need this forward
-** reference.
-*/
-static int clearPendingTerms(fulltext_vtab *v);
-
/*
-** Free the memory used to contain a fulltext_vtab structure.
+** This function is used to create delta-encoded serialized lists of FTS3
+** varints. Each call to this function appends a single varint to a list.
*/
-static void fulltext_vtab_destroy(fulltext_vtab *v){
- int iStmt, i;
-
- FTSTRACE(("FTS3 Destroy %p\n", v));
- for( iStmt=0; iStmt<MAX_STMT; iStmt++ ){
- if( v->pFulltextStatements[iStmt]!=NULL ){
- sqlite3_finalize(v->pFulltextStatements[iStmt]);
- v->pFulltextStatements[iStmt] = NULL;
- }
- }
-
- for( i=0; i<MERGE_COUNT; i++ ){
- if( v->pLeafSelectStmts[i]!=NULL ){
- sqlite3_finalize(v->pLeafSelectStmts[i]);
- v->pLeafSelectStmts[i] = NULL;
- }
- }
+static void fts3PutDeltaVarint(
+ char **pp, /* IN/OUT: Output pointer */
+ sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */
+ sqlite3_int64 iVal /* Write this value to the list */
+){
+ assert( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) );
+ *pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev);
+ *piPrev = iVal;
+}
- if( v->pTokenizer!=NULL ){
- v->pTokenizer->pModule->xDestroy(v->pTokenizer);
- v->pTokenizer = NULL;
+/*
+** When this function is called, *ppPoslist is assumed to point to the
+** start of a position-list. After it returns, *ppPoslist points to the
+** first byte after the position-list.
+**
+** A position list is list of positions (delta encoded) and columns for
+** a single document record of a doclist. So, in other words, this
+** routine advances *ppPoslist so that it points to the next docid in
+** the doclist, or to the first byte past the end of the doclist.
+**
+** If pp is not NULL, then the contents of the position list are copied
+** to *pp. *pp is set to point to the first byte past the last byte copied
+** before this function returns.
+*/
+static void fts3PoslistCopy(char **pp, char **ppPoslist){
+ char *pEnd = *ppPoslist;
+ char c = 0;
+
+ /* The end of a position list is marked by a zero encoded as an FTS3
+ ** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by
+ ** a byte with the 0x80 bit set, then it is not a varint 0, but the tail
+ ** of some other, multi-byte, value.
+ **
+ ** The following while-loop moves pEnd to point to the first byte that is not
+ ** immediately preceded by a byte with the 0x80 bit set. Then increments
+ ** pEnd once more so that it points to the byte immediately following the
+ ** last byte in the position-list.
+ */
+ while( *pEnd | c ){
+ c = *pEnd++ & 0x80;
+ testcase( c!=0 && (*pEnd)==0 );
}
+ pEnd++; /* Advance past the POS_END terminator byte */
- clearPendingTerms(v);
-
- sqlite3_free(v->azColumn);
- for(i = 0; i < v->nColumn; ++i) {
- sqlite3_free(v->azContentColumn[i]);
+ if( pp ){
+ int n = (int)(pEnd - *ppPoslist);
+ char *p = *pp;
+ memcpy(p, *ppPoslist, n);
+ p += n;
+ *pp = p;
}
- sqlite3_free(v->azContentColumn);
- sqlite3_free(v);
+ *ppPoslist = pEnd;
}
/*
-** Token types for parsing the arguments to xConnect or xCreate.
-*/
-#define TOKEN_EOF 0 /* End of file */
-#define TOKEN_SPACE 1 /* Any kind of whitespace */
-#define TOKEN_ID 2 /* An identifier */
-#define TOKEN_STRING 3 /* A string literal */
-#define TOKEN_PUNCT 4 /* A single punctuation character */
-
-/*
-** If X is a character that can be used in an identifier then
-** ftsIdChar(X) will be true. Otherwise it is false.
+** When this function is called, *ppPoslist is assumed to point to the
+** start of a column-list. After it returns, *ppPoslist points to the
+** to the terminator (POS_COLUMN or POS_END) byte of the column-list.
+**
+** A column-list is list of delta-encoded positions for a single column
+** within a single document within a doclist.
**
-** For ASCII, any character with the high-order bit set is
-** allowed in an identifier. For 7-bit characters,
-** isFtsIdChar[X] must be 1.
+** The column-list is terminated either by a POS_COLUMN varint (1) or
+** a POS_END varint (0). This routine leaves *ppPoslist pointing to
+** the POS_COLUMN or POS_END that terminates the column-list.
**
-** Ticket #1066. the SQL standard does not allow '$' in the
-** middle of identfiers. But many SQL implementations do.
-** SQLite will allow '$' in identifiers for compatibility.
-** But the feature is undocumented.
+** If pp is not NULL, then the contents of the column-list are copied
+** to *pp. *pp is set to point to the first byte past the last byte copied
+** before this function returns. The POS_COLUMN or POS_END terminator
+** is not copied into *pp.
*/
-static const char isFtsIdChar[] = {
-/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */
- 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */
- 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */
- 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */
- 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */
- 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */
- 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */
-};
-#define ftsIdChar(C) (((c=C)&0x80)!=0 || (c>0x1f && isFtsIdChar[c-0x20]))
-
+static void fts3ColumnlistCopy(char **pp, char **ppPoslist){
+ char *pEnd = *ppPoslist;
+ char c = 0;
-/*
-** Return the length of the token that begins at z[0].
-** Store the token type in *tokenType before returning.
-*/
-static int ftsGetToken(const char *z, int *tokenType){
- int i, c;
- switch( *z ){
- case 0: {
- *tokenType = TOKEN_EOF;
- return 0;
- }
- case ' ': case '\t': case '\n': case '\f': case '\r': {
- for(i=1; safe_isspace(z[i]); i++){}
- *tokenType = TOKEN_SPACE;
- return i;
- }
- case '`':
- case '\'':
- case '"': {
- int delim = z[0];
- for(i=1; (c=z[i])!=0; i++){
- if( c==delim ){
- if( z[i+1]==delim ){
- i++;
- }else{
- break;
- }
- }
- }
- *tokenType = TOKEN_STRING;
- return i + (c!=0);
- }
- case '[': {
- for(i=1, c=z[0]; c!=']' && (c=z[i])!=0; i++){}
- *tokenType = TOKEN_ID;
- return i;
- }
- default: {
- if( !ftsIdChar(*z) ){
- break;
- }
- for(i=1; ftsIdChar(z[i]); i++){}
- *tokenType = TOKEN_ID;
- return i;
- }
+ /* A column-list is terminated by either a 0x01 or 0x00 byte that is
+ ** not part of a multi-byte varint.
+ */
+ while( 0xFE & (*pEnd | c) ){
+ c = *pEnd++ & 0x80;
+ testcase( c!=0 && ((*pEnd)&0xfe)==0 );
}
- *tokenType = TOKEN_PUNCT;
- return 1;
+ if( pp ){
+ int n = (int)(pEnd - *ppPoslist);
+ char *p = *pp;
+ memcpy(p, *ppPoslist, n);
+ p += n;
+ *pp = p;
+ }
+ *ppPoslist = pEnd;
}
/*
-** A token extracted from a string is an instance of the following
-** structure.
+** Value used to signify the end of an position-list. This is safe because
+** it is not possible to have a document with 2^31 terms.
*/
-typedef struct FtsToken {
- const char *z; /* Pointer to token text. Not '\000' terminated */
- short int n; /* Length of the token text in bytes. */
-} FtsToken;
+#define POSITION_LIST_END 0x7fffffff
/*
-** Given a input string (which is really one of the argv[] parameters
-** passed into xConnect or xCreate) split the string up into tokens.
-** Return an array of pointers to '\000' terminated strings, one string
-** for each non-whitespace token.
-**
-** The returned array is terminated by a single NULL pointer.
-**
-** Space to hold the returned array is obtained from a single
-** malloc and should be freed by passing the return value to free().
-** The individual strings within the token list are all a part of
-** the single memory allocation and will all be freed at once.
-*/
-static char **tokenizeString(const char *z, int *pnToken){
- int nToken = 0;
- FtsToken *aToken = sqlite3_malloc( strlen(z) * sizeof(aToken[0]) );
- int n = 1;
- int e, i;
- int totalSize = 0;
- char **azToken;
- char *zCopy;
- while( n>0 ){
- n = ftsGetToken(z, &e);
- if( e!=TOKEN_SPACE ){
- aToken[nToken].z = z;
- aToken[nToken].n = n;
- nToken++;
- totalSize += n+1;
- }
- z += n;
- }
- azToken = (char**)sqlite3_malloc( nToken*sizeof(char*) + totalSize );
- zCopy = (char*)&azToken[nToken];
- nToken--;
- for(i=0; i<nToken; i++){
- azToken[i] = zCopy;
- n = aToken[i].n;
- memcpy(zCopy, aToken[i].z, n);
- zCopy[n] = 0;
- zCopy += n+1;
+** This function is used to help parse position-lists. When this function is
+** called, *pp may point to the start of the next varint in the position-list
+** being parsed, or it may point to 1 byte past the end of the position-list
+** (in which case **pp will be a terminator bytes POS_END (0) or
+** (1)).
+**
+** If *pp points past the end of the current position-list, set *pi to
+** POSITION_LIST_END and return. Otherwise, read the next varint from *pp,
+** increment the current value of *pi by the value read, and set *pp to
+** point to the next value before returning.
+**
+** Before calling this routine *pi must be initialized to the value of
+** the previous position, or zero if we are reading the first position
+** in the position-list. Because positions are delta-encoded, the value
+** of the previous position is needed in order to compute the value of
+** the next position.
+*/
+static void fts3ReadNextPos(
+ char **pp, /* IN/OUT: Pointer into position-list buffer */
+ sqlite3_int64 *pi /* IN/OUT: Value read from position-list */
+){
+ if( (**pp)&0xFE ){
+ fts3GetDeltaVarint(pp, pi);
+ *pi -= 2;
+ }else{
+ *pi = POSITION_LIST_END;
}
- azToken[nToken] = 0;
- sqlite3_free(aToken);
- *pnToken = nToken;
- return azToken;
}
/*
-** Convert an SQL-style quoted string into a normal string by removing
-** the quote characters. The conversion is done in-place. If the
-** input does not begin with a quote character, then this routine
-** is a no-op.
-**
-** Examples:
+** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by
+** the value of iCol encoded as a varint to *pp. This will start a new
+** column list.
**
-** "abc" becomes abc
-** 'xyz' becomes xyz
-** [pqr] becomes pqr
-** `mno` becomes mno
+** Set *pp to point to the byte just after the last byte written before
+** returning (do not modify it if iCol==0). Return the total number of bytes
+** written (0 if iCol==0).
*/
-static void dequoteString(char *z){
- int quote;
- int i, j;
- if( z==0 ) return;
- quote = z[0];
- switch( quote ){
- case '\'': break;
- case '"': break;
- case '`': break; /* For MySQL compatibility */
- case '[': quote = ']'; break; /* For MS SqlServer compatibility */
- default: return;
- }
- for(i=1, j=0; z[i]; i++){
- if( z[i]==quote ){
- if( z[i+1]==quote ){
- z[j++] = quote;
- i++;
- }else{
- z[j++] = 0;
- break;
- }
- }else{
- z[j++] = z[i];
- }
+static int fts3PutColNumber(char **pp, int iCol){
+ int n = 0; /* Number of bytes written */
+ if( iCol ){
+ char *p = *pp; /* Output pointer */
+ n = 1 + sqlite3Fts3PutVarint(&p[1], iCol);
+ *p = 0x01;
+ *pp = &p[n];
}
+ return n;
}
/*
-** The input azIn is a NULL-terminated list of tokens. Remove the first
-** token and all punctuation tokens. Remove the quotes from
-** around string literal tokens.
-**
-** Example:
-**
-** input: tokenize chinese ( 'simplifed' , 'mixed' )
-** output: chinese simplifed mixed
-**
-** Another example:
-**
-** input: delimiters ( '[' , ']' , '...' )
-** output: [ ] ...
-*/
-static void tokenListToIdList(char **azIn){
- int i, j;
- if( azIn ){
- for(i=0, j=-1; azIn[i]; i++){
- if( safe_isalnum(azIn[i][0]) || azIn[i][1] ){
- dequoteString(azIn[i]);
- if( j>=0 ){
- azIn[j] = azIn[i];
+** Compute the union of two position lists. The output written
+** into *pp contains all positions of both *pp1 and *pp2 in sorted
+** order and with any duplicates removed. All pointers are
+** updated appropriately. The caller is responsible for insuring
+** that there is enough space in *pp to hold the complete output.
+*/
+static void fts3PoslistMerge(
+ char **pp, /* Output buffer */
+ char **pp1, /* Left input list */
+ char **pp2 /* Right input list */
+){
+ char *p = *pp;
+ char *p1 = *pp1;
+ char *p2 = *pp2;
+
+ while( *p1 || *p2 ){
+ int iCol1; /* The current column index in pp1 */
+ int iCol2; /* The current column index in pp2 */
+
+ if( *p1==POS_COLUMN ) sqlite3Fts3GetVarint32(&p1[1], &iCol1);
+ else if( *p1==POS_END ) iCol1 = POSITION_LIST_END;
+ else iCol1 = 0;
+
+ if( *p2==POS_COLUMN ) sqlite3Fts3GetVarint32(&p2[1], &iCol2);
+ else if( *p2==POS_END ) iCol2 = POSITION_LIST_END;
+ else iCol2 = 0;
+
+ if( iCol1==iCol2 ){
+ sqlite3_int64 i1 = 0; /* Last position from pp1 */
+ sqlite3_int64 i2 = 0; /* Last position from pp2 */
+ sqlite3_int64 iPrev = 0;
+ int n = fts3PutColNumber(&p, iCol1);
+ p1 += n;
+ p2 += n;
+
+ /* At this point, both p1 and p2 point to the start of column-lists
+ ** for the same column (the column with index iCol1 and iCol2).
+ ** A column-list is a list of non-negative delta-encoded varints, each
+ ** incremented by 2 before being stored. Each list is terminated by a
+ ** POS_END (0) or POS_COLUMN (1). The following block merges the two lists
+ ** and writes the results to buffer p. p is left pointing to the byte
+ ** after the list written. No terminator (POS_END or POS_COLUMN) is
+ ** written to the output.
+ */
+ fts3GetDeltaVarint(&p1, &i1);
+ fts3GetDeltaVarint(&p2, &i2);
+ do {
+ fts3PutDeltaVarint(&p, &iPrev, (i1<i2) ? i1 : i2);
+ iPrev -= 2;
+ if( i1==i2 ){
+ fts3ReadNextPos(&p1, &i1);
+ fts3ReadNextPos(&p2, &i2);
+ }else if( i1<i2 ){
+ fts3ReadNextPos(&p1, &i1);
+ }else{
+ fts3ReadNextPos(&p2, &i2);
}
- j++;
- }
- }
- azIn[j] = 0;
- }
-}
-
-
-/*
-** Find the first alphanumeric token in the string zIn. Null-terminate
-** this token. Remove any quotation marks. And return a pointer to
-** the result.
-*/
-static char *firstToken(char *zIn, char **pzTail){
- int n, ttype;
- while(1){
- n = ftsGetToken(zIn, &ttype);
- if( ttype==TOKEN_SPACE ){
- zIn += n;
- }else if( ttype==TOKEN_EOF ){
- *pzTail = zIn;
- return 0;
+ }while( i1!=POSITION_LIST_END || i2!=POSITION_LIST_END );
+ }else if( iCol1<iCol2 ){
+ p1 += fts3PutColNumber(&p, iCol1);
+ fts3ColumnlistCopy(&p, &p1);
}else{
- zIn[n] = 0;
- *pzTail = &zIn[1];
- dequoteString(zIn);
- return zIn;
+ p2 += fts3PutColNumber(&p, iCol2);
+ fts3ColumnlistCopy(&p, &p2);
}
}
- /*NOTREACHED*/
-}
-/* Return true if...
-**
-** * s begins with the string t, ignoring case
-** * s is longer than t
-** * The first character of s beyond t is not a alphanumeric
-**
-** Ignore leading space in *s.
-**
-** To put it another way, return true if the first token of
-** s[] is t[].
-*/
-static int startsWith(const char *s, const char *t){
- while( safe_isspace(*s) ){ s++; }
- while( *t ){
- if( safe_tolower(*s++)!=safe_tolower(*t++) ) return 0;
- }
- return *s!='_' && !safe_isalnum(*s);
+ *p++ = POS_END;
+ *pp = p;
+ *pp1 = p1 + 1;
+ *pp2 = p2 + 1;
}
/*
-** An instance of this structure defines the "spec" of a
-** full text index. This structure is populated by parseSpec
-** and use by fulltextConnect and fulltextCreate.
-*/
-typedef struct TableSpec {
- const char *zDb; /* Logical database name */
- const char *zName; /* Name of the full-text index */
- int nColumn; /* Number of columns to be indexed */
- char **azColumn; /* Original names of columns to be indexed */
- char **azContentColumn; /* Column names for %_content */
- char **azTokenizer; /* Name of tokenizer and its arguments */
-} TableSpec;
+** nToken==1 searches for adjacent positions.
+*/
+static int fts3PoslistPhraseMerge(
+ char **pp, /* Output buffer */
+ int nToken, /* Maximum difference in token positions */
+ int isSaveLeft, /* Save the left position */
+ char **pp1, /* Left input list */
+ char **pp2 /* Right input list */
+){
+ char *p = (pp ? *pp : 0);
+ char *p1 = *pp1;
+ char *p2 = *pp2;
+
+ int iCol1 = 0;
+ int iCol2 = 0;
+ assert( *p1!=0 && *p2!=0 );
+ if( *p1==POS_COLUMN ){
+ p1++;
+ p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
+ }
+ if( *p2==POS_COLUMN ){
+ p2++;
+ p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
+ }
+
+ while( 1 ){
+ if( iCol1==iCol2 ){
+ char *pSave = p;
+ sqlite3_int64 iPrev = 0;
+ sqlite3_int64 iPos1 = 0;
+ sqlite3_int64 iPos2 = 0;
+
+ if( pp && iCol1 ){
+ *p++ = POS_COLUMN;
+ p += sqlite3Fts3PutVarint(p, iCol1);
+ }
-/*
-** Reclaim all of the memory used by a TableSpec
-*/
-static void clearTableSpec(TableSpec *p) {
- sqlite3_free(p->azColumn);
- sqlite3_free(p->azContentColumn);
- sqlite3_free(p->azTokenizer);
-}
+ assert( *p1!=POS_END && *p1!=POS_COLUMN );
+ assert( *p2!=POS_END && *p2!=POS_COLUMN );
+ fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
+ fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
-/* Parse a CREATE VIRTUAL TABLE statement, which looks like this:
- *
- * CREATE VIRTUAL TABLE email
- * USING fts3(subject, body, tokenize mytokenizer(myarg))
- *
- * We return parsed information in a TableSpec structure.
- *
- */
-static int parseSpec(TableSpec *pSpec, int argc, const char *const*argv,
- char**pzErr){
- int i, n;
- char *z, *zDummy;
- char **azArg;
- const char *zTokenizer = 0; /* argv[] entry describing the tokenizer */
-
- assert( argc>=3 );
- /* Current interface:
- ** argv[0] - module name
- ** argv[1] - database name
- ** argv[2] - table name
- ** argv[3..] - columns, optionally followed by tokenizer specification
- ** and snippet delimiters specification.
- */
+ while( 1 ){
+ if( iPos2>iPos1 && iPos2<=iPos1+nToken ){
+ sqlite3_int64 iSave;
+ if( !pp ){
+ fts3PoslistCopy(0, &p2);
+ fts3PoslistCopy(0, &p1);
+ *pp1 = p1;
+ *pp2 = p2;
+ return 1;
+ }
+ iSave = isSaveLeft ? iPos1 : iPos2;
+ fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2;
+ pSave = 0;
+ }
+ if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){
+ if( (*p2&0xFE)==0 ) break;
+ fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
+ }else{
+ if( (*p1&0xFE)==0 ) break;
+ fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
+ }
+ }
- /* Make a copy of the complete argv[][] array in a single allocation.
- ** The argv[][] array is read-only and transient. We can write to the
- ** copy in order to modify things and the copy is persistent.
- */
- CLEAR(pSpec);
- for(i=n=0; i<argc; i++){
- n += strlen(argv[i]) + 1;
- }
- azArg = sqlite3_malloc( sizeof(char*)*argc + n );
- if( azArg==0 ){
- return SQLITE_NOMEM;
- }
- z = (char*)&azArg[argc];
- for(i=0; i<argc; i++){
- azArg[i] = z;
- strcpy(z, argv[i]);
- z += strlen(z)+1;
- }
+ if( pSave ){
+ assert( pp && p );
+ p = pSave;
+ }
- /* Identify the column names and the tokenizer and delimiter arguments
- ** in the argv[][] array.
- */
- pSpec->zDb = azArg[1];
- pSpec->zName = azArg[2];
- pSpec->nColumn = 0;
- pSpec->azColumn = azArg;
- zTokenizer = "tokenize simple";
- for(i=3; i<argc; ++i){
- if( startsWith(azArg[i],"tokenize") ){
- zTokenizer = azArg[i];
- }else{
- z = azArg[pSpec->nColumn] = firstToken(azArg[i], &zDummy);
- pSpec->nColumn++;
+ fts3ColumnlistCopy(0, &p1);
+ fts3ColumnlistCopy(0, &p2);
+ assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 );
+ if( 0==*p1 || 0==*p2 ) break;
+
+ p1++;
+ p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
+ p2++;
+ p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
}
- }
- if( pSpec->nColumn==0 ){
- azArg[0] = "content";
- pSpec->nColumn = 1;
- }
- /*
- ** Construct the list of content column names.
- **
- ** Each content column name will be of the form cNNAAAA
- ** where NN is the column number and AAAA is the sanitized
- ** column name. "sanitized" means that special characters are
- ** converted to "_". The cNN prefix guarantees that all column
- ** names are unique.
- **
- ** The AAAA suffix is not strictly necessary. It is included
- ** for the convenience of people who might examine the generated
- ** %_content table and wonder what the columns are used for.
- */
- pSpec->azContentColumn = sqlite3_malloc( pSpec->nColumn * sizeof(char *) );
- if( pSpec->azContentColumn==0 ){
- clearTableSpec(pSpec);
- return SQLITE_NOMEM;
- }
- for(i=0; i<pSpec->nColumn; i++){
- char *p;
- pSpec->azContentColumn[i] = sqlite3_mprintf("c%d%s", i, azArg[i]);
- for (p = pSpec->azContentColumn[i]; *p ; ++p) {
- if( !safe_isalnum(*p) ) *p = '_';
+ /* Advance pointer p1 or p2 (whichever corresponds to the smaller of
+ ** iCol1 and iCol2) so that it points to either the 0x00 that marks the
+ ** end of the position list, or the 0x01 that precedes the next
+ ** column-number in the position list.
+ */
+ else if( iCol1<iCol2 ){
+ fts3ColumnlistCopy(0, &p1);
+ if( 0==*p1 ) break;
+ p1++;
+ p1 += sqlite3Fts3GetVarint32(p1, &iCol1);
+ }else{
+ fts3ColumnlistCopy(0, &p2);
+ if( 0==*p2 ) break;
+ p2++;
+ p2 += sqlite3Fts3GetVarint32(p2, &iCol2);
}
}
- /*
- ** Parse the tokenizer specification string.
- */
- pSpec->azTokenizer = tokenizeString(zTokenizer, &n);
- tokenListToIdList(pSpec->azTokenizer);
-
- return SQLITE_OK;
+ fts3PoslistCopy(0, &p2);
+ fts3PoslistCopy(0, &p1);
+ *pp1 = p1;
+ *pp2 = p2;
+ if( !pp || *pp==p ){
+ return 0;
+ }
+ *p++ = 0x00;
+ *pp = p;
+ return 1;
}
/*
-** Generate a CREATE TABLE statement that describes the schema of
-** the virtual table. Return a pointer to this schema string.
-**
-** Space is obtained from sqlite3_mprintf() and should be freed
-** using sqlite3_free().
-*/
-static char *fulltextSchema(
- int nColumn, /* Number of columns */
- const char *const* azColumn, /* List of columns */
- const char *zTableName /* Name of the table */
+** Merge two position-lists as required by the NEAR operator.
+*/
+static int fts3PoslistNearMerge(
+ char **pp, /* Output buffer */
+ char *aTmp, /* Temporary buffer space */
+ int nRight, /* Maximum difference in token positions */
+ int nLeft, /* Maximum difference in token positions */
+ char **pp1, /* IN/OUT: Left input list */
+ char **pp2 /* IN/OUT: Right input list */
){
- int i;
- char *zSchema, *zNext;
- const char *zSep = "(";
- zSchema = sqlite3_mprintf("CREATE TABLE x");
- for(i=0; i<nColumn; i++){
- zNext = sqlite3_mprintf("%s%s%Q", zSchema, zSep, azColumn[i]);
- sqlite3_free(zSchema);
- zSchema = zNext;
- zSep = ",";
+ char *p1 = *pp1;
+ char *p2 = *pp2;
+
+ if( !pp ){
+ if( fts3PoslistPhraseMerge(0, nRight, 0, pp1, pp2) ) return 1;
+ *pp1 = p1;
+ *pp2 = p2;
+ return fts3PoslistPhraseMerge(0, nLeft, 0, pp2, pp1);
+ }else{
+ char *pTmp1 = aTmp;
+ char *pTmp2;
+ char *aTmp2;
+ int res = 1;
+
+ fts3PoslistPhraseMerge(&pTmp1, nRight, 0, pp1, pp2);
+ aTmp2 = pTmp2 = pTmp1;
+ *pp1 = p1;
+ *pp2 = p2;
+ fts3PoslistPhraseMerge(&pTmp2, nLeft, 1, pp2, pp1);
+ if( pTmp1!=aTmp && pTmp2!=aTmp2 ){
+ fts3PoslistMerge(pp, &aTmp, &aTmp2);
+ }else if( pTmp1!=aTmp ){
+ fts3PoslistCopy(pp, &aTmp);
+ }else if( pTmp2!=aTmp2 ){
+ fts3PoslistCopy(pp, &aTmp2);
+ }else{
+ res = 0;
+ }
+
+ return res;
}
- zNext = sqlite3_mprintf("%s,%Q HIDDEN", zSchema, zTableName);
- sqlite3_free(zSchema);
- zSchema = zNext;
- zNext = sqlite3_mprintf("%s,docid HIDDEN)", zSchema);
- sqlite3_free(zSchema);
- return zNext;
}
/*
-** Build a new sqlite3_vtab structure that will describe the
-** fulltext index defined by spec.
+** Values that may be used as the first parameter to fts3DoclistMerge().
*/
-static int constructVtab(
- sqlite3 *db, /* The SQLite database connection */
- fts3Hash *pHash, /* Hash table containing tokenizers */
- TableSpec *spec, /* Parsed spec information from parseSpec() */
- sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */
- char **pzErr /* Write any error message here */
-){
- int rc;
- int n;
- fulltext_vtab *v = 0;
- const sqlite3_tokenizer_module *m = NULL;
- char *schema;
-
- char const *zTok; /* Name of tokenizer to use for this fts table */
- int nTok; /* Length of zTok, including nul terminator */
-
- v = (fulltext_vtab *) sqlite3_malloc(sizeof(fulltext_vtab));
- if( v==0 ) return SQLITE_NOMEM;
- CLEAR(v);
- /* sqlite will initialize v->base */
- v->db = db;
- v->zDb = spec->zDb; /* Freed when azColumn is freed */
- v->zName = spec->zName; /* Freed when azColumn is freed */
- v->nColumn = spec->nColumn;
- v->azContentColumn = spec->azContentColumn;
- spec->azContentColumn = 0;
- v->azColumn = spec->azColumn;
- spec->azColumn = 0;
-
- if( spec->azTokenizer==0 ){
- return SQLITE_NOMEM;
- }
-
- zTok = spec->azTokenizer[0];
- if( !zTok ){
- zTok = "simple";
- }
- nTok = strlen(zTok)+1;
+#define MERGE_NOT 2 /* D + D -> D */
+#define MERGE_AND 3 /* D + D -> D */
+#define MERGE_OR 4 /* D + D -> D */
+#define MERGE_POS_OR 5 /* P + P -> P */
+#define MERGE_PHRASE 6 /* P + P -> D */
+#define MERGE_POS_PHRASE 7 /* P + P -> P */
+#define MERGE_NEAR 8 /* P + P -> D */
+#define MERGE_POS_NEAR 9 /* P + P -> P */
- m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zTok, nTok);
- if( !m ){
- *pzErr = sqlite3_mprintf("unknown tokenizer: %s", spec->azTokenizer[0]);
- rc = SQLITE_ERROR;
- goto err;
- }
+/*
+** Merge the two doclists passed in buffer a1 (size n1 bytes) and a2
+** (size n2 bytes). The output is written to pre-allocated buffer aBuffer,
+** which is guaranteed to be large enough to hold the results. The number
+** of bytes written to aBuffer is stored in *pnBuffer before returning.
+**
+** If successful, SQLITE_OK is returned. Otherwise, if a malloc error
+** occurs while allocating a temporary buffer as part of the merge operation,
+** SQLITE_NOMEM is returned.
+*/
+static int fts3DoclistMerge(
+ int mergetype, /* One of the MERGE_XXX constants */
+ int nParam1, /* Used by MERGE_NEAR and MERGE_POS_NEAR */
+ int nParam2, /* Used by MERGE_NEAR and MERGE_POS_NEAR */
+ char *aBuffer, /* Pre-allocated output buffer */
+ int *pnBuffer, /* OUT: Bytes written to aBuffer */
+ char *a1, /* Buffer containing first doclist */
+ int n1, /* Size of buffer a1 */
+ char *a2, /* Buffer containing second doclist */
+ int n2 /* Size of buffer a2 */
+){
+ sqlite3_int64 i1 = 0;
+ sqlite3_int64 i2 = 0;
+ sqlite3_int64 iPrev = 0;
+
+ char *p = aBuffer;
+ char *p1 = a1;
+ char *p2 = a2;
+ char *pEnd1 = &a1[n1];
+ char *pEnd2 = &a2[n2];
+
+ assert( mergetype==MERGE_OR || mergetype==MERGE_POS_OR
+ || mergetype==MERGE_AND || mergetype==MERGE_NOT
+ || mergetype==MERGE_PHRASE || mergetype==MERGE_POS_PHRASE
+ || mergetype==MERGE_NEAR || mergetype==MERGE_POS_NEAR
+ );
- for(n=0; spec->azTokenizer[n]; n++){}
- if( n ){
- rc = m->xCreate(n-1, (const char*const*)&spec->azTokenizer[1],
- &v->pTokenizer);
- }else{
- rc = m->xCreate(0, 0, &v->pTokenizer);
+ if( !aBuffer ){
+ *pnBuffer = 0;
+ return SQLITE_NOMEM;
}
- if( rc!=SQLITE_OK ) goto err;
- v->pTokenizer->pModule = m;
-
- /* TODO: verify the existence of backing tables foo_content, foo_term */
-
- schema = fulltextSchema(v->nColumn, (const char*const*)v->azColumn,
- spec->zName);
- rc = sqlite3_declare_vtab(db, schema);
- sqlite3_free(schema);
- if( rc!=SQLITE_OK ) goto err;
- memset(v->pFulltextStatements, 0, sizeof(v->pFulltextStatements));
-
- /* Indicate that the buffer is not live. */
- v->nPendingData = -1;
-
- *ppVTab = &v->base;
- FTSTRACE(("FTS3 Connect %p\n", v));
-
- return rc;
+ /* Read the first docid from each doclist */
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+
+ switch( mergetype ){
+ case MERGE_OR:
+ case MERGE_POS_OR:
+ while( p1 || p2 ){
+ if( p2 && p1 && i1==i2 ){
+ fts3PutDeltaVarint(&p, &iPrev, i1);
+ if( mergetype==MERGE_POS_OR ) fts3PoslistMerge(&p, &p1, &p2);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }else if( !p2 || (p1 && i1<i2) ){
+ fts3PutDeltaVarint(&p, &iPrev, i1);
+ if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p1);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ }else{
+ fts3PutDeltaVarint(&p, &iPrev, i2);
+ if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p2);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }
+ }
+ break;
-err:
- fulltext_vtab_destroy(v);
- return rc;
-}
+ case MERGE_AND:
+ while( p1 && p2 ){
+ if( i1==i2 ){
+ fts3PutDeltaVarint(&p, &iPrev, i1);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }else if( i1<i2 ){
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ }else{
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }
+ }
+ break;
-static int fulltextConnect(
- sqlite3 *db,
- void *pAux,
- int argc, const char *const*argv,
- sqlite3_vtab **ppVTab,
- char **pzErr
-){
- TableSpec spec;
- int rc = parseSpec(&spec, argc, argv, pzErr);
- if( rc!=SQLITE_OK ) return rc;
+ case MERGE_NOT:
+ while( p1 ){
+ if( p2 && i1==i2 ){
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }else if( !p2 || i1<i2 ){
+ fts3PutDeltaVarint(&p, &iPrev, i1);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ }else{
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }
+ }
+ break;
- rc = constructVtab(db, (fts3Hash *)pAux, &spec, ppVTab, pzErr);
- clearTableSpec(&spec);
- return rc;
-}
+ case MERGE_POS_PHRASE:
+ case MERGE_PHRASE: {
+ char **ppPos = (mergetype==MERGE_PHRASE ? 0 : &p);
+ while( p1 && p2 ){
+ if( i1==i2 ){
+ char *pSave = p;
+ sqlite3_int64 iPrevSave = iPrev;
+ fts3PutDeltaVarint(&p, &iPrev, i1);
+ if( 0==fts3PoslistPhraseMerge(ppPos, 1, 0, &p1, &p2) ){
+ p = pSave;
+ iPrev = iPrevSave;
+ }
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }else if( i1<i2 ){
+ fts3PoslistCopy(0, &p1);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ }else{
+ fts3PoslistCopy(0, &p2);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }
+ }
+ break;
+ }
-/* The %_content table holds the text of each document, with
-** the docid column exposed as the SQLite rowid for the table.
-*/
-/* TODO(shess) This comment needs elaboration to match the updated
-** code. Work it into the top-of-file comment at that time.
-*/
-static int fulltextCreate(sqlite3 *db, void *pAux,
- int argc, const char * const *argv,
- sqlite3_vtab **ppVTab, char **pzErr){
- int rc;
- TableSpec spec;
- StringBuffer schema;
- FTSTRACE(("FTS3 Create\n"));
+ default: assert( mergetype==MERGE_POS_NEAR || mergetype==MERGE_NEAR ); {
+ char *aTmp = 0;
+ char **ppPos = 0;
- rc = parseSpec(&spec, argc, argv, pzErr);
- if( rc!=SQLITE_OK ) return rc;
+ if( mergetype==MERGE_POS_NEAR ){
+ ppPos = &p;
+ aTmp = sqlite3_malloc(2*(n1+n2+1));
+ if( !aTmp ){
+ return SQLITE_NOMEM;
+ }
+ }
- initStringBuffer(&schema);
- append(&schema, "CREATE TABLE %_content(");
- append(&schema, " docid INTEGER PRIMARY KEY,");
- appendList(&schema, spec.nColumn, spec.azContentColumn);
- append(&schema, ")");
- rc = sql_exec(db, spec.zDb, spec.zName, stringBufferData(&schema));
- stringBufferDestroy(&schema);
- if( rc!=SQLITE_OK ) goto out;
-
- rc = sql_exec(db, spec.zDb, spec.zName,
- "create table %_segments("
- " blockid INTEGER PRIMARY KEY,"
- " block blob"
- ");"
- );
- if( rc!=SQLITE_OK ) goto out;
-
- rc = sql_exec(db, spec.zDb, spec.zName,
- "create table %_segdir("
- " level integer,"
- " idx integer,"
- " start_block integer,"
- " leaves_end_block integer,"
- " end_block integer,"
- " root blob,"
- " primary key(level, idx)"
- ");");
- if( rc!=SQLITE_OK ) goto out;
-
- rc = constructVtab(db, (fts3Hash *)pAux, &spec, ppVTab, pzErr);
-
-out:
- clearTableSpec(&spec);
- return rc;
-}
+ while( p1 && p2 ){
+ if( i1==i2 ){
+ char *pSave = p;
+ sqlite3_int64 iPrevSave = iPrev;
+ fts3PutDeltaVarint(&p, &iPrev, i1);
-/* Decide how to handle an SQL query. */
-static int fulltextBestIndex(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
- fulltext_vtab *v = (fulltext_vtab *)pVTab;
- int i;
- FTSTRACE(("FTS3 BestIndex\n"));
-
- for(i=0; i<pInfo->nConstraint; ++i){
- const struct sqlite3_index_constraint *pConstraint;
- pConstraint = &pInfo->aConstraint[i];
- if( pConstraint->usable ) {
- if( (pConstraint->iColumn==-1 || pConstraint->iColumn==v->nColumn+1) &&
- pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){
- pInfo->idxNum = QUERY_DOCID; /* lookup by docid */
- FTSTRACE(("FTS3 QUERY_DOCID\n"));
- } else if( pConstraint->iColumn>=0 && pConstraint->iColumn<=v->nColumn &&
- pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH ){
- /* full-text search */
- pInfo->idxNum = QUERY_FULLTEXT + pConstraint->iColumn;
- FTSTRACE(("FTS3 QUERY_FULLTEXT %d\n", pConstraint->iColumn));
- } else continue;
-
- pInfo->aConstraintUsage[i].argvIndex = 1;
- pInfo->aConstraintUsage[i].omit = 1;
-
- /* An arbitrary value for now.
- * TODO: Perhaps docid matches should be considered cheaper than
- * full-text searches. */
- pInfo->estimatedCost = 1.0;
+ if( !fts3PoslistNearMerge(ppPos, aTmp, nParam1, nParam2, &p1, &p2) ){
+ iPrev = iPrevSave;
+ p = pSave;
+ }
- return SQLITE_OK;
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }else if( i1<i2 ){
+ fts3PoslistCopy(0, &p1);
+ fts3GetDeltaVarint2(&p1, pEnd1, &i1);
+ }else{
+ fts3PoslistCopy(0, &p2);
+ fts3GetDeltaVarint2(&p2, pEnd2, &i2);
+ }
+ }
+ sqlite3_free(aTmp);
+ break;
}
}
- pInfo->idxNum = QUERY_GENERIC;
- return SQLITE_OK;
-}
-static int fulltextDisconnect(sqlite3_vtab *pVTab){
- FTSTRACE(("FTS3 Disconnect %p\n", pVTab));
- fulltext_vtab_destroy((fulltext_vtab *)pVTab);
+ *pnBuffer = (int)(p-aBuffer);
return SQLITE_OK;
}
-static int fulltextDestroy(sqlite3_vtab *pVTab){
- fulltext_vtab *v = (fulltext_vtab *)pVTab;
- int rc;
-
- FTSTRACE(("FTS3 Destroy %p\n", pVTab));
- rc = sql_exec(v->db, v->zDb, v->zName,
- "drop table if exists %_content;"
- "drop table if exists %_segments;"
- "drop table if exists %_segdir;"
- );
- if( rc!=SQLITE_OK ) return rc;
-
- fulltext_vtab_destroy((fulltext_vtab *)pVTab);
- return SQLITE_OK;
-}
+/*
+** A pointer to an instance of this structure is used as the context
+** argument to sqlite3Fts3SegReaderIterate()
+*/
+typedef struct TermSelect TermSelect;
+struct TermSelect {
+ int isReqPos;
+ char *aaOutput[16]; /* Malloc'd output buffer */
+ int anOutput[16]; /* Size of output in bytes */
+};
-static int fulltextOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
- fulltext_cursor *c;
+/*
+** Merge all doclists in the TermSelect.aaOutput[] array into a single
+** doclist stored in TermSelect.aaOutput[0]. If successful, delete all
+** other doclists (except the aaOutput[0] one) and return SQLITE_OK.
+**
+** If an OOM error occurs, return SQLITE_NOMEM. In this case it is
+** the responsibility of the caller to free any doclists left in the
+** TermSelect.aaOutput[] array.
+*/
+static int fts3TermSelectMerge(TermSelect *pTS){
+ int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR);
+ char *aOut = 0;
+ int nOut = 0;
+ int i;
- c = (fulltext_cursor *) sqlite3_malloc(sizeof(fulltext_cursor));
- if( c ){
- memset(c, 0, sizeof(fulltext_cursor));
- /* sqlite will initialize c->base */
- *ppCursor = &c->base;
- FTSTRACE(("FTS3 Open %p: %p\n", pVTab, c));
- return SQLITE_OK;
- }else{
- return SQLITE_NOMEM;
+ /* Loop through the doclists in the aaOutput[] array. Merge them all
+ ** into a single doclist.
+ */
+ for(i=0; i<SizeofArray(pTS->aaOutput); i++){
+ if( pTS->aaOutput[i] ){
+ if( !aOut ){
+ aOut = pTS->aaOutput[i];
+ nOut = pTS->anOutput[i];
+ pTS->aaOutput[0] = 0;
+ }else{
+ int nNew = nOut + pTS->anOutput[i];
+ char *aNew = sqlite3_malloc(nNew);
+ if( !aNew ){
+ sqlite3_free(aOut);
+ return SQLITE_NOMEM;
+ }
+ fts3DoclistMerge(mergetype, 0, 0,
+ aNew, &nNew, pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut
+ );
+ sqlite3_free(pTS->aaOutput[i]);
+ sqlite3_free(aOut);
+ pTS->aaOutput[i] = 0;
+ aOut = aNew;
+ nOut = nNew;
+ }
+ }
}
-}
-/* Free all of the dynamically allocated memory held by the
-** Snippet
-*/
-static void snippetClear(Snippet *p){
- sqlite3_free(p->aMatch);
- sqlite3_free(p->zOffset);
- sqlite3_free(p->zSnippet);
- CLEAR(p);
+ pTS->aaOutput[0] = aOut;
+ pTS->anOutput[0] = nOut;
+ return SQLITE_OK;
}
/*
-** Append a single entry to the p->aMatch[] log.
-*/
-static void snippetAppendMatch(
- Snippet *p, /* Append the entry to this snippet */
- int iCol, int iTerm, /* The column and query term */
- int iToken, /* Matching token in document */
- int iStart, int nByte /* Offset and size of the match */
+** This function is used as the sqlite3Fts3SegReaderIterate() callback when
+** querying the full-text index for a doclist associated with a term or
+** term-prefix.
+*/
+static int fts3TermSelectCb(
+ Fts3Table *p, /* Virtual table object */
+ void *pContext, /* Pointer to TermSelect structure */
+ char *zTerm,
+ int nTerm,
+ char *aDoclist,
+ int nDoclist
){
- int i;
- struct snippetMatch *pMatch;
- if( p->nMatch+1>=p->nAlloc ){
- p->nAlloc = p->nAlloc*2 + 10;
- p->aMatch = sqlite3_realloc(p->aMatch, p->nAlloc*sizeof(p->aMatch[0]) );
- if( p->aMatch==0 ){
- p->nMatch = 0;
- p->nAlloc = 0;
- return;
- }
- }
- i = p->nMatch++;
- pMatch = &p->aMatch[i];
- pMatch->iCol = iCol;
- pMatch->iTerm = iTerm;
- pMatch->iToken = iToken;
- pMatch->iStart = iStart;
- pMatch->nByte = nByte;
-}
+ TermSelect *pTS = (TermSelect *)pContext;
-/*
-** Sizing information for the circular buffer used in snippetOffsetsOfColumn()
-*/
-#define FTS3_ROTOR_SZ (32)
-#define FTS3_ROTOR_MASK (FTS3_ROTOR_SZ-1)
+ UNUSED_PARAMETER(p);
+ UNUSED_PARAMETER(zTerm);
+ UNUSED_PARAMETER(nTerm);
-/*
-** Function to iterate through the tokens of a compiled expression.
-**
-** Except, skip all tokens on the right-hand side of a NOT operator.
-** This function is used to find tokens as part of snippet and offset
-** generation and we do nt want snippets and offsets to report matches
-** for tokens on the RHS of a NOT.
-*/
-static int fts3NextExprToken(Fts3Expr **ppExpr, int *piToken){
- Fts3Expr *p = *ppExpr;
- int iToken = *piToken;
- if( iToken<0 ){
- /* In this case the expression p is the root of an expression tree.
- ** Move to the first token in the expression tree.
+ if( pTS->aaOutput[0]==0 ){
+ /* If this is the first term selected, copy the doclist to the output
+ ** buffer using memcpy(). TODO: Add a way to transfer control of the
+ ** aDoclist buffer from the caller so as to avoid the memcpy().
*/
- while( p->pLeft ){
- p = p->pLeft;
+ pTS->aaOutput[0] = sqlite3_malloc(nDoclist);
+ pTS->anOutput[0] = nDoclist;
+ if( pTS->aaOutput[0] ){
+ memcpy(pTS->aaOutput[0], aDoclist, nDoclist);
+ }else{
+ return SQLITE_NOMEM;
}
- iToken = 0;
}else{
- assert(p && p->eType==FTSQUERY_PHRASE );
- if( iToken<(p->pPhrase->nToken-1) ){
- iToken++;
- }else{
- iToken = 0;
- while( p->pParent && p->pParent->pLeft!=p ){
- assert( p->pParent->pRight==p );
- p = p->pParent;
+ int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR);
+ char *aMerge = aDoclist;
+ int nMerge = nDoclist;
+ int iOut;
+
+ for(iOut=0; iOut<SizeofArray(pTS->aaOutput); iOut++){
+ char *aNew;
+ int nNew;
+ if( pTS->aaOutput[iOut]==0 ){
+ assert( iOut>0 );
+ pTS->aaOutput[iOut] = aMerge;
+ pTS->anOutput[iOut] = nMerge;
+ break;
}
- p = p->pParent;
- if( p ){
- assert( p->pRight!=0 );
- p = p->pRight;
- while( p->pLeft ){
- p = p->pLeft;
+
+ nNew = nMerge + pTS->anOutput[iOut];
+ aNew = sqlite3_malloc(nNew);
+ if( !aNew ){
+ if( aMerge!=aDoclist ){
+ sqlite3_free(aMerge);
}
+ return SQLITE_NOMEM;
+ }
+ fts3DoclistMerge(mergetype, 0, 0,
+ aNew, &nNew, pTS->aaOutput[iOut], pTS->anOutput[iOut], aMerge, nMerge
+ );
+
+ if( iOut>0 ) sqlite3_free(aMerge);
+ sqlite3_free(pTS->aaOutput[iOut]);
+ pTS->aaOutput[iOut] = 0;
+
+ aMerge = aNew;
+ nMerge = nNew;
+ if( (iOut+1)==SizeofArray(pTS->aaOutput) ){
+ pTS->aaOutput[iOut] = aMerge;
+ pTS->anOutput[iOut] = nMerge;
}
}
}
-
- *ppExpr = p;
- *piToken = iToken;
- return p?1:0;
-}
-
-/*
-** Return TRUE if the expression node pExpr is located beneath the
-** RHS of a NOT operator.
-*/
-static int fts3ExprBeneathNot(Fts3Expr *p){
- Fts3Expr *pParent;
- while( p ){
- pParent = p->pParent;
- if( pParent && pParent->eType==FTSQUERY_NOT && pParent->pRight==p ){
- return 1;
- }
- p = pParent;
- }
- return 0;
+ return SQLITE_OK;
}
/*
-** Add entries to pSnippet->aMatch[] for every match that occurs against
-** document zDoc[0..nDoc-1] which is stored in column iColumn.
-*/
-static void snippetOffsetsOfColumn(
- fulltext_cursor *pCur, /* The fulltest search cursor */
- Snippet *pSnippet, /* The Snippet object to be filled in */
- int iColumn, /* Index of fulltext table column */
- const char *zDoc, /* Text of the fulltext table column */
- int nDoc /* Length of zDoc in bytes */
+** This function retreives the doclist for the specified term (or term
+** prefix) from the database.
+**
+** The returned doclist may be in one of two formats, depending on the
+** value of parameter isReqPos. If isReqPos is zero, then the doclist is
+** a sorted list of delta-compressed docids (a bare doclist). If isReqPos
+** is non-zero, then the returned list is in the same format as is stored
+** in the database without the found length specifier at the start of on-disk
+** doclists.
+*/
+static int fts3TermSelect(
+ Fts3Table *p, /* Virtual table handle */
+ int iColumn, /* Column to query (or -ve for all columns) */
+ const char *zTerm, /* Term to query for */
+ int nTerm, /* Size of zTerm in bytes */
+ int isPrefix, /* True for a prefix search */
+ int isReqPos, /* True to include position lists in output */
+ int *pnOut, /* OUT: Size of buffer at *ppOut */
+ char **ppOut /* OUT: Malloced result buffer */
){
- const sqlite3_tokenizer_module *pTModule; /* The tokenizer module */
- sqlite3_tokenizer *pTokenizer; /* The specific tokenizer */
- sqlite3_tokenizer_cursor *pTCursor; /* Tokenizer cursor */
- fulltext_vtab *pVtab; /* The full text index */
- int nColumn; /* Number of columns in the index */
- int i, j; /* Loop counters */
- int rc; /* Return code */
- unsigned int match, prevMatch; /* Phrase search bitmasks */
- const char *zToken; /* Next token from the tokenizer */
- int nToken; /* Size of zToken */
- int iBegin, iEnd, iPos; /* Offsets of beginning and end */
-
- /* The following variables keep a circular buffer of the last
- ** few tokens */
- unsigned int iRotor = 0; /* Index of current token */
- int iRotorBegin[FTS3_ROTOR_SZ]; /* Beginning offset of token */
- int iRotorLen[FTS3_ROTOR_SZ]; /* Length of token */
-
- pVtab = cursor_vtab(pCur);
- nColumn = pVtab->nColumn;
- pTokenizer = pVtab->pTokenizer;
- pTModule = pTokenizer->pModule;
- rc = pTModule->xOpen(pTokenizer, zDoc, nDoc, &pTCursor);
- if( rc ) return;
- pTCursor->pTokenizer = pTokenizer;
-
- prevMatch = 0;
- while( !pTModule->xNext(pTCursor, &zToken, &nToken, &iBegin, &iEnd, &iPos) ){
- Fts3Expr *pIter = pCur->pExpr;
- int iIter = -1;
- iRotorBegin[iRotor&FTS3_ROTOR_MASK] = iBegin;
- iRotorLen[iRotor&FTS3_ROTOR_MASK] = iEnd-iBegin;
- match = 0;
- for(i=0; i<(FTS3_ROTOR_SZ-1) && fts3NextExprToken(&pIter, &iIter); i++){
- int nPhrase; /* Number of tokens in current phrase */
- struct PhraseToken *pToken; /* Current token */
- int iCol; /* Column index */
-
- if( fts3ExprBeneathNot(pIter) ) continue;
- nPhrase = pIter->pPhrase->nToken;
- pToken = &pIter->pPhrase->aToken[iIter];
- iCol = pIter->pPhrase->iColumn;
- if( iCol>=0 && iCol<nColumn && iCol!=iColumn ) continue;
- if( pToken->n>nToken ) continue;
- if( !pToken->isPrefix && pToken->n<nToken ) continue;
- assert( pToken->n<=nToken );
- if( memcmp(pToken->z, zToken, pToken->n) ) continue;
- if( iIter>0 && (prevMatch & (1<<i))==0 ) continue;
- match |= 1<<i;
- if( i==(FTS3_ROTOR_SZ-2) || nPhrase==iIter+1 ){
- for(j=nPhrase-1; j>=0; j--){
- int k = (iRotor-j) & FTS3_ROTOR_MASK;
- snippetAppendMatch(pSnippet, iColumn, i-j, iPos-j,
- iRotorBegin[k], iRotorLen[k]);
- }
+ int i;
+ TermSelect tsc;
+ Fts3SegFilter filter; /* Segment term filter configuration */
+ Fts3SegReader **apSegment; /* Array of segments to read data from */
+ int nSegment = 0; /* Size of apSegment array */
+ int nAlloc = 16; /* Allocated size of segment array */
+ int rc; /* Return code */
+ sqlite3_stmt *pStmt = 0; /* SQL statement to scan %_segdir table */
+ int iAge = 0; /* Used to assign ages to segments */
+
+ apSegment = (Fts3SegReader **)sqlite3_malloc(sizeof(Fts3SegReader*)*nAlloc);
+ if( !apSegment ) return SQLITE_NOMEM;
+ rc = sqlite3Fts3SegReaderPending(p, zTerm, nTerm, isPrefix, &apSegment[0]);
+ if( rc!=SQLITE_OK ) goto finished;
+ if( apSegment[0] ){
+ nSegment = 1;
+ }
+
+ /* Loop through the entire %_segdir table. For each segment, create a
+ ** Fts3SegReader to iterate through the subset of the segment leaves
+ ** that may contain a term that matches zTerm/nTerm. For non-prefix
+ ** searches, this is always a single leaf. For prefix searches, this
+ ** may be a contiguous block of leaves.
+ **
+ ** The code in this loop does not actually load any leaves into memory
+ ** (unless the root node happens to be a leaf). It simply examines the
+ ** b-tree structure to determine which leaves need to be inspected.
+ */
+ rc = sqlite3Fts3AllSegdirs(p, &pStmt);
+ while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){
+ Fts3SegReader *pNew = 0;
+ int nRoot = sqlite3_column_bytes(pStmt, 4);
+ char const *zRoot = sqlite3_column_blob(pStmt, 4);
+ if( sqlite3_column_int64(pStmt, 1)==0 ){
+ /* The entire segment is stored on the root node (which must be a
+ ** leaf). Do not bother inspecting any data in this case, just
+ ** create a Fts3SegReader to scan the single leaf.
+ */
+ rc = sqlite3Fts3SegReaderNew(p, iAge, 0, 0, 0, zRoot, nRoot, &pNew);
+ }else{
+ int rc2; /* Return value of sqlite3Fts3ReadBlock() */
+ sqlite3_int64 i1; /* Blockid of leaf that may contain zTerm */
+ rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &i1);
+ if( rc==SQLITE_OK ){
+ sqlite3_int64 i2 = sqlite3_column_int64(pStmt, 2);
+ rc = sqlite3Fts3SegReaderNew(p, iAge, i1, i2, 0, 0, 0, &pNew);
}
- }
- prevMatch = match<<1;
- iRotor++;
- }
- pTModule->xClose(pTCursor);
-}
-/*
-** Remove entries from the pSnippet structure to account for the NEAR
-** operator. When this is called, pSnippet contains the list of token
-** offsets produced by treating all NEAR operators as AND operators.
-** This function removes any entries that should not be present after
-** accounting for the NEAR restriction. For example, if the queried
-** document is:
-**
-** "A B C D E A"
-**
-** and the query is:
-**
-** A NEAR/0 E
-**
-** then when this function is called the Snippet contains token offsets
-** 0, 4 and 5. This function removes the "0" entry (because the first A
-** is not near enough to an E).
-**
-** When this function is called, the value pointed to by parameter piLeft is
-** the integer id of the left-most token in the expression tree headed by
-** pExpr. This function increments *piLeft by the total number of tokens
-** in the expression tree headed by pExpr.
-**
-** Return 1 if any trimming occurs. Return 0 if no trimming is required.
-*/
-static int trimSnippetOffsets(
- Fts3Expr *pExpr, /* The search expression */
- Snippet *pSnippet, /* The set of snippet offsets to be trimmed */
- int *piLeft /* Index of left-most token in pExpr */
-){
- if( pExpr ){
- if( trimSnippetOffsets(pExpr->pLeft, pSnippet, piLeft) ){
- return 1;
+ /* The following call to ReadBlock() serves to reset the SQL statement
+ ** used to retrieve blocks of data from the %_segments table. If it is
+ ** not reset here, then it may remain classified as an active statement
+ ** by SQLite, which may lead to "DROP TABLE" or "DETACH" commands
+ ** failing.
+ */
+ rc2 = sqlite3Fts3ReadBlock(p, 0, 0, 0);
+ if( rc==SQLITE_OK ){
+ rc = rc2;
+ }
}
+ iAge++;
- switch( pExpr->eType ){
- case FTSQUERY_PHRASE:
- *piLeft += pExpr->pPhrase->nToken;
- break;
- case FTSQUERY_NEAR: {
- /* The right-hand-side of a NEAR operator is always a phrase. The
- ** left-hand-side is either a phrase or an expression tree that is
- ** itself headed by a NEAR operator. The following initializations
- ** set local variable iLeft to the token number of the left-most
- ** token in the right-hand phrase, and iRight to the right most
- ** token in the same phrase. For example, if we had:
- **
- ** <col> MATCH '"abc def" NEAR/2 "ghi jkl"'
- **
- ** then iLeft will be set to 2 (token number of ghi) and nToken will
- ** be set to 4.
- */
- Fts3Expr *pLeft = pExpr->pLeft;
- Fts3Expr *pRight = pExpr->pRight;
- int iLeft = *piLeft;
- int nNear = pExpr->nNear;
- int nToken = pRight->pPhrase->nToken;
- int jj, ii;
- if( pLeft->eType==FTSQUERY_NEAR ){
- pLeft = pLeft->pRight;
- }
- assert( pRight->eType==FTSQUERY_PHRASE );
- assert( pLeft->eType==FTSQUERY_PHRASE );
- nToken += pLeft->pPhrase->nToken;
-
- for(ii=0; ii<pSnippet->nMatch; ii++){
- struct snippetMatch *p = &pSnippet->aMatch[ii];
- if( p->iTerm==iLeft ){
- int isOk = 0;
- /* Snippet ii is an occurence of query term iLeft in the document.
- ** It occurs at position (p->iToken) of the document. We now
- ** search for an instance of token (iLeft-1) somewhere in the
- ** range (p->iToken - nNear)...(p->iToken + nNear + nToken) within
- ** the set of snippetMatch structures. If one is found, proceed.
- ** If one cannot be found, then remove snippets ii..(ii+N-1)
- ** from the matching snippets, where N is the number of tokens
- ** in phrase pRight->pPhrase.
- */
- for(jj=0; isOk==0 && jj<pSnippet->nMatch; jj++){
- struct snippetMatch *p2 = &pSnippet->aMatch[jj];
- if( p2->iTerm==(iLeft-1) ){
- if( p2->iToken>=(p->iToken-nNear-1)
- && p2->iToken<(p->iToken+nNear+nToken)
- ){
- isOk = 1;
- }
- }
- }
- if( !isOk ){
- int kk;
- for(kk=0; kk<pRight->pPhrase->nToken; kk++){
- pSnippet->aMatch[kk+ii].iTerm = -2;
- }
- return 1;
- }
- }
- if( p->iTerm==(iLeft-1) ){
- int isOk = 0;
- for(jj=0; isOk==0 && jj<pSnippet->nMatch; jj++){
- struct snippetMatch *p2 = &pSnippet->aMatch[jj];
- if( p2->iTerm==iLeft ){
- if( p2->iToken<=(p->iToken+nNear+1)
- && p2->iToken>(p->iToken-nNear-nToken)
- ){
- isOk = 1;
- }
- }
- }
- if( !isOk ){
- int kk;
- for(kk=0; kk<pLeft->pPhrase->nToken; kk++){
- pSnippet->aMatch[ii-kk].iTerm = -2;
- }
- return 1;
- }
- }
+ /* If a new Fts3SegReader was allocated, add it to the apSegment array. */
+ assert( pNew!=0 || rc!=SQLITE_OK );
+ if( pNew ){
+ if( nSegment==nAlloc ){
+ Fts3SegReader **pArray;
+ nAlloc += 16;
+ pArray = (Fts3SegReader **)sqlite3_realloc(
+ apSegment, nAlloc*sizeof(Fts3SegReader *)
+ );
+ if( !pArray ){
+ sqlite3Fts3SegReaderFree(p, pNew);
+ rc = SQLITE_NOMEM;
+ goto finished;
}
- break;
+ apSegment = pArray;
}
+ apSegment[nSegment++] = pNew;
}
-
- if( trimSnippetOffsets(pExpr->pRight, pSnippet, piLeft) ){
- return 1;
- }
- }
- return 0;
-}
-
-/*
-** Compute all offsets for the current row of the query.
-** If the offsets have already been computed, this routine is a no-op.
-*/
-static void snippetAllOffsets(fulltext_cursor *p){
- int nColumn;
- int iColumn, i;
- int iFirst, iLast;
- int iTerm = 0;
- fulltext_vtab *pFts = cursor_vtab(p);
-
- if( p->snippet.nMatch || p->pExpr==0 ){
- return;
- }
- nColumn = pFts->nColumn;
- iColumn = (p->iCursorType - QUERY_FULLTEXT);
- if( iColumn<0 || iColumn>=nColumn ){
- /* Look for matches over all columns of the full-text index */
- iFirst = 0;
- iLast = nColumn-1;
- }else{
- /* Look for matches in the iColumn-th column of the index only */
- iFirst = iColumn;
- iLast = iColumn;
}
- for(i=iFirst; i<=iLast; i++){
- const char *zDoc;
- int nDoc;
- zDoc = (const char*)sqlite3_column_text(p->pStmt, i+1);
- nDoc = sqlite3_column_bytes(p->pStmt, i+1);
- snippetOffsetsOfColumn(p, &p->snippet, i, zDoc, nDoc);
+ if( rc!=SQLITE_DONE ){
+ assert( rc!=SQLITE_OK );
+ goto finished;
}
- while( trimSnippetOffsets(p->pExpr, &p->snippet, &iTerm) ){
- iTerm = 0;
- }
-}
+ memset(&tsc, 0, sizeof(TermSelect));
+ tsc.isReqPos = isReqPos;
-/*
-** Convert the information in the aMatch[] array of the snippet
-** into the string zOffset[0..nOffset-1]. This string is used as
-** the return of the SQL offsets() function.
-*/
-static void snippetOffsetText(Snippet *p){
- int i;
- int cnt = 0;
- StringBuffer sb;
- char zBuf[200];
- if( p->zOffset ) return;
- initStringBuffer(&sb);
- for(i=0; i<p->nMatch; i++){
- struct snippetMatch *pMatch = &p->aMatch[i];
- if( pMatch->iTerm>=0 ){
- /* If snippetMatch.iTerm is less than 0, then the match was
- ** discarded as part of processing the NEAR operator (see the
- ** trimSnippetOffsetsForNear() function for details). Ignore
- ** it in this case
- */
- zBuf[0] = ' ';
- sqlite3_snprintf(sizeof(zBuf)-1, &zBuf[cnt>0], "%d %d %d %d",
- pMatch->iCol, pMatch->iTerm, pMatch->iStart, pMatch->nByte);
- append(&sb, zBuf);
- cnt++;
- }
- }
- p->zOffset = stringBufferData(&sb);
- p->nOffset = stringBufferLength(&sb);
-}
+ filter.flags = FTS3_SEGMENT_IGNORE_EMPTY
+ | (isPrefix ? FTS3_SEGMENT_PREFIX : 0)
+ | (isReqPos ? FTS3_SEGMENT_REQUIRE_POS : 0)
+ | (iColumn<p->nColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0);
+ filter.iCol = iColumn;
+ filter.zTerm = zTerm;
+ filter.nTerm = nTerm;
-/*
-** zDoc[0..nDoc-1] is phrase of text. aMatch[0..nMatch-1] are a set
-** of matching words some of which might be in zDoc. zDoc is column
-** number iCol.
-**
-** iBreak is suggested spot in zDoc where we could begin or end an
-** excerpt. Return a value similar to iBreak but possibly adjusted
-** to be a little left or right so that the break point is better.
-*/
-static int wordBoundary(
- int iBreak, /* The suggested break point */
- const char *zDoc, /* Document text */
- int nDoc, /* Number of bytes in zDoc[] */
- struct snippetMatch *aMatch, /* Matching words */
- int nMatch, /* Number of entries in aMatch[] */
- int iCol /* The column number for zDoc[] */
-){
- int i;
- if( iBreak<=10 ){
- return 0;
- }
- if( iBreak>=nDoc-10 ){
- return nDoc;
+ rc = sqlite3Fts3SegReaderIterate(p, apSegment, nSegment, &filter,
+ fts3TermSelectCb, (void *)&tsc
+ );
+ if( rc==SQLITE_OK ){
+ rc = fts3TermSelectMerge(&tsc);
}
- for(i=0; i<nMatch && aMatch[i].iCol<iCol; i++){}
- while( i<nMatch && aMatch[i].iStart+aMatch[i].nByte<iBreak ){ i++; }
- if( i<nMatch ){
- if( aMatch[i].iStart<iBreak+10 ){
- return aMatch[i].iStart;
- }
- if( i>0 && aMatch[i-1].iStart+aMatch[i-1].nByte>=iBreak ){
- return aMatch[i-1].iStart;
+
+ if( rc==SQLITE_OK ){
+ *ppOut = tsc.aaOutput[0];
+ *pnOut = tsc.anOutput[0];
+ }else{
+ for(i=0; i<SizeofArray(tsc.aaOutput); i++){
+ sqlite3_free(tsc.aaOutput[i]);
}
}
- for(i=1; i<=10; i++){
- if( safe_isspace(zDoc[iBreak-i]) ){
- return iBreak - i + 1;
- }
- if( safe_isspace(zDoc[iBreak+i]) ){
- return iBreak + i + 1;
- }
+
+finished:
+ sqlite3_reset(pStmt);
+ for(i=0; i<nSegment; i++){
+ sqlite3Fts3SegReaderFree(p, apSegment[i]);
}
- return iBreak;
+ sqlite3_free(apSegment);
+ return rc;
}
-
-/*
-** Allowed values for Snippet.aMatch[].snStatus
-*/
-#define SNIPPET_IGNORE 0 /* It is ok to omit this match from the snippet */
-#define SNIPPET_DESIRED 1 /* We want to include this match in the snippet */
-
-/*
-** Generate the text of a snippet.
+/*
+** Return a DocList corresponding to the phrase *pPhrase.
*/
-static void snippetText(
- fulltext_cursor *pCursor, /* The cursor we need the snippet for */
- const char *zStartMark, /* Markup to appear before each match */
- const char *zEndMark, /* Markup to appear after each match */
- const char *zEllipsis /* Ellipsis mark */
+static int fts3PhraseSelect(
+ Fts3Table *p, /* Virtual table handle */
+ Fts3Phrase *pPhrase, /* Phrase to return a doclist for */
+ int isReqPos, /* True if output should contain positions */
+ char **paOut, /* OUT: Pointer to malloc'd result buffer */
+ int *pnOut /* OUT: Size of buffer at *paOut */
){
- int i, j;
- struct snippetMatch *aMatch;
- int nMatch;
- int nDesired;
- StringBuffer sb;
- int tailCol;
- int tailOffset;
- int iCol;
- int nDoc;
- const char *zDoc;
- int iStart, iEnd;
- int tailEllipsis = 0;
- int iMatch;
-
+ char *pOut = 0;
+ int nOut = 0;
+ int rc = SQLITE_OK;
+ int ii;
+ int iCol = pPhrase->iColumn;
+ int isTermPos = (pPhrase->nToken>1 || isReqPos);
- sqlite3_free(pCursor->snippet.zSnippet);
- pCursor->snippet.zSnippet = 0;
- aMatch = pCursor->snippet.aMatch;
- nMatch = pCursor->snippet.nMatch;
- initStringBuffer(&sb);
+ for(ii=0; ii<pPhrase->nToken; ii++){
+ struct PhraseToken *pTok = &pPhrase->aToken[ii];
+ char *z = pTok->z; /* Next token of the phrase */
+ int n = pTok->n; /* Size of z in bytes */
+ int isPrefix = pTok->isPrefix;/* True if token is a prefix */
+ char *pList; /* Pointer to token doclist */
+ int nList; /* Size of buffer at pList */
- for(i=0; i<nMatch; i++){
- aMatch[i].snStatus = SNIPPET_IGNORE;
- }
- nDesired = 0;
- for(i=0; i<FTS3_ROTOR_SZ; i++){
- for(j=0; j<nMatch; j++){
- if( aMatch[j].iTerm==i ){
- aMatch[j].snStatus = SNIPPET_DESIRED;
- nDesired++;
- break;
- }
- }
- }
+ rc = fts3TermSelect(p, iCol, z, n, isPrefix, isTermPos, &nList, &pList);
+ if( rc!=SQLITE_OK ) break;
- iMatch = 0;
- tailCol = -1;
- tailOffset = 0;
- for(i=0; i<nMatch && nDesired>0; i++){
- if( aMatch[i].snStatus!=SNIPPET_DESIRED ) continue;
- nDesired--;
- iCol = aMatch[i].iCol;
- zDoc = (const char*)sqlite3_column_text(pCursor->pStmt, iCol+1);
- nDoc = sqlite3_column_bytes(pCursor->pStmt, iCol+1);
- iStart = aMatch[i].iStart - 40;
- iStart = wordBoundary(iStart, zDoc, nDoc, aMatch, nMatch, iCol);
- if( iStart<=10 ){
- iStart = 0;
- }
- if( iCol==tailCol && iStart<=tailOffset+20 ){
- iStart = tailOffset;
- }
- if( (iCol!=tailCol && tailCol>=0) || iStart!=tailOffset ){
- trimWhiteSpace(&sb);
- appendWhiteSpace(&sb);
- append(&sb, zEllipsis);
- appendWhiteSpace(&sb);
- }
- iEnd = aMatch[i].iStart + aMatch[i].nByte + 40;
- iEnd = wordBoundary(iEnd, zDoc, nDoc, aMatch, nMatch, iCol);
- if( iEnd>=nDoc-10 ){
- iEnd = nDoc;
- tailEllipsis = 0;
+ if( ii==0 ){
+ pOut = pList;
+ nOut = nList;
}else{
- tailEllipsis = 1;
- }
- while( iMatch<nMatch && aMatch[iMatch].iCol<iCol ){ iMatch++; }
- while( iStart<iEnd ){
- while( iMatch<nMatch && aMatch[iMatch].iStart<iStart
- && aMatch[iMatch].iCol<=iCol ){
- iMatch++;
- }
- if( iMatch<nMatch && aMatch[iMatch].iStart<iEnd
- && aMatch[iMatch].iCol==iCol ){
- nappend(&sb, &zDoc[iStart], aMatch[iMatch].iStart - iStart);
- iStart = aMatch[iMatch].iStart;
- append(&sb, zStartMark);
- nappend(&sb, &zDoc[iStart], aMatch[iMatch].nByte);
- append(&sb, zEndMark);
- iStart += aMatch[iMatch].nByte;
- for(j=iMatch+1; j<nMatch; j++){
- if( aMatch[j].iTerm==aMatch[iMatch].iTerm
- && aMatch[j].snStatus==SNIPPET_DESIRED ){
- nDesired--;
- aMatch[j].snStatus = SNIPPET_IGNORE;
- }
- }
- }else{
- nappend(&sb, &zDoc[iStart], iEnd - iStart);
- iStart = iEnd;
+ /* Merge the new term list and the current output. If this is the
+ ** last term in the phrase, and positions are not required in the
+ ** output of this function, the positions can be dropped as part
+ ** of this merge. Either way, the result of this merge will be
+ ** smaller than nList bytes. The code in fts3DoclistMerge() is written
+ ** so that it is safe to use pList as the output as well as an input
+ ** in this case.
+ */
+ int mergetype = MERGE_POS_PHRASE;
+ if( ii==pPhrase->nToken-1 && !isReqPos ){
+ mergetype = MERGE_PHRASE;
}
+ fts3DoclistMerge(mergetype, 0, 0, pList, &nOut, pOut, nOut, pList, nList);
+ sqlite3_free(pOut);
+ pOut = pList;
}
- tailCol = iCol;
- tailOffset = iEnd;
- }
- trimWhiteSpace(&sb);
- if( tailEllipsis ){
- appendWhiteSpace(&sb);
- append(&sb, zEllipsis);
+ assert( nOut==0 || pOut!=0 );
}
- pCursor->snippet.zSnippet = stringBufferData(&sb);
- pCursor->snippet.nSnippet = stringBufferLength(&sb);
-}
-
-/*
-** Close the cursor. For additional information see the documentation
-** on the xClose method of the virtual table interface.
-*/
-static int fulltextClose(sqlite3_vtab_cursor *pCursor){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
- FTSTRACE(("FTS3 Close %p\n", c));
- sqlite3_finalize(c->pStmt);
- sqlite3Fts3ExprFree(c->pExpr);
- snippetClear(&c->snippet);
- if( c->result.nData!=0 ){
- dlrDestroy(&c->reader);
+ if( rc==SQLITE_OK ){
+ *paOut = pOut;
+ *pnOut = nOut;
+ }else{
+ sqlite3_free(pOut);
}
- dataBufferDestroy(&c->result);
- sqlite3_free(c);
- return SQLITE_OK;
+ return rc;
}
-static int fulltextNext(sqlite3_vtab_cursor *pCursor){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
+static int fts3NearMerge(
+ int mergetype, /* MERGE_POS_NEAR or MERGE_NEAR */
+ int nNear, /* Parameter to NEAR operator */
+ int nTokenLeft, /* Number of tokens in LHS phrase arg */
+ char *aLeft, /* Doclist for LHS (incl. positions) */
+ int nLeft, /* Size of LHS doclist in bytes */
+ int nTokenRight, /* As nTokenLeft */
+ char *aRight, /* As aLeft */
+ int nRight, /* As nRight */
+ char **paOut, /* OUT: Results of merge (malloced) */
+ int *pnOut /* OUT: Sized of output buffer */
+){
+ char *aOut;
int rc;
- FTSTRACE(("FTS3 Next %p\n", pCursor));
- snippetClear(&c->snippet);
- if( c->iCursorType < QUERY_FULLTEXT ){
- /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
- rc = sqlite3_step(c->pStmt);
- switch( rc ){
- case SQLITE_ROW:
- c->eof = 0;
- return SQLITE_OK;
- case SQLITE_DONE:
- c->eof = 1;
- return SQLITE_OK;
- default:
- c->eof = 1;
- return rc;
- }
- } else { /* full-text query */
- rc = sqlite3_reset(c->pStmt);
- if( rc!=SQLITE_OK ) return rc;
+ assert( mergetype==MERGE_POS_NEAR || MERGE_NEAR );
- if( c->result.nData==0 || dlrAtEnd(&c->reader) ){
- c->eof = 1;
- return SQLITE_OK;
- }
- rc = sqlite3_bind_int64(c->pStmt, 1, dlrDocid(&c->reader));
- if( rc!=SQLITE_OK ) return rc;
- rc = dlrStep(&c->reader);
- if( rc!=SQLITE_OK ) return rc;
- /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
- rc = sqlite3_step(c->pStmt);
- if( rc==SQLITE_ROW ){ /* the case we expect */
- c->eof = 0;
- return SQLITE_OK;
+ aOut = sqlite3_malloc(nLeft+nRight+1);
+ if( aOut==0 ){
+ rc = SQLITE_NOMEM;
+ }else{
+ rc = fts3DoclistMerge(mergetype, nNear+nTokenRight, nNear+nTokenLeft,
+ aOut, pnOut, aLeft, nLeft, aRight, nRight
+ );
+ if( rc!=SQLITE_OK ){
+ sqlite3_free(aOut);
+ aOut = 0;
}
- /* Corrupt if the index refers to missing document. */
- if( rc==SQLITE_DONE ) return SQLITE_CORRUPT_BKPT;
-
- return rc;
}
-}
-
-
-/* TODO(shess) If we pushed LeafReader to the top of the file, or to
-** another file, term_select() could be pushed above
-** docListOfTerm().
-*/
-static int termSelect(fulltext_vtab *v, int iColumn,
- const char *pTerm, int nTerm, int isPrefix,
- DocListType iType, DataBuffer *out);
-/*
-** Return a DocList corresponding to the phrase *pPhrase.
-**
-** The resulting DL_DOCIDS doclist is stored in pResult, which is
-** overwritten.
-*/
-static int docListOfPhrase(
- fulltext_vtab *pTab, /* The full text index */
- Fts3Phrase *pPhrase, /* Phrase to return a doclist corresponding to */
- DocListType eListType, /* Either DL_DOCIDS or DL_POSITIONS */
- DataBuffer *pResult /* Write the result here */
-){
- int ii;
- int rc = SQLITE_OK;
- int iCol = pPhrase->iColumn;
- DocListType eType = eListType;
- assert( eType==DL_POSITIONS || eType==DL_DOCIDS );
- if( pPhrase->nToken>1 ){
- eType = DL_POSITIONS;
- }
+ *paOut = aOut;
+ return rc;
+}
- /* This code should never be called with buffered updates. */
- assert( pTab->nPendingData<0 );
+int sqlite3Fts3ExprNearTrim(Fts3Expr *pLeft, Fts3Expr *pRight, int nNear){
+ int rc;
+ if( pLeft->aDoclist==0 || pRight->aDoclist==0 ){
+ sqlite3_free(pLeft->aDoclist);
+ sqlite3_free(pRight->aDoclist);
+ pRight->aDoclist = 0;
+ pLeft->aDoclist = 0;
+ rc = SQLITE_OK;
+ }else{
+ char *aOut;
+ int nOut;
- for(ii=0; rc==SQLITE_OK && ii<pPhrase->nToken; ii++){
- DataBuffer tmp;
- struct PhraseToken *p = &pPhrase->aToken[ii];
- rc = termSelect(pTab, iCol, p->z, p->n, p->isPrefix, eType, &tmp);
- if( rc==SQLITE_OK ){
- if( ii==0 ){
- *pResult = tmp;
- }else{
- DataBuffer res = *pResult;
- dataBufferInit(pResult, 0);
- if( ii==(pPhrase->nToken-1) ){
- eType = eListType;
- }
- rc = docListPhraseMerge(
- res.pData, res.nData, tmp.pData, tmp.nData, 0, 0, eType, pResult
- );
- dataBufferDestroy(&res);
- dataBufferDestroy(&tmp);
- if( rc!= SQLITE_OK ) return rc;
- }
- }
+ rc = fts3NearMerge(MERGE_POS_NEAR, nNear,
+ pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist,
+ pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist,
+ &aOut, &nOut
+ );
+ if( rc!=SQLITE_OK ) return rc;
+ sqlite3_free(pRight->aDoclist);
+ pRight->aDoclist = aOut;
+ pRight->nDoclist = nOut;
+
+ rc = fts3NearMerge(MERGE_POS_NEAR, nNear,
+ pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist,
+ pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist,
+ &aOut, &nOut
+ );
+ sqlite3_free(pLeft->aDoclist);
+ pLeft->aDoclist = aOut;
+ pLeft->nDoclist = nOut;
}
-
return rc;
}
/*
-** Evaluate the full-text expression pExpr against fts3 table pTab. Write
-** the results into pRes.
+** Evaluate the full-text expression pExpr against fts3 table pTab. Store
+** the resulting doclist in *paOut and *pnOut. This routine mallocs for
+** the space needed to store the output. The caller is responsible for
+** freeing the space when it has finished.
*/
static int evalFts3Expr(
- fulltext_vtab *pTab, /* Fts3 Virtual table object */
- Fts3Expr *pExpr, /* Parsed fts3 expression */
- DataBuffer *pRes /* OUT: Write results of the expression here */
+ Fts3Table *p, /* Virtual table handle */
+ Fts3Expr *pExpr, /* Parsed fts3 expression */
+ char **paOut, /* OUT: Pointer to malloc'd result buffer */
+ int *pnOut, /* OUT: Size of buffer at *paOut */
+ int isReqPos /* Require positions in output buffer */
){
- int rc = SQLITE_OK;
+ int rc = SQLITE_OK; /* Return code */
- /* Initialize the output buffer. If this is an empty query (pExpr==0),
- ** this is all that needs to be done. Empty queries produce empty
- ** result sets.
- */
- dataBufferInit(pRes, 0);
+ /* Zero the output parameters. */
+ *paOut = 0;
+ *pnOut = 0;
if( pExpr ){
+ assert( pExpr->eType==FTSQUERY_PHRASE
+ || pExpr->eType==FTSQUERY_NEAR
+ || isReqPos==0
+ );
if( pExpr->eType==FTSQUERY_PHRASE ){
- DocListType eType = DL_DOCIDS;
- if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){
- eType = DL_POSITIONS;
- }
- rc = docListOfPhrase(pTab, pExpr->pPhrase, eType, pRes);
+ rc = fts3PhraseSelect(p, pExpr->pPhrase,
+ isReqPos || (pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR),
+ paOut, pnOut
+ );
}else{
- DataBuffer lhs;
- DataBuffer rhs;
+ char *aLeft;
+ char *aRight;
+ int nLeft;
+ int nRight;
- dataBufferInit(&rhs, 0);
- if( SQLITE_OK==(rc = evalFts3Expr(pTab, pExpr->pLeft, &lhs))
- && SQLITE_OK==(rc = evalFts3Expr(pTab, pExpr->pRight, &rhs))
+ if( 0==(rc = evalFts3Expr(p, pExpr->pRight, &aRight, &nRight, isReqPos))
+ && 0==(rc = evalFts3Expr(p, pExpr->pLeft, &aLeft, &nLeft, isReqPos))
){
+ assert( pExpr->eType==FTSQUERY_NEAR || pExpr->eType==FTSQUERY_OR
+ || pExpr->eType==FTSQUERY_AND || pExpr->eType==FTSQUERY_NOT
+ );
switch( pExpr->eType ){
case FTSQUERY_NEAR: {
- int nToken;
Fts3Expr *pLeft;
- DocListType eType = DL_DOCIDS;
+ Fts3Expr *pRight;
+ int mergetype = isReqPos ? MERGE_POS_NEAR : MERGE_NEAR;
+
if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){
- eType = DL_POSITIONS;
+ mergetype = MERGE_POS_NEAR;
}
pLeft = pExpr->pLeft;
while( pLeft->eType==FTSQUERY_NEAR ){
pLeft=pLeft->pRight;
}
- assert( pExpr->pRight->eType==FTSQUERY_PHRASE );
+ pRight = pExpr->pRight;
+ assert( pRight->eType==FTSQUERY_PHRASE );
assert( pLeft->eType==FTSQUERY_PHRASE );
- nToken = pLeft->pPhrase->nToken + pExpr->pRight->pPhrase->nToken;
- rc = docListPhraseMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData,
- pExpr->nNear+1, nToken, eType, pRes
+
+ rc = fts3NearMerge(mergetype, pExpr->nNear,
+ pLeft->pPhrase->nToken, aLeft, nLeft,
+ pRight->pPhrase->nToken, aRight, nRight,
+ paOut, pnOut
);
+ sqlite3_free(aLeft);
break;
}
- case FTSQUERY_NOT: {
- rc = docListExceptMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData,pRes);
- break;
- }
- case FTSQUERY_AND: {
- rc = docListAndMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData, pRes);
+
+ case FTSQUERY_OR: {
+ /* Allocate a buffer for the output. The maximum size is the
+ ** sum of the sizes of the two input buffers. The +1 term is
+ ** so that a buffer of zero bytes is never allocated - this can
+ ** cause fts3DoclistMerge() to incorrectly return SQLITE_NOMEM.
+ */
+ char *aBuffer = sqlite3_malloc(nRight+nLeft+1);
+ rc = fts3DoclistMerge(MERGE_OR, 0, 0, aBuffer, pnOut,
+ aLeft, nLeft, aRight, nRight
+ );
+ *paOut = aBuffer;
+ sqlite3_free(aLeft);
break;
}
- case FTSQUERY_OR: {
- rc = docListOrMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData, pRes);
+
+ default: {
+ assert( FTSQUERY_NOT==MERGE_NOT && FTSQUERY_AND==MERGE_AND );
+ fts3DoclistMerge(pExpr->eType, 0, 0, aLeft, pnOut,
+ aLeft, nLeft, aRight, nRight
+ );
+ *paOut = aLeft;
break;
}
}
}
- dataBufferDestroy(&lhs);
- dataBufferDestroy(&rhs);
+ sqlite3_free(aRight);
}
}
return rc;
}
-/* TODO(shess) Refactor the code to remove this forward decl. */
-static int flushPendingTerms(fulltext_vtab *v);
-
-/* Perform a full-text query using the search expression in
-** zInput[0..nInput-1]. Return a list of matching documents
-** in pResult.
-**
-** Queries must match column iColumn. Or if iColumn>=nColumn
-** they are allowed to match against any column.
-*/
-static int fulltextQuery(
- fulltext_vtab *v, /* The full text index */
- int iColumn, /* Match against this column by default */
- const char *zInput, /* The query string */
- int nInput, /* Number of bytes in zInput[] */
- DataBuffer *pResult, /* Write the result doclist here */
- Fts3Expr **ppExpr /* Put parsed query string here */
-){
- int rc;
-
- /* TODO(shess) Instead of flushing pendingTerms, we could query for
- ** the relevant term and merge the doclist into what we receive from
- ** the database. Wait and see if this is a common issue, first.
- **
- ** A good reason not to flush is to not generate update-related
- ** error codes from here.
- */
-
- /* Flush any buffered updates before executing the query. */
- rc = flushPendingTerms(v);
- if( rc!=SQLITE_OK ){
- return rc;
- }
-
- /* Parse the query passed to the MATCH operator. */
- rc = sqlite3Fts3ExprParse(v->pTokenizer,
- v->azColumn, v->nColumn, iColumn, zInput, nInput, ppExpr
- );
- if( rc!=SQLITE_OK ){
- assert( 0==(*ppExpr) );
- return rc;
- }
-
- return evalFts3Expr(v, *ppExpr, pResult);
-}
-
/*
** This is the xFilter interface for the virtual table. See
** the virtual table xFilter method documentation for additional
** information.
**
-** If idxNum==QUERY_GENERIC then do a full table scan against
+** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against
** the %_content table.
**
-** If idxNum==QUERY_DOCID then do a docid lookup for a single entry
+** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry
** in the %_content table.
**
-** If idxNum>=QUERY_FULLTEXT then use the full text index. The
+** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The
** column on the left-hand side of the MATCH operator is column
-** number idxNum-QUERY_FULLTEXT, 0 indexed. argv[0] is the right-hand
+** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand
** side of the MATCH operator.
*/
/* TODO(shess) Upgrade the cursor initialization and destruction to
-** account for fulltextFilter() being called multiple times on the
-** same cursor. The current solution is very fragile. Apply fix to
+** account for fts3FilterMethod() being called multiple times on the
+** same cursor. The current solution is very fragile. Apply fix to
** fts3 as appropriate.
*/
-static int fulltextFilter(
- sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */
- int idxNum, const char *idxStr, /* Which indexing scheme to use */
- int argc, sqlite3_value **argv /* Arguments for the indexing scheme */
+static int fts3FilterMethod(
+ sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */
+ int idxNum, /* Strategy index */
+ const char *idxStr, /* Unused */
+ int nVal, /* Number of elements in apVal */
+ sqlite3_value **apVal /* Arguments for the indexing scheme */
){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
- fulltext_vtab *v = cursor_vtab(c);
- int rc;
-
- FTSTRACE(("FTS3 Filter %p\n",pCursor));
-
- /* If the cursor has a statement that was not prepared according to
- ** idxNum, clear it. I believe all calls to fulltextFilter with a
- ** given cursor will have the same idxNum , but in this case it's
- ** easy to be safe.
+ const char *azSql[] = {
+ "SELECT * FROM %Q.'%q_content' WHERE docid = ?", /* non-full-table-scan */
+ "SELECT * FROM %Q.'%q_content'", /* full-table-scan */
+ };
+ int rc; /* Return code */
+ char *zSql; /* SQL statement used to access %_content */
+ Fts3Table *p = (Fts3Table *)pCursor->pVtab;
+ Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
+
+ UNUSED_PARAMETER(idxStr);
+ UNUSED_PARAMETER(nVal);
+
+ assert( idxNum>=0 && idxNum<=(FTS3_FULLTEXT_SEARCH+p->nColumn) );
+ assert( nVal==0 || nVal==1 );
+ assert( (nVal==0)==(idxNum==FTS3_FULLSCAN_SEARCH) );
+
+ /* In case the cursor has been used before, clear it now. */
+ sqlite3_finalize(pCsr->pStmt);
+ sqlite3_free(pCsr->aDoclist);
+ sqlite3Fts3ExprFree(pCsr->pExpr);
+ memset(&pCursor[1], 0, sizeof(Fts3Cursor)-sizeof(sqlite3_vtab_cursor));
+
+ /* Compile a SELECT statement for this cursor. For a full-table-scan, the
+ ** statement loops through all rows of the %_content table. For a
+ ** full-text query or docid lookup, the statement retrieves a single
+ ** row by docid.
*/
- if( c->pStmt && c->iCursorType!=idxNum ){
- sqlite3_finalize(c->pStmt);
- c->pStmt = NULL;
- }
-
- /* Get a fresh statement appropriate to idxNum. */
- /* TODO(shess): Add a prepared-statement cache in the vt structure.
- ** The cache must handle multiple open cursors. Easier to cache the
- ** statement variants at the vt to reduce malloc/realloc/free here.
- ** Or we could have a StringBuffer variant which allowed stack
- ** construction for small values.
- */
- if( !c->pStmt ){
- StringBuffer sb;
- initStringBuffer(&sb);
- append(&sb, "SELECT docid, ");
- appendList(&sb, v->nColumn, v->azContentColumn);
- append(&sb, " FROM %_content");
- if( idxNum!=QUERY_GENERIC ) append(&sb, " WHERE docid = ?");
- rc = sql_prepare(v->db, v->zDb, v->zName, &c->pStmt,
- stringBufferData(&sb));
- stringBufferDestroy(&sb);
- if( rc!=SQLITE_OK ) return rc;
- c->iCursorType = idxNum;
+ zSql = sqlite3_mprintf(azSql[idxNum==FTS3_FULLSCAN_SEARCH], p->zDb, p->zName);
+ if( !zSql ){
+ rc = SQLITE_NOMEM;
}else{
- sqlite3_reset(c->pStmt);
- assert( c->iCursorType==idxNum );
- }
-
- switch( idxNum ){
- case QUERY_GENERIC:
- break;
-
- case QUERY_DOCID:
- rc = sqlite3_bind_int64(c->pStmt, 1, sqlite3_value_int64(argv[0]));
- if( rc!=SQLITE_OK ) return rc;
- break;
-
- default: /* full-text search */
- {
- int iCol = idxNum-QUERY_FULLTEXT;
- const char *zQuery = (const char *)sqlite3_value_text(argv[0]);
- assert( idxNum<=QUERY_FULLTEXT+v->nColumn);
- assert( argc==1 );
- if( c->result.nData!=0 ){
- /* This case happens if the same cursor is used repeatedly. */
- dlrDestroy(&c->reader);
- dataBufferReset(&c->result);
- }else{
- dataBufferInit(&c->result, 0);
- }
- rc = fulltextQuery(v, iCol, zQuery, -1, &c->result, &c->pExpr);
- if( rc!=SQLITE_OK ) return rc;
- if( c->result.nData!=0 ){
- dlrInit(&c->reader, DL_DOCIDS, c->result.pData, c->result.nData);
- }
- break;
- }
- }
-
- return fulltextNext(pCursor);
-}
-
-/* This is the xEof method of the virtual table. The SQLite core
-** calls this routine to find out if it has reached the end of
-** a query's results set.
-*/
-static int fulltextEof(sqlite3_vtab_cursor *pCursor){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
- return c->eof;
-}
-
-/* This is the xColumn method of the virtual table. The SQLite
-** core calls this method during a query when it needs the value
-** of a column from the virtual table. This method needs to use
-** one of the sqlite3_result_*() routines to store the requested
-** value back in the pContext.
-*/
-static int fulltextColumn(sqlite3_vtab_cursor *pCursor,
- sqlite3_context *pContext, int idxCol){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
- fulltext_vtab *v = cursor_vtab(c);
-
- if( idxCol<v->nColumn ){
- sqlite3_value *pVal = sqlite3_column_value(c->pStmt, idxCol+1);
- sqlite3_result_value(pContext, pVal);
- }else if( idxCol==v->nColumn ){
- /* The extra column whose name is the same as the table.
- ** Return a blob which is a pointer to the cursor
- */
- sqlite3_result_blob(pContext, &c, sizeof(c), SQLITE_TRANSIENT);
- }else if( idxCol==v->nColumn+1 ){
- /* The docid column, which is an alias for rowid. */
- sqlite3_value *pVal = sqlite3_column_value(c->pStmt, 0);
- sqlite3_result_value(pContext, pVal);
- }
- return SQLITE_OK;
-}
-
-/* This is the xRowid method. The SQLite core calls this routine to
-** retrieve the rowid for the current row of the result set. fts3
-** exposes %_content.docid as the rowid for the virtual table. The
-** rowid should be written to *pRowid.
-*/
-static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
- fulltext_cursor *c = (fulltext_cursor *) pCursor;
-
- *pRowid = sqlite3_column_int64(c->pStmt, 0);
- return SQLITE_OK;
-}
-
-/* Add all terms in [zText] to pendingTerms table. If [iColumn] > 0,
-** we also store positions and offsets in the hash table using that
-** column number.
-*/
-static int buildTerms(fulltext_vtab *v, sqlite_int64 iDocid,
- const char *zText, int iColumn){
- sqlite3_tokenizer *pTokenizer = v->pTokenizer;
- sqlite3_tokenizer_cursor *pCursor;
- const char *pToken;
- int nTokenBytes;
- int iStartOffset, iEndOffset, iPosition;
- int rc;
-
- rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor);
- if( rc!=SQLITE_OK ) return rc;
-
- pCursor->pTokenizer = pTokenizer;
- while( SQLITE_OK==(rc=pTokenizer->pModule->xNext(pCursor,
- &pToken, &nTokenBytes,
- &iStartOffset, &iEndOffset,
- &iPosition)) ){
- DLCollector *p;
- int nData; /* Size of doclist before our update. */
-
- /* Positions can't be negative; we use -1 as a terminator
- * internally. Token can't be NULL or empty. */
- if( iPosition<0 || pToken == NULL || nTokenBytes == 0 ){
- rc = SQLITE_ERROR;
- break;
- }
-
- p = fts3HashFind(&v->pendingTerms, pToken, nTokenBytes);
- if( p==NULL ){
- nData = 0;
- p = dlcNew(iDocid, DL_DEFAULT);
- fts3HashInsert(&v->pendingTerms, pToken, nTokenBytes, p);
-
- /* Overhead for our hash table entry, the key, and the value. */
- v->nPendingData += sizeof(struct fts3HashElem)+sizeof(*p)+nTokenBytes;
- }else{
- nData = p->b.nData;
- if( p->dlw.iPrevDocid!=iDocid ) dlcNext(p, iDocid);
- }
- if( iColumn>=0 ){
- dlcAddPos(p, iColumn, iPosition, iStartOffset, iEndOffset);
- }
-
- /* Accumulate data added by dlcNew or dlcNext, and dlcAddPos. */
- v->nPendingData += p->b.nData-nData;
- }
-
- /* TODO(shess) Check return? Should this be able to cause errors at
- ** this point? Actually, same question about sqlite3_finalize(),
- ** though one could argue that failure there means that the data is
- ** not durable. *ponder*
- */
- pTokenizer->pModule->xClose(pCursor);
- if( SQLITE_DONE == rc ) return SQLITE_OK;
- return rc;
-}
-
-/* Add doclists for all terms in [pValues] to pendingTerms table. */
-static int insertTerms(fulltext_vtab *v, sqlite_int64 iDocid,
- sqlite3_value **pValues){
- int i;
- for(i = 0; i < v->nColumn ; ++i){
- char *zText = (char*)sqlite3_value_text(pValues[i]);
- int rc = buildTerms(v, iDocid, zText, i);
- if( rc!=SQLITE_OK ) return rc;
- }
- return SQLITE_OK;
-}
-
-/* Add empty doclists for all terms in the given row's content to
-** pendingTerms.
-*/
-static int deleteTerms(fulltext_vtab *v, sqlite_int64 iDocid){
- const char **pValues;
- int i, rc;
-
- /* TODO(shess) Should we allow such tables at all? */
- if( DL_DEFAULT==DL_DOCIDS ) return SQLITE_ERROR;
-
- rc = content_select(v, iDocid, &pValues);
- if( rc!=SQLITE_OK ) return rc;
-
- for(i = 0 ; i < v->nColumn; ++i) {
- rc = buildTerms(v, iDocid, pValues[i], -1);
- if( rc!=SQLITE_OK ) break;
+ rc = sqlite3_prepare_v2(p->db, zSql, -1, &pCsr->pStmt, 0);
+ sqlite3_free(zSql);
}
-
- freeStringArray(v->nColumn, pValues);
- return SQLITE_OK;
-}
-
-/* TODO(shess) Refactor the code to remove this forward decl. */
-static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid);
-
-/* Insert a row into the %_content table; set *piDocid to be the ID of the
-** new row. Add doclists for terms to pendingTerms.
-*/
-static int index_insert(fulltext_vtab *v, sqlite3_value *pRequestDocid,
- sqlite3_value **pValues, sqlite_int64 *piDocid){
- int rc;
-
- rc = content_insert(v, pRequestDocid, pValues); /* execute an SQL INSERT */
- if( rc!=SQLITE_OK ) return rc;
-
- /* docid column is an alias for rowid. */
- *piDocid = sqlite3_last_insert_rowid(v->db);
- rc = initPendingTerms(v, *piDocid);
- if( rc!=SQLITE_OK ) return rc;
-
- return insertTerms(v, *piDocid, pValues);
-}
-
-/* Delete a row from the %_content table; add empty doclists for terms
-** to pendingTerms.
-*/
-static int index_delete(fulltext_vtab *v, sqlite_int64 iRow){
- int rc = initPendingTerms(v, iRow);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = deleteTerms(v, iRow);
- if( rc!=SQLITE_OK ) return rc;
-
- return content_delete(v, iRow); /* execute an SQL DELETE */
-}
-
-/* Update a row in the %_content table; add delete doclists to
-** pendingTerms for old terms not in the new data, add insert doclists
-** to pendingTerms for terms in the new data.
-*/
-static int index_update(fulltext_vtab *v, sqlite_int64 iRow,
- sqlite3_value **pValues){
- int rc = initPendingTerms(v, iRow);
if( rc!=SQLITE_OK ) return rc;
+ pCsr->eSearch = (i16)idxNum;
- /* Generate an empty doclist for each term that previously appeared in this
- * row. */
- rc = deleteTerms(v, iRow);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = content_update(v, pValues, iRow); /* execute an SQL UPDATE */
- if( rc!=SQLITE_OK ) return rc;
-
- /* Now add positions for terms which appear in the updated row. */
- return insertTerms(v, iRow, pValues);
-}
-
-/*******************************************************************/
-/* InteriorWriter is used to collect terms and block references into
-** interior nodes in %_segments. See commentary at top of file for
-** format.
-*/
-
-/* How large interior nodes can grow. */
-#define INTERIOR_MAX 2048
+ if( idxNum==FTS3_DOCID_SEARCH ){
+ rc = sqlite3_bind_value(pCsr->pStmt, 1, apVal[0]);
+ }else if( idxNum!=FTS3_FULLSCAN_SEARCH ){
+ int iCol = idxNum-FTS3_FULLTEXT_SEARCH;
+ const char *zQuery = (const char *)sqlite3_value_text(apVal[0]);
-/* Minimum number of terms per interior node (except the root). This
-** prevents large terms from making the tree too skinny - must be >0
-** so that the tree always makes progress. Note that the min tree
-** fanout will be INTERIOR_MIN_TERMS+1.
-*/
-#define INTERIOR_MIN_TERMS 7
-#if INTERIOR_MIN_TERMS<1
-# error INTERIOR_MIN_TERMS must be greater than 0.
-#endif
-
-/* ROOT_MAX controls how much data is stored inline in the segment
-** directory.
-*/
-/* TODO(shess) Push ROOT_MAX down to whoever is writing things. It's
-** only here so that interiorWriterRootInfo() and leafWriterRootInfo()
-** can both see it, but if the caller passed it in, we wouldn't even
-** need a define.
-*/
-#define ROOT_MAX 1024
-#if ROOT_MAX<VARINT_MAX*2
-# error ROOT_MAX must have enough space for a header.
-#endif
-
-/* InteriorBlock stores a linked-list of interior blocks while a lower
-** layer is being constructed.
-*/
-typedef struct InteriorBlock {
- DataBuffer term; /* Leftmost term in block's subtree. */
- DataBuffer data; /* Accumulated data for the block. */
- struct InteriorBlock *next;
-} InteriorBlock;
-
-static InteriorBlock *interiorBlockNew(int iHeight, sqlite_int64 iChildBlock,
- const char *pTerm, int nTerm){
- InteriorBlock *block = sqlite3_malloc(sizeof(InteriorBlock));
- char c[VARINT_MAX+VARINT_MAX];
- int n;
-
- if( block ){
- memset(block, 0, sizeof(*block));
- dataBufferInit(&block->term, 0);
- dataBufferReplace(&block->term, pTerm, nTerm);
-
- n = fts3PutVarint(c, iHeight);
- n += fts3PutVarint(c+n, iChildBlock);
- dataBufferInit(&block->data, INTERIOR_MAX);
- dataBufferReplace(&block->data, c, n);
- }
- return block;
-}
-
-#ifndef NDEBUG
-/* Verify that the data is readable as an interior node. */
-static void interiorBlockValidate(InteriorBlock *pBlock){
- const char *pData = pBlock->data.pData;
- int nData = pBlock->data.nData;
- int n, iDummy;
- sqlite_int64 iBlockid;
-
- assert( nData>0 );
- assert( pData!=0 );
- assert( pData+nData>pData );
-
- /* Must lead with height of node as a varint(n), n>0 */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n<nData );
- pData += n;
- nData -= n;
-
- /* Must contain iBlockid. */
- n = fts3GetVarint(pData, &iBlockid);
- assert( n>0 );
- assert( n<=nData );
- pData += n;
- nData -= n;
-
- /* Zero or more terms of positive length */
- if( nData!=0 ){
- /* First term is not delta-encoded. */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0);
- assert( n+iDummy<=nData );
- pData += n+iDummy;
- nData -= n+iDummy;
-
- /* Following terms delta-encoded. */
- while( nData!=0 ){
- /* Length of shared prefix. */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>=0 );
- assert( n<nData );
- pData += n;
- nData -= n;
-
- /* Length and data of distinct suffix. */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0);
- assert( n+iDummy<=nData );
- pData += n+iDummy;
- nData -= n+iDummy;
+ if( zQuery==0 && sqlite3_value_type(apVal[0])!=SQLITE_NULL ){
+ return SQLITE_NOMEM;
}
- }
-}
-#define ASSERT_VALID_INTERIOR_BLOCK(x) interiorBlockValidate(x)
-#else
-#define ASSERT_VALID_INTERIOR_BLOCK(x) assert( 1 )
-#endif
-
-typedef struct InteriorWriter {
- int iHeight; /* from 0 at leaves. */
- InteriorBlock *first, *last;
- struct InteriorWriter *parentWriter;
- DataBuffer term; /* Last term written to block "last". */
- sqlite_int64 iOpeningChildBlock; /* First child block in block "last". */
-#ifndef NDEBUG
- sqlite_int64 iLastChildBlock; /* for consistency checks. */
-#endif
-} InteriorWriter;
-
-/* Initialize an interior node where pTerm[nTerm] marks the leftmost
-** term in the tree. iChildBlock is the leftmost child block at the
-** next level down the tree.
-*/
-static void interiorWriterInit(int iHeight, const char *pTerm, int nTerm,
- sqlite_int64 iChildBlock,
- InteriorWriter *pWriter){
- InteriorBlock *block;
- assert( iHeight>0 );
- CLEAR(pWriter);
-
- pWriter->iHeight = iHeight;
- pWriter->iOpeningChildBlock = iChildBlock;
-#ifndef NDEBUG
- pWriter->iLastChildBlock = iChildBlock;
-#endif
- block = interiorBlockNew(iHeight, iChildBlock, pTerm, nTerm);
- pWriter->last = pWriter->first = block;
- ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
- dataBufferInit(&pWriter->term, 0);
-}
-
-/* Append the child node rooted at iChildBlock to the interior node,
-** with pTerm[nTerm] as the leftmost term in iChildBlock's subtree.
-*/
-static void interiorWriterAppend(InteriorWriter *pWriter,
- const char *pTerm, int nTerm,
- sqlite_int64 iChildBlock){
- char c[VARINT_MAX+VARINT_MAX];
- int n, nPrefix = 0;
-
- ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
-
- /* The first term written into an interior node is actually
- ** associated with the second child added (the first child was added
- ** in interiorWriterInit, or in the if clause at the bottom of this
- ** function). That term gets encoded straight up, with nPrefix left
- ** at 0.
- */
- if( pWriter->term.nData==0 ){
- n = fts3PutVarint(c, nTerm);
- }else{
- while( nPrefix<pWriter->term.nData &&
- pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){
- nPrefix++;
+ rc = sqlite3Fts3ExprParse(p->pTokenizer, p->azColumn, p->nColumn,
+ iCol, zQuery, -1, &pCsr->pExpr
+ );
+ if( rc!=SQLITE_OK ){
+ if( rc==SQLITE_ERROR ){
+ p->base.zErrMsg = sqlite3_mprintf("malformed MATCH expression: [%s]",
+ zQuery);
+ }
+ return rc;
}
- n = fts3PutVarint(c, nPrefix);
- n += fts3PutVarint(c+n, nTerm-nPrefix);
- }
-
-#ifndef NDEBUG
- pWriter->iLastChildBlock++;
-#endif
- assert( pWriter->iLastChildBlock==iChildBlock );
-
- /* Overflow to a new block if the new term makes the current block
- ** too big, and the current block already has enough terms.
- */
- if( pWriter->last->data.nData+n+nTerm-nPrefix>INTERIOR_MAX &&
- iChildBlock-pWriter->iOpeningChildBlock>INTERIOR_MIN_TERMS ){
- pWriter->last->next = interiorBlockNew(pWriter->iHeight, iChildBlock,
- pTerm, nTerm);
- pWriter->last = pWriter->last->next;
- pWriter->iOpeningChildBlock = iChildBlock;
- dataBufferReset(&pWriter->term);
- }else{
- dataBufferAppend2(&pWriter->last->data, c, n,
- pTerm+nPrefix, nTerm-nPrefix);
- dataBufferReplace(&pWriter->term, pTerm, nTerm);
- }
- ASSERT_VALID_INTERIOR_BLOCK(pWriter->last);
-}
-
-/* Free the space used by pWriter, including the linked-list of
-** InteriorBlocks, and parentWriter, if present.
-*/
-static int interiorWriterDestroy(InteriorWriter *pWriter){
- InteriorBlock *block = pWriter->first;
-
- while( block!=NULL ){
- InteriorBlock *b = block;
- block = block->next;
- dataBufferDestroy(&b->term);
- dataBufferDestroy(&b->data);
- sqlite3_free(b);
- }
- if( pWriter->parentWriter!=NULL ){
- interiorWriterDestroy(pWriter->parentWriter);
- sqlite3_free(pWriter->parentWriter);
- }
- dataBufferDestroy(&pWriter->term);
- SCRAMBLE(pWriter);
- return SQLITE_OK;
-}
-
-/* If pWriter can fit entirely in ROOT_MAX, return it as the root info
-** directly, leaving *piEndBlockid unchanged. Otherwise, flush
-** pWriter to %_segments, building a new layer of interior nodes, and
-** recursively ask for their root into.
-*/
-static int interiorWriterRootInfo(fulltext_vtab *v, InteriorWriter *pWriter,
- char **ppRootInfo, int *pnRootInfo,
- sqlite_int64 *piEndBlockid){
- InteriorBlock *block = pWriter->first;
- sqlite_int64 iBlockid = 0;
- int rc;
-
- /* If we can fit the segment inline */
- if( block==pWriter->last && block->data.nData<ROOT_MAX ){
- *ppRootInfo = block->data.pData;
- *pnRootInfo = block->data.nData;
- return SQLITE_OK;
- }
-
- /* Flush the first block to %_segments, and create a new level of
- ** interior node.
- */
- ASSERT_VALID_INTERIOR_BLOCK(block);
- rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid);
- if( rc!=SQLITE_OK ) return rc;
- *piEndBlockid = iBlockid;
-
- pWriter->parentWriter = sqlite3_malloc(sizeof(*pWriter->parentWriter));
- interiorWriterInit(pWriter->iHeight+1,
- block->term.pData, block->term.nData,
- iBlockid, pWriter->parentWriter);
-
- /* Flush additional blocks and append to the higher interior
- ** node.
- */
- for(block=block->next; block!=NULL; block=block->next){
- ASSERT_VALID_INTERIOR_BLOCK(block);
- rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid);
+ rc = sqlite3Fts3ReadLock(p);
if( rc!=SQLITE_OK ) return rc;
- *piEndBlockid = iBlockid;
-
- interiorWriterAppend(pWriter->parentWriter,
- block->term.pData, block->term.nData, iBlockid);
- }
-
- /* Parent node gets the chance to be the root. */
- return interiorWriterRootInfo(v, pWriter->parentWriter,
- ppRootInfo, pnRootInfo, piEndBlockid);
-}
-
-/****************************************************************/
-/* InteriorReader is used to read off the data from an interior node
-** (see comment at top of file for the format).
-*/
-typedef struct InteriorReader {
- const char *pData;
- int nData;
-
- DataBuffer term; /* previous term, for decoding term delta. */
- sqlite_int64 iBlockid;
-} InteriorReader;
-
-static void interiorReaderDestroy(InteriorReader *pReader){
- dataBufferDestroy(&pReader->term);
- SCRAMBLE(pReader);
-}
-
-static int interiorReaderInit(const char *pData, int nData,
- InteriorReader *pReader){
- int n, nTerm;
-
- /* These conditions are checked and met by the callers. */
- assert( nData>0 );
- assert( pData[0]!='\0' );
-
- CLEAR(pReader);
-
- /* Decode the base blockid, and set the cursor to the first term. */
- n = fts3GetVarintSafe(pData+1, &pReader->iBlockid, nData-1);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- pReader->pData = pData+1+n;
- pReader->nData = nData-(1+n);
-
- /* A single-child interior node (such as when a leaf node was too
- ** large for the segment directory) won't have any terms.
- ** Otherwise, decode the first term.
- */
- if( pReader->nData==0 ){
- dataBufferInit(&pReader->term, 0);
- }else{
- n = fts3GetVarint32Safe(pReader->pData, &nTerm, pReader->nData);
- if( !n || nTerm<0 || nTerm>pReader->nData-n) return SQLITE_CORRUPT_BKPT;
- dataBufferInit(&pReader->term, nTerm);
- dataBufferReplace(&pReader->term, pReader->pData+n, nTerm);
- pReader->pData += n+nTerm;
- pReader->nData -= n+nTerm;
+ rc = evalFts3Expr(p, pCsr->pExpr, &pCsr->aDoclist, &pCsr->nDoclist, 0);
+ pCsr->pNextId = pCsr->aDoclist;
+ pCsr->iPrevId = 0;
}
- return SQLITE_OK;
-}
-
-static int interiorReaderAtEnd(InteriorReader *pReader){
- return pReader->term.nData<=0;
-}
-
-static sqlite_int64 interiorReaderCurrentBlockid(InteriorReader *pReader){
- return pReader->iBlockid;
-}
-static int interiorReaderTermBytes(InteriorReader *pReader){
- assert( !interiorReaderAtEnd(pReader) );
- return pReader->term.nData;
-}
-static const char *interiorReaderTerm(InteriorReader *pReader){
- assert( !interiorReaderAtEnd(pReader) );
- return pReader->term.pData;
-}
-
-/* Step forward to the next term in the node. */
-static int interiorReaderStep(InteriorReader *pReader){
- assert( !interiorReaderAtEnd(pReader) );
-
- /* If the last term has been read, signal eof, else construct the
- ** next term.
- */
- if( pReader->nData==0 ){
- dataBufferReset(&pReader->term);
- }else{
- int n, nPrefix, nSuffix;
-
- n = fts3GetVarint32Safe(pReader->pData, &nPrefix, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- pReader->nData -= n;
- pReader->pData += n;
- n = fts3GetVarint32Safe(pReader->pData, &nSuffix, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- pReader->nData -= n;
- pReader->pData += n;
- if( nSuffix<0 || nSuffix>pReader->nData ) return SQLITE_CORRUPT_BKPT;
- if( nPrefix<0 || nPrefix>pReader->term.nData ) return SQLITE_CORRUPT_BKPT;
-
- /* Truncate the current term and append suffix data. */
- pReader->term.nData = nPrefix;
- dataBufferAppend(&pReader->term, pReader->pData, nSuffix);
-
- pReader->pData += nSuffix;
- pReader->nData -= nSuffix;
- }
- pReader->iBlockid++;
- return SQLITE_OK;
-}
-
-/* Compare the current term to pTerm[nTerm], returning strcmp-style
-** results. If isPrefix, equality means equal through nTerm bytes.
-*/
-static int interiorReaderTermCmp(InteriorReader *pReader,
- const char *pTerm, int nTerm, int isPrefix){
- const char *pReaderTerm = interiorReaderTerm(pReader);
- int nReaderTerm = interiorReaderTermBytes(pReader);
- int c, n = nReaderTerm<nTerm ? nReaderTerm : nTerm;
-
- if( n==0 ){
- if( nReaderTerm>0 ) return -1;
- if( nTerm>0 ) return 1;
- return 0;
- }
-
- c = memcmp(pReaderTerm, pTerm, n);
- if( c!=0 ) return c;
- if( isPrefix && n==nTerm ) return 0;
- return nReaderTerm - nTerm;
+ if( rc!=SQLITE_OK ) return rc;
+ return fts3NextMethod(pCursor);
}
-/****************************************************************/
-/* LeafWriter is used to collect terms and associated doclist data
-** into leaf blocks in %_segments (see top of file for format info).
-** Expected usage is:
-**
-** LeafWriter writer;
-** leafWriterInit(0, 0, &writer);
-** while( sorted_terms_left_to_process ){
-** // data is doclist data for that term.
-** rc = leafWriterStep(v, &writer, pTerm, nTerm, pData, nData);
-** if( rc!=SQLITE_OK ) goto err;
-** }
-** rc = leafWriterFinalize(v, &writer);
-**err:
-** leafWriterDestroy(&writer);
-** return rc;
-**
-** leafWriterStep() may write a collected leaf out to %_segments.
-** leafWriterFinalize() finishes writing any buffered data and stores
-** a root node in %_segdir. leafWriterDestroy() frees all buffers and
-** InteriorWriters allocated as part of writing this segment.
-**
-** TODO(shess) Document leafWriterStepMerge().
+/*
+** This is the xEof method of the virtual table. SQLite calls this
+** routine to find out if it has reached the end of a result set.
*/
-
-/* Put terms with data this big in their own block. */
-#define STANDALONE_MIN 1024
-
-/* Keep leaf blocks below this size. */
-#define LEAF_MAX 2048
-
-typedef struct LeafWriter {
- int iLevel;
- int idx;
- sqlite_int64 iStartBlockid; /* needed to create the root info */
- sqlite_int64 iEndBlockid; /* when we're done writing. */
-
- DataBuffer term; /* previous encoded term */
- DataBuffer data; /* encoding buffer */
-
- /* bytes of first term in the current node which distinguishes that
- ** term from the last term of the previous node.
- */
- int nTermDistinct;
-
- InteriorWriter parentWriter; /* if we overflow */
- int has_parent;
-} LeafWriter;
-
-static void leafWriterInit(int iLevel, int idx, LeafWriter *pWriter){
- CLEAR(pWriter);
- pWriter->iLevel = iLevel;
- pWriter->idx = idx;
-
- dataBufferInit(&pWriter->term, 32);
-
- /* Start out with a reasonably sized block, though it can grow. */
- dataBufferInit(&pWriter->data, LEAF_MAX);
-}
-
-#ifndef NDEBUG
-/* Verify that the data is readable as a leaf node. */
-static void leafNodeValidate(const char *pData, int nData){
- int n, iDummy;
-
- if( nData==0 ) return;
- assert( nData>0 );
- assert( pData!=0 );
- assert( pData+nData>pData );
-
- /* Must lead with a varint(0) */
- n = fts3GetVarint32(pData, &iDummy);
- assert( iDummy==0 );
- assert( n>0 );
- assert( n<nData );
- pData += n;
- nData -= n;
-
- /* Leading term length and data must fit in buffer. */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0 );
- assert( n+iDummy<nData );
- pData += n+iDummy;
- nData -= n+iDummy;
-
- /* Leading term's doclist length and data must fit. */
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0 );
- assert( n+iDummy<=nData );
- ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL);
- pData += n+iDummy;
- nData -= n+iDummy;
-
- /* Verify that trailing terms and doclists also are readable. */
- while( nData!=0 ){
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>=0 );
- assert( n<nData );
- pData += n;
- nData -= n;
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0 );
- assert( n+iDummy<nData );
- pData += n+iDummy;
- nData -= n+iDummy;
-
- n = fts3GetVarint32(pData, &iDummy);
- assert( n>0 );
- assert( iDummy>0 );
- assert( n+iDummy>0 );
- assert( n+iDummy<=nData );
- ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL);
- pData += n+iDummy;
- nData -= n+iDummy;
- }
+static int fts3EofMethod(sqlite3_vtab_cursor *pCursor){
+ return ((Fts3Cursor *)pCursor)->isEof;
}
-#define ASSERT_VALID_LEAF_NODE(p, n) leafNodeValidate(p, n)
-#else
-#define ASSERT_VALID_LEAF_NODE(p, n) assert( 1 )
-#endif
-/* Flush the current leaf node to %_segments, and adding the resulting
-** blockid and the starting term to the interior node which will
-** contain it.
+/*
+** This is the xRowid method. The SQLite core calls this routine to
+** retrieve the rowid for the current row of the result set. fts3
+** exposes %_content.docid as the rowid for the virtual table. The
+** rowid should be written to *pRowid.
*/
-static int leafWriterInternalFlush(fulltext_vtab *v, LeafWriter *pWriter,
- int iData, int nData){
- sqlite_int64 iBlockid = 0;
- const char *pStartingTerm;
- int nStartingTerm, rc, n;
-
- /* Must have the leading varint(0) flag, plus at least some
- ** valid-looking data.
- */
- assert( nData>2 );
- assert( iData>=0 );
- assert( iData+nData<=pWriter->data.nData );
- ASSERT_VALID_LEAF_NODE(pWriter->data.pData+iData, nData);
-
- rc = block_insert(v, pWriter->data.pData+iData, nData, &iBlockid);
- if( rc!=SQLITE_OK ) return rc;
- assert( iBlockid!=0 );
-
- /* Reconstruct the first term in the leaf for purposes of building
- ** the interior node.
- */
- n = fts3GetVarint32(pWriter->data.pData+iData+1, &nStartingTerm);
- pStartingTerm = pWriter->data.pData+iData+1+n;
- assert( pWriter->data.nData>iData+1+n+nStartingTerm );
- assert( pWriter->nTermDistinct>0 );
- assert( pWriter->nTermDistinct<=nStartingTerm );
- nStartingTerm = pWriter->nTermDistinct;
-
- if( pWriter->has_parent ){
- interiorWriterAppend(&pWriter->parentWriter,
- pStartingTerm, nStartingTerm, iBlockid);
- }else{
- interiorWriterInit(1, pStartingTerm, nStartingTerm, iBlockid,
- &pWriter->parentWriter);
- pWriter->has_parent = 1;
- }
-
- /* Track the span of this segment's leaf nodes. */
- if( pWriter->iEndBlockid==0 ){
- pWriter->iEndBlockid = pWriter->iStartBlockid = iBlockid;
+static int fts3RowidMethod(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
+ Fts3Cursor *pCsr = (Fts3Cursor *) pCursor;
+ if( pCsr->aDoclist ){
+ *pRowid = pCsr->iPrevId;
}else{
- pWriter->iEndBlockid++;
- assert( iBlockid==pWriter->iEndBlockid );
+ *pRowid = sqlite3_column_int64(pCsr->pStmt, 0);
}
-
return SQLITE_OK;
}
-static int leafWriterFlush(fulltext_vtab *v, LeafWriter *pWriter){
- int rc = leafWriterInternalFlush(v, pWriter, 0, pWriter->data.nData);
- if( rc!=SQLITE_OK ) return rc;
-
- /* Re-initialize the output buffer. */
- dataBufferReset(&pWriter->data);
- return SQLITE_OK;
-}
-
-/* Fetch the root info for the segment. If the entire leaf fits
-** within ROOT_MAX, then it will be returned directly, otherwise it
-** will be flushed and the root info will be returned from the
-** interior node. *piEndBlockid is set to the blockid of the last
-** interior or leaf node written to disk (0 if none are written at
-** all).
-*/
-static int leafWriterRootInfo(fulltext_vtab *v, LeafWriter *pWriter,
- char **ppRootInfo, int *pnRootInfo,
- sqlite_int64 *piEndBlockid){
- /* we can fit the segment entirely inline */
- if( !pWriter->has_parent && pWriter->data.nData<ROOT_MAX ){
- *ppRootInfo = pWriter->data.pData;
- *pnRootInfo = pWriter->data.nData;
- *piEndBlockid = 0;
- return SQLITE_OK;
- }
-
- /* Flush remaining leaf data. */
- if( pWriter->data.nData>0 ){
- int rc = leafWriterFlush(v, pWriter);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- /* We must have flushed a leaf at some point. */
- assert( pWriter->has_parent );
-
- /* Tenatively set the end leaf blockid as the end blockid. If the
- ** interior node can be returned inline, this will be the final
- ** blockid, otherwise it will be overwritten by
- ** interiorWriterRootInfo().
- */
- *piEndBlockid = pWriter->iEndBlockid;
-
- return interiorWriterRootInfo(v, &pWriter->parentWriter,
- ppRootInfo, pnRootInfo, piEndBlockid);
-}
-
-/* Collect the rootInfo data and store it into the segment directory.
-** This has the effect of flushing the segment's leaf data to
-** %_segments, and also flushing any interior nodes to %_segments.
+/*
+** This is the xColumn method, called by SQLite to request a value from
+** the row that the supplied cursor currently points to.
*/
-static int leafWriterFinalize(fulltext_vtab *v, LeafWriter *pWriter){
- sqlite_int64 iEndBlockid;
- char *pRootInfo;
- int rc, nRootInfo;
-
- rc = leafWriterRootInfo(v, pWriter, &pRootInfo, &nRootInfo, &iEndBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- /* Don't bother storing an entirely empty segment. */
- if( iEndBlockid==0 && nRootInfo==0 ) return SQLITE_OK;
-
- return segdir_set(v, pWriter->iLevel, pWriter->idx,
- pWriter->iStartBlockid, pWriter->iEndBlockid,
- iEndBlockid, pRootInfo, nRootInfo);
-}
+static int fts3ColumnMethod(
+ sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */
+ sqlite3_context *pContext, /* Context for sqlite3_result_xxx() calls */
+ int iCol /* Index of column to read value from */
+){
+ int rc; /* Return Code */
+ Fts3Cursor *pCsr = (Fts3Cursor *) pCursor;
+ Fts3Table *p = (Fts3Table *)pCursor->pVtab;
-static void leafWriterDestroy(LeafWriter *pWriter){
- if( pWriter->has_parent ) interiorWriterDestroy(&pWriter->parentWriter);
- dataBufferDestroy(&pWriter->term);
- dataBufferDestroy(&pWriter->data);
-}
+ /* The column value supplied by SQLite must be in range. */
+ assert( iCol>=0 && iCol<=p->nColumn+1 );
-/* Encode a term into the leafWriter, delta-encoding as appropriate.
-** Returns the length of the new term which distinguishes it from the
-** previous term, which can be used to set nTermDistinct when a node
-** boundary is crossed.
-*/
-static int leafWriterEncodeTerm(LeafWriter *pWriter,
- const char *pTerm, int nTerm){
- char c[VARINT_MAX+VARINT_MAX];
- int n, nPrefix = 0;
-
- assert( nTerm>0 );
- while( nPrefix<pWriter->term.nData &&
- pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){
- nPrefix++;
- /* Failing this implies that the terms weren't in order. */
- assert( nPrefix<nTerm );
- }
-
- if( pWriter->data.nData==0 ){
- /* Encode the node header and leading term as:
- ** varint(0)
- ** varint(nTerm)
- ** char pTerm[nTerm]
+ if( iCol==p->nColumn+1 ){
+ /* This call is a request for the "docid" column. Since "docid" is an
+ ** alias for "rowid", use the xRowid() method to obtain the value.
*/
- n = fts3PutVarint(c, '\0');
- n += fts3PutVarint(c+n, nTerm);
- dataBufferAppend2(&pWriter->data, c, n, pTerm, nTerm);
- }else{
- /* Delta-encode the term as:
- ** varint(nPrefix)
- ** varint(nSuffix)
- ** char pTermSuffix[nSuffix]
+ sqlite3_int64 iRowid;
+ rc = fts3RowidMethod(pCursor, &iRowid);
+ sqlite3_result_int64(pContext, iRowid);
+ }else if( iCol==p->nColumn ){
+ /* The extra column whose name is the same as the table.
+ ** Return a blob which is a pointer to the cursor.
*/
- n = fts3PutVarint(c, nPrefix);
- n += fts3PutVarint(c+n, nTerm-nPrefix);
- dataBufferAppend2(&pWriter->data, c, n, pTerm+nPrefix, nTerm-nPrefix);
- }
- dataBufferReplace(&pWriter->term, pTerm, nTerm);
-
- return nPrefix+1;
-}
-
-/* Used to avoid a memmove when a large amount of doclist data is in
-** the buffer. This constructs a node and term header before
-** iDoclistData and flushes the resulting complete node using
-** leafWriterInternalFlush().
-*/
-static int leafWriterInlineFlush(fulltext_vtab *v, LeafWriter *pWriter,
- const char *pTerm, int nTerm,
- int iDoclistData){
- char c[VARINT_MAX+VARINT_MAX];
- int iData, n = fts3PutVarint(c, 0);
- n += fts3PutVarint(c+n, nTerm);
-
- /* There should always be room for the header. Even if pTerm shared
- ** a substantial prefix with the previous term, the entire prefix
- ** could be constructed from earlier data in the doclist, so there
- ** should be room.
- */
- assert( iDoclistData>=n+nTerm );
-
- iData = iDoclistData-(n+nTerm);
- memcpy(pWriter->data.pData+iData, c, n);
- memcpy(pWriter->data.pData+iData+n, pTerm, nTerm);
-
- return leafWriterInternalFlush(v, pWriter, iData, pWriter->data.nData-iData);
-}
-
-/* Push pTerm[nTerm] along with the doclist data to the leaf layer of
-** %_segments.
-*/
-static int leafWriterStepMerge(fulltext_vtab *v, LeafWriter *pWriter,
- const char *pTerm, int nTerm,
- DLReader *pReaders, int nReaders){
- char c[VARINT_MAX+VARINT_MAX];
- int iTermData = pWriter->data.nData, iDoclistData;
- int i, nData, n, nActualData, nActual, rc, nTermDistinct;
-
- ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData);
- nTermDistinct = leafWriterEncodeTerm(pWriter, pTerm, nTerm);
-
- /* Remember nTermDistinct if opening a new node. */
- if( iTermData==0 ) pWriter->nTermDistinct = nTermDistinct;
-
- iDoclistData = pWriter->data.nData;
-
- /* Estimate the length of the merged doclist so we can leave space
- ** to encode it.
- */
- for(i=0, nData=0; i<nReaders; i++){
- nData += dlrAllDataBytes(&pReaders[i]);
- }
- n = fts3PutVarint(c, nData);
- dataBufferAppend(&pWriter->data, c, n);
-
- rc = docListMerge(&pWriter->data, pReaders, nReaders);
- if( rc!=SQLITE_OK ) return rc;
- ASSERT_VALID_DOCLIST(DL_DEFAULT,
- pWriter->data.pData+iDoclistData+n,
- pWriter->data.nData-iDoclistData-n, NULL);
-
- /* The actual amount of doclist data at this point could be smaller
- ** than the length we encoded. Additionally, the space required to
- ** encode this length could be smaller. For small doclists, this is
- ** not a big deal, we can just use memmove() to adjust things.
- */
- nActualData = pWriter->data.nData-(iDoclistData+n);
- nActual = fts3PutVarint(c, nActualData);
- assert( nActualData<=nData );
- assert( nActual<=n );
-
- /* If the new doclist is big enough for force a standalone leaf
- ** node, we can immediately flush it inline without doing the
- ** memmove().
- */
- /* TODO(shess) This test matches leafWriterStep(), which does this
- ** test before it knows the cost to varint-encode the term and
- ** doclist lengths. At some point, change to
- ** pWriter->data.nData-iTermData>STANDALONE_MIN.
- */
- if( nTerm+nActualData>STANDALONE_MIN ){
- /* Push leaf node from before this term. */
- if( iTermData>0 ){
- rc = leafWriterInternalFlush(v, pWriter, 0, iTermData);
- if( rc!=SQLITE_OK ) return rc;
-
- pWriter->nTermDistinct = nTermDistinct;
+ sqlite3_result_blob(pContext, &pCsr, sizeof(pCsr), SQLITE_TRANSIENT);
+ rc = SQLITE_OK;
+ }else{
+ rc = fts3CursorSeek(0, pCsr);
+ if( rc==SQLITE_OK ){
+ sqlite3_result_value(pContext, sqlite3_column_value(pCsr->pStmt, iCol+1));
}
-
- /* Fix the encoded doclist length. */
- iDoclistData += n - nActual;
- memcpy(pWriter->data.pData+iDoclistData, c, nActual);
-
- /* Push the standalone leaf node. */
- rc = leafWriterInlineFlush(v, pWriter, pTerm, nTerm, iDoclistData);
- if( rc!=SQLITE_OK ) return rc;
-
- /* Leave the node empty. */
- dataBufferReset(&pWriter->data);
-
- return rc;
- }
-
- /* At this point, we know that the doclist was small, so do the
- ** memmove if indicated.
- */
- if( nActual<n ){
- memmove(pWriter->data.pData+iDoclistData+nActual,
- pWriter->data.pData+iDoclistData+n,
- pWriter->data.nData-(iDoclistData+n));
- pWriter->data.nData -= n-nActual;
- }
-
- /* Replace written length with actual length. */
- memcpy(pWriter->data.pData+iDoclistData, c, nActual);
-
- /* If the node is too large, break things up. */
- /* TODO(shess) This test matches leafWriterStep(), which does this
- ** test before it knows the cost to varint-encode the term and
- ** doclist lengths. At some point, change to
- ** pWriter->data.nData>LEAF_MAX.
- */
- if( iTermData+nTerm+nActualData>LEAF_MAX ){
- /* Flush out the leading data as a node */
- rc = leafWriterInternalFlush(v, pWriter, 0, iTermData);
- if( rc!=SQLITE_OK ) return rc;
-
- pWriter->nTermDistinct = nTermDistinct;
-
- /* Rebuild header using the current term */
- n = fts3PutVarint(pWriter->data.pData, 0);
- n += fts3PutVarint(pWriter->data.pData+n, nTerm);
- memcpy(pWriter->data.pData+n, pTerm, nTerm);
- n += nTerm;
-
- /* There should always be room, because the previous encoding
- ** included all data necessary to construct the term.
- */
- assert( n<iDoclistData );
- /* So long as STANDALONE_MIN is half or less of LEAF_MAX, the
- ** following memcpy() is safe (as opposed to needing a memmove).
- */
- assert( 2*STANDALONE_MIN<=LEAF_MAX );
- assert( n+pWriter->data.nData-iDoclistData<iDoclistData );
- memcpy(pWriter->data.pData+n,
- pWriter->data.pData+iDoclistData,
- pWriter->data.nData-iDoclistData);
- pWriter->data.nData -= iDoclistData-n;
}
- ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData);
-
- return SQLITE_OK;
-}
-
-/* Push pTerm[nTerm] along with the doclist data to the leaf layer of
-** %_segments.
-*/
-/* TODO(shess) Revise writeZeroSegment() so that doclists are
-** constructed directly in pWriter->data.
-*/
-static int leafWriterStep(fulltext_vtab *v, LeafWriter *pWriter,
- const char *pTerm, int nTerm,
- const char *pData, int nData){
- int rc;
- DLReader reader;
-
- rc = dlrInit(&reader, DL_DEFAULT, pData, nData);
- if( rc!=SQLITE_OK ) return rc;
- rc = leafWriterStepMerge(v, pWriter, pTerm, nTerm, &reader, 1);
- dlrDestroy(&reader);
-
return rc;
}
-
-/****************************************************************/
-/* LeafReader is used to iterate over an individual leaf node. */
-typedef struct LeafReader {
- DataBuffer term; /* copy of current term. */
-
- const char *pData; /* data for current term. */
- int nData;
-} LeafReader;
-
-static void leafReaderDestroy(LeafReader *pReader){
- dataBufferDestroy(&pReader->term);
- SCRAMBLE(pReader);
-}
-
-static int leafReaderAtEnd(LeafReader *pReader){
- return pReader->nData<=0;
-}
-
-/* Access the current term. */
-static int leafReaderTermBytes(LeafReader *pReader){
- return pReader->term.nData;
-}
-static const char *leafReaderTerm(LeafReader *pReader){
- assert( pReader->term.nData>0 );
- return pReader->term.pData;
-}
-
-/* Access the doclist data for the current term. */
-static int leafReaderDataBytes(LeafReader *pReader){
- int nData;
- assert( pReader->term.nData>0 );
- fts3GetVarint32(pReader->pData, &nData);
- return nData;
-}
-static const char *leafReaderData(LeafReader *pReader){
- int n, nData;
- assert( pReader->term.nData>0 );
- n = fts3GetVarint32Safe(pReader->pData, &nData, pReader->nData);
- if( !n || nData>pReader->nData-n ) return NULL;
- return pReader->pData+n;
-}
-
-static int leafReaderInit(const char *pData, int nData,
- LeafReader *pReader){
- int nTerm, n;
-
- /* All callers check this precondition. */
- assert( nData>0 );
- assert( pData[0]=='\0' );
-
- CLEAR(pReader);
-
- /* Read the first term, skipping the header byte. */
- n = fts3GetVarint32Safe(pData+1, &nTerm, nData-1);
- if( !n || nTerm<0 || nTerm>nData-1-n ) return SQLITE_CORRUPT_BKPT;
- dataBufferInit(&pReader->term, nTerm);
- dataBufferReplace(&pReader->term, pData+1+n, nTerm);
-
- /* Position after the first term. */
- pReader->pData = pData+1+n+nTerm;
- pReader->nData = nData-1-n-nTerm;
- return SQLITE_OK;
-}
-
-/* Step the reader forward to the next term. */
-static int leafReaderStep(LeafReader *pReader){
- int n, nData, nPrefix, nSuffix;
- assert( !leafReaderAtEnd(pReader) );
-
- /* Skip previous entry's data block. */
- n = fts3GetVarint32Safe(pReader->pData, &nData, pReader->nData);
- if( !n || nData<0 || nData>pReader->nData-n ) return SQLITE_CORRUPT_BKPT;
- pReader->pData += n+nData;
- pReader->nData -= n+nData;
-
- if( !leafReaderAtEnd(pReader) ){
- /* Construct the new term using a prefix from the old term plus a
- ** suffix from the leaf data.
- */
- n = fts3GetVarint32Safe(pReader->pData, &nPrefix, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- pReader->nData -= n;
- pReader->pData += n;
- n = fts3GetVarint32Safe(pReader->pData, &nSuffix, pReader->nData);
- if( !n ) return SQLITE_CORRUPT_BKPT;
- pReader->nData -= n;
- pReader->pData += n;
- if( nSuffix<0 || nSuffix>pReader->nData ) return SQLITE_CORRUPT_BKPT;
- if( nPrefix<0 || nPrefix>pReader->term.nData ) return SQLITE_CORRUPT_BKPT;
- pReader->term.nData = nPrefix;
- dataBufferAppend(&pReader->term, pReader->pData, nSuffix);
-
- pReader->pData += nSuffix;
- pReader->nData -= nSuffix;
- }
- return SQLITE_OK;
-}
-
-/* strcmp-style comparison of pReader's current term against pTerm.
-** If isPrefix, equality means equal through nTerm bytes.
-*/
-static int leafReaderTermCmp(LeafReader *pReader,
- const char *pTerm, int nTerm, int isPrefix){
- int c, n = pReader->term.nData<nTerm ? pReader->term.nData : nTerm;
- if( n==0 ){
- if( pReader->term.nData>0 ) return -1;
- if(nTerm>0 ) return 1;
- return 0;
- }
-
- c = memcmp(pReader->term.pData, pTerm, n);
- if( c!=0 ) return c;
- if( isPrefix && n==nTerm ) return 0;
- return pReader->term.nData - nTerm;
-}
-
-
-/****************************************************************/
-/* LeavesReader wraps LeafReader to allow iterating over the entire
-** leaf layer of the tree.
-*/
-typedef struct LeavesReader {
- int idx; /* Index within the segment. */
-
- sqlite3_stmt *pStmt; /* Statement we're streaming leaves from. */
- int eof; /* we've seen SQLITE_DONE from pStmt. */
-
- LeafReader leafReader; /* reader for the current leaf. */
- DataBuffer rootData; /* root data for inline. */
-} LeavesReader;
-
-/* Access the current term. */
-static int leavesReaderTermBytes(LeavesReader *pReader){
- assert( !pReader->eof );
- return leafReaderTermBytes(&pReader->leafReader);
-}
-static const char *leavesReaderTerm(LeavesReader *pReader){
- assert( !pReader->eof );
- return leafReaderTerm(&pReader->leafReader);
-}
-
-/* Access the doclist data for the current term. */
-static int leavesReaderDataBytes(LeavesReader *pReader){
- assert( !pReader->eof );
- return leafReaderDataBytes(&pReader->leafReader);
-}
-static const char *leavesReaderData(LeavesReader *pReader){
- assert( !pReader->eof );
- return leafReaderData(&pReader->leafReader);
-}
-
-static int leavesReaderAtEnd(LeavesReader *pReader){
- return pReader->eof;
-}
-
-/* loadSegmentLeaves() may not read all the way to SQLITE_DONE, thus
-** leaving the statement handle open, which locks the table.
-*/
-/* TODO(shess) This "solution" is not satisfactory. Really, there
-** should be check-in function for all statement handles which
-** arranges to call sqlite3_reset(). This most likely will require
-** modification to control flow all over the place, though, so for now
-** just punt.
-**
-** Note the the current system assumes that segment merges will run to
-** completion, which is why this particular probably hasn't arisen in
-** this case. Probably a brittle assumption.
-*/
-static int leavesReaderReset(LeavesReader *pReader){
- return sqlite3_reset(pReader->pStmt);
-}
-
-static void leavesReaderDestroy(LeavesReader *pReader){
- /* If idx is -1, that means we're using a non-cached statement
- ** handle in the optimize() case, so we need to release it.
- */
- if( pReader->pStmt!=NULL && pReader->idx==-1 ){
- sqlite3_finalize(pReader->pStmt);
- }
- leafReaderDestroy(&pReader->leafReader);
- dataBufferDestroy(&pReader->rootData);
- SCRAMBLE(pReader);
+/*
+** This function is the implementation of the xUpdate callback used by
+** FTS3 virtual tables. It is invoked by SQLite each time a row is to be
+** inserted, updated or deleted.
+*/
+static int fts3UpdateMethod(
+ sqlite3_vtab *pVtab, /* Virtual table handle */
+ int nArg, /* Size of argument array */
+ sqlite3_value **apVal, /* Array of arguments */
+ sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */
+){
+ return sqlite3Fts3UpdateMethod(pVtab, nArg, apVal, pRowid);
}
-/* Initialize pReader with the given root data (if iStartBlockid==0
-** the leaf data was entirely contained in the root), or from the
-** stream of blocks between iStartBlockid and iEndBlockid, inclusive.
+/*
+** Implementation of xSync() method. Flush the contents of the pending-terms
+** hash-table to the database.
*/
-static int leavesReaderInit(fulltext_vtab *v,
- int idx,
- sqlite_int64 iStartBlockid,
- sqlite_int64 iEndBlockid,
- const char *pRootData, int nRootData,
- LeavesReader *pReader){
- CLEAR(pReader);
- pReader->idx = idx;
-
- dataBufferInit(&pReader->rootData, 0);
- if( iStartBlockid==0 ){
- int rc;
- /* Corrupt if this can't be a leaf node. */
- if( pRootData==NULL || nRootData<1 || pRootData[0]!='\0' ){
- return SQLITE_CORRUPT_BKPT;
- }
- /* Entire leaf level fit in root data. */
- dataBufferReplace(&pReader->rootData, pRootData, nRootData);
- rc = leafReaderInit(pReader->rootData.pData, pReader->rootData.nData,
- &pReader->leafReader);
- if( rc!=SQLITE_OK ){
- dataBufferDestroy(&pReader->rootData);
- return rc;
- }
- }else{
- sqlite3_stmt *s;
- int rc = sql_get_leaf_statement(v, idx, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iStartBlockid);
- if( rc!=SQLITE_OK ) goto err;
-
- rc = sqlite3_bind_int64(s, 2, iEndBlockid);
- if( rc!=SQLITE_OK ) goto err;
-
- rc = sqlite3_step(s);
-
- /* Corrupt if interior node referenced missing leaf node. */
- if( rc==SQLITE_DONE ){
- rc = SQLITE_CORRUPT_BKPT;
- goto err;
- }
-
- if( rc!=SQLITE_ROW ) goto err;
- rc = SQLITE_OK;
-
- /* Corrupt if leaf data isn't a blob. */
- if( sqlite3_column_type(s, 0)!=SQLITE_BLOB ){
- rc = SQLITE_CORRUPT_BKPT;
- }else{
- const char *pLeafData = sqlite3_column_blob(s, 0);
- int nLeafData = sqlite3_column_bytes(s, 0);
-
- /* Corrupt if this can't be a leaf node. */
- if( pLeafData==NULL || nLeafData<1 || pLeafData[0]!='\0' ){
- rc = SQLITE_CORRUPT_BKPT;
- }else{
- rc = leafReaderInit(pLeafData, nLeafData, &pReader->leafReader);
- }
- }
-
- err:
- if( rc!=SQLITE_OK ){
- if( idx==-1 ){
- sqlite3_finalize(s);
- }else{
- sqlite3_reset(s);
- }
- return rc;
- }
-
- pReader->pStmt = s;
- }
- return SQLITE_OK;
+static int fts3SyncMethod(sqlite3_vtab *pVtab){
+ return sqlite3Fts3PendingTermsFlush((Fts3Table *)pVtab);
}
-/* Step the current leaf forward to the next term. If we reach the
-** end of the current leaf, step forward to the next leaf block.
+/*
+** Implementation of xBegin() method. This is a no-op.
*/
-static int leavesReaderStep(fulltext_vtab *v, LeavesReader *pReader){
- int rc;
- assert( !leavesReaderAtEnd(pReader) );
- rc = leafReaderStep(&pReader->leafReader);
- if( rc!=SQLITE_OK ) return rc;
-
- if( leafReaderAtEnd(&pReader->leafReader) ){
- if( pReader->rootData.pData ){
- pReader->eof = 1;
- return SQLITE_OK;
- }
- rc = sqlite3_step(pReader->pStmt);
- if( rc!=SQLITE_ROW ){
- pReader->eof = 1;
- return rc==SQLITE_DONE ? SQLITE_OK : rc;
- }
-
- /* Corrupt if leaf data isn't a blob. */
- if( sqlite3_column_type(pReader->pStmt, 0)!=SQLITE_BLOB ){
- return SQLITE_CORRUPT_BKPT;
- }else{
- LeafReader tmp;
- const char *pLeafData = sqlite3_column_blob(pReader->pStmt, 0);
- int nLeafData = sqlite3_column_bytes(pReader->pStmt, 0);
-
- /* Corrupt if this can't be a leaf node. */
- if( pLeafData==NULL || nLeafData<1 || pLeafData[0]!='\0' ){
- return SQLITE_CORRUPT_BKPT;
- }
-
- rc = leafReaderInit(pLeafData, nLeafData, &tmp);
- if( rc!=SQLITE_OK ) return rc;
- leafReaderDestroy(&pReader->leafReader);
- pReader->leafReader = tmp;
- }
- }
+static int fts3BeginMethod(sqlite3_vtab *pVtab){
+ UNUSED_PARAMETER(pVtab);
+ assert( ((Fts3Table *)pVtab)->nPendingData==0 );
return SQLITE_OK;
}
-/* Order LeavesReaders by their term, ignoring idx. Readers at eof
-** always sort to the end.
-*/
-static int leavesReaderTermCmp(LeavesReader *lr1, LeavesReader *lr2){
- if( leavesReaderAtEnd(lr1) ){
- if( leavesReaderAtEnd(lr2) ) return 0;
- return 1;
- }
- if( leavesReaderAtEnd(lr2) ) return -1;
-
- return leafReaderTermCmp(&lr1->leafReader,
- leavesReaderTerm(lr2), leavesReaderTermBytes(lr2),
- 0);
-}
-
-/* Similar to leavesReaderTermCmp(), with additional ordering by idx
-** so that older segments sort before newer segments.
-*/
-static int leavesReaderCmp(LeavesReader *lr1, LeavesReader *lr2){
- int c = leavesReaderTermCmp(lr1, lr2);
- if( c!=0 ) return c;
- return lr1->idx-lr2->idx;
-}
-
-/* Assume that pLr[1]..pLr[nLr] are sorted. Bubble pLr[0] into its
-** sorted position.
-*/
-static void leavesReaderReorder(LeavesReader *pLr, int nLr){
- while( nLr>1 && leavesReaderCmp(pLr, pLr+1)>0 ){
- LeavesReader tmp = pLr[0];
- pLr[0] = pLr[1];
- pLr[1] = tmp;
- nLr--;
- pLr++;
- }
-}
-
-/* Initializes pReaders with the segments from level iLevel, returning
-** the number of segments in *piReaders. Leaves pReaders in sorted
-** order.
+/*
+** Implementation of xCommit() method. This is a no-op. The contents of
+** the pending-terms hash-table have already been flushed into the database
+** by fts3SyncMethod().
*/
-static int leavesReadersInit(fulltext_vtab *v, int iLevel,
- LeavesReader *pReaders, int *piReaders){
- sqlite3_stmt *s;
- int i, rc = sql_get_statement(v, SEGDIR_SELECT_LEVEL_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int(s, 1, iLevel);
- if( rc!=SQLITE_OK ) return rc;
-
- i = 0;
- while( (rc = sqlite3_step(s))==SQLITE_ROW ){
- sqlite_int64 iStart = sqlite3_column_int64(s, 0);
- sqlite_int64 iEnd = sqlite3_column_int64(s, 1);
- const char *pRootData = sqlite3_column_blob(s, 2);
- int nRootData = sqlite3_column_bytes(s, 2);
- sqlite_int64 iIndex = sqlite3_column_int64(s, 3);
-
- /* Corrupt if we get back different types than we stored. */
- /* Also corrupt if the index is not sequential starting at 0. */
- if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 1)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 2)!=SQLITE_BLOB ||
- i!=iIndex ||
- i>=MERGE_COUNT ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
-
- rc = leavesReaderInit(v, i, iStart, iEnd, pRootData, nRootData,
- &pReaders[i]);
- if( rc!=SQLITE_OK ) break;
-
- i++;
- }
- if( rc!=SQLITE_DONE ){
- while( i-->0 ){
- leavesReaderDestroy(&pReaders[i]);
- }
- sqlite3_reset(s); /* So we don't leave a lock. */
- return rc;
- }
-
- *piReaders = i;
-
- /* Leave our results sorted by term, then age. */
- while( i-- ){
- leavesReaderReorder(pReaders+i, *piReaders-i);
- }
+static int fts3CommitMethod(sqlite3_vtab *pVtab){
+ UNUSED_PARAMETER(pVtab);
+ assert( ((Fts3Table *)pVtab)->nPendingData==0 );
return SQLITE_OK;
}
-/* Merge doclists from pReaders[nReaders] into a single doclist, which
-** is written to pWriter. Assumes pReaders is ordered oldest to
-** newest.
-*/
-/* TODO(shess) Consider putting this inline in segmentMerge(). */
-static int leavesReadersMerge(fulltext_vtab *v,
- LeavesReader *pReaders, int nReaders,
- LeafWriter *pWriter){
- DLReader dlReaders[MERGE_COUNT];
- const char *pTerm = leavesReaderTerm(pReaders);
- int i, nTerm = leavesReaderTermBytes(pReaders);
- int rc;
-
- assert( nReaders<=MERGE_COUNT );
-
- for(i=0; i<nReaders; i++){
- const char *pData = leavesReaderData(pReaders+i);
- if( pData==NULL ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
- rc = dlrInit(&dlReaders[i], DL_DEFAULT,
- pData,
- leavesReaderDataBytes(pReaders+i));
- if( rc!=SQLITE_OK ) break;
- }
- if( rc!=SQLITE_OK ){
- while( i-->0 ){
- dlrDestroy(&dlReaders[i]);
- }
- return rc;
- }
-
- return leafWriterStepMerge(v, pWriter, pTerm, nTerm, dlReaders, nReaders);
-}
-
-/* Forward ref due to mutual recursion with segdirNextIndex(). */
-static int segmentMerge(fulltext_vtab *v, int iLevel);
-
-/* Put the next available index at iLevel into *pidx. If iLevel
-** already has MERGE_COUNT segments, they are merged to a higher
-** level to make room.
+/*
+** Implementation of xRollback(). Discard the contents of the pending-terms
+** hash-table. Any changes made to the database are reverted by SQLite.
*/
-static int segdirNextIndex(fulltext_vtab *v, int iLevel, int *pidx){
- int rc = segdir_max_index(v, iLevel, pidx);
- if( rc==SQLITE_DONE ){ /* No segments at iLevel. */
- *pidx = 0;
- }else if( rc==SQLITE_ROW ){
- if( *pidx==(MERGE_COUNT-1) ){
- rc = segmentMerge(v, iLevel);
- if( rc!=SQLITE_OK ) return rc;
- *pidx = 0;
- }else{
- (*pidx)++;
- }
- }else{
- return rc;
- }
+static int fts3RollbackMethod(sqlite3_vtab *pVtab){
+ sqlite3Fts3PendingTermsClear((Fts3Table *)pVtab);
return SQLITE_OK;
}
-/* Merge MERGE_COUNT segments at iLevel into a new segment at
-** iLevel+1. If iLevel+1 is already full of segments, those will be
-** merged to make room.
+/*
+** Load the doclist associated with expression pExpr to pExpr->aDoclist.
+** The loaded doclist contains positions as well as the document ids.
+** This is used by the matchinfo(), snippet() and offsets() auxillary
+** functions.
*/
-static int segmentMerge(fulltext_vtab *v, int iLevel){
- LeafWriter writer;
- LeavesReader lrs[MERGE_COUNT];
- int i, rc, idx = 0;
-
- /* Determine the next available segment index at the next level,
- ** merging as necessary.
- */
- rc = segdirNextIndex(v, iLevel+1, &idx);
- if( rc!=SQLITE_OK ) return rc;
-
- /* TODO(shess) This assumes that we'll always see exactly
- ** MERGE_COUNT segments to merge at a given level. That will be
- ** broken if we allow the developer to request preemptive or
- ** deferred merging.
- */
- memset(&lrs, '\0', sizeof(lrs));
- rc = leavesReadersInit(v, iLevel, lrs, &i);
- if( rc!=SQLITE_OK ) return rc;
-
- leafWriterInit(iLevel+1, idx, &writer);
-
- if( i!=MERGE_COUNT ){
- rc = SQLITE_CORRUPT_BKPT;
- goto err;
- }
-
- /* Since leavesReaderReorder() pushes readers at eof to the end,
- ** when the first reader is empty, all will be empty.
- */
- while( !leavesReaderAtEnd(lrs) ){
- /* Figure out how many readers share their next term. */
- for(i=1; i<MERGE_COUNT && !leavesReaderAtEnd(lrs+i); i++){
- if( 0!=leavesReaderTermCmp(lrs, lrs+i) ) break;
- }
-
- rc = leavesReadersMerge(v, lrs, i, &writer);
- if( rc!=SQLITE_OK ) goto err;
-
- /* Step forward those that were merged. */
- while( i-->0 ){
- rc = leavesReaderStep(v, lrs+i);
- if( rc!=SQLITE_OK ) goto err;
-
- /* Reorder by term, then by age. */
- leavesReaderReorder(lrs+i, MERGE_COUNT-i);
- }
- }
-
- for(i=0; i<MERGE_COUNT; i++){
- leavesReaderDestroy(&lrs[i]);
- }
-
- rc = leafWriterFinalize(v, &writer);
- leafWriterDestroy(&writer);
- if( rc!=SQLITE_OK ) return rc;
-
- /* Delete the merged segment data. */
- return segdir_delete(v, iLevel);
-
- err:
- for(i=0; i<MERGE_COUNT; i++){
- leavesReaderDestroy(&lrs[i]);
- }
- leafWriterDestroy(&writer);
- return rc;
-}
-
-/* Accumulate the union of *acc and *pData into *acc. */
-static int docListAccumulateUnion(DataBuffer *acc,
- const char *pData, int nData) {
- DataBuffer tmp = *acc;
- int rc;
- dataBufferInit(acc, tmp.nData+nData);
- rc = docListUnion(tmp.pData, tmp.nData, pData, nData, acc);
- dataBufferDestroy(&tmp);
- return rc;
+int sqlite3Fts3ExprLoadDoclist(Fts3Table *pTab, Fts3Expr *pExpr){
+ return evalFts3Expr(pTab, pExpr, &pExpr->aDoclist, &pExpr->nDoclist, 1);
}
-/* TODO(shess) It might be interesting to explore different merge
-** strategies, here. For instance, since this is a sorted merge, we
-** could easily merge many doclists in parallel. With some
-** comprehension of the storage format, we could merge all of the
-** doclists within a leaf node directly from the leaf node's storage.
-** It may be worthwhile to merge smaller doclists before larger
-** doclists, since they can be traversed more quickly - but the
-** results may have less overlap, making them more expensive in a
-** different way.
-*/
-
-/* Scan pReader for pTerm/nTerm, and merge the term's doclist over
-** *out (any doclists with duplicate docids overwrite those in *out).
-** Internal function for loadSegmentLeaf().
-*/
-static int loadSegmentLeavesInt(fulltext_vtab *v, LeavesReader *pReader,
- const char *pTerm, int nTerm, int isPrefix,
- DataBuffer *out){
- /* doclist data is accumulated into pBuffers similar to how one does
- ** increment in binary arithmetic. If index 0 is empty, the data is
- ** stored there. If there is data there, it is merged and the
- ** results carried into position 1, with further merge-and-carry
- ** until an empty position is found.
- */
- DataBuffer *pBuffers = NULL;
- int nBuffers = 0, nMaxBuffers = 0, rc;
-
- assert( nTerm>0 );
-
- for(rc=SQLITE_OK; rc==SQLITE_OK && !leavesReaderAtEnd(pReader);
- rc=leavesReaderStep(v, pReader)){
- /* TODO(shess) Really want leavesReaderTermCmp(), but that name is
- ** already taken to compare the terms of two LeavesReaders. Think
- ** on a better name. [Meanwhile, break encapsulation rather than
- ** use a confusing name.]
- */
- int c = leafReaderTermCmp(&pReader->leafReader, pTerm, nTerm, isPrefix);
- if( c>0 ) break; /* Past any possible matches. */
- if( c==0 ){
- int iBuffer, nData;
- const char *pData = leavesReaderData(pReader);
- if( pData==NULL ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
- nData = leavesReaderDataBytes(pReader);
-
- /* Find the first empty buffer. */
- for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){
- if( 0==pBuffers[iBuffer].nData ) break;
- }
-
- /* Out of buffers, add an empty one. */
- if( iBuffer==nBuffers ){
- if( nBuffers==nMaxBuffers ){
- DataBuffer *p;
- nMaxBuffers += 20;
-
- /* Manual realloc so we can handle NULL appropriately. */
- p = sqlite3_malloc(nMaxBuffers*sizeof(*pBuffers));
- if( p==NULL ){
- rc = SQLITE_NOMEM;
- break;
- }
-
- if( nBuffers>0 ){
- assert(pBuffers!=NULL);
- memcpy(p, pBuffers, nBuffers*sizeof(*pBuffers));
- sqlite3_free(pBuffers);
- }
- pBuffers = p;
+/*
+** After ExprLoadDoclist() (see above) has been called, this function is
+** used to iterate/search through the position lists that make up the doclist
+** stored in pExpr->aDoclist.
+*/
+char *sqlite3Fts3FindPositions(
+ Fts3Expr *pExpr, /* Access this expressions doclist */
+ sqlite3_int64 iDocid, /* Docid associated with requested pos-list */
+ int iCol /* Column of requested pos-list */
+){
+ assert( pExpr->isLoaded );
+ if( pExpr->aDoclist ){
+ char *pEnd = &pExpr->aDoclist[pExpr->nDoclist];
+ char *pCsr = pExpr->pCurrent;
+
+ assert( pCsr );
+ while( pCsr<pEnd ){
+ if( pExpr->iCurrent<iDocid ){
+ fts3PoslistCopy(0, &pCsr);
+ if( pCsr<pEnd ){
+ fts3GetDeltaVarint(&pCsr, &pExpr->iCurrent);
}
- dataBufferInit(&(pBuffers[nBuffers]), 0);
- nBuffers++;
- }
-
- /* At this point, must have an empty at iBuffer. */
- assert(iBuffer<nBuffers && pBuffers[iBuffer].nData==0);
-
- /* If empty was first buffer, no need for merge logic. */
- if( iBuffer==0 ){
- dataBufferReplace(&(pBuffers[0]), pData, nData);
+ pExpr->pCurrent = pCsr;
}else{
- /* pAcc is the empty buffer the merged data will end up in. */
- DataBuffer *pAcc = &(pBuffers[iBuffer]);
- DataBuffer *p = &(pBuffers[0]);
-
- /* Handle position 0 specially to avoid need to prime pAcc
- ** with pData/nData.
- */
- dataBufferSwap(p, pAcc);
- rc = docListAccumulateUnion(pAcc, pData, nData);
- if( rc!=SQLITE_OK ) goto err;
-
- /* Accumulate remaining doclists into pAcc. */
- for(++p; p<pAcc; ++p){
- rc = docListAccumulateUnion(pAcc, p->pData, p->nData);
- if( rc!=SQLITE_OK ) goto err;
-
- /* dataBufferReset() could allow a large doclist to blow up
- ** our memory requirements.
- */
- if( p->nCapacity<1024 ){
- dataBufferReset(p);
- }else{
- dataBufferDestroy(p);
- dataBufferInit(p, 0);
+ if( pExpr->iCurrent==iDocid ){
+ int iThis = 0;
+ if( iCol<0 ){
+ /* If iCol is negative, return a pointer to the start of the
+ ** position-list (instead of a pointer to the start of a list
+ ** of offsets associated with a specific column).
+ */
+ return pCsr;
}
+ while( iThis<iCol ){
+ fts3ColumnlistCopy(0, &pCsr);
+ if( *pCsr==0x00 ) return 0;
+ pCsr++;
+ pCsr += sqlite3Fts3GetVarint32(pCsr, &iThis);
+ }
+ if( iCol==iThis && (*pCsr&0xFE) ) return pCsr;
}
+ return 0;
}
}
}
- /* Union all the doclists together into *out. */
- /* TODO(shess) What if *out is big? Sigh. */
- if( rc==SQLITE_OK && nBuffers>0 ){
- int iBuffer;
- for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){
- if( pBuffers[iBuffer].nData>0 ){
- if( out->nData==0 ){
- dataBufferSwap(out, &(pBuffers[iBuffer]));
- }else{
- rc = docListAccumulateUnion(out, pBuffers[iBuffer].pData,
- pBuffers[iBuffer].nData);
- if( rc!=SQLITE_OK ) break;
- }
- }
- }
- }
-
-err:
- while( nBuffers-- ){
- dataBufferDestroy(&(pBuffers[nBuffers]));
- }
- if( pBuffers!=NULL ) sqlite3_free(pBuffers);
-
- return rc;
-}
-
-/* Call loadSegmentLeavesInt() with pData/nData as input. */
-static int loadSegmentLeaf(fulltext_vtab *v, const char *pData, int nData,
- const char *pTerm, int nTerm, int isPrefix,
- DataBuffer *out){
- LeavesReader reader;
- int rc;
-
- assert( nData>1 );
- assert( *pData=='\0' );
- rc = leavesReaderInit(v, 0, 0, 0, pData, nData, &reader);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out);
- leavesReaderReset(&reader);
- leavesReaderDestroy(&reader);
- return rc;
-}
-
-/* Call loadSegmentLeavesInt() with the leaf nodes from iStartLeaf to
-** iEndLeaf (inclusive) as input, and merge the resulting doclist into
-** out.
-*/
-static int loadSegmentLeaves(fulltext_vtab *v,
- sqlite_int64 iStartLeaf, sqlite_int64 iEndLeaf,
- const char *pTerm, int nTerm, int isPrefix,
- DataBuffer *out){
- int rc;
- LeavesReader reader;
-
- assert( iStartLeaf<=iEndLeaf );
- rc = leavesReaderInit(v, 0, iStartLeaf, iEndLeaf, NULL, 0, &reader);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out);
- leavesReaderReset(&reader);
- leavesReaderDestroy(&reader);
- return rc;
-}
-
-/* Taking pData/nData as an interior node, find the sequence of child
-** nodes which could include pTerm/nTerm/isPrefix. Note that the
-** interior node terms logically come between the blocks, so there is
-** one more blockid than there are terms (that block contains terms >=
-** the last interior-node term).
-*/
-/* TODO(shess) The calling code may already know that the end child is
-** not worth calculating, because the end may be in a later sibling
-** node. Consider whether breaking symmetry is worthwhile. I suspect
-** it is not worthwhile.
-*/
-static int getChildrenContaining(const char *pData, int nData,
- const char *pTerm, int nTerm, int isPrefix,
- sqlite_int64 *piStartChild,
- sqlite_int64 *piEndChild){
- InteriorReader reader;
- int rc;
-
- assert( nData>1 );
- assert( *pData!='\0' );
- rc = interiorReaderInit(pData, nData, &reader);
- if( rc!=SQLITE_OK ) return rc;
-
- /* Scan for the first child which could contain pTerm/nTerm. */
- while( !interiorReaderAtEnd(&reader) ){
- if( interiorReaderTermCmp(&reader, pTerm, nTerm, 0)>0 ) break;
- rc = interiorReaderStep(&reader);
- if( rc!=SQLITE_OK ){
- interiorReaderDestroy(&reader);
- return rc;
- }
- }
- *piStartChild = interiorReaderCurrentBlockid(&reader);
-
- /* Keep scanning to find a term greater than our term, using prefix
- ** comparison if indicated. If isPrefix is false, this will be the
- ** same blockid as the starting block.
- */
- while( !interiorReaderAtEnd(&reader) ){
- if( interiorReaderTermCmp(&reader, pTerm, nTerm, isPrefix)>0 ) break;
- rc = interiorReaderStep(&reader);
- if( rc!=SQLITE_OK ){
- interiorReaderDestroy(&reader);
- return rc;
- }
- }
- *piEndChild = interiorReaderCurrentBlockid(&reader);
-
- interiorReaderDestroy(&reader);
-
- /* Children must ascend, and if !prefix, both must be the same. */
- assert( *piEndChild>=*piStartChild );
- assert( isPrefix || *piStartChild==*piEndChild );
- return rc;
-}
-
-/* Read block at iBlockid and pass it with other params to
-** getChildrenContaining().
-*/
-static int loadAndGetChildrenContaining(
- fulltext_vtab *v,
- sqlite_int64 iBlockid,
- const char *pTerm, int nTerm, int isPrefix,
- sqlite_int64 *piStartChild, sqlite_int64 *piEndChild
-){
- sqlite3_stmt *s = NULL;
- int rc;
-
- assert( iBlockid!=0 );
- assert( pTerm!=NULL );
- assert( nTerm!=0 ); /* TODO(shess) Why not allow this? */
- assert( piStartChild!=NULL );
- assert( piEndChild!=NULL );
-
- rc = sql_get_statement(v, BLOCK_SELECT_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_bind_int64(s, 1, iBlockid);
- if( rc!=SQLITE_OK ) return rc;
-
- rc = sqlite3_step(s);
- /* Corrupt if interior node references missing child node. */
- if( rc==SQLITE_DONE ) return SQLITE_CORRUPT_BKPT;
- if( rc!=SQLITE_ROW ) return rc;
-
- /* Corrupt if child node isn't a blob. */
- if( sqlite3_column_type(s, 0)!=SQLITE_BLOB ){
- sqlite3_reset(s); /* So we don't leave a lock. */
- return SQLITE_CORRUPT_BKPT;
- }else{
- const char *pData = sqlite3_column_blob(s, 0);
- int nData = sqlite3_column_bytes(s, 0);
-
- /* Corrupt if child is not a valid interior node. */
- if( pData==NULL || nData<1 || pData[0]=='\0' ){
- sqlite3_reset(s); /* So we don't leave a lock. */
- return SQLITE_CORRUPT_BKPT;
- }
-
- rc = getChildrenContaining(pData, nData, pTerm, nTerm,
- isPrefix, piStartChild, piEndChild);
- if( rc!=SQLITE_OK ){
- sqlite3_reset(s);
- return rc;
- }
- }
-
- /* We expect only one row. We must execute another sqlite3_step()
- * to complete the iteration; otherwise the table will remain
- * locked. */
- rc = sqlite3_step(s);
- if( rc==SQLITE_ROW ) return SQLITE_ERROR;
- if( rc!=SQLITE_DONE ) return rc;
-
- return SQLITE_OK;
-}
-
-/* Traverse the tree represented by pData[nData] looking for
-** pTerm[nTerm], placing its doclist into *out. This is internal to
-** loadSegment() to make error-handling cleaner.
-*/
-static int loadSegmentInt(fulltext_vtab *v, const char *pData, int nData,
- sqlite_int64 iLeavesEnd,
- const char *pTerm, int nTerm, int isPrefix,
- DataBuffer *out){
- /* Special case where root is a leaf. */
- if( *pData=='\0' ){
- return loadSegmentLeaf(v, pData, nData, pTerm, nTerm, isPrefix, out);
- }else{
- int rc;
- sqlite_int64 iStartChild, iEndChild;
-
- /* Process pData as an interior node, then loop down the tree
- ** until we find the set of leaf nodes to scan for the term.
- */
- rc = getChildrenContaining(pData, nData, pTerm, nTerm, isPrefix,
- &iStartChild, &iEndChild);
- if( rc!=SQLITE_OK ) return rc;
- while( iStartChild>iLeavesEnd ){
- sqlite_int64 iNextStart, iNextEnd;
- rc = loadAndGetChildrenContaining(v, iStartChild, pTerm, nTerm, isPrefix,
- &iNextStart, &iNextEnd);
- if( rc!=SQLITE_OK ) return rc;
-
- /* If we've branched, follow the end branch, too. */
- if( iStartChild!=iEndChild ){
- sqlite_int64 iDummy;
- rc = loadAndGetChildrenContaining(v, iEndChild, pTerm, nTerm, isPrefix,
- &iDummy, &iNextEnd);
- if( rc!=SQLITE_OK ) return rc;
- }
-
- assert( iNextStart<=iNextEnd );
- iStartChild = iNextStart;
- iEndChild = iNextEnd;
- }
- assert( iStartChild<=iLeavesEnd );
- assert( iEndChild<=iLeavesEnd );
-
- /* Scan through the leaf segments for doclists. */
- return loadSegmentLeaves(v, iStartChild, iEndChild,
- pTerm, nTerm, isPrefix, out);
- }
-}
-
-/* Call loadSegmentInt() to collect the doclist for pTerm/nTerm, then
-** merge its doclist over *out (any duplicate doclists read from the
-** segment rooted at pData will overwrite those in *out).
-*/
-/* TODO(shess) Consider changing this to determine the depth of the
-** leaves using either the first characters of interior nodes (when
-** ==1, we're one level above the leaves), or the first character of
-** the root (which will describe the height of the tree directly).
-** Either feels somewhat tricky to me.
-*/
-/* TODO(shess) The current merge is likely to be slow for large
-** doclists (though it should process from newest/smallest to
-** oldest/largest, so it may not be that bad). It might be useful to
-** modify things to allow for N-way merging. This could either be
-** within a segment, with pairwise merges across segments, or across
-** all segments at once.
-*/
-static int loadSegment(fulltext_vtab *v, const char *pData, int nData,
- sqlite_int64 iLeavesEnd,
- const char *pTerm, int nTerm, int isPrefix,
- DataBuffer *out){
- DataBuffer result;
- int rc;
-
- /* Corrupt if segment root can't be valid. */
- if( pData==NULL || nData<1 ) return SQLITE_CORRUPT_BKPT;
-
- /* This code should never be called with buffered updates. */
- assert( v->nPendingData<0 );
-
- dataBufferInit(&result, 0);
- rc = loadSegmentInt(v, pData, nData, iLeavesEnd,
- pTerm, nTerm, isPrefix, &result);
- if( rc==SQLITE_OK && result.nData>0 ){
- if( out->nData==0 ){
- DataBuffer tmp = *out;
- *out = result;
- result = tmp;
- }else{
- DataBuffer merged;
- DLReader readers[2];
-
- rc = dlrInit(&readers[0], DL_DEFAULT, out->pData, out->nData);
- if( rc==SQLITE_OK ){
- rc = dlrInit(&readers[1], DL_DEFAULT, result.pData, result.nData);
- if( rc==SQLITE_OK ){
- dataBufferInit(&merged, out->nData+result.nData);
- rc = docListMerge(&merged, readers, 2);
- dataBufferDestroy(out);
- *out = merged;
- dlrDestroy(&readers[1]);
- }
- dlrDestroy(&readers[0]);
- }
- }
- }
-
- dataBufferDestroy(&result);
- return rc;
+ return 0;
}
-/* Scan the database and merge together the posting lists for the term
-** into *out.
-*/
-static int termSelect(
- fulltext_vtab *v,
- int iColumn,
- const char *pTerm, int nTerm, /* Term to query for */
- int isPrefix, /* True for a prefix search */
- DocListType iType,
- DataBuffer *out /* Write results here */
+/*
+** Helper function used by the implementation of the overloaded snippet(),
+** offsets() and optimize() SQL functions.
+**
+** If the value passed as the third argument is a blob of size
+** sizeof(Fts3Cursor*), then the blob contents are copied to the
+** output variable *ppCsr and SQLITE_OK is returned. Otherwise, an error
+** message is written to context pContext and SQLITE_ERROR returned. The
+** string passed via zFunc is used as part of the error message.
+*/
+static int fts3FunctionArg(
+ sqlite3_context *pContext, /* SQL function call context */
+ const char *zFunc, /* Function name */
+ sqlite3_value *pVal, /* argv[0] passed to function */
+ Fts3Cursor **ppCsr /* OUT: Store cursor handle here */
){
- DataBuffer doclist;
- sqlite3_stmt *s;
- int rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
- if( rc!=SQLITE_OK ) return rc;
-
- /* This code should never be called with buffered updates. */
- assert( v->nPendingData<0 );
-
- dataBufferInit(&doclist, 0);
- dataBufferInit(out, 0);
-
- /* Traverse the segments from oldest to newest so that newer doclist
- ** elements for given docids overwrite older elements.
- */
- while( (rc = sqlite3_step(s))==SQLITE_ROW ){
- const char *pData = sqlite3_column_blob(s, 2);
- const int nData = sqlite3_column_bytes(s, 2);
- const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1);
-
- /* Corrupt if we get back different types than we stored. */
- if( sqlite3_column_type(s, 1)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 2)!=SQLITE_BLOB ){
- rc = SQLITE_CORRUPT_BKPT;
- goto err;
- }
-
- rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, isPrefix,
- &doclist);
- if( rc!=SQLITE_OK ) goto err;
- }
- if( rc==SQLITE_DONE ){
- rc = SQLITE_OK;
- if( doclist.nData!=0 ){
- /* TODO(shess) The old term_select_all() code applied the column
- ** restrict as we merged segments, leading to smaller buffers.
- ** This is probably worthwhile to bring back, once the new storage
- ** system is checked in.
- */
- if( iColumn==v->nColumn) iColumn = -1;
- rc = docListTrim(DL_DEFAULT, doclist.pData, doclist.nData,
- iColumn, iType, out);
- }
- }
-
- err:
- sqlite3_reset(s); /* So we don't leave a lock. */
- dataBufferDestroy(&doclist);
- return rc;
-}
-
-/****************************************************************/
-/* Used to hold hashtable data for sorting. */
-typedef struct TermData {
- const char *pTerm;
- int nTerm;
- DLCollector *pCollector;
-} TermData;
-
-/* Orders TermData elements in strcmp fashion ( <0 for less-than, 0
-** for equal, >0 for greater-than).
-*/
-static int termDataCmp(const void *av, const void *bv){
- const TermData *a = (const TermData *)av;
- const TermData *b = (const TermData *)bv;
- int n = a->nTerm<b->nTerm ? a->nTerm : b->nTerm;
- int c = memcmp(a->pTerm, b->pTerm, n);
- if( c!=0 ) return c;
- return a->nTerm-b->nTerm;
-}
-
-/* Order pTerms data by term, then write a new level 0 segment using
-** LeafWriter.
-*/
-static int writeZeroSegment(fulltext_vtab *v, fts3Hash *pTerms){
- fts3HashElem *e;
- int idx, rc, i, n;
- TermData *pData;
- LeafWriter writer;
- DataBuffer dl;
-
- /* Determine the next index at level 0, merging as necessary. */
- rc = segdirNextIndex(v, 0, &idx);
- if( rc!=SQLITE_OK ) return rc;
-
- n = fts3HashCount(pTerms);
- pData = sqlite3_malloc(n*sizeof(TermData));
-
- for(i = 0, e = fts3HashFirst(pTerms); e; i++, e = fts3HashNext(e)){
- assert( i<n );
- pData[i].pTerm = fts3HashKey(e);
- pData[i].nTerm = fts3HashKeysize(e);
- pData[i].pCollector = fts3HashData(e);
- }
- assert( i==n );
-
- /* TODO(shess) Should we allow user-defined collation sequences,
- ** here? I think we only need that once we support prefix searches.
- */
- if( n>1 ) qsort(pData, n, sizeof(*pData), termDataCmp);
-
- /* TODO(shess) Refactor so that we can write directly to the segment
- ** DataBuffer, as happens for segment merges.
- */
- leafWriterInit(0, idx, &writer);
- dataBufferInit(&dl, 0);
- for(i=0; i<n; i++){
- dataBufferReset(&dl);
- dlcAddDoclist(pData[i].pCollector, &dl);
- rc = leafWriterStep(v, &writer,
- pData[i].pTerm, pData[i].nTerm, dl.pData, dl.nData);
- if( rc!=SQLITE_OK ) goto err;
- }
- rc = leafWriterFinalize(v, &writer);
-
- err:
- dataBufferDestroy(&dl);
- sqlite3_free(pData);
- leafWriterDestroy(&writer);
- return rc;
-}
-
-/* If pendingTerms has data, free it. */
-static int clearPendingTerms(fulltext_vtab *v){
- if( v->nPendingData>=0 ){
- fts3HashElem *e;
- for(e=fts3HashFirst(&v->pendingTerms); e; e=fts3HashNext(e)){
- dlcDelete(fts3HashData(e));
- }
- fts3HashClear(&v->pendingTerms);
- v->nPendingData = -1;
- }
- return SQLITE_OK;
-}
-
-/* If pendingTerms has data, flush it to a level-zero segment, and
-** free it.
-*/
-static int flushPendingTerms(fulltext_vtab *v){
- if( v->nPendingData>=0 ){
- int rc = writeZeroSegment(v, &v->pendingTerms);
- if( rc==SQLITE_OK ) clearPendingTerms(v);
- return rc;
- }
- return SQLITE_OK;
-}
-
-/* If pendingTerms is "too big", or docid is out of order, flush it.
-** Regardless, be certain that pendingTerms is initialized for use.
-*/
-static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid){
- /* TODO(shess) Explore whether partially flushing the buffer on
- ** forced-flush would provide better performance. I suspect that if
- ** we ordered the doclists by size and flushed the largest until the
- ** buffer was half empty, that would let the less frequent terms
- ** generate longer doclists.
- */
- if( iDocid<=v->iPrevDocid || v->nPendingData>kPendingThreshold ){
- int rc = flushPendingTerms(v);
- if( rc!=SQLITE_OK ) return rc;
- }
- if( v->nPendingData<0 ){
- fts3HashInit(&v->pendingTerms, FTS3_HASH_STRING, 1);
- v->nPendingData = 0;
+ Fts3Cursor *pRet;
+ if( sqlite3_value_type(pVal)!=SQLITE_BLOB
+ || sqlite3_value_bytes(pVal)!=sizeof(Fts3Cursor *)
+ ){
+ char *zErr = sqlite3_mprintf("illegal first argument to %s", zFunc);
+ sqlite3_result_error(pContext, zErr, -1);
+ sqlite3_free(zErr);
+ return SQLITE_ERROR;
}
- v->iPrevDocid = iDocid;
+ memcpy(&pRet, sqlite3_value_blob(pVal), sizeof(Fts3Cursor *));
+ *ppCsr = pRet;
return SQLITE_OK;
}
-/* This function implements the xUpdate callback; it is the top-level entry
- * point for inserting, deleting or updating a row in a full-text table. */
-static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg,
- sqlite_int64 *pRowid){
- fulltext_vtab *v = (fulltext_vtab *) pVtab;
- int rc;
-
- FTSTRACE(("FTS3 Update %p\n", pVtab));
-
- if( nArg<2 ){
- rc = index_delete(v, sqlite3_value_int64(ppArg[0]));
- if( rc==SQLITE_OK ){
- /* If we just deleted the last row in the table, clear out the
- ** index data.
- */
- rc = content_exists(v);
- if( rc==SQLITE_ROW ){
- rc = SQLITE_OK;
- }else if( rc==SQLITE_DONE ){
- /* Clear the pending terms so we don't flush a useless level-0
- ** segment when the transaction closes.
- */
- rc = clearPendingTerms(v);
- if( rc==SQLITE_OK ){
- rc = segdir_delete_all(v);
- }
- }
- }
- } else if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){
- /* An update:
- * ppArg[0] = old rowid
- * ppArg[1] = new rowid
- * ppArg[2..2+v->nColumn-1] = values
- * ppArg[2+v->nColumn] = value for magic column (we ignore this)
- * ppArg[2+v->nColumn+1] = value for docid
- */
- sqlite_int64 rowid = sqlite3_value_int64(ppArg[0]);
- if( sqlite3_value_type(ppArg[1]) != SQLITE_INTEGER ||
- sqlite3_value_int64(ppArg[1]) != rowid ){
- rc = SQLITE_ERROR; /* we don't allow changing the rowid */
- }else if( sqlite3_value_type(ppArg[2+v->nColumn+1]) != SQLITE_INTEGER ||
- sqlite3_value_int64(ppArg[2+v->nColumn+1]) != rowid ){
- rc = SQLITE_ERROR; /* we don't allow changing the docid */
- }else{
- assert( nArg==2+v->nColumn+2);
- rc = index_update(v, rowid, &ppArg[2]);
- }
- } else {
- /* An insert:
- * ppArg[1] = requested rowid
- * ppArg[2..2+v->nColumn-1] = values
- * ppArg[2+v->nColumn] = value for magic column (we ignore this)
- * ppArg[2+v->nColumn+1] = value for docid
- */
- sqlite3_value *pRequestDocid = ppArg[2+v->nColumn+1];
- assert( nArg==2+v->nColumn+2);
- if( SQLITE_NULL != sqlite3_value_type(pRequestDocid) &&
- SQLITE_NULL != sqlite3_value_type(ppArg[1]) ){
- /* TODO(shess) Consider allowing this to work if the values are
- ** identical. I'm inclined to discourage that usage, though,
- ** given that both rowid and docid are special columns. Better
- ** would be to define one or the other as the default winner,
- ** but should it be fts3-centric (docid) or SQLite-centric
- ** (rowid)?
- */
- rc = SQLITE_ERROR;
- }else{
- if( SQLITE_NULL == sqlite3_value_type(pRequestDocid) ){
- pRequestDocid = ppArg[1];
- }
- rc = index_insert(v, pRequestDocid, &ppArg[2], pRowid);
- }
- }
-
- return rc;
-}
-
-static int fulltextSync(sqlite3_vtab *pVtab){
- FTSTRACE(("FTS3 xSync()\n"));
- return flushPendingTerms((fulltext_vtab *)pVtab);
-}
-
-static int fulltextBegin(sqlite3_vtab *pVtab){
- fulltext_vtab *v = (fulltext_vtab *) pVtab;
- FTSTRACE(("FTS3 xBegin()\n"));
-
- /* Any buffered updates should have been cleared by the previous
- ** transaction.
- */
- assert( v->nPendingData<0 );
- return clearPendingTerms(v);
-}
-
-static int fulltextCommit(sqlite3_vtab *pVtab){
- fulltext_vtab *v = (fulltext_vtab *) pVtab;
- FTSTRACE(("FTS3 xCommit()\n"));
-
- /* Buffered updates should have been cleared by fulltextSync(). */
- assert( v->nPendingData<0 );
- return clearPendingTerms(v);
-}
-
-static int fulltextRollback(sqlite3_vtab *pVtab){
- FTSTRACE(("FTS3 xRollback()\n"));
- return clearPendingTerms((fulltext_vtab *)pVtab);
-}
-
/*
** Implementation of the snippet() function for FTS3
*/
-static void snippetFunc(
- sqlite3_context *pContext,
- int argc,
- sqlite3_value **argv
-){
- fulltext_cursor *pCursor;
- if( argc<1 ) return;
- if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
- sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
- sqlite3_result_error(pContext, "illegal first argument to html_snippet",-1);
- }else{
- const char *zStart = "<b>";
- const char *zEnd = "</b>";
- const char *zEllipsis = "<b>...</b>";
- memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
- if( argc>=2 ){
- zStart = (const char*)sqlite3_value_text(argv[1]);
- if( argc>=3 ){
- zEnd = (const char*)sqlite3_value_text(argv[2]);
- if( argc>=4 ){
- zEllipsis = (const char*)sqlite3_value_text(argv[3]);
- }
- }
- }
- snippetAllOffsets(pCursor);
- snippetText(pCursor, zStart, zEnd, zEllipsis);
- sqlite3_result_text(pContext, pCursor->snippet.zSnippet,
- pCursor->snippet.nSnippet, SQLITE_STATIC);
- }
-}
-
-/*
-** Implementation of the offsets() function for FTS3
-*/
-static void snippetOffsetsFunc(
- sqlite3_context *pContext,
- int argc,
- sqlite3_value **argv
+static void fts3SnippetFunc(
+ sqlite3_context *pContext, /* SQLite function call context */
+ int nVal, /* Size of apVal[] array */
+ sqlite3_value **apVal /* Array of arguments */
){
- fulltext_cursor *pCursor;
- if( argc<1 ) return;
- if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
- sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
- sqlite3_result_error(pContext, "illegal first argument to offsets",-1);
- }else{
- memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
- snippetAllOffsets(pCursor);
- snippetOffsetText(&pCursor->snippet);
- sqlite3_result_text(pContext,
- pCursor->snippet.zOffset, pCursor->snippet.nOffset,
- SQLITE_STATIC);
- }
-}
-
-/* OptLeavesReader is nearly identical to LeavesReader, except that
-** where LeavesReader is geared towards the merging of complete
-** segment levels (with exactly MERGE_COUNT segments), OptLeavesReader
-** is geared towards implementation of the optimize() function, and
-** can merge all segments simultaneously. This version may be
-** somewhat less efficient than LeavesReader because it merges into an
-** accumulator rather than doing an N-way merge, but since segment
-** size grows exponentially (so segment count logrithmically) this is
-** probably not an immediate problem.
-*/
-/* TODO(shess): Prove that assertion, or extend the merge code to
-** merge tree fashion (like the prefix-searching code does).
-*/
-/* TODO(shess): OptLeavesReader and LeavesReader could probably be
-** merged with little or no loss of performance for LeavesReader. The
-** merged code would need to handle >MERGE_COUNT segments, and would
-** also need to be able to optionally optimize away deletes.
-*/
-typedef struct OptLeavesReader {
- /* Segment number, to order readers by age. */
- int segment;
- LeavesReader reader;
-} OptLeavesReader;
-
-static int optLeavesReaderAtEnd(OptLeavesReader *pReader){
- return leavesReaderAtEnd(&pReader->reader);
-}
-static int optLeavesReaderTermBytes(OptLeavesReader *pReader){
- return leavesReaderTermBytes(&pReader->reader);
-}
-static const char *optLeavesReaderData(OptLeavesReader *pReader){
- return leavesReaderData(&pReader->reader);
-}
-static int optLeavesReaderDataBytes(OptLeavesReader *pReader){
- return leavesReaderDataBytes(&pReader->reader);
-}
-static const char *optLeavesReaderTerm(OptLeavesReader *pReader){
- return leavesReaderTerm(&pReader->reader);
-}
-static int optLeavesReaderStep(fulltext_vtab *v, OptLeavesReader *pReader){
- return leavesReaderStep(v, &pReader->reader);
-}
-static int optLeavesReaderTermCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){
- return leavesReaderTermCmp(&lr1->reader, &lr2->reader);
-}
-/* Order by term ascending, segment ascending (oldest to newest), with
-** exhausted readers to the end.
-*/
-static int optLeavesReaderCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){
- int c = optLeavesReaderTermCmp(lr1, lr2);
- if( c!=0 ) return c;
- return lr1->segment-lr2->segment;
-}
-/* Bubble pLr[0] to appropriate place in pLr[1..nLr-1]. Assumes that
-** pLr[1..nLr-1] is already sorted.
-*/
-static void optLeavesReaderReorder(OptLeavesReader *pLr, int nLr){
- while( nLr>1 && optLeavesReaderCmp(pLr, pLr+1)>0 ){
- OptLeavesReader tmp = pLr[0];
- pLr[0] = pLr[1];
- pLr[1] = tmp;
- nLr--;
- pLr++;
- }
-}
-
-/* optimize() helper function. Put the readers in order and iterate
-** through them, merging doclists for matching terms into pWriter.
-** Returns SQLITE_OK on success, or the SQLite error code which
-** prevented success.
-*/
-static int optimizeInternal(fulltext_vtab *v,
- OptLeavesReader *readers, int nReaders,
- LeafWriter *pWriter){
- int i, rc = SQLITE_OK;
- DataBuffer doclist, merged, tmp;
- const char *pData;
-
- /* Order the readers. */
- i = nReaders;
- while( i-- > 0 ){
- optLeavesReaderReorder(&readers[i], nReaders-i);
- }
-
- dataBufferInit(&doclist, LEAF_MAX);
- dataBufferInit(&merged, LEAF_MAX);
-
- /* Exhausted readers bubble to the end, so when the first reader is
- ** at eof, all are at eof.
+ Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
+ const char *zStart = "<b>";
+ const char *zEnd = "</b>";
+ const char *zEllipsis = "<b>...</b>";
+ int iCol = -1;
+ int nToken = 15; /* Default number of tokens in snippet */
+
+ /* There must be at least one argument passed to this function (otherwise
+ ** the non-overloaded version would have been called instead of this one).
*/
- while( !optLeavesReaderAtEnd(&readers[0]) ){
+ assert( nVal>=1 );
- /* Figure out how many readers share the next term. */
- for(i=1; i<nReaders && !optLeavesReaderAtEnd(&readers[i]); i++){
- if( 0!=optLeavesReaderTermCmp(&readers[0], &readers[i]) ) break;
- }
-
- pData = optLeavesReaderData(&readers[0]);
- if( pData==NULL ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
-
- /* Special-case for no merge. */
- if( i==1 ){
- /* Trim deletions from the doclist. */
- dataBufferReset(&merged);
- rc = docListTrim(DL_DEFAULT, pData,
- optLeavesReaderDataBytes(&readers[0]),
- -1, DL_DEFAULT, &merged);
- if( rc!=SQLITE_OK ) break;
- }else{
- DLReader dlReaders[MERGE_COUNT];
- int iReader, nReaders;
-
- /* Prime the pipeline with the first reader's doclist. After
- ** one pass index 0 will reference the accumulated doclist.
- */
- rc = dlrInit(&dlReaders[0], DL_DEFAULT,
- pData,
- optLeavesReaderDataBytes(&readers[0]));
- if( rc!=SQLITE_OK ) break;
- iReader = 1;
-
- assert( iReader<i ); /* Must execute the loop at least once. */
- while( iReader<i ){
- /* Merge 16 inputs per pass. */
- for( nReaders=1; iReader<i && nReaders<MERGE_COUNT;
- iReader++, nReaders++ ){
- pData = optLeavesReaderData(&readers[iReader]);
- if( pData==NULL ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
- rc = dlrInit(&dlReaders[nReaders], DL_DEFAULT, pData,
- optLeavesReaderDataBytes(&readers[iReader]));
- if( rc!=SQLITE_OK ) break;
- }
-
- /* Merge doclists and swap result into accumulator. */
- if( rc==SQLITE_OK ){
- dataBufferReset(&merged);
- rc = docListMerge(&merged, dlReaders, nReaders);
- tmp = merged;
- merged = doclist;
- doclist = tmp;
- }
-
- while( nReaders-- > 0 ){
- dlrDestroy(&dlReaders[nReaders]);
- }
-
- if( rc!=SQLITE_OK ) goto err;
-
- /* Accumulated doclist to reader 0 for next pass. */
- rc = dlrInit(&dlReaders[0], DL_DEFAULT, doclist.pData, doclist.nData);
- if( rc!=SQLITE_OK ) goto err;
- }
-
- /* Destroy reader that was left in the pipeline. */
- dlrDestroy(&dlReaders[0]);
-
- /* Trim deletions from the doclist. */
- dataBufferReset(&merged);
- rc = docListTrim(DL_DEFAULT, doclist.pData, doclist.nData,
- -1, DL_DEFAULT, &merged);
- if( rc!=SQLITE_OK ) goto err;
- }
-
- /* Only pass doclists with hits (skip if all hits deleted). */
- if( merged.nData>0 ){
- rc = leafWriterStep(v, pWriter,
- optLeavesReaderTerm(&readers[0]),
- optLeavesReaderTermBytes(&readers[0]),
- merged.pData, merged.nData);
- if( rc!=SQLITE_OK ) goto err;
- }
-
- /* Step merged readers to next term and reorder. */
- while( i-- > 0 ){
- rc = optLeavesReaderStep(v, &readers[i]);
- if( rc!=SQLITE_OK ) goto err;
-
- optLeavesReaderReorder(&readers[i], nReaders-i);
- }
- }
-
- err:
- dataBufferDestroy(&doclist);
- dataBufferDestroy(&merged);
- return rc;
-}
-
-/* Implement optimize() function for FTS3. optimize(t) merges all
-** segments in the fts index into a single segment. 't' is the magic
-** table-named column.
-*/
-static void optimizeFunc(sqlite3_context *pContext,
- int argc, sqlite3_value **argv){
- fulltext_cursor *pCursor;
- if( argc>1 ){
- sqlite3_result_error(pContext, "excess arguments to optimize()",-1);
- }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
- sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
- sqlite3_result_error(pContext, "illegal first argument to optimize",-1);
- }else{
- fulltext_vtab *v;
- int i, rc, iMaxLevel;
- OptLeavesReader *readers;
- int nReaders;
- LeafWriter writer;
- sqlite3_stmt *s;
-
- memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
- v = cursor_vtab(pCursor);
-
- /* Flush any buffered updates before optimizing. */
- rc = flushPendingTerms(v);
- if( rc!=SQLITE_OK ) goto err;
-
- rc = segdir_count(v, &nReaders, &iMaxLevel);
- if( rc!=SQLITE_OK ) goto err;
- if( nReaders==0 || nReaders==1 ){
- sqlite3_result_text(pContext, "Index already optimal", -1,
- SQLITE_STATIC);
- return;
- }
-
- rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
- if( rc!=SQLITE_OK ) goto err;
-
- readers = sqlite3_malloc(nReaders*sizeof(readers[0]));
- if( readers==NULL ) goto err;
-
- /* Note that there will already be a segment at this position
- ** until we call segdir_delete() on iMaxLevel.
- */
- leafWriterInit(iMaxLevel, 0, &writer);
-
- i = 0;
- while( (rc = sqlite3_step(s))==SQLITE_ROW ){
- sqlite_int64 iStart = sqlite3_column_int64(s, 0);
- sqlite_int64 iEnd = sqlite3_column_int64(s, 1);
- const char *pRootData = sqlite3_column_blob(s, 2);
- int nRootData = sqlite3_column_bytes(s, 2);
-
- /* Corrupt if we get back different types than we stored. */
- if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 1)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 2)!=SQLITE_BLOB ){
- rc = SQLITE_CORRUPT_BKPT;
- break;
- }
-
- assert( i<nReaders );
- rc = leavesReaderInit(v, -1, iStart, iEnd, pRootData, nRootData,
- &readers[i].reader);
- if( rc!=SQLITE_OK ) break;
-
- readers[i].segment = i;
- i++;
- }
-
- /* If we managed to successfully read them all, optimize them. */
- if( rc==SQLITE_DONE ){
- assert( i==nReaders );
- rc = optimizeInternal(v, readers, nReaders, &writer);
- }else{
- sqlite3_reset(s); /* So we don't leave a lock. */
- }
-
- while( i-- > 0 ){
- leavesReaderDestroy(&readers[i].reader);
- }
- sqlite3_free(readers);
-
- /* If we've successfully gotten to here, delete the old segments
- ** and flush the interior structure of the new segment.
- */
- if( rc==SQLITE_OK ){
- for( i=0; i<=iMaxLevel; i++ ){
- rc = segdir_delete(v, i);
- if( rc!=SQLITE_OK ) break;
- }
-
- if( rc==SQLITE_OK ) rc = leafWriterFinalize(v, &writer);
- }
-
- leafWriterDestroy(&writer);
-
- if( rc!=SQLITE_OK ) goto err;
-
- sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC);
+ if( nVal>6 ){
+ sqlite3_result_error(pContext,
+ "wrong number of arguments to function snippet()", -1);
return;
-
- /* TODO(shess): Error-handling needs to be improved along the
- ** lines of the dump_ functions.
- */
- err:
- {
- char buf[512];
- sqlite3_snprintf(sizeof(buf), buf, "Error in optimize: %s",
- sqlite3_errmsg(sqlite3_context_db_handle(pContext)));
- sqlite3_result_error(pContext, buf, -1);
- }
}
-}
-
-#ifdef SQLITE_TEST
-/* Generate an error of the form "<prefix>: <msg>". If msg is NULL,
-** pull the error from the context's db handle.
-*/
-static void generateError(sqlite3_context *pContext,
- const char *prefix, const char *msg){
- char buf[512];
- if( msg==NULL ) msg = sqlite3_errmsg(sqlite3_context_db_handle(pContext));
- sqlite3_snprintf(sizeof(buf), buf, "%s: %s", prefix, msg);
- sqlite3_result_error(pContext, buf, -1);
-}
-
-/* Helper function to collect the set of terms in the segment into
-** pTerms. The segment is defined by the leaf nodes between
-** iStartBlockid and iEndBlockid, inclusive, or by the contents of
-** pRootData if iStartBlockid is 0 (in which case the entire segment
-** fit in a leaf).
-*/
-static int collectSegmentTerms(fulltext_vtab *v, sqlite3_stmt *s,
- fts3Hash *pTerms){
- const sqlite_int64 iStartBlockid = sqlite3_column_int64(s, 0);
- const sqlite_int64 iEndBlockid = sqlite3_column_int64(s, 1);
- const char *pRootData = sqlite3_column_blob(s, 2);
- const int nRootData = sqlite3_column_bytes(s, 2);
- int rc;
- LeavesReader reader;
+ if( fts3FunctionArg(pContext, "snippet", apVal[0], &pCsr) ) return;
- /* Corrupt if we get back different types than we stored. */
- if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 1)!=SQLITE_INTEGER ||
- sqlite3_column_type(s, 2)!=SQLITE_BLOB ){
- return SQLITE_CORRUPT_BKPT;
+ switch( nVal ){
+ case 6: nToken = sqlite3_value_int(apVal[5]);
+ case 5: iCol = sqlite3_value_int(apVal[4]);
+ case 4: zEllipsis = (const char*)sqlite3_value_text(apVal[3]);
+ case 3: zEnd = (const char*)sqlite3_value_text(apVal[2]);
+ case 2: zStart = (const char*)sqlite3_value_text(apVal[1]);
}
-
- rc = leavesReaderInit(v, 0, iStartBlockid, iEndBlockid,
- pRootData, nRootData, &reader);
- if( rc!=SQLITE_OK ) return rc;
-
- while( rc==SQLITE_OK && !leavesReaderAtEnd(&reader) ){
- const char *pTerm = leavesReaderTerm(&reader);
- const int nTerm = leavesReaderTermBytes(&reader);
- void *oldValue = sqlite3Fts3HashFind(pTerms, pTerm, nTerm);
- void *newValue = (void *)((char *)oldValue+1);
-
- /* From the comment before sqlite3Fts3HashInsert in fts3_hash.c,
- ** the data value passed is returned in case of malloc failure.
- */
- if( newValue==sqlite3Fts3HashInsert(pTerms, pTerm, nTerm, newValue) ){
- rc = SQLITE_NOMEM;
- }else{
- rc = leavesReaderStep(v, &reader);
- }
+ if( !zEllipsis || !zEnd || !zStart ){
+ sqlite3_result_error_nomem(pContext);
+ }else if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){
+ sqlite3Fts3Snippet(pContext, pCsr, zStart, zEnd, zEllipsis, iCol, nToken);
}
-
- leavesReaderDestroy(&reader);
- return rc;
}
-/* Helper function to build the result string for dump_terms(). */
-static int generateTermsResult(sqlite3_context *pContext, fts3Hash *pTerms){
- int iTerm, nTerms, nResultBytes, iByte;
- char *result;
- TermData *pData;
- fts3HashElem *e;
-
- /* Iterate pTerms to generate an array of terms in pData for
- ** sorting.
- */
- nTerms = fts3HashCount(pTerms);
- assert( nTerms>0 );
- pData = sqlite3_malloc(nTerms*sizeof(TermData));
- if( pData==NULL ) return SQLITE_NOMEM;
-
- nResultBytes = 0;
- for(iTerm = 0, e = fts3HashFirst(pTerms); e; iTerm++, e = fts3HashNext(e)){
- nResultBytes += fts3HashKeysize(e)+1; /* Term plus trailing space */
- assert( iTerm<nTerms );
- pData[iTerm].pTerm = fts3HashKey(e);
- pData[iTerm].nTerm = fts3HashKeysize(e);
- pData[iTerm].pCollector = fts3HashData(e); /* unused */
- }
- assert( iTerm==nTerms );
-
- assert( nResultBytes>0 ); /* nTerms>0, nResultsBytes must be, too. */
- result = sqlite3_malloc(nResultBytes);
- if( result==NULL ){
- sqlite3_free(pData);
- return SQLITE_NOMEM;
- }
-
- if( nTerms>1 ) qsort(pData, nTerms, sizeof(*pData), termDataCmp);
-
- /* Read the terms in order to build the result. */
- iByte = 0;
- for(iTerm=0; iTerm<nTerms; ++iTerm){
- memcpy(result+iByte, pData[iTerm].pTerm, pData[iTerm].nTerm);
- iByte += pData[iTerm].nTerm;
- result[iByte++] = ' ';
- }
- assert( iByte==nResultBytes );
- assert( result[nResultBytes-1]==' ' );
- result[nResultBytes-1] = '\0';
-
- /* Passes away ownership of result. */
- sqlite3_result_text(pContext, result, nResultBytes-1, sqlite3_free);
- sqlite3_free(pData);
- return SQLITE_OK;
-}
-
-/* Implements dump_terms() for use in inspecting the fts3 index from
-** tests. TEXT result containing the ordered list of terms joined by
-** spaces. dump_terms(t, level, idx) dumps the terms for the segment
-** specified by level, idx (in %_segdir), while dump_terms(t) dumps
-** all terms in the index. In both cases t is the fts table's magic
-** table-named column.
+/*
+** Implementation of the offsets() function for FTS3
*/
-static void dumpTermsFunc(
- sqlite3_context *pContext,
- int argc, sqlite3_value **argv
+static void fts3OffsetsFunc(
+ sqlite3_context *pContext, /* SQLite function call context */
+ int nVal, /* Size of argument array */
+ sqlite3_value **apVal /* Array of arguments */
){
- fulltext_cursor *pCursor;
- if( argc!=3 && argc!=1 ){
- generateError(pContext, "dump_terms", "incorrect arguments");
- }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
- sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
- generateError(pContext, "dump_terms", "illegal first argument");
- }else{
- fulltext_vtab *v;
- fts3Hash terms;
- sqlite3_stmt *s = NULL;
- int rc;
-
- memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
- v = cursor_vtab(pCursor);
+ Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
- /* If passed only the cursor column, get all segments. Otherwise
- ** get the segment described by the following two arguments.
- */
- if( argc==1 ){
- rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s);
- }else{
- rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s);
- if( rc==SQLITE_OK ){
- rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[1]));
- if( rc==SQLITE_OK ){
- rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[2]));
- }
- }
- }
+ UNUSED_PARAMETER(nVal);
- if( rc!=SQLITE_OK ){
- generateError(pContext, "dump_terms", NULL);
- return;
- }
-
- /* Collect the terms for each segment. */
- sqlite3Fts3HashInit(&terms, FTS3_HASH_STRING, 1);
- while( (rc = sqlite3_step(s))==SQLITE_ROW ){
- rc = collectSegmentTerms(v, s, &terms);
- if( rc!=SQLITE_OK ) break;
- }
-
- if( rc!=SQLITE_DONE ){
- sqlite3_reset(s);
- generateError(pContext, "dump_terms", NULL);
- }else{
- const int nTerms = fts3HashCount(&terms);
- if( nTerms>0 ){
- rc = generateTermsResult(pContext, &terms);
- if( rc==SQLITE_NOMEM ){
- generateError(pContext, "dump_terms", "out of memory");
- }else{
- assert( rc==SQLITE_OK );
- }
- }else if( argc==3 ){
- /* The specific segment asked for could not be found. */
- generateError(pContext, "dump_terms", "segment not found");
- }else{
- /* No segments found. */
- /* TODO(shess): It should be impossible to reach this. This
- ** case can only happen for an empty table, in which case
- ** SQLite has no rows to call this function on.
- */
- sqlite3_result_null(pContext);
- }
- }
- sqlite3Fts3HashClear(&terms);
+ assert( nVal==1 );
+ if( fts3FunctionArg(pContext, "offsets", apVal[0], &pCsr) ) return;
+ assert( pCsr );
+ if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){
+ sqlite3Fts3Offsets(pContext, pCsr);
}
}
-/* Expand the DL_DEFAULT doclist in pData into a text result in
-** pContext.
+/*
+** Implementation of the special optimize() function for FTS3. This
+** function merges all segments in the database to a single segment.
+** Example usage is:
+**
+** SELECT optimize(t) FROM t LIMIT 1;
+**
+** where 't' is the name of an FTS3 table.
*/
-static void createDoclistResult(sqlite3_context *pContext,
- const char *pData, int nData){
- DataBuffer dump;
- DLReader dlReader;
- int rc;
+static void fts3OptimizeFunc(
+ sqlite3_context *pContext, /* SQLite function call context */
+ int nVal, /* Size of argument array */
+ sqlite3_value **apVal /* Array of arguments */
+){
+ int rc; /* Return code */
+ Fts3Table *p; /* Virtual table handle */
+ Fts3Cursor *pCursor; /* Cursor handle passed through apVal[0] */
- assert( pData!=NULL && nData>0 );
+ UNUSED_PARAMETER(nVal);
- rc = dlrInit(&dlReader, DL_DEFAULT, pData, nData);
- if( rc!=SQLITE_OK ) return rc;
- dataBufferInit(&dump, 0);
- for( ; rc==SQLITE_OK && !dlrAtEnd(&dlReader); rc = dlrStep(&dlReader) ){
- char buf[256];
- PLReader plReader;
+ assert( nVal==1 );
+ if( fts3FunctionArg(pContext, "optimize", apVal[0], &pCursor) ) return;
+ p = (Fts3Table *)pCursor->base.pVtab;
+ assert( p );
- rc = plrInit(&plReader, &dlReader);
- if( rc!=SQLITE_OK ) break;
- if( DL_DEFAULT==DL_DOCIDS || plrAtEnd(&plReader) ){
- sqlite3_snprintf(sizeof(buf), buf, "[%lld] ", dlrDocid(&dlReader));
- dataBufferAppend(&dump, buf, strlen(buf));
- }else{
- int iColumn = plrColumn(&plReader);
-
- sqlite3_snprintf(sizeof(buf), buf, "[%lld %d[",
- dlrDocid(&dlReader), iColumn);
- dataBufferAppend(&dump, buf, strlen(buf));
-
- for( ; !plrAtEnd(&plReader); rc = plrStep(&plReader) ){
- if( rc!=SQLITE_OK ) break;
- if( plrColumn(&plReader)!=iColumn ){
- iColumn = plrColumn(&plReader);
- sqlite3_snprintf(sizeof(buf), buf, "] %d[", iColumn);
- assert( dump.nData>0 );
- dump.nData--; /* Overwrite trailing space. */
- assert( dump.pData[dump.nData]==' ');
- dataBufferAppend(&dump, buf, strlen(buf));
- }
- if( DL_DEFAULT==DL_POSITIONS_OFFSETS ){
- sqlite3_snprintf(sizeof(buf), buf, "%d,%d,%d ",
- plrPosition(&plReader),
- plrStartOffset(&plReader), plrEndOffset(&plReader));
- }else if( DL_DEFAULT==DL_POSITIONS ){
- sqlite3_snprintf(sizeof(buf), buf, "%d ", plrPosition(&plReader));
- }else{
- assert( NULL=="Unhandled DL_DEFAULT value");
- }
- dataBufferAppend(&dump, buf, strlen(buf));
- }
- plrDestroy(&plReader);
- if( rc!= SQLITE_OK ) break;
+ rc = sqlite3Fts3Optimize(p);
- assert( dump.nData>0 );
- dump.nData--; /* Overwrite trailing space. */
- assert( dump.pData[dump.nData]==' ');
- dataBufferAppend(&dump, "]] ", 3);
- }
- }
- dlrDestroy(&dlReader);
- if( rc!=SQLITE_OK ){
- dataBufferDestroy(&dump);
- return rc;
+ switch( rc ){
+ case SQLITE_OK:
+ sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC);
+ break;
+ case SQLITE_DONE:
+ sqlite3_result_text(pContext, "Index already optimal", -1, SQLITE_STATIC);
+ break;
+ default:
+ sqlite3_result_error_code(pContext, rc);
+ break;
}
-
- assert( dump.nData>0 );
- dump.nData--; /* Overwrite trailing space. */
- assert( dump.pData[dump.nData]==' ');
- dump.pData[dump.nData] = '\0';
- assert( dump.nData>0 );
-
- /* Passes ownership of dump's buffer to pContext. */
- sqlite3_result_text(pContext, dump.pData, dump.nData, sqlite3_free);
- dump.pData = NULL;
- dump.nData = dump.nCapacity = 0;
- return SQLITE_OK;
}
-/* Implements dump_doclist() for use in inspecting the fts3 index from
-** tests. TEXT result containing a string representation of the
-** doclist for the indicated term. dump_doclist(t, term, level, idx)
-** dumps the doclist for term from the segment specified by level, idx
-** (in %_segdir), while dump_doclist(t, term) dumps the logical
-** doclist for the term across all segments. The per-segment doclist
-** can contain deletions, while the full-index doclist will not
-** (deletions are omitted).
-**
-** Result formats differ with the setting of DL_DEFAULTS. Examples:
-**
-** DL_DOCIDS: [1] [3] [7]
-** DL_POSITIONS: [1 0[0 4] 1[17]] [3 1[5]]
-** DL_POSITIONS_OFFSETS: [1 0[0,0,3 4,23,26] 1[17,102,105]] [3 1[5,20,23]]
-**
-** In each case the number after the outer '[' is the docid. In the
-** latter two cases, the number before the inner '[' is the column
-** associated with the values within. For DL_POSITIONS the numbers
-** within are the positions, for DL_POSITIONS_OFFSETS they are the
-** position, the start offset, and the end offset.
+/*
+** Implementation of the matchinfo() function for FTS3
*/
-static void dumpDoclistFunc(
- sqlite3_context *pContext,
- int argc, sqlite3_value **argv
+static void fts3MatchinfoFunc(
+ sqlite3_context *pContext, /* SQLite function call context */
+ int nVal, /* Size of argument array */
+ sqlite3_value **apVal /* Array of arguments */
){
- fulltext_cursor *pCursor;
- if( argc!=2 && argc!=4 ){
- generateError(pContext, "dump_doclist", "incorrect arguments");
- }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
- sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
- generateError(pContext, "dump_doclist", "illegal first argument");
- }else if( sqlite3_value_text(argv[1])==NULL ||
- sqlite3_value_text(argv[1])[0]=='\0' ){
- generateError(pContext, "dump_doclist", "empty second argument");
- }else{
- const char *pTerm = (const char *)sqlite3_value_text(argv[1]);
- const int nTerm = strlen(pTerm);
- fulltext_vtab *v;
- int rc;
- DataBuffer doclist;
-
- memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
- v = cursor_vtab(pCursor);
-
- dataBufferInit(&doclist, 0);
-
- /* termSelect() yields the same logical doclist that queries are
- ** run against.
- */
- if( argc==2 ){
- rc = termSelect(v, v->nColumn, pTerm, nTerm, 0, DL_DEFAULT, &doclist);
- }else{
- sqlite3_stmt *s = NULL;
-
- /* Get our specific segment's information. */
- rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s);
- if( rc==SQLITE_OK ){
- rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[2]));
- if( rc==SQLITE_OK ){
- rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[3]));
- }
- }
-
- if( rc==SQLITE_OK ){
- rc = sqlite3_step(s);
-
- if( rc==SQLITE_DONE ){
- dataBufferDestroy(&doclist);
- generateError(pContext, "dump_doclist", "segment not found");
- return;
- }
-
- /* Found a segment, load it into doclist. */
- if( rc==SQLITE_ROW ){
- const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1);
- const char *pData = sqlite3_column_blob(s, 2);
- const int nData = sqlite3_column_bytes(s, 2);
-
- /* loadSegment() is used by termSelect() to load each
- ** segment's data.
- */
- rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, 0,
- &doclist);
- if( rc==SQLITE_OK ){
- rc = sqlite3_step(s);
-
- /* Should not have more than one matching segment. */
- if( rc!=SQLITE_DONE ){
- sqlite3_reset(s);
- dataBufferDestroy(&doclist);
- generateError(pContext, "dump_doclist", "invalid segdir");
- return;
- }
- rc = SQLITE_OK;
- }
- }
- }
-
- sqlite3_reset(s);
- }
+ Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
- if( rc==SQLITE_OK ){
- if( doclist.nData>0 ){
- createDoclistResult(pContext, doclist.pData, doclist.nData);
- }else{
- /* TODO(shess): This can happen if the term is not present, or
- ** if all instances of the term have been deleted and this is
- ** an all-index dump. It may be interesting to distinguish
- ** these cases.
- */
- sqlite3_result_text(pContext, "", 0, SQLITE_STATIC);
- }
- }else if( rc==SQLITE_NOMEM ){
- /* Handle out-of-memory cases specially because if they are
- ** generated in fts3 code they may not be reflected in the db
- ** handle.
- */
- /* TODO(shess): Handle this more comprehensively.
- ** sqlite3ErrStr() has what I need, but is internal.
- */
- generateError(pContext, "dump_doclist", "out of memory");
- }else{
- generateError(pContext, "dump_doclist", NULL);
- }
+ if( nVal!=1 ){
+ sqlite3_result_error(pContext,
+ "wrong number of arguments to function matchinfo()", -1);
+ return;
+ }
- dataBufferDestroy(&doclist);
+ if( SQLITE_OK==fts3FunctionArg(pContext, "matchinfo", apVal[0], &pCsr) ){
+ sqlite3Fts3Matchinfo(pContext, pCsr);
}
}
-#endif
/*
** This routine implements the xFindFunction method for the FTS3
** virtual table.
*/
-static int fulltextFindFunction(
- sqlite3_vtab *pVtab,
- int nArg,
- const char *zName,
- void (**pxFunc)(sqlite3_context*,int,sqlite3_value**),
- void **ppArg
+static int fts3FindFunctionMethod(
+ sqlite3_vtab *pVtab, /* Virtual table handle */
+ int nArg, /* Number of SQL function arguments */
+ const char *zName, /* Name of SQL function */
+ void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), /* OUT: Result */
+ void **ppArg /* Unused */
){
- if( strcmp(zName,"snippet")==0 ){
- *pxFunc = snippetFunc;
- return 1;
- }else if( strcmp(zName,"offsets")==0 ){
- *pxFunc = snippetOffsetsFunc;
- return 1;
- }else if( strcmp(zName,"optimize")==0 ){
- *pxFunc = optimizeFunc;
- return 1;
-#ifdef SQLITE_TEST
- /* NOTE(shess): These functions are present only for testing
- ** purposes. No particular effort is made to optimize their
- ** execution or how they build their results.
- */
- }else if( strcmp(zName,"dump_terms")==0 ){
- /* fprintf(stderr, "Found dump_terms\n"); */
- *pxFunc = dumpTermsFunc;
- return 1;
- }else if( strcmp(zName,"dump_doclist")==0 ){
- /* fprintf(stderr, "Found dump_doclist\n"); */
- *pxFunc = dumpDoclistFunc;
- return 1;
-#endif
+ struct Overloaded {
+ const char *zName;
+ void (*xFunc)(sqlite3_context*,int,sqlite3_value**);
+ } aOverload[] = {
+ { "snippet", fts3SnippetFunc },
+ { "offsets", fts3OffsetsFunc },
+ { "optimize", fts3OptimizeFunc },
+ { "matchinfo", fts3MatchinfoFunc },
+ };
+ int i; /* Iterator variable */
+
+ UNUSED_PARAMETER(pVtab);
+ UNUSED_PARAMETER(nArg);
+ UNUSED_PARAMETER(ppArg);
+
+ for(i=0; i<SizeofArray(aOverload); i++){
+ if( strcmp(zName, aOverload[i].zName)==0 ){
+ *pxFunc = aOverload[i].xFunc;
+ return 1;
+ }
}
+
+ /* No function of the specified name was found. Return 0. */
return 0;
}
/*
-** Rename an fts3 table.
+** Implementation of FTS3 xRename method. Rename an fts3 table.
*/
-static int fulltextRename(
- sqlite3_vtab *pVtab,
- const char *zName
+static int fts3RenameMethod(
+ sqlite3_vtab *pVtab, /* Virtual table handle */
+ const char *zName /* New name of table */
){
- fulltext_vtab *p = (fulltext_vtab *)pVtab;
- int rc = SQLITE_NOMEM;
- char *zSql = sqlite3_mprintf(
- "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';"
- "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';"
- "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';"
- , p->zDb, p->zName, zName
- , p->zDb, p->zName, zName
- , p->zDb, p->zName, zName
+ Fts3Table *p = (Fts3Table *)pVtab;
+ sqlite3 *db = p->db; /* Database connection */
+ int rc; /* Return Code */
+
+ rc = sqlite3Fts3PendingTermsFlush(p);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+
+ fts3DbExec(&rc, db,
+ "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';",
+ p->zDb, p->zName, zName
);
- if( zSql ){
- rc = sqlite3_exec(p->db, zSql, 0, 0, 0);
- sqlite3_free(zSql);
+ if( rc==SQLITE_ERROR ) rc = SQLITE_OK;
+ if( p->bHasDocsize ){
+ fts3DbExec(&rc, db,
+ "ALTER TABLE %Q.'%q_docsize' RENAME TO '%q_docsize';",
+ p->zDb, p->zName, zName
+ );
+ fts3DbExec(&rc, db,
+ "ALTER TABLE %Q.'%q_stat' RENAME TO '%q_stat';",
+ p->zDb, p->zName, zName
+ );
}
+ fts3DbExec(&rc, db,
+ "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';",
+ p->zDb, p->zName, zName
+ );
+ fts3DbExec(&rc, db,
+ "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';",
+ p->zDb, p->zName, zName
+ );
return rc;
}
static const sqlite3_module fts3Module = {
/* iVersion */ 0,
- /* xCreate */ fulltextCreate,
- /* xConnect */ fulltextConnect,
- /* xBestIndex */ fulltextBestIndex,
- /* xDisconnect */ fulltextDisconnect,
- /* xDestroy */ fulltextDestroy,
- /* xOpen */ fulltextOpen,
+ /* xCreate */ fts3CreateMethod,
+ /* xConnect */ fts3ConnectMethod,
+ /* xBestIndex */ fts3BestIndexMethod,
+ /* xDisconnect */ fts3DisconnectMethod,
+ /* xDestroy */ fts3DestroyMethod,
+ /* xOpen */ fts3OpenMethod,
/* xClose */ fulltextClose,
- /* xFilter */ fulltextFilter,
- /* xNext */ fulltextNext,
- /* xEof */ fulltextEof,
- /* xColumn */ fulltextColumn,
- /* xRowid */ fulltextRowid,
- /* xUpdate */ fulltextUpdate,
- /* xBegin */ fulltextBegin,
- /* xSync */ fulltextSync,
- /* xCommit */ fulltextCommit,
- /* xRollback */ fulltextRollback,
- /* xFindFunction */ fulltextFindFunction,
- /* xRename */ fulltextRename,
+ /* xFilter */ fts3FilterMethod,
+ /* xNext */ fts3NextMethod,
+ /* xEof */ fts3EofMethod,
+ /* xColumn */ fts3ColumnMethod,
+ /* xRowid */ fts3RowidMethod,
+ /* xUpdate */ fts3UpdateMethod,
+ /* xBegin */ fts3BeginMethod,
+ /* xSync */ fts3SyncMethod,
+ /* xCommit */ fts3CommitMethod,
+ /* xRollback */ fts3RollbackMethod,
+ /* xFindFunction */ fts3FindFunctionMethod,
+ /* xRename */ fts3RenameMethod,
};
+/*
+** This function is registered as the module destructor (called when an
+** FTS3 enabled database connection is closed). It frees the memory
+** allocated for the tokenizer hash table.
+*/
static void hashDestroy(void *p){
- fts3Hash *pHash = (fts3Hash *)p;
+ Fts3Hash *pHash = (Fts3Hash *)p;
sqlite3Fts3HashClear(pHash);
sqlite3_free(pHash);
}
@@ -7318,7 +2669,7 @@ static void hashDestroy(void *p){
** used to retrieve the respective implementations.
**
** Calling sqlite3Fts3SimpleTokenizerModule() sets the value pointed
-** to by the argument to point a the "simple" tokenizer implementation.
+** to by the argument to point to the "simple" tokenizer implementation.
** Function ...PorterTokenizerModule() sets *pModule to point to the
** porter tokenizer/stemmer implementation.
*/
@@ -7326,8 +2677,6 @@ void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule);
void sqlite3Fts3PorterTokenizerModule(sqlite3_tokenizer_module const**ppModule);
void sqlite3Fts3IcuTokenizerModule(sqlite3_tokenizer_module const**ppModule);
-int sqlite3Fts3InitHashTable(sqlite3 *, fts3Hash *, const char *);
-
/*
** Initialise the fts3 extension. If this extension is built as part
** of the sqlite library, then this function is called directly by
@@ -7336,19 +2685,20 @@ int sqlite3Fts3InitHashTable(sqlite3 *, fts3Hash *, const char *);
*/
int sqlite3Fts3Init(sqlite3 *db){
int rc = SQLITE_OK;
- fts3Hash *pHash = 0;
+ Fts3Hash *pHash = 0;
const sqlite3_tokenizer_module *pSimple = 0;
const sqlite3_tokenizer_module *pPorter = 0;
- const sqlite3_tokenizer_module *pIcu = 0;
- sqlite3Fts3SimpleTokenizerModule(&pSimple);
- sqlite3Fts3PorterTokenizerModule(&pPorter);
#ifdef SQLITE_ENABLE_ICU
+ const sqlite3_tokenizer_module *pIcu = 0;
sqlite3Fts3IcuTokenizerModule(&pIcu);
#endif
+ sqlite3Fts3SimpleTokenizerModule(&pSimple);
+ sqlite3Fts3PorterTokenizerModule(&pPorter);
+
/* Allocate and initialise the hash-table used to store tokenizers. */
- pHash = sqlite3_malloc(sizeof(fts3Hash));
+ pHash = sqlite3_malloc(sizeof(Fts3Hash));
if( !pHash ){
rc = SQLITE_NOMEM;
}else{
@@ -7359,14 +2709,18 @@ int sqlite3Fts3Init(sqlite3 *db){
if( rc==SQLITE_OK ){
if( sqlite3Fts3HashInsert(pHash, "simple", 7, (void *)pSimple)
|| sqlite3Fts3HashInsert(pHash, "porter", 7, (void *)pPorter)
+#ifdef SQLITE_ENABLE_ICU
|| (pIcu && sqlite3Fts3HashInsert(pHash, "icu", 4, (void *)pIcu))
+#endif
){
rc = SQLITE_NOMEM;
}
}
#ifdef SQLITE_TEST
- sqlite3Fts3ExprInitTestInterface(db);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3Fts3ExprInitTestInterface(db);
+ }
#endif
/* Create the virtual table wrapper around the hash-table and overload
@@ -7374,22 +2728,21 @@ int sqlite3Fts3Init(sqlite3 *db){
** module with sqlite.
*/
if( SQLITE_OK==rc
-#if CHROMIUM_FTS3_CHANGES && !SQLITE_TEST
- /* fts3_tokenizer() disabled for security reasons. */
-#else
&& SQLITE_OK==(rc = sqlite3Fts3InitHashTable(db, pHash, "fts3_tokenizer"))
-#endif
&& SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1))
- && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", -1))
- && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", -1))
-#ifdef SQLITE_TEST
- && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_terms", -1))
- && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_doclist", -1))
-#endif
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", 1))
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", -1))
+ && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", 1))
){
- return sqlite3_create_module_v2(
+ rc = sqlite3_create_module_v2(
db, "fts3", &fts3Module, (void *)pHash, hashDestroy
);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_create_module_v2(
+ db, "fts4", &fts3Module, (void *)pHash, 0
+ );
+ }
+ return rc;
}
/* An error has occurred. Delete the hash table and return the error code. */
@@ -7412,4 +2765,4 @@ int sqlite3_extension_init(
}
#endif
-#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */
+#endif
diff --git a/third_party/sqlite/src/ext/fts3/fts3Int.h b/third_party/sqlite/src/ext/fts3/fts3Int.h
new file mode 100644
index 0000000..8ed31aed
--- /dev/null
+++ b/third_party/sqlite/src/ext/fts3/fts3Int.h
@@ -0,0 +1,326 @@
+/*
+** 2009 Nov 12
+**
+** The author disclaims copyright to this source code. In place of
+** a legal notice, here is a blessing:
+**
+** May you do good and not evil.
+** May you find forgiveness for yourself and forgive others.
+** May you share freely, never taking more than you give.
+**
+******************************************************************************
+**
+*/
+
+#ifndef _FTSINT_H
+#define _FTSINT_H
+
+#if !defined(NDEBUG) && !defined(SQLITE_DEBUG)
+# define NDEBUG 1
+#endif
+
+#include "sqlite3.h"
+#include "fts3_tokenizer.h"
+#include "fts3_hash.h"
+
+/*
+** This constant controls how often segments are merged. Once there are
+** FTS3_MERGE_COUNT segments of level N, they are merged into a single
+** segment of level N+1.
+*/
+#define FTS3_MERGE_COUNT 16
+
+/*
+** This is the maximum amount of data (in bytes) to store in the
+** Fts3Table.pendingTerms hash table. Normally, the hash table is
+** populated as documents are inserted/updated/deleted in a transaction
+** and used to create a new segment when the transaction is committed.
+** However if this limit is reached midway through a transaction, a new
+** segment is created and the hash table cleared immediately.
+*/
+#define FTS3_MAX_PENDING_DATA (1*1024*1024)
+
+/*
+** Macro to return the number of elements in an array. SQLite has a
+** similar macro called ArraySize(). Use a different name to avoid
+** a collision when building an amalgamation with built-in FTS3.
+*/
+#define SizeofArray(X) ((int)(sizeof(X)/sizeof(X[0])))
+
+/*
+** Maximum length of a varint encoded integer. The varint format is different
+** from that used by SQLite, so the maximum length is 10, not 9.
+*/
+#define FTS3_VARINT_MAX 10
+
+/*
+** The testcase() macro is only used by the amalgamation. If undefined,
+** make it a no-op.
+*/
+#ifndef testcase
+# define testcase(X)
+#endif
+
+/*
+** Terminator values for position-lists and column-lists.
+*/
+#define POS_COLUMN (1) /* Column-list terminator */
+#define POS_END (0) /* Position-list terminator */
+
+/*
+** This section provides definitions to allow the
+** FTS3 extension to be compiled outside of the
+** amalgamation.
+*/
+#ifndef SQLITE_AMALGAMATION
+/*
+** Macros indicating that conditional expressions are always true or
+** false.
+*/
+# define ALWAYS(x) (x)
+# define NEVER(X) (x)
+/*
+** Internal types used by SQLite.
+*/
+typedef unsigned char u8; /* 1-byte (or larger) unsigned integer */
+typedef short int i16; /* 2-byte (or larger) signed integer */
+typedef unsigned int u32; /* 4-byte unsigned integer */
+typedef sqlite3_uint64 u64; /* 8-byte unsigned integer */
+/*
+** Macro used to suppress compiler warnings for unused parameters.
+*/
+#define UNUSED_PARAMETER(x) (void)(x)
+#endif
+
+typedef struct Fts3Table Fts3Table;
+typedef struct Fts3Cursor Fts3Cursor;
+typedef struct Fts3Expr Fts3Expr;
+typedef struct Fts3Phrase Fts3Phrase;
+typedef struct Fts3SegReader Fts3SegReader;
+typedef struct Fts3SegFilter Fts3SegFilter;
+
+/*
+** A connection to a fulltext index is an instance of the following
+** structure. The xCreate and xConnect methods create an instance
+** of this structure and xDestroy and xDisconnect free that instance.
+** All other methods receive a pointer to the structure as one of their
+** arguments.
+*/
+struct Fts3Table {
+ sqlite3_vtab base; /* Base class used by SQLite core */
+ sqlite3 *db; /* The database connection */
+ const char *zDb; /* logical database name */
+ const char *zName; /* virtual table name */
+ int nColumn; /* number of named columns in virtual table */
+ char **azColumn; /* column names. malloced */
+ sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */
+
+ /* Precompiled statements used by the implementation. Each of these
+ ** statements is run and reset within a single virtual table API call.
+ */
+ sqlite3_stmt *aStmt[25];
+
+ /* Pointer to string containing the SQL:
+ **
+ ** "SELECT block FROM %_segments WHERE blockid BETWEEN ? AND ?
+ ** ORDER BY blockid"
+ */
+ char *zSelectLeaves;
+ int nLeavesStmt; /* Valid statements in aLeavesStmt */
+ int nLeavesTotal; /* Total number of prepared leaves stmts */
+ int nLeavesAlloc; /* Allocated size of aLeavesStmt */
+ sqlite3_stmt **aLeavesStmt; /* Array of prepared zSelectLeaves stmts */
+
+ int nNodeSize; /* Soft limit for node size */
+ u8 bHasContent; /* True if %_content table exists */
+ u8 bHasDocsize; /* True if %_docsize table exists */
+
+ /* The following hash table is used to buffer pending index updates during
+ ** transactions. Variable nPendingData estimates the memory size of the
+ ** pending data, including hash table overhead, but not malloc overhead.
+ ** When nPendingData exceeds nMaxPendingData, the buffer is flushed
+ ** automatically. Variable iPrevDocid is the docid of the most recently
+ ** inserted record.
+ */
+ int nMaxPendingData;
+ int nPendingData;
+ sqlite_int64 iPrevDocid;
+ Fts3Hash pendingTerms;
+};
+
+/*
+** When the core wants to read from the virtual table, it creates a
+** virtual table cursor (an instance of the following structure) using
+** the xOpen method. Cursors are destroyed using the xClose method.
+*/
+struct Fts3Cursor {
+ sqlite3_vtab_cursor base; /* Base class used by SQLite core */
+ i16 eSearch; /* Search strategy (see below) */
+ u8 isEof; /* True if at End Of Results */
+ u8 isRequireSeek; /* True if must seek pStmt to %_content row */
+ sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */
+ Fts3Expr *pExpr; /* Parsed MATCH query string */
+ sqlite3_int64 iPrevId; /* Previous id read from aDoclist */
+ char *pNextId; /* Pointer into the body of aDoclist */
+ char *aDoclist; /* List of docids for full-text queries */
+ int nDoclist; /* Size of buffer at aDoclist */
+ int isMatchinfoNeeded; /* True when aMatchinfo[] needs filling in */
+ u32 *aMatchinfo; /* Information about most recent match */
+};
+
+/*
+** The Fts3Cursor.eSearch member is always set to one of the following.
+** Actualy, Fts3Cursor.eSearch can be greater than or equal to
+** FTS3_FULLTEXT_SEARCH. If so, then Fts3Cursor.eSearch - 2 is the index
+** of the column to be searched. For example, in
+**
+** CREATE VIRTUAL TABLE ex1 USING fts3(a,b,c,d);
+** SELECT docid FROM ex1 WHERE b MATCH 'one two three';
+**
+** Because the LHS of the MATCH operator is 2nd column "b",
+** Fts3Cursor.eSearch will be set to FTS3_FULLTEXT_SEARCH+1. (+0 for a,
+** +1 for b, +2 for c, +3 for d.) If the LHS of MATCH were "ex1"
+** indicating that all columns should be searched,
+** then eSearch would be set to FTS3_FULLTEXT_SEARCH+4.
+*/
+#define FTS3_FULLSCAN_SEARCH 0 /* Linear scan of %_content table */
+#define FTS3_DOCID_SEARCH 1 /* Lookup by rowid on %_content table */
+#define FTS3_FULLTEXT_SEARCH 2 /* Full-text index search */
+
+/*
+** A "phrase" is a sequence of one or more tokens that must match in
+** sequence. A single token is the base case and the most common case.
+** For a sequence of tokens contained in "...", nToken will be the number
+** of tokens in the string.
+*/
+struct Fts3Phrase {
+ int nToken; /* Number of tokens in the phrase */
+ int iColumn; /* Index of column this phrase must match */
+ int isNot; /* Phrase prefixed by unary not (-) operator */
+ struct PhraseToken {
+ char *z; /* Text of the token */
+ int n; /* Number of bytes in buffer pointed to by z */
+ int isPrefix; /* True if token ends in with a "*" character */
+ } aToken[1]; /* One entry for each token in the phrase */
+};
+
+/*
+** A tree of these objects forms the RHS of a MATCH operator.
+**
+** If Fts3Expr.eType is either FTSQUERY_NEAR or FTSQUERY_PHRASE and isLoaded
+** is true, then aDoclist points to a malloced buffer, size nDoclist bytes,
+** containing the results of the NEAR or phrase query in FTS3 doclist
+** format. As usual, the initial "Length" field found in doclists stored
+** on disk is omitted from this buffer.
+**
+** Variable pCurrent always points to the start of a docid field within
+** aDoclist. Since the doclist is usually scanned in docid order, this can
+** be used to accelerate seeking to the required docid within the doclist.
+*/
+struct Fts3Expr {
+ int eType; /* One of the FTSQUERY_XXX values defined below */
+ int nNear; /* Valid if eType==FTSQUERY_NEAR */
+ Fts3Expr *pParent; /* pParent->pLeft==this or pParent->pRight==this */
+ Fts3Expr *pLeft; /* Left operand */
+ Fts3Expr *pRight; /* Right operand */
+ Fts3Phrase *pPhrase; /* Valid if eType==FTSQUERY_PHRASE */
+
+ int isLoaded; /* True if aDoclist/nDoclist are initialized. */
+ char *aDoclist; /* Buffer containing doclist */
+ int nDoclist; /* Size of aDoclist in bytes */
+
+ sqlite3_int64 iCurrent;
+ char *pCurrent;
+};
+
+/*
+** Candidate values for Fts3Query.eType. Note that the order of the first
+** four values is in order of precedence when parsing expressions. For
+** example, the following:
+**
+** "a OR b AND c NOT d NEAR e"
+**
+** is equivalent to:
+**
+** "a OR (b AND (c NOT (d NEAR e)))"
+*/
+#define FTSQUERY_NEAR 1
+#define FTSQUERY_NOT 2
+#define FTSQUERY_AND 3
+#define FTSQUERY_OR 4
+#define FTSQUERY_PHRASE 5
+
+
+/* fts3_init.c */
+int sqlite3Fts3DeleteVtab(int, sqlite3_vtab *);
+int sqlite3Fts3InitVtab(int, sqlite3*, void*, int, const char*const*,
+ sqlite3_vtab **, char **);
+
+/* fts3_write.c */
+int sqlite3Fts3UpdateMethod(sqlite3_vtab*,int,sqlite3_value**,sqlite3_int64*);
+int sqlite3Fts3PendingTermsFlush(Fts3Table *);
+void sqlite3Fts3PendingTermsClear(Fts3Table *);
+int sqlite3Fts3Optimize(Fts3Table *);
+int sqlite3Fts3SegReaderNew(Fts3Table *,int, sqlite3_int64,
+ sqlite3_int64, sqlite3_int64, const char *, int, Fts3SegReader**);
+int sqlite3Fts3SegReaderPending(Fts3Table*,const char*,int,int,Fts3SegReader**);
+void sqlite3Fts3SegReaderFree(Fts3Table *, Fts3SegReader *);
+int sqlite3Fts3SegReaderIterate(
+ Fts3Table *, Fts3SegReader **, int, Fts3SegFilter *,
+ int (*)(Fts3Table *, void *, char *, int, char *, int), void *
+);
+int sqlite3Fts3ReadBlock(Fts3Table*, sqlite3_int64, char const**, int*);
+int sqlite3Fts3AllSegdirs(Fts3Table*, sqlite3_stmt **);
+int sqlite3Fts3MatchinfoDocsizeLocal(Fts3Cursor*, u32*);
+int sqlite3Fts3MatchinfoDocsizeGlobal(Fts3Cursor*, u32*);
+int sqlite3Fts3ReadLock(Fts3Table *);
+
+/* Flags allowed as part of the 4th argument to SegmentReaderIterate() */
+#define FTS3_SEGMENT_REQUIRE_POS 0x00000001
+#define FTS3_SEGMENT_IGNORE_EMPTY 0x00000002
+#define FTS3_SEGMENT_COLUMN_FILTER 0x00000004
+#define FTS3_SEGMENT_PREFIX 0x00000008
+
+/* Type passed as 4th argument to SegmentReaderIterate() */
+struct Fts3SegFilter {
+ const char *zTerm;
+ int nTerm;
+ int iCol;
+ int flags;
+};
+
+/* fts3.c */
+int sqlite3Fts3PutVarint(char *, sqlite3_int64);
+int sqlite3Fts3GetVarint(const char *, sqlite_int64 *);
+int sqlite3Fts3GetVarint32(const char *, int *);
+int sqlite3Fts3VarintLen(sqlite3_uint64);
+void sqlite3Fts3Dequote(char *);
+
+char *sqlite3Fts3FindPositions(Fts3Expr *, sqlite3_int64, int);
+int sqlite3Fts3ExprLoadDoclist(Fts3Table *, Fts3Expr *);
+int sqlite3Fts3ExprNearTrim(Fts3Expr *, Fts3Expr *, int);
+
+/* fts3_tokenizer.c */
+const char *sqlite3Fts3NextToken(const char *, int *);
+int sqlite3Fts3InitHashTable(sqlite3 *, Fts3Hash *, const char *);
+int sqlite3Fts3InitTokenizer(Fts3Hash *pHash,
+ const char *, sqlite3_tokenizer **, const char **, char **
+);
+
+/* fts3_snippet.c */
+void sqlite3Fts3Offsets(sqlite3_context*, Fts3Cursor*);
+void sqlite3Fts3Snippet(sqlite3_context *, Fts3Cursor *, const char *,
+ const char *, const char *, int, int
+);
+void sqlite3Fts3Matchinfo(sqlite3_context *, Fts3Cursor *);
+
+/* fts3_expr.c */
+int sqlite3Fts3ExprParse(sqlite3_tokenizer *,
+ char **, int, int, const char *, int, Fts3Expr **
+);
+void sqlite3Fts3ExprFree(Fts3Expr *);
+#ifdef SQLITE_TEST
+int sqlite3Fts3ExprInitTestInterface(sqlite3 *db);
+#endif
+
+#endif /* _FTSINT_H */
diff --git a/third_party/sqlite/src/ext/fts3/fts3_expr.c b/third_party/sqlite/src/ext/fts3/fts3_expr.c
index bfca3e1..008ba81 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_expr.c
+++ b/third_party/sqlite/src/ext/fts3/fts3_expr.c
@@ -13,8 +13,7 @@
** This module contains code that implements a parser for fts3 query strings
** (the right-hand argument to the MATCH operator). Because the supported
** syntax is relatively simple, the whole tokenizer/parser system is
-** hand-coded. The public interface to this module is declared in source
-** code file "fts3_expr.h".
+** hand-coded.
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
@@ -40,7 +39,29 @@
** to zero causes the module to use the old syntax. If it is set to
** non-zero the new syntax is activated. This is so both syntaxes can
** be tested using a single build of testfixture.
+**
+** The following describes the syntax supported by the fts3 MATCH
+** operator in a similar format to that used by the lemon parser
+** generator. This module does not use actually lemon, it uses a
+** custom parser.
+**
+** query ::= andexpr (OR andexpr)*.
+**
+** andexpr ::= notexpr (AND? notexpr)*.
+**
+** notexpr ::= nearexpr (NOT nearexpr|-TOKEN)*.
+** notexpr ::= LP query RP.
+**
+** nearexpr ::= phrase (NEAR distance_opt nearexpr)*.
+**
+** distance_opt ::= .
+** distance_opt ::= / INTEGER.
+**
+** phrase ::= TOKEN.
+** phrase ::= COLUMN:TOKEN.
+** phrase ::= "TOKEN TOKEN TOKEN...".
*/
+
#ifdef SQLITE_TEST
int sqlite3_fts3_enable_parentheses = 0;
#else
@@ -56,8 +77,7 @@ int sqlite3_fts3_enable_parentheses = 0;
*/
#define SQLITE_FTS3_DEFAULT_NEAR_PARAM 10
-#include "fts3_expr.h"
-#include "sqlite3.h"
+#include "fts3Int.h"
#include <string.h>
#include <assert.h>
@@ -160,7 +180,7 @@ static int getNextToken(
** Enlarge a memory allocation. If an out-of-memory allocation occurs,
** then free the old allocation.
*/
-void *fts3ReallocOrFree(void *pOrig, int nNew){
+static void *fts3ReallocOrFree(void *pOrig, int nNew){
void *pRet = sqlite3_realloc(pOrig, nNew);
if( !pRet ){
sqlite3_free(pOrig);
@@ -231,7 +251,7 @@ static int getNextString(
if( rc==SQLITE_DONE ){
int jj;
- char *zNew;
+ char *zNew = NULL;
int nNew = 0;
int nByte = sizeof(Fts3Expr) + sizeof(Fts3Phrase);
nByte += (p?(p->pPhrase->nToken-1):0) * sizeof(struct PhraseToken);
@@ -290,7 +310,7 @@ static int getNextNode(
int *pnConsumed /* OUT: Number of bytes consumed */
){
static const struct Fts3Keyword {
- char z[4]; /* Keyword text */
+ char *z; /* Keyword text */
unsigned char n; /* Length of the keyword */
unsigned char parenOnly; /* Only valid in paren mode */
unsigned char eType; /* Keyword code */
@@ -353,11 +373,14 @@ static int getNextNode(
|| cNext=='"' || cNext=='(' || cNext==')' || cNext==0
){
pRet = (Fts3Expr *)sqlite3_malloc(sizeof(Fts3Expr));
+ if( !pRet ){
+ return SQLITE_NOMEM;
+ }
memset(pRet, 0, sizeof(Fts3Expr));
pRet->eType = pKey->eType;
pRet->nNear = nNear;
*ppExpr = pRet;
- *pnConsumed = (zInput - z) + nKey;
+ *pnConsumed = (int)((zInput - z) + nKey);
return SQLITE_OK;
}
@@ -377,14 +400,14 @@ static int getNextNode(
if( rc==SQLITE_OK && !*ppExpr ){
rc = SQLITE_DONE;
}
- *pnConsumed = (zInput - z) + 1 + nConsumed;
+ *pnConsumed = (int)((zInput - z) + 1 + nConsumed);
return rc;
}
/* Check for a close bracket. */
if( *zInput==')' ){
pParse->nNest--;
- *pnConsumed = (zInput - z) + 1;
+ *pnConsumed = (int)((zInput - z) + 1);
return SQLITE_DONE;
}
}
@@ -396,7 +419,7 @@ static int getNextNode(
*/
if( *zInput=='"' ){
for(ii=1; ii<nInput && zInput[ii]!='"'; ii++);
- *pnConsumed = (zInput - z) + ii + 1;
+ *pnConsumed = (int)((zInput - z) + ii + 1);
if( ii==nInput ){
return SQLITE_ERROR;
}
@@ -419,12 +442,12 @@ static int getNextNode(
iColLen = 0;
for(ii=0; ii<pParse->nCol; ii++){
const char *zStr = pParse->azCol[ii];
- int nStr = strlen(zStr);
+ int nStr = (int)strlen(zStr);
if( nInput>nStr && zInput[nStr]==':'
&& sqlite3_strnicmp(zStr, zInput, nStr)==0
){
iCol = ii;
- iColLen = ((zInput - z) + nStr + 1);
+ iColLen = (int)((zInput - z) + nStr + 1);
break;
}
}
@@ -690,7 +713,7 @@ int sqlite3Fts3ExprParse(
return SQLITE_OK;
}
if( n<0 ){
- n = strlen(z);
+ n = (int)strlen(z);
}
rc = fts3ExprParse(&sParse, z, n, ppExpr, &nParsed);
@@ -711,6 +734,7 @@ void sqlite3Fts3ExprFree(Fts3Expr *p){
if( p ){
sqlite3Fts3ExprFree(p->pLeft);
sqlite3Fts3ExprFree(p->pRight);
+ sqlite3_free(p->aDoclist);
sqlite3_free(p);
}
}
@@ -745,7 +769,7 @@ static int queryTestTokenizer(
sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC);
if( SQLITE_ROW==sqlite3_step(pStmt) ){
if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){
- memcpy(pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp));
+ memcpy((void *)pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp));
}
}
@@ -889,8 +913,8 @@ exprtest_out:
** Register the query expression parser test function fts3_exprtest()
** with database connection db.
*/
-void sqlite3Fts3ExprInitTestInterface(sqlite3* db){
- sqlite3_create_function(
+int sqlite3Fts3ExprInitTestInterface(sqlite3* db){
+ return sqlite3_create_function(
db, "fts3_exprtest", -1, SQLITE_UTF8, 0, fts3ExprTest, 0, 0
);
}
diff --git a/third_party/sqlite/src/ext/fts3/fts3_expr.h b/third_party/sqlite/src/ext/fts3/fts3_expr.h
deleted file mode 100644
index a48dee6..0000000
--- a/third_party/sqlite/src/ext/fts3/fts3_expr.h
+++ /dev/null
@@ -1,96 +0,0 @@
-/*
-** 2008 Nov 28
-**
-** The author disclaims copyright to this source code. In place of
-** a legal notice, here is a blessing:
-**
-** May you do good and not evil.
-** May you find forgiveness for yourself and forgive others.
-** May you share freely, never taking more than you give.
-**
-******************************************************************************
-**
-*/
-
-#include "fts3_tokenizer.h"
-#include "sqlite3.h"
-
-/*
-** The following describes the syntax supported by the fts3 MATCH
-** operator in a similar format to that used by the lemon parser
-** generator. This module does not use actually lemon, it uses a
-** custom parser.
-**
-** query ::= andexpr (OR andexpr)*.
-**
-** andexpr ::= notexpr (AND? notexpr)*.
-**
-** notexpr ::= nearexpr (NOT nearexpr|-TOKEN)*.
-** notexpr ::= LP query RP.
-**
-** nearexpr ::= phrase (NEAR distance_opt nearexpr)*.
-**
-** distance_opt ::= .
-** distance_opt ::= / INTEGER.
-**
-** phrase ::= TOKEN.
-** phrase ::= COLUMN:TOKEN.
-** phrase ::= "TOKEN TOKEN TOKEN...".
-*/
-
-typedef struct Fts3Expr Fts3Expr;
-typedef struct Fts3Phrase Fts3Phrase;
-
-/*
-** A "phrase" is a sequence of one or more tokens that must match in
-** sequence. A single token is the base case and the most common case.
-** For a sequence of tokens contained in "...", nToken will be the number
-** of tokens in the string.
-*/
-struct Fts3Phrase {
- int nToken; /* Number of tokens in the phrase */
- int iColumn; /* Index of column this phrase must match */
- int isNot; /* Phrase prefixed by unary not (-) operator */
- struct PhraseToken {
- char *z; /* Text of the token */
- int n; /* Number of bytes in buffer pointed to by z */
- int isPrefix; /* True if token ends in with a "*" character */
- } aToken[1]; /* One entry for each token in the phrase */
-};
-
-/*
-** A tree of these objects forms the RHS of a MATCH operator.
-*/
-struct Fts3Expr {
- int eType; /* One of the FTSQUERY_XXX values defined below */
- int nNear; /* Valid if eType==FTSQUERY_NEAR */
- Fts3Expr *pParent; /* pParent->pLeft==this or pParent->pRight==this */
- Fts3Expr *pLeft; /* Left operand */
- Fts3Expr *pRight; /* Right operand */
- Fts3Phrase *pPhrase; /* Valid if eType==FTSQUERY_PHRASE */
-};
-
-int sqlite3Fts3ExprParse(sqlite3_tokenizer *, char **, int, int,
- const char *, int, Fts3Expr **);
-void sqlite3Fts3ExprFree(Fts3Expr *);
-
-/*
-** Candidate values for Fts3Query.eType. Note that the order of the first
-** four values is in order of precedence when parsing expressions. For
-** example, the following:
-**
-** "a OR b AND c NOT d NEAR e"
-**
-** is equivalent to:
-**
-** "a OR (b AND (c NOT (d NEAR e)))"
-*/
-#define FTSQUERY_NEAR 1
-#define FTSQUERY_NOT 2
-#define FTSQUERY_AND 3
-#define FTSQUERY_OR 4
-#define FTSQUERY_PHRASE 5
-
-#ifdef SQLITE_TEST
-void sqlite3Fts3ExprInitTestInterface(sqlite3 *db);
-#endif
diff --git a/third_party/sqlite/src/ext/fts3/fts3_hash.c b/third_party/sqlite/src/ext/fts3/fts3_hash.c
index ee30117..98be529 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_hash.c
+++ b/third_party/sqlite/src/ext/fts3/fts3_hash.c
@@ -56,7 +56,7 @@ static void fts3HashFree(void *p){
** true if the hash table should make its own private copy of keys and
** false if it should just use the supplied pointer.
*/
-void sqlite3Fts3HashInit(fts3Hash *pNew, int keyClass, int copyKey){
+void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey){
assert( pNew!=0 );
assert( keyClass>=FTS3_HASH_STRING && keyClass<=FTS3_HASH_BINARY );
pNew->keyClass = keyClass;
@@ -71,8 +71,8 @@ void sqlite3Fts3HashInit(fts3Hash *pNew, int keyClass, int copyKey){
** Call this routine to delete a hash table or to reset a hash table
** to the empty state.
*/
-void sqlite3Fts3HashClear(fts3Hash *pH){
- fts3HashElem *elem; /* For looping over all elements of the table */
+void sqlite3Fts3HashClear(Fts3Hash *pH){
+ Fts3HashElem *elem; /* For looping over all elements of the table */
assert( pH!=0 );
elem = pH->first;
@@ -81,7 +81,7 @@ void sqlite3Fts3HashClear(fts3Hash *pH){
pH->ht = 0;
pH->htsize = 0;
while( elem ){
- fts3HashElem *next_elem = elem->next;
+ Fts3HashElem *next_elem = elem->next;
if( pH->copyKey && elem->pKey ){
fts3HashFree(elem->pKey);
}
@@ -164,11 +164,11 @@ static int (*ftsCompareFunction(int keyClass))(const void*,int,const void*,int){
/* Link an element into the hash table
*/
static void fts3HashInsertElement(
- fts3Hash *pH, /* The complete hash table */
+ Fts3Hash *pH, /* The complete hash table */
struct _fts3ht *pEntry, /* The entry into which pNew is inserted */
- fts3HashElem *pNew /* The element to be inserted */
+ Fts3HashElem *pNew /* The element to be inserted */
){
- fts3HashElem *pHead; /* First element already in pEntry */
+ Fts3HashElem *pHead; /* First element already in pEntry */
pHead = pEntry->chain;
if( pHead ){
pNew->next = pHead;
@@ -190,15 +190,17 @@ static void fts3HashInsertElement(
/* Resize the hash table so that it cantains "new_size" buckets.
** "new_size" must be a power of 2. The hash table might fail
** to resize if sqliteMalloc() fails.
+**
+** Return non-zero if a memory allocation error occurs.
*/
-static void fts3Rehash(fts3Hash *pH, int new_size){
+static int fts3Rehash(Fts3Hash *pH, int new_size){
struct _fts3ht *new_ht; /* The new hash table */
- fts3HashElem *elem, *next_elem; /* For looping over existing elements */
+ Fts3HashElem *elem, *next_elem; /* For looping over existing elements */
int (*xHash)(const void*,int); /* The hash function */
assert( (new_size & (new_size-1))==0 );
new_ht = (struct _fts3ht *)fts3HashMalloc( new_size*sizeof(struct _fts3ht) );
- if( new_ht==0 ) return;
+ if( new_ht==0 ) return 1;
fts3HashFree(pH->ht);
pH->ht = new_ht;
pH->htsize = new_size;
@@ -208,19 +210,20 @@ static void fts3Rehash(fts3Hash *pH, int new_size){
next_elem = elem->next;
fts3HashInsertElement(pH, &new_ht[h], elem);
}
+ return 0;
}
/* This function (for internal use only) locates an element in an
** hash table that matches the given key. The hash for this key has
** already been computed and is passed as the 4th parameter.
*/
-static fts3HashElem *fts3FindElementByHash(
- const fts3Hash *pH, /* The pH to be searched */
+static Fts3HashElem *fts3FindElementByHash(
+ const Fts3Hash *pH, /* The pH to be searched */
const void *pKey, /* The key we are searching for */
int nKey,
int h /* The hash for this key. */
){
- fts3HashElem *elem; /* Used to loop thru the element list */
+ Fts3HashElem *elem; /* Used to loop thru the element list */
int count; /* Number of elements left to test */
int (*xCompare)(const void*,int,const void*,int); /* comparison function */
@@ -243,8 +246,8 @@ static fts3HashElem *fts3FindElementByHash(
** element and a hash on the element's key.
*/
static void fts3RemoveElementByHash(
- fts3Hash *pH, /* The pH containing "elem" */
- fts3HashElem* elem, /* The element to be removed from the pH */
+ Fts3Hash *pH, /* The pH containing "elem" */
+ Fts3HashElem* elem, /* The element to be removed from the pH */
int h /* Hash value for the element */
){
struct _fts3ht *pEntry;
@@ -276,13 +279,12 @@ static void fts3RemoveElementByHash(
}
}
-/* Attempt to locate an element of the hash table pH with a key
-** that matches pKey,nKey. Return the data for this element if it is
-** found, or NULL if there is no match.
-*/
-void *sqlite3Fts3HashFind(const fts3Hash *pH, const void *pKey, int nKey){
- int h; /* A hash on key */
- fts3HashElem *elem; /* The element that matches key */
+Fts3HashElem *sqlite3Fts3HashFindElem(
+ const Fts3Hash *pH,
+ const void *pKey,
+ int nKey
+){
+ int h; /* A hash on key */
int (*xHash)(const void*,int); /* The hash function */
if( pH==0 || pH->ht==0 ) return 0;
@@ -290,8 +292,19 @@ void *sqlite3Fts3HashFind(const fts3Hash *pH, const void *pKey, int nKey){
assert( xHash!=0 );
h = (*xHash)(pKey,nKey);
assert( (pH->htsize & (pH->htsize-1))==0 );
- elem = fts3FindElementByHash(pH,pKey,nKey, h & (pH->htsize-1));
- return elem ? elem->data : 0;
+ return fts3FindElementByHash(pH,pKey,nKey, h & (pH->htsize-1));
+}
+
+/*
+** Attempt to locate an element of the hash table pH with a key
+** that matches pKey,nKey. Return the data for this element if it is
+** found, or NULL if there is no match.
+*/
+void *sqlite3Fts3HashFind(const Fts3Hash *pH, const void *pKey, int nKey){
+ Fts3HashElem *pElem; /* The element that matches key (if any) */
+
+ pElem = sqlite3Fts3HashFindElem(pH, pKey, nKey);
+ return pElem ? pElem->data : 0;
}
/* Insert an element into the hash table pH. The key is pKey,nKey
@@ -310,15 +323,15 @@ void *sqlite3Fts3HashFind(const fts3Hash *pH, const void *pKey, int nKey){
** element corresponding to "key" is removed from the hash table.
*/
void *sqlite3Fts3HashInsert(
- fts3Hash *pH, /* The hash table to insert into */
+ Fts3Hash *pH, /* The hash table to insert into */
const void *pKey, /* The key */
int nKey, /* Number of bytes in the key */
void *data /* The data */
){
int hraw; /* Raw hash value of the key */
int h; /* the hash of the key modulo hash table size */
- fts3HashElem *elem; /* Used to loop thru the element list */
- fts3HashElem *new_elem; /* New element added to the pH */
+ Fts3HashElem *elem; /* Used to loop thru the element list */
+ Fts3HashElem *new_elem; /* New element added to the pH */
int (*xHash)(const void*,int); /* The hash function */
assert( pH!=0 );
@@ -338,14 +351,14 @@ void *sqlite3Fts3HashInsert(
return old_data;
}
if( data==0 ) return 0;
- if( pH->htsize==0 ){
- fts3Rehash(pH,8);
- if( pH->htsize==0 ){
- pH->count = 0;
- return data;
- }
+ if( (pH->htsize==0 && fts3Rehash(pH,8))
+ || (pH->count>=pH->htsize && fts3Rehash(pH, pH->htsize*2))
+ ){
+ pH->count = 0;
+ return data;
}
- new_elem = (fts3HashElem*)fts3HashMalloc( sizeof(fts3HashElem) );
+ assert( pH->htsize>0 );
+ new_elem = (Fts3HashElem*)fts3HashMalloc( sizeof(Fts3HashElem) );
if( new_elem==0 ) return data;
if( pH->copyKey && pKey!=0 ){
new_elem->pKey = fts3HashMalloc( nKey );
@@ -359,9 +372,6 @@ void *sqlite3Fts3HashInsert(
}
new_elem->nKey = nKey;
pH->count++;
- if( pH->count > pH->htsize ){
- fts3Rehash(pH,pH->htsize*2);
- }
assert( pH->htsize>0 );
assert( (pH->htsize & (pH->htsize-1))==0 );
h = hraw & (pH->htsize-1);
diff --git a/third_party/sqlite/src/ext/fts3/fts3_hash.h b/third_party/sqlite/src/ext/fts3/fts3_hash.h
index e01954e..399f515 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_hash.h
+++ b/third_party/sqlite/src/ext/fts3/fts3_hash.h
@@ -18,8 +18,8 @@
#define _FTS3_HASH_H_
/* Forward declarations of structures. */
-typedef struct fts3Hash fts3Hash;
-typedef struct fts3HashElem fts3HashElem;
+typedef struct Fts3Hash Fts3Hash;
+typedef struct Fts3HashElem Fts3HashElem;
/* A complete hash table is an instance of the following structure.
** The internals of this structure are intended to be opaque -- client
@@ -29,15 +29,15 @@ typedef struct fts3HashElem fts3HashElem;
** accessing this structure are really macros, so we can't really make
** this structure opaque.
*/
-struct fts3Hash {
+struct Fts3Hash {
char keyClass; /* HASH_INT, _POINTER, _STRING, _BINARY */
char copyKey; /* True if copy of key made on insert */
int count; /* Number of entries in this table */
- fts3HashElem *first; /* The first element of the array */
+ Fts3HashElem *first; /* The first element of the array */
int htsize; /* Number of buckets in the hash table */
struct _fts3ht { /* the hash table */
int count; /* Number of entries with this hash */
- fts3HashElem *chain; /* Pointer to first entry with this hash */
+ Fts3HashElem *chain; /* Pointer to first entry with this hash */
} *ht;
};
@@ -47,8 +47,8 @@ struct fts3Hash {
** Again, this structure is intended to be opaque, but it can't really
** be opaque because it is used by macros.
*/
-struct fts3HashElem {
- fts3HashElem *next, *prev; /* Next and previous elements in the table */
+struct Fts3HashElem {
+ Fts3HashElem *next, *prev; /* Next and previous elements in the table */
void *data; /* Data associated with this element */
void *pKey; int nKey; /* Key associated with this element */
};
@@ -71,25 +71,27 @@ struct fts3HashElem {
/*
** Access routines. To delete, insert a NULL pointer.
*/
-void sqlite3Fts3HashInit(fts3Hash*, int keytype, int copyKey);
-void *sqlite3Fts3HashInsert(fts3Hash*, const void *pKey, int nKey, void *pData);
-void *sqlite3Fts3HashFind(const fts3Hash*, const void *pKey, int nKey);
-void sqlite3Fts3HashClear(fts3Hash*);
+void sqlite3Fts3HashInit(Fts3Hash *pNew, char keyClass, char copyKey);
+void *sqlite3Fts3HashInsert(Fts3Hash*, const void *pKey, int nKey, void *pData);
+void *sqlite3Fts3HashFind(const Fts3Hash*, const void *pKey, int nKey);
+void sqlite3Fts3HashClear(Fts3Hash*);
+Fts3HashElem *sqlite3Fts3HashFindElem(const Fts3Hash *, const void *, int);
/*
** Shorthand for the functions above
*/
-#define fts3HashInit sqlite3Fts3HashInit
-#define fts3HashInsert sqlite3Fts3HashInsert
-#define fts3HashFind sqlite3Fts3HashFind
-#define fts3HashClear sqlite3Fts3HashClear
+#define fts3HashInit sqlite3Fts3HashInit
+#define fts3HashInsert sqlite3Fts3HashInsert
+#define fts3HashFind sqlite3Fts3HashFind
+#define fts3HashClear sqlite3Fts3HashClear
+#define fts3HashFindElem sqlite3Fts3HashFindElem
/*
** Macros for looping over all elements of a hash table. The idiom is
** like this:
**
-** fts3Hash h;
-** fts3HashElem *p;
+** Fts3Hash h;
+** Fts3HashElem *p;
** ...
** for(p=fts3HashFirst(&h); p; p=fts3HashNext(p)){
** SomeStructure *pData = fts3HashData(p);
diff --git a/third_party/sqlite/src/ext/fts3/fts3_porter.c b/third_party/sqlite/src/ext/fts3/fts3_porter.c
index 001060c..5963abc 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_porter.c
+++ b/third_party/sqlite/src/ext/fts3/fts3_porter.c
@@ -24,6 +24,7 @@
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
+#include "fts3Int.h"
#include <assert.h>
#include <stdlib.h>
@@ -53,10 +54,6 @@ typedef struct porter_tokenizer_cursor {
} porter_tokenizer_cursor;
-/* Forward declaration */
-static const sqlite3_tokenizer_module porterTokenizerModule;
-
-
/*
** Create a new tokenizer instance.
*/
@@ -65,6 +62,10 @@ static int porterCreate(
sqlite3_tokenizer **ppTokenizer
){
porter_tokenizer *t;
+
+ UNUSED_PARAMETER(argc);
+ UNUSED_PARAMETER(argv);
+
t = (porter_tokenizer *) sqlite3_malloc(sizeof(*t));
if( t==NULL ) return SQLITE_NOMEM;
memset(t, 0, sizeof(*t));
@@ -93,6 +94,8 @@ static int porterOpen(
){
porter_tokenizer_cursor *c;
+ UNUSED_PARAMETER(pTokenizer);
+
c = (porter_tokenizer_cursor *) sqlite3_malloc(sizeof(*c));
if( c==NULL ) return SQLITE_NOMEM;
@@ -233,7 +236,7 @@ static int hasVowel(const char *z){
** the first two characters of z[].
*/
static int doubleConsonant(const char *z){
- return isConsonant(z) && z[0]==z[1] && isConsonant(z+1);
+ return isConsonant(z) && z[0]==z[1];
}
/*
@@ -246,10 +249,10 @@ static int doubleConsonant(const char *z){
*/
static int star_oh(const char *z){
return
- z[0]!=0 && isConsonant(z) &&
+ isConsonant(z) &&
z[0]!='w' && z[0]!='x' && z[0]!='y' &&
- z[1]!=0 && isVowel(z+1) &&
- z[2]!=0 && isConsonant(z+2);
+ isVowel(z+1) &&
+ isConsonant(z+2);
}
/*
@@ -293,7 +296,7 @@ static void copy_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){
int i, mx, j;
int hasDigit = 0;
for(i=0; i<nIn; i++){
- int c = zIn[i];
+ char c = zIn[i];
if( c>='A' && c<='Z' ){
zOut[i] = c - 'A' + 'a';
}else{
@@ -337,7 +340,7 @@ static void copy_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){
** no chance of overflowing the zOut buffer.
*/
static void porter_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){
- int i, j, c;
+ int i, j;
char zReverse[28];
char *z, *z2;
if( nIn<3 || nIn>=sizeof(zReverse)-7 ){
@@ -347,7 +350,7 @@ static void porter_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){
return;
}
for(i=0, j=sizeof(zReverse)-6; i<nIn; i++, j--){
- c = zIn[i];
+ char c = zIn[i];
if( c>='A' && c<='Z' ){
zReverse[j] = c + 'a' - 'A';
}else if( c>='a' && c<='z' ){
@@ -546,7 +549,7 @@ static void porter_stemmer(const char *zIn, int nIn, char *zOut, int *pnOut){
/* z[] is now the stemmed word in reverse order. Flip it back
** around into forward order and return.
*/
- *pnOut = i = strlen(z);
+ *pnOut = i = (int)strlen(z);
zOut[i] = 0;
while( *z ){
zOut[--i] = *(z++);
@@ -601,9 +604,11 @@ static int porterNext(
if( c->iOffset>iStartOffset ){
int n = c->iOffset-iStartOffset;
if( n>c->nAllocated ){
+ char *pNew;
c->nAllocated = n+20;
- c->zToken = sqlite3_realloc(c->zToken, c->nAllocated);
- if( c->zToken==NULL ) return SQLITE_NOMEM;
+ pNew = sqlite3_realloc(c->zToken, c->nAllocated);
+ if( !pNew ) return SQLITE_NOMEM;
+ c->zToken = pNew;
}
porter_stemmer(&z[iStartOffset], n, c->zToken, pnBytes);
*pzToken = c->zToken;
diff --git a/third_party/sqlite/src/ext/fts3/fts3_snippet.c b/third_party/sqlite/src/ext/fts3/fts3_snippet.c
new file mode 100644
index 0000000..d67f7ac
--- /dev/null
+++ b/third_party/sqlite/src/ext/fts3/fts3_snippet.c
@@ -0,0 +1,1205 @@
+/*
+** 2009 Oct 23
+**
+** The author disclaims copyright to this source code. In place of
+** a legal notice, here is a blessing:
+**
+** May you do good and not evil.
+** May you find forgiveness for yourself and forgive others.
+** May you share freely, never taking more than you give.
+**
+******************************************************************************
+*/
+
+#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
+
+#include "fts3Int.h"
+#include <string.h>
+#include <assert.h>
+
+
+/*
+** Used as an fts3ExprIterate() context when loading phrase doclists to
+** Fts3Expr.aDoclist[]/nDoclist.
+*/
+typedef struct LoadDoclistCtx LoadDoclistCtx;
+struct LoadDoclistCtx {
+ Fts3Table *pTab; /* FTS3 Table */
+ int nPhrase; /* Number of phrases seen so far */
+ int nToken; /* Number of tokens seen so far */
+};
+
+/*
+** The following types are used as part of the implementation of the
+** fts3BestSnippet() routine.
+*/
+typedef struct SnippetIter SnippetIter;
+typedef struct SnippetPhrase SnippetPhrase;
+typedef struct SnippetFragment SnippetFragment;
+
+struct SnippetIter {
+ Fts3Cursor *pCsr; /* Cursor snippet is being generated from */
+ int iCol; /* Extract snippet from this column */
+ int nSnippet; /* Requested snippet length (in tokens) */
+ int nPhrase; /* Number of phrases in query */
+ SnippetPhrase *aPhrase; /* Array of size nPhrase */
+ int iCurrent; /* First token of current snippet */
+};
+
+struct SnippetPhrase {
+ int nToken; /* Number of tokens in phrase */
+ char *pList; /* Pointer to start of phrase position list */
+ int iHead; /* Next value in position list */
+ char *pHead; /* Position list data following iHead */
+ int iTail; /* Next value in trailing position list */
+ char *pTail; /* Position list data following iTail */
+};
+
+struct SnippetFragment {
+ int iCol; /* Column snippet is extracted from */
+ int iPos; /* Index of first token in snippet */
+ u64 covered; /* Mask of query phrases covered */
+ u64 hlmask; /* Mask of snippet terms to highlight */
+};
+
+/*
+** This type is used as an fts3ExprIterate() context object while
+** accumulating the data returned by the matchinfo() function.
+*/
+typedef struct MatchInfo MatchInfo;
+struct MatchInfo {
+ Fts3Cursor *pCursor; /* FTS3 Cursor */
+ int nCol; /* Number of columns in table */
+ u32 *aMatchinfo; /* Pre-allocated buffer */
+};
+
+
+
+/*
+** The snippet() and offsets() functions both return text values. An instance
+** of the following structure is used to accumulate those values while the
+** functions are running. See fts3StringAppend() for details.
+*/
+typedef struct StrBuffer StrBuffer;
+struct StrBuffer {
+ char *z; /* Pointer to buffer containing string */
+ int n; /* Length of z in bytes (excl. nul-term) */
+ int nAlloc; /* Allocated size of buffer z in bytes */
+};
+
+
+/*
+** This function is used to help iterate through a position-list. A position
+** list is a list of unique integers, sorted from smallest to largest. Each
+** element of the list is represented by an FTS3 varint that takes the value
+** of the difference between the current element and the previous one plus
+** two. For example, to store the position-list:
+**
+** 4 9 113
+**
+** the three varints:
+**
+** 6 7 106
+**
+** are encoded.
+**
+** When this function is called, *pp points to the start of an element of
+** the list. *piPos contains the value of the previous entry in the list.
+** After it returns, *piPos contains the value of the next element of the
+** list and *pp is advanced to the following varint.
+*/
+static void fts3GetDeltaPosition(char **pp, int *piPos){
+ int iVal;
+ *pp += sqlite3Fts3GetVarint32(*pp, &iVal);
+ *piPos += (iVal-2);
+}
+
+/*
+** Helper function for fts3ExprIterate() (see below).
+*/
+static int fts3ExprIterate2(
+ Fts3Expr *pExpr, /* Expression to iterate phrases of */
+ int *piPhrase, /* Pointer to phrase counter */
+ int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */
+ void *pCtx /* Second argument to pass to callback */
+){
+ int rc; /* Return code */
+ int eType = pExpr->eType; /* Type of expression node pExpr */
+
+ if( eType!=FTSQUERY_PHRASE ){
+ assert( pExpr->pLeft && pExpr->pRight );
+ rc = fts3ExprIterate2(pExpr->pLeft, piPhrase, x, pCtx);
+ if( rc==SQLITE_OK && eType!=FTSQUERY_NOT ){
+ rc = fts3ExprIterate2(pExpr->pRight, piPhrase, x, pCtx);
+ }
+ }else{
+ rc = x(pExpr, *piPhrase, pCtx);
+ (*piPhrase)++;
+ }
+ return rc;
+}
+
+/*
+** Iterate through all phrase nodes in an FTS3 query, except those that
+** are part of a sub-tree that is the right-hand-side of a NOT operator.
+** For each phrase node found, the supplied callback function is invoked.
+**
+** If the callback function returns anything other than SQLITE_OK,
+** the iteration is abandoned and the error code returned immediately.
+** Otherwise, SQLITE_OK is returned after a callback has been made for
+** all eligible phrase nodes.
+*/
+static int fts3ExprIterate(
+ Fts3Expr *pExpr, /* Expression to iterate phrases of */
+ int (*x)(Fts3Expr*,int,void*), /* Callback function to invoke for phrases */
+ void *pCtx /* Second argument to pass to callback */
+){
+ int iPhrase = 0; /* Variable used as the phrase counter */
+ return fts3ExprIterate2(pExpr, &iPhrase, x, pCtx);
+}
+
+/*
+** The argument to this function is always a phrase node. Its doclist
+** (Fts3Expr.aDoclist[]) and the doclists associated with all phrase nodes
+** to the left of this one in the query tree have already been loaded.
+**
+** If this phrase node is part of a series of phrase nodes joined by
+** NEAR operators (and is not the left-most of said series), then elements are
+** removed from the phrases doclist consistent with the NEAR restriction. If
+** required, elements may be removed from the doclists of phrases to the
+** left of this one that are part of the same series of NEAR operator
+** connected phrases.
+**
+** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK.
+*/
+static int fts3ExprNearTrim(Fts3Expr *pExpr){
+ int rc = SQLITE_OK;
+ Fts3Expr *pParent = pExpr->pParent;
+
+ assert( pExpr->eType==FTSQUERY_PHRASE );
+ while( rc==SQLITE_OK
+ && pParent
+ && pParent->eType==FTSQUERY_NEAR
+ && pParent->pRight==pExpr
+ ){
+ /* This expression (pExpr) is the right-hand-side of a NEAR operator.
+ ** Find the expression to the left of the same operator.
+ */
+ int nNear = pParent->nNear;
+ Fts3Expr *pLeft = pParent->pLeft;
+
+ if( pLeft->eType!=FTSQUERY_PHRASE ){
+ assert( pLeft->eType==FTSQUERY_NEAR );
+ assert( pLeft->pRight->eType==FTSQUERY_PHRASE );
+ pLeft = pLeft->pRight;
+ }
+
+ rc = sqlite3Fts3ExprNearTrim(pLeft, pExpr, nNear);
+
+ pExpr = pLeft;
+ pParent = pExpr->pParent;
+ }
+
+ return rc;
+}
+
+/*
+** This is an fts3ExprIterate() callback used while loading the doclists
+** for each phrase into Fts3Expr.aDoclist[]/nDoclist. See also
+** fts3ExprLoadDoclists().
+*/
+static int fts3ExprLoadDoclistsCb1(Fts3Expr *pExpr, int iPhrase, void *ctx){
+ int rc = SQLITE_OK;
+ LoadDoclistCtx *p = (LoadDoclistCtx *)ctx;
+
+ UNUSED_PARAMETER(iPhrase);
+
+ p->nPhrase++;
+ p->nToken += pExpr->pPhrase->nToken;
+
+ if( pExpr->isLoaded==0 ){
+ rc = sqlite3Fts3ExprLoadDoclist(p->pTab, pExpr);
+ pExpr->isLoaded = 1;
+ if( rc==SQLITE_OK ){
+ rc = fts3ExprNearTrim(pExpr);
+ }
+ }
+
+ return rc;
+}
+
+/*
+** This is an fts3ExprIterate() callback used while loading the doclists
+** for each phrase into Fts3Expr.aDoclist[]/nDoclist. See also
+** fts3ExprLoadDoclists().
+*/
+static int fts3ExprLoadDoclistsCb2(Fts3Expr *pExpr, int iPhrase, void *ctx){
+ UNUSED_PARAMETER(iPhrase);
+ UNUSED_PARAMETER(ctx);
+ if( pExpr->aDoclist ){
+ pExpr->pCurrent = pExpr->aDoclist;
+ pExpr->iCurrent = 0;
+ pExpr->pCurrent += sqlite3Fts3GetVarint(pExpr->pCurrent, &pExpr->iCurrent);
+ }
+ return SQLITE_OK;
+}
+
+/*
+** Load the doclists for each phrase in the query associated with FTS3 cursor
+** pCsr.
+**
+** If pnPhrase is not NULL, then *pnPhrase is set to the number of matchable
+** phrases in the expression (all phrases except those directly or
+** indirectly descended from the right-hand-side of a NOT operator). If
+** pnToken is not NULL, then it is set to the number of tokens in all
+** matchable phrases of the expression.
+*/
+static int fts3ExprLoadDoclists(
+ Fts3Cursor *pCsr, /* Fts3 cursor for current query */
+ int *pnPhrase, /* OUT: Number of phrases in query */
+ int *pnToken /* OUT: Number of tokens in query */
+){
+ int rc; /* Return Code */
+ LoadDoclistCtx sCtx = {0,0,0}; /* Context for fts3ExprIterate() */
+ sCtx.pTab = (Fts3Table *)pCsr->base.pVtab;
+ rc = fts3ExprIterate(pCsr->pExpr, fts3ExprLoadDoclistsCb1, (void *)&sCtx);
+ if( rc==SQLITE_OK ){
+ (void)fts3ExprIterate(pCsr->pExpr, fts3ExprLoadDoclistsCb2, 0);
+ }
+ if( pnPhrase ) *pnPhrase = sCtx.nPhrase;
+ if( pnToken ) *pnToken = sCtx.nToken;
+ return rc;
+}
+
+/*
+** Advance the position list iterator specified by the first two
+** arguments so that it points to the first element with a value greater
+** than or equal to parameter iNext.
+*/
+static void fts3SnippetAdvance(char **ppIter, int *piIter, int iNext){
+ char *pIter = *ppIter;
+ if( pIter ){
+ int iIter = *piIter;
+
+ while( iIter<iNext ){
+ if( 0==(*pIter & 0xFE) ){
+ iIter = -1;
+ pIter = 0;
+ break;
+ }
+ fts3GetDeltaPosition(&pIter, &iIter);
+ }
+
+ *piIter = iIter;
+ *ppIter = pIter;
+ }
+}
+
+/*
+** Advance the snippet iterator to the next candidate snippet.
+*/
+static int fts3SnippetNextCandidate(SnippetIter *pIter){
+ int i; /* Loop counter */
+
+ if( pIter->iCurrent<0 ){
+ /* The SnippetIter object has just been initialized. The first snippet
+ ** candidate always starts at offset 0 (even if this candidate has a
+ ** score of 0.0).
+ */
+ pIter->iCurrent = 0;
+
+ /* Advance the 'head' iterator of each phrase to the first offset that
+ ** is greater than or equal to (iNext+nSnippet).
+ */
+ for(i=0; i<pIter->nPhrase; i++){
+ SnippetPhrase *pPhrase = &pIter->aPhrase[i];
+ fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, pIter->nSnippet);
+ }
+ }else{
+ int iStart;
+ int iEnd = 0x7FFFFFFF;
+
+ for(i=0; i<pIter->nPhrase; i++){
+ SnippetPhrase *pPhrase = &pIter->aPhrase[i];
+ if( pPhrase->pHead && pPhrase->iHead<iEnd ){
+ iEnd = pPhrase->iHead;
+ }
+ }
+ if( iEnd==0x7FFFFFFF ){
+ return 1;
+ }
+
+ pIter->iCurrent = iStart = iEnd - pIter->nSnippet + 1;
+ for(i=0; i<pIter->nPhrase; i++){
+ SnippetPhrase *pPhrase = &pIter->aPhrase[i];
+ fts3SnippetAdvance(&pPhrase->pHead, &pPhrase->iHead, iEnd+1);
+ fts3SnippetAdvance(&pPhrase->pTail, &pPhrase->iTail, iStart);
+ }
+ }
+
+ return 0;
+}
+
+/*
+** Retrieve information about the current candidate snippet of snippet
+** iterator pIter.
+*/
+static void fts3SnippetDetails(
+ SnippetIter *pIter, /* Snippet iterator */
+ u64 mCovered, /* Bitmask of phrases already covered */
+ int *piToken, /* OUT: First token of proposed snippet */
+ int *piScore, /* OUT: "Score" for this snippet */
+ u64 *pmCover, /* OUT: Bitmask of phrases covered */
+ u64 *pmHighlight /* OUT: Bitmask of terms to highlight */
+){
+ int iStart = pIter->iCurrent; /* First token of snippet */
+ int iScore = 0; /* Score of this snippet */
+ int i; /* Loop counter */
+ u64 mCover = 0; /* Mask of phrases covered by this snippet */
+ u64 mHighlight = 0; /* Mask of tokens to highlight in snippet */
+
+ for(i=0; i<pIter->nPhrase; i++){
+ SnippetPhrase *pPhrase = &pIter->aPhrase[i];
+ if( pPhrase->pTail ){
+ char *pCsr = pPhrase->pTail;
+ int iCsr = pPhrase->iTail;
+
+ while( iCsr<(iStart+pIter->nSnippet) ){
+ int j;
+ u64 mPhrase = (u64)1 << i;
+ u64 mPos = (u64)1 << (iCsr - iStart);
+ assert( iCsr>=iStart );
+ if( (mCover|mCovered)&mPhrase ){
+ iScore++;
+ }else{
+ iScore += 1000;
+ }
+ mCover |= mPhrase;
+
+ for(j=0; j<pPhrase->nToken; j++){
+ mHighlight |= (mPos>>j);
+ }
+
+ if( 0==(*pCsr & 0x0FE) ) break;
+ fts3GetDeltaPosition(&pCsr, &iCsr);
+ }
+ }
+ }
+
+ /* Set the output variables before returning. */
+ *piToken = iStart;
+ *piScore = iScore;
+ *pmCover = mCover;
+ *pmHighlight = mHighlight;
+}
+
+/*
+** This function is an fts3ExprIterate() callback used by fts3BestSnippet().
+** Each invocation populates an element of the SnippetIter.aPhrase[] array.
+*/
+static int fts3SnippetFindPositions(Fts3Expr *pExpr, int iPhrase, void *ctx){
+ SnippetIter *p = (SnippetIter *)ctx;
+ SnippetPhrase *pPhrase = &p->aPhrase[iPhrase];
+ char *pCsr;
+
+ pPhrase->nToken = pExpr->pPhrase->nToken;
+
+ pCsr = sqlite3Fts3FindPositions(pExpr, p->pCsr->iPrevId, p->iCol);
+ if( pCsr ){
+ int iFirst = 0;
+ pPhrase->pList = pCsr;
+ fts3GetDeltaPosition(&pCsr, &iFirst);
+ pPhrase->pHead = pCsr;
+ pPhrase->pTail = pCsr;
+ pPhrase->iHead = iFirst;
+ pPhrase->iTail = iFirst;
+ }else{
+ assert( pPhrase->pList==0 && pPhrase->pHead==0 && pPhrase->pTail==0 );
+ }
+
+ return SQLITE_OK;
+}
+
+/*
+** Select the fragment of text consisting of nFragment contiguous tokens
+** from column iCol that represent the "best" snippet. The best snippet
+** is the snippet with the highest score, where scores are calculated
+** by adding:
+**
+** (a) +1 point for each occurence of a matchable phrase in the snippet.
+**
+** (b) +1000 points for the first occurence of each matchable phrase in
+** the snippet for which the corresponding mCovered bit is not set.
+**
+** The selected snippet parameters are stored in structure *pFragment before
+** returning. The score of the selected snippet is stored in *piScore
+** before returning.
+*/
+static int fts3BestSnippet(
+ int nSnippet, /* Desired snippet length */
+ Fts3Cursor *pCsr, /* Cursor to create snippet for */
+ int iCol, /* Index of column to create snippet from */
+ u64 mCovered, /* Mask of phrases already covered */
+ u64 *pmSeen, /* IN/OUT: Mask of phrases seen */
+ SnippetFragment *pFragment, /* OUT: Best snippet found */
+ int *piScore /* OUT: Score of snippet pFragment */
+){
+ int rc; /* Return Code */
+ int nList; /* Number of phrases in expression */
+ SnippetIter sIter; /* Iterates through snippet candidates */
+ int nByte; /* Number of bytes of space to allocate */
+ int iBestScore = -1; /* Best snippet score found so far */
+ int i; /* Loop counter */
+
+ memset(&sIter, 0, sizeof(sIter));
+
+ /* Iterate through the phrases in the expression to count them. The same
+ ** callback makes sure the doclists are loaded for each phrase.
+ */
+ rc = fts3ExprLoadDoclists(pCsr, &nList, 0);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+
+ /* Now that it is known how many phrases there are, allocate and zero
+ ** the required space using malloc().
+ */
+ nByte = sizeof(SnippetPhrase) * nList;
+ sIter.aPhrase = (SnippetPhrase *)sqlite3_malloc(nByte);
+ if( !sIter.aPhrase ){
+ return SQLITE_NOMEM;
+ }
+ memset(sIter.aPhrase, 0, nByte);
+
+ /* Initialize the contents of the SnippetIter object. Then iterate through
+ ** the set of phrases in the expression to populate the aPhrase[] array.
+ */
+ sIter.pCsr = pCsr;
+ sIter.iCol = iCol;
+ sIter.nSnippet = nSnippet;
+ sIter.nPhrase = nList;
+ sIter.iCurrent = -1;
+ (void)fts3ExprIterate(pCsr->pExpr, fts3SnippetFindPositions, (void *)&sIter);
+
+ /* Set the *pmSeen output variable. */
+ for(i=0; i<nList; i++){
+ if( sIter.aPhrase[i].pHead ){
+ *pmSeen |= (u64)1 << i;
+ }
+ }
+
+ /* Loop through all candidate snippets. Store the best snippet in
+ ** *pFragment. Store its associated 'score' in iBestScore.
+ */
+ pFragment->iCol = iCol;
+ while( !fts3SnippetNextCandidate(&sIter) ){
+ int iPos;
+ int iScore;
+ u64 mCover;
+ u64 mHighlight;
+ fts3SnippetDetails(&sIter, mCovered, &iPos, &iScore, &mCover, &mHighlight);
+ assert( iScore>=0 );
+ if( iScore>iBestScore ){
+ pFragment->iPos = iPos;
+ pFragment->hlmask = mHighlight;
+ pFragment->covered = mCover;
+ iBestScore = iScore;
+ }
+ }
+
+ sqlite3_free(sIter.aPhrase);
+ *piScore = iBestScore;
+ return SQLITE_OK;
+}
+
+
+/*
+** Append a string to the string-buffer passed as the first argument.
+**
+** If nAppend is negative, then the length of the string zAppend is
+** determined using strlen().
+*/
+static int fts3StringAppend(
+ StrBuffer *pStr, /* Buffer to append to */
+ const char *zAppend, /* Pointer to data to append to buffer */
+ int nAppend /* Size of zAppend in bytes (or -1) */
+){
+ if( nAppend<0 ){
+ nAppend = (int)strlen(zAppend);
+ }
+
+ /* If there is insufficient space allocated at StrBuffer.z, use realloc()
+ ** to grow the buffer until so that it is big enough to accomadate the
+ ** appended data.
+ */
+ if( pStr->n+nAppend+1>=pStr->nAlloc ){
+ int nAlloc = pStr->nAlloc+nAppend+100;
+ char *zNew = sqlite3_realloc(pStr->z, nAlloc);
+ if( !zNew ){
+ return SQLITE_NOMEM;
+ }
+ pStr->z = zNew;
+ pStr->nAlloc = nAlloc;
+ }
+
+ /* Append the data to the string buffer. */
+ memcpy(&pStr->z[pStr->n], zAppend, nAppend);
+ pStr->n += nAppend;
+ pStr->z[pStr->n] = '\0';
+
+ return SQLITE_OK;
+}
+
+/*
+** The fts3BestSnippet() function often selects snippets that end with a
+** query term. That is, the final term of the snippet is always a term
+** that requires highlighting. For example, if 'X' is a highlighted term
+** and '.' is a non-highlighted term, BestSnippet() may select:
+**
+** ........X.....X
+**
+** This function "shifts" the beginning of the snippet forward in the
+** document so that there are approximately the same number of
+** non-highlighted terms to the right of the final highlighted term as there
+** are to the left of the first highlighted term. For example, to this:
+**
+** ....X.....X....
+**
+** This is done as part of extracting the snippet text, not when selecting
+** the snippet. Snippet selection is done based on doclists only, so there
+** is no way for fts3BestSnippet() to know whether or not the document
+** actually contains terms that follow the final highlighted term.
+*/
+static int fts3SnippetShift(
+ Fts3Table *pTab, /* FTS3 table snippet comes from */
+ int nSnippet, /* Number of tokens desired for snippet */
+ const char *zDoc, /* Document text to extract snippet from */
+ int nDoc, /* Size of buffer zDoc in bytes */
+ int *piPos, /* IN/OUT: First token of snippet */
+ u64 *pHlmask /* IN/OUT: Mask of tokens to highlight */
+){
+ u64 hlmask = *pHlmask; /* Local copy of initial highlight-mask */
+
+ if( hlmask ){
+ int nLeft; /* Tokens to the left of first highlight */
+ int nRight; /* Tokens to the right of last highlight */
+ int nDesired; /* Ideal number of tokens to shift forward */
+
+ for(nLeft=0; !(hlmask & ((u64)1 << nLeft)); nLeft++);
+ for(nRight=0; !(hlmask & ((u64)1 << (nSnippet-1-nRight))); nRight++);
+ nDesired = (nLeft-nRight)/2;
+
+ /* Ideally, the start of the snippet should be pushed forward in the
+ ** document nDesired tokens. This block checks if there are actually
+ ** nDesired tokens to the right of the snippet. If so, *piPos and
+ ** *pHlMask are updated to shift the snippet nDesired tokens to the
+ ** right. Otherwise, the snippet is shifted by the number of tokens
+ ** available.
+ */
+ if( nDesired>0 ){
+ int nShift; /* Number of tokens to shift snippet by */
+ int iCurrent = 0; /* Token counter */
+ int rc; /* Return Code */
+ sqlite3_tokenizer_module *pMod;
+ sqlite3_tokenizer_cursor *pC;
+ pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule;
+
+ /* Open a cursor on zDoc/nDoc. Check if there are (nSnippet+nDesired)
+ ** or more tokens in zDoc/nDoc.
+ */
+ rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ pC->pTokenizer = pTab->pTokenizer;
+ while( rc==SQLITE_OK && iCurrent<(nSnippet+nDesired) ){
+ const char *ZDUMMY; int DUMMY1, DUMMY2, DUMMY3;
+ rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &DUMMY2, &DUMMY3, &iCurrent);
+ }
+ pMod->xClose(pC);
+ if( rc!=SQLITE_OK && rc!=SQLITE_DONE ){ return rc; }
+
+ nShift = (rc==SQLITE_DONE)+iCurrent-nSnippet;
+ assert( nShift<=nDesired );
+ if( nShift>0 ){
+ *piPos += nShift;
+ *pHlmask = hlmask >> nShift;
+ }
+ }
+ }
+ return SQLITE_OK;
+}
+
+/*
+** Extract the snippet text for fragment pFragment from cursor pCsr and
+** append it to string buffer pOut.
+*/
+static int fts3SnippetText(
+ Fts3Cursor *pCsr, /* FTS3 Cursor */
+ SnippetFragment *pFragment, /* Snippet to extract */
+ int iFragment, /* Fragment number */
+ int isLast, /* True for final fragment in snippet */
+ int nSnippet, /* Number of tokens in extracted snippet */
+ const char *zOpen, /* String inserted before highlighted term */
+ const char *zClose, /* String inserted after highlighted term */
+ const char *zEllipsis, /* String inserted between snippets */
+ StrBuffer *pOut /* Write output here */
+){
+ Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
+ int rc; /* Return code */
+ const char *zDoc; /* Document text to extract snippet from */
+ int nDoc; /* Size of zDoc in bytes */
+ int iCurrent = 0; /* Current token number of document */
+ int iEnd = 0; /* Byte offset of end of current token */
+ int isShiftDone = 0; /* True after snippet is shifted */
+ int iPos = pFragment->iPos; /* First token of snippet */
+ u64 hlmask = pFragment->hlmask; /* Highlight-mask for snippet */
+ int iCol = pFragment->iCol+1; /* Query column to extract text from */
+ sqlite3_tokenizer_module *pMod; /* Tokenizer module methods object */
+ sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor open on zDoc/nDoc */
+ const char *ZDUMMY; /* Dummy argument used with tokenizer */
+ int DUMMY1; /* Dummy argument used with tokenizer */
+
+ zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol);
+ if( zDoc==0 ){
+ if( sqlite3_column_type(pCsr->pStmt, iCol)!=SQLITE_NULL ){
+ return SQLITE_NOMEM;
+ }
+ return SQLITE_OK;
+ }
+ nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol);
+
+ /* Open a token cursor on the document. */
+ pMod = (sqlite3_tokenizer_module *)pTab->pTokenizer->pModule;
+ rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ pC->pTokenizer = pTab->pTokenizer;
+
+ while( rc==SQLITE_OK ){
+ int iBegin; /* Offset in zDoc of start of token */
+ int iFin; /* Offset in zDoc of end of token */
+ int isHighlight; /* True for highlighted terms */
+
+ rc = pMod->xNext(pC, &ZDUMMY, &DUMMY1, &iBegin, &iFin, &iCurrent);
+ if( rc!=SQLITE_OK ){
+ if( rc==SQLITE_DONE ){
+ /* Special case - the last token of the snippet is also the last token
+ ** of the column. Append any punctuation that occurred between the end
+ ** of the previous token and the end of the document to the output.
+ ** Then break out of the loop. */
+ rc = fts3StringAppend(pOut, &zDoc[iEnd], -1);
+ }
+ break;
+ }
+ if( iCurrent<iPos ){ continue; }
+
+ if( !isShiftDone ){
+ int n = nDoc - iBegin;
+ rc = fts3SnippetShift(pTab, nSnippet, &zDoc[iBegin], n, &iPos, &hlmask);
+ isShiftDone = 1;
+
+ /* Now that the shift has been done, check if the initial "..." are
+ ** required. They are required if (a) this is not the first fragment,
+ ** or (b) this fragment does not begin at position 0 of its column.
+ */
+ if( rc==SQLITE_OK && (iPos>0 || iFragment>0) ){
+ rc = fts3StringAppend(pOut, zEllipsis, -1);
+ }
+ if( rc!=SQLITE_OK || iCurrent<iPos ) continue;
+ }
+
+ if( iCurrent>=(iPos+nSnippet) ){
+ if( isLast ){
+ rc = fts3StringAppend(pOut, zEllipsis, -1);
+ }
+ break;
+ }
+
+ /* Set isHighlight to true if this term should be highlighted. */
+ isHighlight = (hlmask & ((u64)1 << (iCurrent-iPos)))!=0;
+
+ if( iCurrent>iPos ) rc = fts3StringAppend(pOut, &zDoc[iEnd], iBegin-iEnd);
+ if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zOpen, -1);
+ if( rc==SQLITE_OK ) rc = fts3StringAppend(pOut, &zDoc[iBegin], iFin-iBegin);
+ if( rc==SQLITE_OK && isHighlight ) rc = fts3StringAppend(pOut, zClose, -1);
+
+ iEnd = iFin;
+ }
+
+ pMod->xClose(pC);
+ return rc;
+}
+
+
+/*
+** This function is used to count the entries in a column-list (a
+** delta-encoded list of term offsets within a single column of a single
+** row). When this function is called, *ppCollist should point to the
+** beginning of the first varint in the column-list (the varint that
+** contains the position of the first matching term in the column data).
+** Before returning, *ppCollist is set to point to the first byte after
+** the last varint in the column-list (either the 0x00 signifying the end
+** of the position-list, or the 0x01 that precedes the column number of
+** the next column in the position-list).
+**
+** The number of elements in the column-list is returned.
+*/
+static int fts3ColumnlistCount(char **ppCollist){
+ char *pEnd = *ppCollist;
+ char c = 0;
+ int nEntry = 0;
+
+ /* A column-list is terminated by either a 0x01 or 0x00. */
+ while( 0xFE & (*pEnd | c) ){
+ c = *pEnd++ & 0x80;
+ if( !c ) nEntry++;
+ }
+
+ *ppCollist = pEnd;
+ return nEntry;
+}
+
+static void fts3LoadColumnlistCounts(char **pp, u32 *aOut, int isGlobal){
+ char *pCsr = *pp;
+ while( *pCsr ){
+ int nHit;
+ sqlite3_int64 iCol = 0;
+ if( *pCsr==0x01 ){
+ pCsr++;
+ pCsr += sqlite3Fts3GetVarint(pCsr, &iCol);
+ }
+ nHit = fts3ColumnlistCount(&pCsr);
+ assert( nHit>0 );
+ if( isGlobal ){
+ aOut[iCol*3+1]++;
+ }
+ aOut[iCol*3] += nHit;
+ }
+ pCsr++;
+ *pp = pCsr;
+}
+
+/*
+** fts3ExprIterate() callback used to collect the "global" matchinfo stats
+** for a single query. The "global" stats are those elements of the matchinfo
+** array that are constant for all rows returned by the current query.
+*/
+static int fts3ExprGlobalMatchinfoCb(
+ Fts3Expr *pExpr, /* Phrase expression node */
+ int iPhrase, /* Phrase number (numbered from zero) */
+ void *pCtx /* Pointer to MatchInfo structure */
+){
+ MatchInfo *p = (MatchInfo *)pCtx;
+ char *pCsr;
+ char *pEnd;
+ const int iStart = 2 + (iPhrase * p->nCol * 3) + 1;
+
+ assert( pExpr->isLoaded );
+
+ /* Fill in the global hit count matrix row for this phrase. */
+ pCsr = pExpr->aDoclist;
+ pEnd = &pExpr->aDoclist[pExpr->nDoclist];
+ while( pCsr<pEnd ){
+ while( *pCsr++ & 0x80 ); /* Skip past docid. */
+ fts3LoadColumnlistCounts(&pCsr, &p->aMatchinfo[iStart], 1);
+ }
+
+ return SQLITE_OK;
+}
+
+/*
+** fts3ExprIterate() callback used to collect the "local" matchinfo stats
+** for a single query. The "local" stats are those elements of the matchinfo
+** array that are different for each row returned by the query.
+*/
+static int fts3ExprLocalMatchinfoCb(
+ Fts3Expr *pExpr, /* Phrase expression node */
+ int iPhrase, /* Phrase number */
+ void *pCtx /* Pointer to MatchInfo structure */
+){
+ MatchInfo *p = (MatchInfo *)pCtx;
+
+ if( pExpr->aDoclist ){
+ char *pCsr;
+ int iStart = 2 + (iPhrase * p->nCol * 3);
+ int i;
+
+ for(i=0; i<p->nCol; i++) p->aMatchinfo[iStart+i*3] = 0;
+
+ pCsr = sqlite3Fts3FindPositions(pExpr, p->pCursor->iPrevId, -1);
+ if( pCsr ){
+ fts3LoadColumnlistCounts(&pCsr, &p->aMatchinfo[iStart], 0);
+ }
+ }
+
+ return SQLITE_OK;
+}
+
+/*
+** Populate pCsr->aMatchinfo[] with data for the current row. The
+** 'matchinfo' data is an array of 32-bit unsigned integers (C type u32).
+*/
+static int fts3GetMatchinfo(Fts3Cursor *pCsr){
+ MatchInfo sInfo;
+ Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
+ int rc = SQLITE_OK;
+
+ sInfo.pCursor = pCsr;
+ sInfo.nCol = pTab->nColumn;
+
+ if( pCsr->aMatchinfo==0 ){
+ /* If Fts3Cursor.aMatchinfo[] is NULL, then this is the first time the
+ ** matchinfo function has been called for this query. In this case
+ ** allocate the array used to accumulate the matchinfo data and
+ ** initialize those elements that are constant for every row.
+ */
+ int nPhrase; /* Number of phrases */
+ int nMatchinfo; /* Number of u32 elements in match-info */
+
+ /* Load doclists for each phrase in the query. */
+ rc = fts3ExprLoadDoclists(pCsr, &nPhrase, 0);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ nMatchinfo = 2 + 3*sInfo.nCol*nPhrase;
+ if( pTab->bHasDocsize ){
+ nMatchinfo += 1 + 2*pTab->nColumn;
+ }
+
+ sInfo.aMatchinfo = (u32 *)sqlite3_malloc(sizeof(u32)*nMatchinfo);
+ if( !sInfo.aMatchinfo ){
+ return SQLITE_NOMEM;
+ }
+ memset(sInfo.aMatchinfo, 0, sizeof(u32)*nMatchinfo);
+
+
+ /* First element of match-info is the number of phrases in the query */
+ sInfo.aMatchinfo[0] = nPhrase;
+ sInfo.aMatchinfo[1] = sInfo.nCol;
+ (void)fts3ExprIterate(pCsr->pExpr, fts3ExprGlobalMatchinfoCb,(void*)&sInfo);
+ if( pTab->bHasDocsize ){
+ int ofst = 2 + 3*sInfo.aMatchinfo[0]*sInfo.aMatchinfo[1];
+ rc = sqlite3Fts3MatchinfoDocsizeGlobal(pCsr, &sInfo.aMatchinfo[ofst]);
+ }
+ pCsr->aMatchinfo = sInfo.aMatchinfo;
+ pCsr->isMatchinfoNeeded = 1;
+ }
+
+ sInfo.aMatchinfo = pCsr->aMatchinfo;
+ if( rc==SQLITE_OK && pCsr->isMatchinfoNeeded ){
+ (void)fts3ExprIterate(pCsr->pExpr, fts3ExprLocalMatchinfoCb, (void*)&sInfo);
+ if( pTab->bHasDocsize ){
+ int ofst = 2 + 3*sInfo.aMatchinfo[0]*sInfo.aMatchinfo[1];
+ rc = sqlite3Fts3MatchinfoDocsizeLocal(pCsr, &sInfo.aMatchinfo[ofst]);
+ }
+ pCsr->isMatchinfoNeeded = 0;
+ }
+
+ return SQLITE_OK;
+}
+
+/*
+** Implementation of snippet() function.
+*/
+void sqlite3Fts3Snippet(
+ sqlite3_context *pCtx, /* SQLite function call context */
+ Fts3Cursor *pCsr, /* Cursor object */
+ const char *zStart, /* Snippet start text - "<b>" */
+ const char *zEnd, /* Snippet end text - "</b>" */
+ const char *zEllipsis, /* Snippet ellipsis text - "<b>...</b>" */
+ int iCol, /* Extract snippet from this column */
+ int nToken /* Approximate number of tokens in snippet */
+){
+ Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
+ int rc = SQLITE_OK;
+ int i;
+ StrBuffer res = {0, 0, 0};
+
+ /* The returned text includes up to four fragments of text extracted from
+ ** the data in the current row. The first iteration of the for(...) loop
+ ** below attempts to locate a single fragment of text nToken tokens in
+ ** size that contains at least one instance of all phrases in the query
+ ** expression that appear in the current row. If such a fragment of text
+ ** cannot be found, the second iteration of the loop attempts to locate
+ ** a pair of fragments, and so on.
+ */
+ int nSnippet = 0; /* Number of fragments in this snippet */
+ SnippetFragment aSnippet[4]; /* Maximum of 4 fragments per snippet */
+ int nFToken = -1; /* Number of tokens in each fragment */
+
+ if( !pCsr->pExpr ){
+ sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC);
+ return;
+ }
+
+ for(nSnippet=1; 1; nSnippet++){
+
+ int iSnip; /* Loop counter 0..nSnippet-1 */
+ u64 mCovered = 0; /* Bitmask of phrases covered by snippet */
+ u64 mSeen = 0; /* Bitmask of phrases seen by BestSnippet() */
+
+ if( nToken>=0 ){
+ nFToken = (nToken+nSnippet-1) / nSnippet;
+ }else{
+ nFToken = -1 * nToken;
+ }
+
+ for(iSnip=0; iSnip<nSnippet; iSnip++){
+ int iBestScore = -1; /* Best score of columns checked so far */
+ int iRead; /* Used to iterate through columns */
+ SnippetFragment *pFragment = &aSnippet[iSnip];
+
+ memset(pFragment, 0, sizeof(*pFragment));
+
+ /* Loop through all columns of the table being considered for snippets.
+ ** If the iCol argument to this function was negative, this means all
+ ** columns of the FTS3 table. Otherwise, only column iCol is considered.
+ */
+ for(iRead=0; iRead<pTab->nColumn; iRead++){
+ SnippetFragment sF;
+ int iS;
+ if( iCol>=0 && iRead!=iCol ) continue;
+
+ /* Find the best snippet of nFToken tokens in column iRead. */
+ rc = fts3BestSnippet(nFToken, pCsr, iRead, mCovered, &mSeen, &sF, &iS);
+ if( rc!=SQLITE_OK ){
+ goto snippet_out;
+ }
+ if( iS>iBestScore ){
+ *pFragment = sF;
+ iBestScore = iS;
+ }
+ }
+
+ mCovered |= pFragment->covered;
+ }
+
+ /* If all query phrases seen by fts3BestSnippet() are present in at least
+ ** one of the nSnippet snippet fragments, break out of the loop.
+ */
+ assert( (mCovered&mSeen)==mCovered );
+ if( mSeen==mCovered || nSnippet==SizeofArray(aSnippet) ) break;
+ }
+
+ assert( nFToken>0 );
+
+ for(i=0; i<nSnippet && rc==SQLITE_OK; i++){
+ rc = fts3SnippetText(pCsr, &aSnippet[i],
+ i, (i==nSnippet-1), nFToken, zStart, zEnd, zEllipsis, &res
+ );
+ }
+
+ snippet_out:
+ if( rc!=SQLITE_OK ){
+ sqlite3_result_error_code(pCtx, rc);
+ sqlite3_free(res.z);
+ }else{
+ sqlite3_result_text(pCtx, res.z, -1, sqlite3_free);
+ }
+}
+
+
+typedef struct TermOffset TermOffset;
+typedef struct TermOffsetCtx TermOffsetCtx;
+
+struct TermOffset {
+ char *pList; /* Position-list */
+ int iPos; /* Position just read from pList */
+ int iOff; /* Offset of this term from read positions */
+};
+
+struct TermOffsetCtx {
+ int iCol; /* Column of table to populate aTerm for */
+ int iTerm;
+ sqlite3_int64 iDocid;
+ TermOffset *aTerm;
+};
+
+/*
+** This function is an fts3ExprIterate() callback used by sqlite3Fts3Offsets().
+*/
+static int fts3ExprTermOffsetInit(Fts3Expr *pExpr, int iPhrase, void *ctx){
+ TermOffsetCtx *p = (TermOffsetCtx *)ctx;
+ int nTerm; /* Number of tokens in phrase */
+ int iTerm; /* For looping through nTerm phrase terms */
+ char *pList; /* Pointer to position list for phrase */
+ int iPos = 0; /* First position in position-list */
+
+ UNUSED_PARAMETER(iPhrase);
+ pList = sqlite3Fts3FindPositions(pExpr, p->iDocid, p->iCol);
+ nTerm = pExpr->pPhrase->nToken;
+ if( pList ){
+ fts3GetDeltaPosition(&pList, &iPos);
+ assert( iPos>=0 );
+ }
+
+ for(iTerm=0; iTerm<nTerm; iTerm++){
+ TermOffset *pT = &p->aTerm[p->iTerm++];
+ pT->iOff = nTerm-iTerm-1;
+ pT->pList = pList;
+ pT->iPos = iPos;
+ }
+
+ return SQLITE_OK;
+}
+
+/*
+** Implementation of offsets() function.
+*/
+void sqlite3Fts3Offsets(
+ sqlite3_context *pCtx, /* SQLite function call context */
+ Fts3Cursor *pCsr /* Cursor object */
+){
+ Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
+ sqlite3_tokenizer_module const *pMod = pTab->pTokenizer->pModule;
+ const char *ZDUMMY; /* Dummy argument used with xNext() */
+ int NDUMMY; /* Dummy argument used with xNext() */
+ int rc; /* Return Code */
+ int nToken; /* Number of tokens in query */
+ int iCol; /* Column currently being processed */
+ StrBuffer res = {0, 0, 0}; /* Result string */
+ TermOffsetCtx sCtx; /* Context for fts3ExprTermOffsetInit() */
+
+ if( !pCsr->pExpr ){
+ sqlite3_result_text(pCtx, "", 0, SQLITE_STATIC);
+ return;
+ }
+
+ memset(&sCtx, 0, sizeof(sCtx));
+ assert( pCsr->isRequireSeek==0 );
+
+ /* Count the number of terms in the query */
+ rc = fts3ExprLoadDoclists(pCsr, 0, &nToken);
+ if( rc!=SQLITE_OK ) goto offsets_out;
+
+ /* Allocate the array of TermOffset iterators. */
+ sCtx.aTerm = (TermOffset *)sqlite3_malloc(sizeof(TermOffset)*nToken);
+ if( 0==sCtx.aTerm ){
+ rc = SQLITE_NOMEM;
+ goto offsets_out;
+ }
+ sCtx.iDocid = pCsr->iPrevId;
+
+ /* Loop through the table columns, appending offset information to
+ ** string-buffer res for each column.
+ */
+ for(iCol=0; iCol<pTab->nColumn; iCol++){
+ sqlite3_tokenizer_cursor *pC; /* Tokenizer cursor */
+ int iStart;
+ int iEnd;
+ int iCurrent;
+ const char *zDoc;
+ int nDoc;
+
+ /* Initialize the contents of sCtx.aTerm[] for column iCol. There is
+ ** no way that this operation can fail, so the return code from
+ ** fts3ExprIterate() can be discarded.
+ */
+ sCtx.iCol = iCol;
+ sCtx.iTerm = 0;
+ (void)fts3ExprIterate(pCsr->pExpr, fts3ExprTermOffsetInit, (void *)&sCtx);
+
+ /* Retreive the text stored in column iCol. If an SQL NULL is stored
+ ** in column iCol, jump immediately to the next iteration of the loop.
+ ** If an OOM occurs while retrieving the data (this can happen if SQLite
+ ** needs to transform the data from utf-16 to utf-8), return SQLITE_NOMEM
+ ** to the caller.
+ */
+ zDoc = (const char *)sqlite3_column_text(pCsr->pStmt, iCol+1);
+ nDoc = sqlite3_column_bytes(pCsr->pStmt, iCol+1);
+ if( zDoc==0 ){
+ if( sqlite3_column_type(pCsr->pStmt, iCol+1)==SQLITE_NULL ){
+ continue;
+ }
+ rc = SQLITE_NOMEM;
+ goto offsets_out;
+ }
+
+ /* Initialize a tokenizer iterator to iterate through column iCol. */
+ rc = pMod->xOpen(pTab->pTokenizer, zDoc, nDoc, &pC);
+ if( rc!=SQLITE_OK ) goto offsets_out;
+ pC->pTokenizer = pTab->pTokenizer;
+
+ rc = pMod->xNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent);
+ while( rc==SQLITE_OK ){
+ int i; /* Used to loop through terms */
+ int iMinPos = 0x7FFFFFFF; /* Position of next token */
+ TermOffset *pTerm = 0; /* TermOffset associated with next token */
+
+ for(i=0; i<nToken; i++){
+ TermOffset *pT = &sCtx.aTerm[i];
+ if( pT->pList && (pT->iPos-pT->iOff)<iMinPos ){
+ iMinPos = pT->iPos-pT->iOff;
+ pTerm = pT;
+ }
+ }
+
+ if( !pTerm ){
+ /* All offsets for this column have been gathered. */
+ break;
+ }else{
+ assert( iCurrent<=iMinPos );
+ if( 0==(0xFE&*pTerm->pList) ){
+ pTerm->pList = 0;
+ }else{
+ fts3GetDeltaPosition(&pTerm->pList, &pTerm->iPos);
+ }
+ while( rc==SQLITE_OK && iCurrent<iMinPos ){
+ rc = pMod->xNext(pC, &ZDUMMY, &NDUMMY, &iStart, &iEnd, &iCurrent);
+ }
+ if( rc==SQLITE_OK ){
+ char aBuffer[64];
+ sqlite3_snprintf(sizeof(aBuffer), aBuffer,
+ "%d %d %d %d ", iCol, pTerm-sCtx.aTerm, iStart, iEnd-iStart
+ );
+ rc = fts3StringAppend(&res, aBuffer, -1);
+ }else if( rc==SQLITE_DONE ){
+ rc = SQLITE_CORRUPT;
+ }
+ }
+ }
+ if( rc==SQLITE_DONE ){
+ rc = SQLITE_OK;
+ }
+
+ pMod->xClose(pC);
+ if( rc!=SQLITE_OK ) goto offsets_out;
+ }
+
+ offsets_out:
+ sqlite3_free(sCtx.aTerm);
+ assert( rc!=SQLITE_DONE );
+ if( rc!=SQLITE_OK ){
+ sqlite3_result_error_code(pCtx, rc);
+ sqlite3_free(res.z);
+ }else{
+ sqlite3_result_text(pCtx, res.z, res.n-1, sqlite3_free);
+ }
+ return;
+}
+
+/*
+** Implementation of matchinfo() function.
+*/
+void sqlite3Fts3Matchinfo(sqlite3_context *pContext, Fts3Cursor *pCsr){
+ int rc;
+ if( !pCsr->pExpr ){
+ sqlite3_result_blob(pContext, "", 0, SQLITE_STATIC);
+ return;
+ }
+ rc = fts3GetMatchinfo(pCsr);
+ if( rc!=SQLITE_OK ){
+ sqlite3_result_error_code(pContext, rc);
+ }else{
+ Fts3Table *pTab = (Fts3Table*)pCsr->base.pVtab;
+ int n = sizeof(u32)*(2+pCsr->aMatchinfo[0]*pCsr->aMatchinfo[1]*3);
+ if( pTab->bHasDocsize ){
+ n += sizeof(u32)*(1 + 2*pTab->nColumn);
+ }
+ sqlite3_result_blob(pContext, pCsr->aMatchinfo, n, SQLITE_TRANSIENT);
+ }
+}
+
+#endif
diff --git a/third_party/sqlite/src/ext/fts3/fts3_tokenizer.c b/third_party/sqlite/src/ext/fts3/fts3_tokenizer.c
index e5de9c9..1121b00 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_tokenizer.c
+++ b/third_party/sqlite/src/ext/fts3/fts3_tokenizer.c
@@ -30,10 +30,10 @@
SQLITE_EXTENSION_INIT1
#endif
-#include "fts3_hash.h"
-#include "fts3_tokenizer.h"
+#include "fts3Int.h"
#include <assert.h>
#include <stddef.h>
+#include <string.h>
/*
** Implementation of the SQL scalar function for accessing the underlying
@@ -60,14 +60,14 @@ static void scalarFunc(
int argc,
sqlite3_value **argv
){
- fts3Hash *pHash;
+ Fts3Hash *pHash;
void *pPtr = 0;
const unsigned char *zName;
int nName;
assert( argc==1 || argc==2 );
- pHash = (fts3Hash *)sqlite3_user_data(context);
+ pHash = (Fts3Hash *)sqlite3_user_data(context);
zName = sqlite3_value_text(argv[0]);
nName = sqlite3_value_bytes(argv[0])+1;
@@ -98,6 +98,127 @@ static void scalarFunc(
sqlite3_result_blob(context, (void *)&pPtr, sizeof(pPtr), SQLITE_TRANSIENT);
}
+static int fts3IsIdChar(char c){
+ static const char isFtsIdChar[] = {
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 0x */
+ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 1x */
+ 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */
+ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */
+ 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */
+ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */
+ 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */
+ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */
+ };
+ return (c&0x80 || isFtsIdChar[(int)(c)]);
+}
+
+const char *sqlite3Fts3NextToken(const char *zStr, int *pn){
+ const char *z1;
+ const char *z2 = 0;
+
+ /* Find the start of the next token. */
+ z1 = zStr;
+ while( z2==0 ){
+ char c = *z1;
+ switch( c ){
+ case '\0': return 0; /* No more tokens here */
+ case '\'':
+ case '"':
+ case '`': {
+ z2 = z1;
+ while( *++z2 && (*z2!=c || *++z2==c) );
+ break;
+ }
+ case '[':
+ z2 = &z1[1];
+ while( *z2 && z2[0]!=']' ) z2++;
+ if( *z2 ) z2++;
+ break;
+
+ default:
+ if( fts3IsIdChar(*z1) ){
+ z2 = &z1[1];
+ while( fts3IsIdChar(*z2) ) z2++;
+ }else{
+ z1++;
+ }
+ }
+ }
+
+ *pn = (int)(z2-z1);
+ return z1;
+}
+
+int sqlite3Fts3InitTokenizer(
+ Fts3Hash *pHash, /* Tokenizer hash table */
+ const char *zArg, /* Possible tokenizer specification */
+ sqlite3_tokenizer **ppTok, /* OUT: Tokenizer (if applicable) */
+ const char **pzTokenizer, /* OUT: Set to zArg if is tokenizer */
+ char **pzErr /* OUT: Set to malloced error message */
+){
+ int rc;
+ char *z = (char *)zArg;
+ int n;
+ char *zCopy;
+ char *zEnd; /* Pointer to nul-term of zCopy */
+ sqlite3_tokenizer_module *m;
+
+ if( !z ){
+ zCopy = sqlite3_mprintf("simple");
+ }else{
+ if( sqlite3_strnicmp(z, "tokenize", 8) || fts3IsIdChar(z[8])){
+ return SQLITE_OK;
+ }
+ zCopy = sqlite3_mprintf("%s", &z[8]);
+ *pzTokenizer = zArg;
+ }
+ if( !zCopy ){
+ return SQLITE_NOMEM;
+ }
+
+ zEnd = &zCopy[strlen(zCopy)];
+
+ z = (char *)sqlite3Fts3NextToken(zCopy, &n);
+ z[n] = '\0';
+ sqlite3Fts3Dequote(z);
+
+ m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, z, (int)strlen(z)+1);
+ if( !m ){
+ *pzErr = sqlite3_mprintf("unknown tokenizer: %s", z);
+ rc = SQLITE_ERROR;
+ }else{
+ char const **aArg = 0;
+ int iArg = 0;
+ z = &z[n+1];
+ while( z<zEnd && (NULL!=(z = (char *)sqlite3Fts3NextToken(z, &n))) ){
+ int nNew = sizeof(char *)*(iArg+1);
+ char const **aNew = (const char **)sqlite3_realloc((void *)aArg, nNew);
+ if( !aNew ){
+ sqlite3_free(zCopy);
+ sqlite3_free((void *)aArg);
+ return SQLITE_NOMEM;
+ }
+ aArg = aNew;
+ aArg[iArg++] = z;
+ z[n] = '\0';
+ sqlite3Fts3Dequote(z);
+ z = &z[n+1];
+ }
+ rc = m->xCreate(iArg, aArg, ppTok);
+ assert( rc!=SQLITE_OK || *ppTok );
+ if( rc!=SQLITE_OK ){
+ *pzErr = sqlite3_mprintf("unknown tokenizer");
+ }else{
+ (*ppTok)->pModule = m;
+ }
+ sqlite3_free((void *)aArg);
+ }
+
+ sqlite3_free(zCopy);
+ return rc;
+}
+
+
#ifdef SQLITE_TEST
#include <tcl.h>
@@ -134,7 +255,7 @@ static void testFunc(
int argc,
sqlite3_value **argv
){
- fts3Hash *pHash;
+ Fts3Hash *pHash;
sqlite3_tokenizer_module *p;
sqlite3_tokenizer *pTokenizer = 0;
sqlite3_tokenizer_cursor *pCsr = 0;
@@ -167,7 +288,7 @@ static void testFunc(
zArg = (const char *)sqlite3_value_text(argv[1]);
}
- pHash = (fts3Hash *)sqlite3_user_data(context);
+ pHash = (Fts3Hash *)sqlite3_user_data(context);
p = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zName, nName+1);
if( !p ){
@@ -258,7 +379,7 @@ int queryTokenizer(
sqlite3_bind_text(pStmt, 1, zName, -1, SQLITE_STATIC);
if( SQLITE_ROW==sqlite3_step(pStmt) ){
if( sqlite3_column_type(pStmt, 0)==SQLITE_BLOB ){
- memcpy(pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp));
+ memcpy((void *)pp, sqlite3_column_blob(pStmt, 0), sizeof(*pp));
}
}
@@ -295,6 +416,9 @@ static void intTestFunc(
const sqlite3_tokenizer_module *p2;
sqlite3 *db = (sqlite3 *)sqlite3_user_data(context);
+ UNUSED_PARAMETER(argc);
+ UNUSED_PARAMETER(argv);
+
/* Test the query function */
sqlite3Fts3SimpleTokenizerModule(&p1);
rc = queryTokenizer(db, "simple", &p2);
@@ -336,16 +460,16 @@ static void intTestFunc(
*/
int sqlite3Fts3InitHashTable(
sqlite3 *db,
- fts3Hash *pHash,
+ Fts3Hash *pHash,
const char *zName
){
int rc = SQLITE_OK;
void *p = (void *)pHash;
const int any = SQLITE_ANY;
- char *zTest = 0;
- char *zTest2 = 0;
#ifdef SQLITE_TEST
+ char *zTest = 0;
+ char *zTest2 = 0;
void *pdb = (void *)db;
zTest = sqlite3_mprintf("%s_test", zName);
zTest2 = sqlite3_mprintf("%s_internal_test", zName);
@@ -354,18 +478,21 @@ int sqlite3Fts3InitHashTable(
}
#endif
- if( rc!=SQLITE_OK
- || (rc = sqlite3_create_function(db, zName, 1, any, p, scalarFunc, 0, 0))
- || (rc = sqlite3_create_function(db, zName, 2, any, p, scalarFunc, 0, 0))
+ if( SQLITE_OK!=rc
+ || SQLITE_OK!=(rc = sqlite3_create_function(db, zName, 1, any, p, scalarFunc, 0, 0))
+ || SQLITE_OK!=(rc = sqlite3_create_function(db, zName, 2, any, p, scalarFunc, 0, 0))
#ifdef SQLITE_TEST
- || (rc = sqlite3_create_function(db, zTest, 2, any, p, testFunc, 0, 0))
- || (rc = sqlite3_create_function(db, zTest, 3, any, p, testFunc, 0, 0))
- || (rc = sqlite3_create_function(db, zTest2, 0, any, pdb, intTestFunc, 0, 0))
+ || SQLITE_OK!=(rc = sqlite3_create_function(db, zTest, 2, any, p, testFunc, 0, 0))
+ || SQLITE_OK!=(rc = sqlite3_create_function(db, zTest, 3, any, p, testFunc, 0, 0))
+ || SQLITE_OK!=(rc = sqlite3_create_function(db, zTest2, 0, any, pdb, intTestFunc, 0, 0))
#endif
- );
+ );
+#ifdef SQLITE_TEST
sqlite3_free(zTest);
sqlite3_free(zTest2);
+#endif
+
return rc;
}
diff --git a/third_party/sqlite/src/ext/fts3/fts3_tokenizer.h b/third_party/sqlite/src/ext/fts3/fts3_tokenizer.h
index 906303d..6156445 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_tokenizer.h
+++ b/third_party/sqlite/src/ext/fts3/fts3_tokenizer.h
@@ -145,4 +145,8 @@ struct sqlite3_tokenizer_cursor {
/* Tokenizer implementations will typically add additional fields */
};
+int fts3_global_term_cnt(int iTerm, int iCol);
+int fts3_term_cnt(int iTerm, int iCol);
+
+
#endif /* _FTS3_TOKENIZER_H_ */
diff --git a/third_party/sqlite/src/ext/fts3/fts3_tokenizer1.c b/third_party/sqlite/src/ext/fts3/fts3_tokenizer1.c
index bbc5696..432c35d 100644
--- a/third_party/sqlite/src/ext/fts3/fts3_tokenizer1.c
+++ b/third_party/sqlite/src/ext/fts3/fts3_tokenizer1.c
@@ -24,6 +24,7 @@
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
+#include "fts3Int.h"
#include <assert.h>
#include <stdlib.h>
@@ -48,9 +49,6 @@ typedef struct simple_tokenizer_cursor {
} simple_tokenizer_cursor;
-/* Forward declaration */
-static const sqlite3_tokenizer_module simpleTokenizerModule;
-
static int simpleDelim(simple_tokenizer *t, unsigned char c){
return c<0x80 && t->delim[c];
}
@@ -77,7 +75,7 @@ static int simpleCreate(
** information on the initial create.
*/
if( argc>1 ){
- int i, n = strlen(argv[1]);
+ int i, n = (int)strlen(argv[1]);
for(i=0; i<n; i++){
unsigned char ch = argv[1][i];
/* We explicitly don't support UTF-8 delimiters for now. */
@@ -91,7 +89,7 @@ static int simpleCreate(
/* Mark non-alphanumeric ASCII characters as delimiters */
int i;
for(i=1; i<0x80; i++){
- t->delim[i] = !fts3_isalnum(i);
+ t->delim[i] = !fts3_isalnum(i) ? -1 : 0;
}
}
@@ -120,6 +118,8 @@ static int simpleOpen(
){
simple_tokenizer_cursor *c;
+ UNUSED_PARAMETER(pTokenizer);
+
c = (simple_tokenizer_cursor *) sqlite3_malloc(sizeof(*c));
if( c==NULL ) return SQLITE_NOMEM;
@@ -184,16 +184,18 @@ static int simpleNext(
if( c->iOffset>iStartOffset ){
int i, n = c->iOffset-iStartOffset;
if( n>c->nTokenAllocated ){
+ char *pNew;
c->nTokenAllocated = n+20;
- c->pToken = sqlite3_realloc(c->pToken, c->nTokenAllocated);
- if( c->pToken==NULL ) return SQLITE_NOMEM;
+ pNew = sqlite3_realloc(c->pToken, c->nTokenAllocated);
+ if( !pNew ) return SQLITE_NOMEM;
+ c->pToken = pNew;
}
for(i=0; i<n; i++){
/* TODO(shess) This needs expansion to handle UTF-8
** case-insensitivity.
*/
unsigned char ch = p[iStartOffset+i];
- c->pToken[i] = (ch>='A' && ch<='Z') ? ch-'A'+'a' : ch;
+ c->pToken[i] = (char)((ch>='A' && ch<='Z') ? ch-'A'+'a' : ch);
}
*ppToken = c->pToken;
*pnBytes = n;
diff --git a/third_party/sqlite/src/ext/fts3/fts3_write.c b/third_party/sqlite/src/ext/fts3/fts3_write.c
new file mode 100644
index 0000000..e434360
--- /dev/null
+++ b/third_party/sqlite/src/ext/fts3/fts3_write.c
@@ -0,0 +1,2622 @@
+/*
+** 2009 Oct 23
+**
+** The author disclaims copyright to this source code. In place of
+** a legal notice, here is a blessing:
+**
+** May you do good and not evil.
+** May you find forgiveness for yourself and forgive others.
+** May you share freely, never taking more than you give.
+**
+******************************************************************************
+**
+** This file is part of the SQLite FTS3 extension module. Specifically,
+** this file contains code to insert, update and delete rows from FTS3
+** tables. It also contains code to merge FTS3 b-tree segments. Some
+** of the sub-routines used to merge segments are also used by the query
+** code in fts3.c.
+*/
+
+#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
+
+#include "fts3Int.h"
+#include <string.h>
+#include <assert.h>
+#include <stdlib.h>
+
+typedef struct PendingList PendingList;
+typedef struct SegmentNode SegmentNode;
+typedef struct SegmentWriter SegmentWriter;
+
+/*
+** Data structure used while accumulating terms in the pending-terms hash
+** table. The hash table entry maps from term (a string) to a malloc'd
+** instance of this structure.
+*/
+struct PendingList {
+ int nData;
+ char *aData;
+ int nSpace;
+ sqlite3_int64 iLastDocid;
+ sqlite3_int64 iLastCol;
+ sqlite3_int64 iLastPos;
+};
+
+/*
+** An instance of this structure is used to iterate through the terms on
+** a contiguous set of segment b-tree leaf nodes. Although the details of
+** this structure are only manipulated by code in this file, opaque handles
+** of type Fts3SegReader* are also used by code in fts3.c to iterate through
+** terms when querying the full-text index. See functions:
+**
+** sqlite3Fts3SegReaderNew()
+** sqlite3Fts3SegReaderFree()
+** sqlite3Fts3SegReaderIterate()
+**
+** Methods used to manipulate Fts3SegReader structures:
+**
+** fts3SegReaderNext()
+** fts3SegReaderFirstDocid()
+** fts3SegReaderNextDocid()
+*/
+struct Fts3SegReader {
+ int iIdx; /* Index within level, or 0x7FFFFFFF for PT */
+ sqlite3_int64 iStartBlock;
+ sqlite3_int64 iEndBlock;
+ sqlite3_stmt *pStmt; /* SQL Statement to access leaf nodes */
+ char *aNode; /* Pointer to node data (or NULL) */
+ int nNode; /* Size of buffer at aNode (or 0) */
+ int nTermAlloc; /* Allocated size of zTerm buffer */
+ Fts3HashElem **ppNextElem;
+
+ /* Variables set by fts3SegReaderNext(). These may be read directly
+ ** by the caller. They are valid from the time SegmentReaderNew() returns
+ ** until SegmentReaderNext() returns something other than SQLITE_OK
+ ** (i.e. SQLITE_DONE).
+ */
+ int nTerm; /* Number of bytes in current term */
+ char *zTerm; /* Pointer to current term */
+ char *aDoclist; /* Pointer to doclist of current entry */
+ int nDoclist; /* Size of doclist in current entry */
+
+ /* The following variables are used to iterate through the current doclist */
+ char *pOffsetList;
+ sqlite3_int64 iDocid;
+};
+
+#define fts3SegReaderIsPending(p) ((p)->ppNextElem!=0)
+
+/*
+** An instance of this structure is used to create a segment b-tree in the
+** database. The internal details of this type are only accessed by the
+** following functions:
+**
+** fts3SegWriterAdd()
+** fts3SegWriterFlush()
+** fts3SegWriterFree()
+*/
+struct SegmentWriter {
+ SegmentNode *pTree; /* Pointer to interior tree structure */
+ sqlite3_int64 iFirst; /* First slot in %_segments written */
+ sqlite3_int64 iFree; /* Next free slot in %_segments */
+ char *zTerm; /* Pointer to previous term buffer */
+ int nTerm; /* Number of bytes in zTerm */
+ int nMalloc; /* Size of malloc'd buffer at zMalloc */
+ char *zMalloc; /* Malloc'd space (possibly) used for zTerm */
+ int nSize; /* Size of allocation at aData */
+ int nData; /* Bytes of data in aData */
+ char *aData; /* Pointer to block from malloc() */
+};
+
+/*
+** Type SegmentNode is used by the following three functions to create
+** the interior part of the segment b+-tree structures (everything except
+** the leaf nodes). These functions and type are only ever used by code
+** within the fts3SegWriterXXX() family of functions described above.
+**
+** fts3NodeAddTerm()
+** fts3NodeWrite()
+** fts3NodeFree()
+*/
+struct SegmentNode {
+ SegmentNode *pParent; /* Parent node (or NULL for root node) */
+ SegmentNode *pRight; /* Pointer to right-sibling */
+ SegmentNode *pLeftmost; /* Pointer to left-most node of this depth */
+ int nEntry; /* Number of terms written to node so far */
+ char *zTerm; /* Pointer to previous term buffer */
+ int nTerm; /* Number of bytes in zTerm */
+ int nMalloc; /* Size of malloc'd buffer at zMalloc */
+ char *zMalloc; /* Malloc'd space (possibly) used for zTerm */
+ int nData; /* Bytes of valid data so far */
+ char *aData; /* Node data */
+};
+
+/*
+** Valid values for the second argument to fts3SqlStmt().
+*/
+#define SQL_DELETE_CONTENT 0
+#define SQL_IS_EMPTY 1
+#define SQL_DELETE_ALL_CONTENT 2
+#define SQL_DELETE_ALL_SEGMENTS 3
+#define SQL_DELETE_ALL_SEGDIR 4
+#define SQL_DELETE_ALL_DOCSIZE 5
+#define SQL_DELETE_ALL_STAT 6
+#define SQL_SELECT_CONTENT_BY_ROWID 7
+#define SQL_NEXT_SEGMENT_INDEX 8
+#define SQL_INSERT_SEGMENTS 9
+#define SQL_NEXT_SEGMENTS_ID 10
+#define SQL_INSERT_SEGDIR 11
+#define SQL_SELECT_LEVEL 12
+#define SQL_SELECT_ALL_LEVEL 13
+#define SQL_SELECT_LEVEL_COUNT 14
+#define SQL_SELECT_SEGDIR_COUNT_MAX 15
+#define SQL_DELETE_SEGDIR_BY_LEVEL 16
+#define SQL_DELETE_SEGMENTS_RANGE 17
+#define SQL_CONTENT_INSERT 18
+#define SQL_GET_BLOCK 19
+#define SQL_DELETE_DOCSIZE 20
+#define SQL_REPLACE_DOCSIZE 21
+#define SQL_SELECT_DOCSIZE 22
+#define SQL_SELECT_DOCTOTAL 23
+#define SQL_REPLACE_DOCTOTAL 24
+
+/*
+** This function is used to obtain an SQLite prepared statement handle
+** for the statement identified by the second argument. If successful,
+** *pp is set to the requested statement handle and SQLITE_OK returned.
+** Otherwise, an SQLite error code is returned and *pp is set to 0.
+**
+** If argument apVal is not NULL, then it must point to an array with
+** at least as many entries as the requested statement has bound
+** parameters. The values are bound to the statements parameters before
+** returning.
+*/
+static int fts3SqlStmt(
+ Fts3Table *p, /* Virtual table handle */
+ int eStmt, /* One of the SQL_XXX constants above */
+ sqlite3_stmt **pp, /* OUT: Statement handle */
+ sqlite3_value **apVal /* Values to bind to statement */
+){
+ const char *azSql[] = {
+/* 0 */ "DELETE FROM %Q.'%q_content' WHERE rowid = ?",
+/* 1 */ "SELECT NOT EXISTS(SELECT docid FROM %Q.'%q_content' WHERE rowid!=?)",
+/* 2 */ "DELETE FROM %Q.'%q_content'",
+/* 3 */ "DELETE FROM %Q.'%q_segments'",
+/* 4 */ "DELETE FROM %Q.'%q_segdir'",
+/* 5 */ "DELETE FROM %Q.'%q_docsize'",
+/* 6 */ "DELETE FROM %Q.'%q_stat'",
+/* 7 */ "SELECT * FROM %Q.'%q_content' WHERE rowid=?",
+/* 8 */ "SELECT (SELECT max(idx) FROM %Q.'%q_segdir' WHERE level = ?) + 1",
+/* 9 */ "INSERT INTO %Q.'%q_segments'(blockid, block) VALUES(?, ?)",
+/* 10 */ "SELECT coalesce((SELECT max(blockid) FROM %Q.'%q_segments') + 1, 1)",
+/* 11 */ "INSERT INTO %Q.'%q_segdir' VALUES(?,?,?,?,?,?)",
+
+ /* Return segments in order from oldest to newest.*/
+/* 12 */ "SELECT idx, start_block, leaves_end_block, end_block, root "
+ "FROM %Q.'%q_segdir' WHERE level = ? ORDER BY idx ASC",
+/* 13 */ "SELECT idx, start_block, leaves_end_block, end_block, root "
+ "FROM %Q.'%q_segdir' ORDER BY level DESC, idx ASC",
+
+/* 14 */ "SELECT count(*) FROM %Q.'%q_segdir' WHERE level = ?",
+/* 15 */ "SELECT count(*), max(level) FROM %Q.'%q_segdir'",
+
+/* 16 */ "DELETE FROM %Q.'%q_segdir' WHERE level = ?",
+/* 17 */ "DELETE FROM %Q.'%q_segments' WHERE blockid BETWEEN ? AND ?",
+/* 18 */ "INSERT INTO %Q.'%q_content' VALUES(%z)",
+/* 19 */ "SELECT block FROM %Q.'%q_segments' WHERE blockid = ?",
+/* 20 */ "DELETE FROM %Q.'%q_docsize' WHERE docid = ?",
+/* 21 */ "REPLACE INTO %Q.'%q_docsize' VALUES(?,?)",
+/* 22 */ "SELECT size FROM %Q.'%q_docsize' WHERE docid=?",
+/* 23 */ "SELECT value FROM %Q.'%q_stat' WHERE id=0",
+/* 24 */ "REPLACE INTO %Q.'%q_stat' VALUES(0,?)",
+ };
+ int rc = SQLITE_OK;
+ sqlite3_stmt *pStmt;
+
+ assert( SizeofArray(azSql)==SizeofArray(p->aStmt) );
+ assert( eStmt<SizeofArray(azSql) && eStmt>=0 );
+
+ pStmt = p->aStmt[eStmt];
+ if( !pStmt ){
+ char *zSql;
+ if( eStmt==SQL_CONTENT_INSERT ){
+ int i; /* Iterator variable */
+ char *zVarlist; /* The "?, ?, ..." string */
+ zVarlist = (char *)sqlite3_malloc(2*p->nColumn+2);
+ if( !zVarlist ){
+ *pp = 0;
+ return SQLITE_NOMEM;
+ }
+ zVarlist[0] = '?';
+ zVarlist[p->nColumn*2+1] = '\0';
+ for(i=1; i<=p->nColumn; i++){
+ zVarlist[i*2-1] = ',';
+ zVarlist[i*2] = '?';
+ }
+ zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName, zVarlist);
+ }else{
+ zSql = sqlite3_mprintf(azSql[eStmt], p->zDb, p->zName);
+ }
+ if( !zSql ){
+ rc = SQLITE_NOMEM;
+ }else{
+ rc = sqlite3_prepare_v2(p->db, zSql, -1, &pStmt, NULL);
+ sqlite3_free(zSql);
+ assert( rc==SQLITE_OK || pStmt==0 );
+ p->aStmt[eStmt] = pStmt;
+ }
+ }
+ if( apVal ){
+ int i;
+ int nParam = sqlite3_bind_parameter_count(pStmt);
+ for(i=0; rc==SQLITE_OK && i<nParam; i++){
+ rc = sqlite3_bind_value(pStmt, i+1, apVal[i]);
+ }
+ }
+ *pp = pStmt;
+ return rc;
+}
+
+/*
+** Similar to fts3SqlStmt(). Except, after binding the parameters in
+** array apVal[] to the SQL statement identified by eStmt, the statement
+** is executed.
+**
+** Returns SQLITE_OK if the statement is successfully executed, or an
+** SQLite error code otherwise.
+*/
+static void fts3SqlExec(
+ int *pRC, /* Result code */
+ Fts3Table *p, /* The FTS3 table */
+ int eStmt, /* Index of statement to evaluate */
+ sqlite3_value **apVal /* Parameters to bind */
+){
+ sqlite3_stmt *pStmt;
+ int rc;
+ if( *pRC ) return;
+ rc = fts3SqlStmt(p, eStmt, &pStmt, apVal);
+ if( rc==SQLITE_OK ){
+ sqlite3_step(pStmt);
+ rc = sqlite3_reset(pStmt);
+ }
+ *pRC = rc;
+}
+
+
+/*
+** Read a single block from the %_segments table. If the specified block
+** does not exist, return SQLITE_CORRUPT. If some other error (malloc, IO
+** etc.) occurs, return the appropriate SQLite error code.
+**
+** Otherwise, if successful, set *pzBlock to point to a buffer containing
+** the block read from the database, and *pnBlock to the size of the read
+** block in bytes.
+**
+** WARNING: The returned buffer is only valid until the next call to
+** sqlite3Fts3ReadBlock().
+*/
+int sqlite3Fts3ReadBlock(
+ Fts3Table *p,
+ sqlite3_int64 iBlock,
+ char const **pzBlock,
+ int *pnBlock
+){
+ sqlite3_stmt *pStmt;
+ int rc = fts3SqlStmt(p, SQL_GET_BLOCK, &pStmt, 0);
+ if( rc!=SQLITE_OK ) return rc;
+ sqlite3_reset(pStmt);
+
+ if( pzBlock ){
+ sqlite3_bind_int64(pStmt, 1, iBlock);
+ rc = sqlite3_step(pStmt);
+ if( rc!=SQLITE_ROW ){
+ return (rc==SQLITE_DONE ? SQLITE_CORRUPT : rc);
+ }
+
+ *pnBlock = sqlite3_column_bytes(pStmt, 0);
+ *pzBlock = (char *)sqlite3_column_blob(pStmt, 0);
+ if( sqlite3_column_type(pStmt, 0)!=SQLITE_BLOB ){
+ return SQLITE_CORRUPT;
+ }
+ }
+ return SQLITE_OK;
+}
+
+/*
+** This function ensures that the caller has obtained a shared-cache
+** table-lock on the %_content table. This is required before reading
+** data from the fts3 table. If this lock is not acquired first, then
+** the caller may end up holding read-locks on the %_segments and %_segdir
+** tables, but no read-lock on the %_content table. If this happens
+** a second connection will be able to write to the fts3 table, but
+** attempting to commit those writes might return SQLITE_LOCKED or
+** SQLITE_LOCKED_SHAREDCACHE (because the commit attempts to obtain
+** write-locks on the %_segments and %_segdir ** tables).
+**
+** We try to avoid this because if FTS3 returns any error when committing
+** a transaction, the whole transaction will be rolled back. And this is
+** not what users expect when they get SQLITE_LOCKED_SHAREDCACHE. It can
+** still happen if the user reads data directly from the %_segments or
+** %_segdir tables instead of going through FTS3 though.
+*/
+int sqlite3Fts3ReadLock(Fts3Table *p){
+ int rc; /* Return code */
+ sqlite3_stmt *pStmt; /* Statement used to obtain lock */
+
+ rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pStmt, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3_bind_null(pStmt, 1);
+ sqlite3_step(pStmt);
+ rc = sqlite3_reset(pStmt);
+ }
+ return rc;
+}
+
+/*
+** Set *ppStmt to a statement handle that may be used to iterate through
+** all rows in the %_segdir table, from oldest to newest. If successful,
+** return SQLITE_OK. If an error occurs while preparing the statement,
+** return an SQLite error code.
+**
+** There is only ever one instance of this SQL statement compiled for
+** each FTS3 table.
+**
+** The statement returns the following columns from the %_segdir table:
+**
+** 0: idx
+** 1: start_block
+** 2: leaves_end_block
+** 3: end_block
+** 4: root
+*/
+int sqlite3Fts3AllSegdirs(Fts3Table *p, sqlite3_stmt **ppStmt){
+ return fts3SqlStmt(p, SQL_SELECT_ALL_LEVEL, ppStmt, 0);
+}
+
+
+/*
+** Append a single varint to a PendingList buffer. SQLITE_OK is returned
+** if successful, or an SQLite error code otherwise.
+**
+** This function also serves to allocate the PendingList structure itself.
+** For example, to create a new PendingList structure containing two
+** varints:
+**
+** PendingList *p = 0;
+** fts3PendingListAppendVarint(&p, 1);
+** fts3PendingListAppendVarint(&p, 2);
+*/
+static int fts3PendingListAppendVarint(
+ PendingList **pp, /* IN/OUT: Pointer to PendingList struct */
+ sqlite3_int64 i /* Value to append to data */
+){
+ PendingList *p = *pp;
+
+ /* Allocate or grow the PendingList as required. */
+ if( !p ){
+ p = sqlite3_malloc(sizeof(*p) + 100);
+ if( !p ){
+ return SQLITE_NOMEM;
+ }
+ p->nSpace = 100;
+ p->aData = (char *)&p[1];
+ p->nData = 0;
+ }
+ else if( p->nData+FTS3_VARINT_MAX+1>p->nSpace ){
+ int nNew = p->nSpace * 2;
+ p = sqlite3_realloc(p, sizeof(*p) + nNew);
+ if( !p ){
+ sqlite3_free(*pp);
+ *pp = 0;
+ return SQLITE_NOMEM;
+ }
+ p->nSpace = nNew;
+ p->aData = (char *)&p[1];
+ }
+
+ /* Append the new serialized varint to the end of the list. */
+ p->nData += sqlite3Fts3PutVarint(&p->aData[p->nData], i);
+ p->aData[p->nData] = '\0';
+ *pp = p;
+ return SQLITE_OK;
+}
+
+/*
+** Add a docid/column/position entry to a PendingList structure. Non-zero
+** is returned if the structure is sqlite3_realloced as part of adding
+** the entry. Otherwise, zero.
+**
+** If an OOM error occurs, *pRc is set to SQLITE_NOMEM before returning.
+** Zero is always returned in this case. Otherwise, if no OOM error occurs,
+** it is set to SQLITE_OK.
+*/
+static int fts3PendingListAppend(
+ PendingList **pp, /* IN/OUT: PendingList structure */
+ sqlite3_int64 iDocid, /* Docid for entry to add */
+ sqlite3_int64 iCol, /* Column for entry to add */
+ sqlite3_int64 iPos, /* Position of term for entry to add */
+ int *pRc /* OUT: Return code */
+){
+ PendingList *p = *pp;
+ int rc = SQLITE_OK;
+
+ assert( !p || p->iLastDocid<=iDocid );
+
+ if( !p || p->iLastDocid!=iDocid ){
+ sqlite3_int64 iDelta = iDocid - (p ? p->iLastDocid : 0);
+ if( p ){
+ assert( p->nData<p->nSpace );
+ assert( p->aData[p->nData]==0 );
+ p->nData++;
+ }
+ if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iDelta)) ){
+ goto pendinglistappend_out;
+ }
+ p->iLastCol = -1;
+ p->iLastPos = 0;
+ p->iLastDocid = iDocid;
+ }
+ if( iCol>0 && p->iLastCol!=iCol ){
+ if( SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, 1))
+ || SQLITE_OK!=(rc = fts3PendingListAppendVarint(&p, iCol))
+ ){
+ goto pendinglistappend_out;
+ }
+ p->iLastCol = iCol;
+ p->iLastPos = 0;
+ }
+ if( iCol>=0 ){
+ assert( iPos>p->iLastPos || (iPos==0 && p->iLastPos==0) );
+ rc = fts3PendingListAppendVarint(&p, 2+iPos-p->iLastPos);
+ if( rc==SQLITE_OK ){
+ p->iLastPos = iPos;
+ }
+ }
+
+ pendinglistappend_out:
+ *pRc = rc;
+ if( p!=*pp ){
+ *pp = p;
+ return 1;
+ }
+ return 0;
+}
+
+/*
+** Tokenize the nul-terminated string zText and add all tokens to the
+** pending-terms hash-table. The docid used is that currently stored in
+** p->iPrevDocid, and the column is specified by argument iCol.
+**
+** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code.
+*/
+static int fts3PendingTermsAdd(
+ Fts3Table *p, /* FTS table into which text will be inserted */
+ const char *zText, /* Text of document to be inseted */
+ int iCol, /* Column number into which text is inserted */
+ u32 *pnWord /* OUT: Number of tokens inserted */
+){
+ int rc;
+ int iStart;
+ int iEnd;
+ int iPos;
+ int nWord = 0;
+
+ char const *zToken;
+ int nToken;
+
+ sqlite3_tokenizer *pTokenizer = p->pTokenizer;
+ sqlite3_tokenizer_module const *pModule = pTokenizer->pModule;
+ sqlite3_tokenizer_cursor *pCsr;
+ int (*xNext)(sqlite3_tokenizer_cursor *pCursor,
+ const char**,int*,int*,int*,int*);
+
+ assert( pTokenizer && pModule );
+
+ rc = pModule->xOpen(pTokenizer, zText, -1, &pCsr);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ pCsr->pTokenizer = pTokenizer;
+
+ xNext = pModule->xNext;
+ while( SQLITE_OK==rc
+ && SQLITE_OK==(rc = xNext(pCsr, &zToken, &nToken, &iStart, &iEnd, &iPos))
+ ){
+ PendingList *pList;
+
+ if( iPos>=nWord ) nWord = iPos+1;
+
+ /* Positions cannot be negative; we use -1 as a terminator internally.
+ ** Tokens must have a non-zero length.
+ */
+ if( iPos<0 || !zToken || nToken<=0 ){
+ rc = SQLITE_ERROR;
+ break;
+ }
+
+ pList = (PendingList *)fts3HashFind(&p->pendingTerms, zToken, nToken);
+ if( pList ){
+ p->nPendingData -= (pList->nData + nToken + sizeof(Fts3HashElem));
+ }
+ if( fts3PendingListAppend(&pList, p->iPrevDocid, iCol, iPos, &rc) ){
+ if( pList==fts3HashInsert(&p->pendingTerms, zToken, nToken, pList) ){
+ /* Malloc failed while inserting the new entry. This can only
+ ** happen if there was no previous entry for this token.
+ */
+ assert( 0==fts3HashFind(&p->pendingTerms, zToken, nToken) );
+ sqlite3_free(pList);
+ rc = SQLITE_NOMEM;
+ }
+ }
+ if( rc==SQLITE_OK ){
+ p->nPendingData += (pList->nData + nToken + sizeof(Fts3HashElem));
+ }
+ }
+
+ pModule->xClose(pCsr);
+ *pnWord = nWord;
+ return (rc==SQLITE_DONE ? SQLITE_OK : rc);
+}
+
+/*
+** Calling this function indicates that subsequent calls to
+** fts3PendingTermsAdd() are to add term/position-list pairs for the
+** contents of the document with docid iDocid.
+*/
+static int fts3PendingTermsDocid(Fts3Table *p, sqlite_int64 iDocid){
+ /* TODO(shess) Explore whether partially flushing the buffer on
+ ** forced-flush would provide better performance. I suspect that if
+ ** we ordered the doclists by size and flushed the largest until the
+ ** buffer was half empty, that would let the less frequent terms
+ ** generate longer doclists.
+ */
+ if( iDocid<=p->iPrevDocid || p->nPendingData>p->nMaxPendingData ){
+ int rc = sqlite3Fts3PendingTermsFlush(p);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+ p->iPrevDocid = iDocid;
+ return SQLITE_OK;
+}
+
+void sqlite3Fts3PendingTermsClear(Fts3Table *p){
+ Fts3HashElem *pElem;
+ for(pElem=fts3HashFirst(&p->pendingTerms); pElem; pElem=fts3HashNext(pElem)){
+ sqlite3_free(fts3HashData(pElem));
+ }
+ fts3HashClear(&p->pendingTerms);
+ p->nPendingData = 0;
+}
+
+/*
+** This function is called by the xUpdate() method as part of an INSERT
+** operation. It adds entries for each term in the new record to the
+** pendingTerms hash table.
+**
+** Argument apVal is the same as the similarly named argument passed to
+** fts3InsertData(). Parameter iDocid is the docid of the new row.
+*/
+static int fts3InsertTerms(Fts3Table *p, sqlite3_value **apVal, u32 *aSz){
+ int i; /* Iterator variable */
+ for(i=2; i<p->nColumn+2; i++){
+ const char *zText = (const char *)sqlite3_value_text(apVal[i]);
+ if( zText ){
+ int rc = fts3PendingTermsAdd(p, zText, i-2, &aSz[i-2]);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ }
+ }
+ return SQLITE_OK;
+}
+
+/*
+** This function is called by the xUpdate() method for an INSERT operation.
+** The apVal parameter is passed a copy of the apVal argument passed by
+** SQLite to the xUpdate() method. i.e:
+**
+** apVal[0] Not used for INSERT.
+** apVal[1] rowid
+** apVal[2] Left-most user-defined column
+** ...
+** apVal[p->nColumn+1] Right-most user-defined column
+** apVal[p->nColumn+2] Hidden column with same name as table
+** apVal[p->nColumn+3] Hidden "docid" column (alias for rowid)
+*/
+static int fts3InsertData(
+ Fts3Table *p, /* Full-text table */
+ sqlite3_value **apVal, /* Array of values to insert */
+ sqlite3_int64 *piDocid /* OUT: Docid for row just inserted */
+){
+ int rc; /* Return code */
+ sqlite3_stmt *pContentInsert; /* INSERT INTO %_content VALUES(...) */
+
+ /* Locate the statement handle used to insert data into the %_content
+ ** table. The SQL for this statement is:
+ **
+ ** INSERT INTO %_content VALUES(?, ?, ?, ...)
+ **
+ ** The statement features N '?' variables, where N is the number of user
+ ** defined columns in the FTS3 table, plus one for the docid field.
+ */
+ rc = fts3SqlStmt(p, SQL_CONTENT_INSERT, &pContentInsert, &apVal[1]);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+
+ /* There is a quirk here. The users INSERT statement may have specified
+ ** a value for the "rowid" field, for the "docid" field, or for both.
+ ** Which is a problem, since "rowid" and "docid" are aliases for the
+ ** same value. For example:
+ **
+ ** INSERT INTO fts3tbl(rowid, docid) VALUES(1, 2);
+ **
+ ** In FTS3, this is an error. It is an error to specify non-NULL values
+ ** for both docid and some other rowid alias.
+ */
+ if( SQLITE_NULL!=sqlite3_value_type(apVal[3+p->nColumn]) ){
+ if( SQLITE_NULL==sqlite3_value_type(apVal[0])
+ && SQLITE_NULL!=sqlite3_value_type(apVal[1])
+ ){
+ /* A rowid/docid conflict. */
+ return SQLITE_ERROR;
+ }
+ rc = sqlite3_bind_value(pContentInsert, 1, apVal[3+p->nColumn]);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+
+ /* Execute the statement to insert the record. Set *piDocid to the
+ ** new docid value.
+ */
+ sqlite3_step(pContentInsert);
+ rc = sqlite3_reset(pContentInsert);
+
+ *piDocid = sqlite3_last_insert_rowid(p->db);
+ return rc;
+}
+
+
+
+/*
+** Remove all data from the FTS3 table. Clear the hash table containing
+** pending terms.
+*/
+static int fts3DeleteAll(Fts3Table *p){
+ int rc = SQLITE_OK; /* Return code */
+
+ /* Discard the contents of the pending-terms hash table. */
+ sqlite3Fts3PendingTermsClear(p);
+
+ /* Delete everything from the %_content, %_segments and %_segdir tables. */
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_CONTENT, 0);
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGMENTS, 0);
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGDIR, 0);
+ if( p->bHasDocsize ){
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_DOCSIZE, 0);
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_STAT, 0);
+ }
+ return rc;
+}
+
+/*
+** The first element in the apVal[] array is assumed to contain the docid
+** (an integer) of a row about to be deleted. Remove all terms from the
+** full-text index.
+*/
+static void fts3DeleteTerms(
+ int *pRC, /* Result code */
+ Fts3Table *p, /* The FTS table to delete from */
+ sqlite3_value **apVal, /* apVal[] contains the docid to be deleted */
+ u32 *aSz /* Sizes of deleted document written here */
+){
+ int rc;
+ sqlite3_stmt *pSelect;
+
+ if( *pRC ) return;
+ rc = fts3SqlStmt(p, SQL_SELECT_CONTENT_BY_ROWID, &pSelect, apVal);
+ if( rc==SQLITE_OK ){
+ if( SQLITE_ROW==sqlite3_step(pSelect) ){
+ int i;
+ for(i=1; i<=p->nColumn; i++){
+ const char *zText = (const char *)sqlite3_column_text(pSelect, i);
+ rc = fts3PendingTermsAdd(p, zText, -1, &aSz[i-1]);
+ if( rc!=SQLITE_OK ){
+ sqlite3_reset(pSelect);
+ *pRC = rc;
+ return;
+ }
+ }
+ }
+ rc = sqlite3_reset(pSelect);
+ }else{
+ sqlite3_reset(pSelect);
+ }
+ *pRC = rc;
+}
+
+/*
+** Forward declaration to account for the circular dependency between
+** functions fts3SegmentMerge() and fts3AllocateSegdirIdx().
+*/
+static int fts3SegmentMerge(Fts3Table *, int);
+
+/*
+** This function allocates a new level iLevel index in the segdir table.
+** Usually, indexes are allocated within a level sequentially starting
+** with 0, so the allocated index is one greater than the value returned
+** by:
+**
+** SELECT max(idx) FROM %_segdir WHERE level = :iLevel
+**
+** However, if there are already FTS3_MERGE_COUNT indexes at the requested
+** level, they are merged into a single level (iLevel+1) segment and the
+** allocated index is 0.
+**
+** If successful, *piIdx is set to the allocated index slot and SQLITE_OK
+** returned. Otherwise, an SQLite error code is returned.
+*/
+static int fts3AllocateSegdirIdx(Fts3Table *p, int iLevel, int *piIdx){
+ int rc; /* Return Code */
+ sqlite3_stmt *pNextIdx; /* Query for next idx at level iLevel */
+ int iNext = 0; /* Result of query pNextIdx */
+
+ /* Set variable iNext to the next available segdir index at level iLevel. */
+ rc = fts3SqlStmt(p, SQL_NEXT_SEGMENT_INDEX, &pNextIdx, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3_bind_int(pNextIdx, 1, iLevel);
+ if( SQLITE_ROW==sqlite3_step(pNextIdx) ){
+ iNext = sqlite3_column_int(pNextIdx, 0);
+ }
+ rc = sqlite3_reset(pNextIdx);
+ }
+
+ if( rc==SQLITE_OK ){
+ /* If iNext is FTS3_MERGE_COUNT, indicating that level iLevel is already
+ ** full, merge all segments in level iLevel into a single iLevel+1
+ ** segment and allocate (newly freed) index 0 at level iLevel. Otherwise,
+ ** if iNext is less than FTS3_MERGE_COUNT, allocate index iNext.
+ */
+ if( iNext>=FTS3_MERGE_COUNT ){
+ rc = fts3SegmentMerge(p, iLevel);
+ *piIdx = 0;
+ }else{
+ *piIdx = iNext;
+ }
+ }
+
+ return rc;
+}
+
+/*
+** Move the iterator passed as the first argument to the next term in the
+** segment. If successful, SQLITE_OK is returned. If there is no next term,
+** SQLITE_DONE. Otherwise, an SQLite error code.
+*/
+static int fts3SegReaderNext(Fts3SegReader *pReader){
+ char *pNext; /* Cursor variable */
+ int nPrefix; /* Number of bytes in term prefix */
+ int nSuffix; /* Number of bytes in term suffix */
+
+ if( !pReader->aDoclist ){
+ pNext = pReader->aNode;
+ }else{
+ pNext = &pReader->aDoclist[pReader->nDoclist];
+ }
+
+ if( !pNext || pNext>=&pReader->aNode[pReader->nNode] ){
+ int rc;
+ if( fts3SegReaderIsPending(pReader) ){
+ Fts3HashElem *pElem = *(pReader->ppNextElem);
+ if( pElem==0 ){
+ pReader->aNode = 0;
+ }else{
+ PendingList *pList = (PendingList *)fts3HashData(pElem);
+ pReader->zTerm = (char *)fts3HashKey(pElem);
+ pReader->nTerm = fts3HashKeysize(pElem);
+ pReader->nNode = pReader->nDoclist = pList->nData + 1;
+ pReader->aNode = pReader->aDoclist = pList->aData;
+ pReader->ppNextElem++;
+ assert( pReader->aNode );
+ }
+ return SQLITE_OK;
+ }
+ if( !pReader->pStmt ){
+ pReader->aNode = 0;
+ return SQLITE_OK;
+ }
+ rc = sqlite3_step(pReader->pStmt);
+ if( rc!=SQLITE_ROW ){
+ pReader->aNode = 0;
+ return (rc==SQLITE_DONE ? SQLITE_OK : rc);
+ }
+ pReader->nNode = sqlite3_column_bytes(pReader->pStmt, 0);
+ pReader->aNode = (char *)sqlite3_column_blob(pReader->pStmt, 0);
+ pNext = pReader->aNode;
+ }
+
+ pNext += sqlite3Fts3GetVarint32(pNext, &nPrefix);
+ pNext += sqlite3Fts3GetVarint32(pNext, &nSuffix);
+
+ if( nPrefix+nSuffix>pReader->nTermAlloc ){
+ int nNew = (nPrefix+nSuffix)*2;
+ char *zNew = sqlite3_realloc(pReader->zTerm, nNew);
+ if( !zNew ){
+ return SQLITE_NOMEM;
+ }
+ pReader->zTerm = zNew;
+ pReader->nTermAlloc = nNew;
+ }
+ memcpy(&pReader->zTerm[nPrefix], pNext, nSuffix);
+ pReader->nTerm = nPrefix+nSuffix;
+ pNext += nSuffix;
+ pNext += sqlite3Fts3GetVarint32(pNext, &pReader->nDoclist);
+ assert( pNext<&pReader->aNode[pReader->nNode] );
+ pReader->aDoclist = pNext;
+ pReader->pOffsetList = 0;
+ return SQLITE_OK;
+}
+
+/*
+** Set the SegReader to point to the first docid in the doclist associated
+** with the current term.
+*/
+static void fts3SegReaderFirstDocid(Fts3SegReader *pReader){
+ int n;
+ assert( pReader->aDoclist );
+ assert( !pReader->pOffsetList );
+ n = sqlite3Fts3GetVarint(pReader->aDoclist, &pReader->iDocid);
+ pReader->pOffsetList = &pReader->aDoclist[n];
+}
+
+/*
+** Advance the SegReader to point to the next docid in the doclist
+** associated with the current term.
+**
+** If arguments ppOffsetList and pnOffsetList are not NULL, then
+** *ppOffsetList is set to point to the first column-offset list
+** in the doclist entry (i.e. immediately past the docid varint).
+** *pnOffsetList is set to the length of the set of column-offset
+** lists, not including the nul-terminator byte. For example:
+*/
+static void fts3SegReaderNextDocid(
+ Fts3SegReader *pReader,
+ char **ppOffsetList,
+ int *pnOffsetList
+){
+ char *p = pReader->pOffsetList;
+ char c = 0;
+
+ /* Pointer p currently points at the first byte of an offset list. The
+ ** following two lines advance it to point one byte past the end of
+ ** the same offset list.
+ */
+ while( *p | c ) c = *p++ & 0x80;
+ p++;
+
+ /* If required, populate the output variables with a pointer to and the
+ ** size of the previous offset-list.
+ */
+ if( ppOffsetList ){
+ *ppOffsetList = pReader->pOffsetList;
+ *pnOffsetList = (int)(p - pReader->pOffsetList - 1);
+ }
+
+ /* If there are no more entries in the doclist, set pOffsetList to
+ ** NULL. Otherwise, set Fts3SegReader.iDocid to the next docid and
+ ** Fts3SegReader.pOffsetList to point to the next offset list before
+ ** returning.
+ */
+ if( p>=&pReader->aDoclist[pReader->nDoclist] ){
+ pReader->pOffsetList = 0;
+ }else{
+ sqlite3_int64 iDelta;
+ pReader->pOffsetList = p + sqlite3Fts3GetVarint(p, &iDelta);
+ pReader->iDocid += iDelta;
+ }
+}
+
+/*
+** Free all allocations associated with the iterator passed as the
+** second argument.
+*/
+void sqlite3Fts3SegReaderFree(Fts3Table *p, Fts3SegReader *pReader){
+ if( pReader ){
+ if( pReader->pStmt ){
+ /* Move the leaf-range SELECT statement to the aLeavesStmt[] array,
+ ** so that it can be reused when required by another query.
+ */
+ assert( p->nLeavesStmt<p->nLeavesTotal );
+ sqlite3_reset(pReader->pStmt);
+ p->aLeavesStmt[p->nLeavesStmt++] = pReader->pStmt;
+ }
+ if( !fts3SegReaderIsPending(pReader) ){
+ sqlite3_free(pReader->zTerm);
+ }
+ sqlite3_free(pReader);
+ }
+}
+
+/*
+** Allocate a new SegReader object.
+*/
+int sqlite3Fts3SegReaderNew(
+ Fts3Table *p, /* Virtual table handle */
+ int iAge, /* Segment "age". */
+ sqlite3_int64 iStartLeaf, /* First leaf to traverse */
+ sqlite3_int64 iEndLeaf, /* Final leaf to traverse */
+ sqlite3_int64 iEndBlock, /* Final block of segment */
+ const char *zRoot, /* Buffer containing root node */
+ int nRoot, /* Size of buffer containing root node */
+ Fts3SegReader **ppReader /* OUT: Allocated Fts3SegReader */
+){
+ int rc = SQLITE_OK; /* Return code */
+ Fts3SegReader *pReader; /* Newly allocated SegReader object */
+ int nExtra = 0; /* Bytes to allocate segment root node */
+
+ if( iStartLeaf==0 ){
+ nExtra = nRoot;
+ }
+
+ pReader = (Fts3SegReader *)sqlite3_malloc(sizeof(Fts3SegReader) + nExtra);
+ if( !pReader ){
+ return SQLITE_NOMEM;
+ }
+ memset(pReader, 0, sizeof(Fts3SegReader));
+ pReader->iStartBlock = iStartLeaf;
+ pReader->iIdx = iAge;
+ pReader->iEndBlock = iEndBlock;
+
+ if( nExtra ){
+ /* The entire segment is stored in the root node. */
+ pReader->aNode = (char *)&pReader[1];
+ pReader->nNode = nRoot;
+ memcpy(pReader->aNode, zRoot, nRoot);
+ }else{
+ /* If the text of the SQL statement to iterate through a contiguous
+ ** set of entries in the %_segments table has not yet been composed,
+ ** compose it now.
+ */
+ if( !p->zSelectLeaves ){
+ p->zSelectLeaves = sqlite3_mprintf(
+ "SELECT block FROM %Q.'%q_segments' WHERE blockid BETWEEN ? AND ? "
+ "ORDER BY blockid", p->zDb, p->zName
+ );
+ if( !p->zSelectLeaves ){
+ rc = SQLITE_NOMEM;
+ goto finished;
+ }
+ }
+
+ /* If there are no free statements in the aLeavesStmt[] array, prepare
+ ** a new statement now. Otherwise, reuse a prepared statement from
+ ** aLeavesStmt[].
+ */
+ if( p->nLeavesStmt==0 ){
+ if( p->nLeavesTotal==p->nLeavesAlloc ){
+ int nNew = p->nLeavesAlloc + 16;
+ sqlite3_stmt **aNew = (sqlite3_stmt **)sqlite3_realloc(
+ p->aLeavesStmt, nNew*sizeof(sqlite3_stmt *)
+ );
+ if( !aNew ){
+ rc = SQLITE_NOMEM;
+ goto finished;
+ }
+ p->nLeavesAlloc = nNew;
+ p->aLeavesStmt = aNew;
+ }
+ rc = sqlite3_prepare_v2(p->db, p->zSelectLeaves, -1, &pReader->pStmt, 0);
+ if( rc!=SQLITE_OK ){
+ goto finished;
+ }
+ p->nLeavesTotal++;
+ }else{
+ pReader->pStmt = p->aLeavesStmt[--p->nLeavesStmt];
+ }
+
+ /* Bind the start and end leaf blockids to the prepared SQL statement. */
+ sqlite3_bind_int64(pReader->pStmt, 1, iStartLeaf);
+ sqlite3_bind_int64(pReader->pStmt, 2, iEndLeaf);
+ }
+ rc = fts3SegReaderNext(pReader);
+
+ finished:
+ if( rc==SQLITE_OK ){
+ *ppReader = pReader;
+ }else{
+ sqlite3Fts3SegReaderFree(p, pReader);
+ }
+ return rc;
+}
+
+/*
+** This is a comparison function used as a qsort() callback when sorting
+** an array of pending terms by term. This occurs as part of flushing
+** the contents of the pending-terms hash table to the database.
+*/
+static int fts3CompareElemByTerm(const void *lhs, const void *rhs){
+ char *z1 = fts3HashKey(*(Fts3HashElem **)lhs);
+ char *z2 = fts3HashKey(*(Fts3HashElem **)rhs);
+ int n1 = fts3HashKeysize(*(Fts3HashElem **)lhs);
+ int n2 = fts3HashKeysize(*(Fts3HashElem **)rhs);
+
+ int n = (n1<n2 ? n1 : n2);
+ int c = memcmp(z1, z2, n);
+ if( c==0 ){
+ c = n1 - n2;
+ }
+ return c;
+}
+
+/*
+** This function is used to allocate an Fts3SegReader that iterates through
+** a subset of the terms stored in the Fts3Table.pendingTerms array.
+*/
+int sqlite3Fts3SegReaderPending(
+ Fts3Table *p, /* Virtual table handle */
+ const char *zTerm, /* Term to search for */
+ int nTerm, /* Size of buffer zTerm */
+ int isPrefix, /* True for a term-prefix query */
+ Fts3SegReader **ppReader /* OUT: SegReader for pending-terms */
+){
+ Fts3SegReader *pReader = 0; /* Fts3SegReader object to return */
+ Fts3HashElem **aElem = 0; /* Array of term hash entries to scan */
+ int nElem = 0; /* Size of array at aElem */
+ int rc = SQLITE_OK; /* Return Code */
+
+ if( isPrefix ){
+ int nAlloc = 0; /* Size of allocated array at aElem */
+ Fts3HashElem *pE = 0; /* Iterator variable */
+
+ for(pE=fts3HashFirst(&p->pendingTerms); pE; pE=fts3HashNext(pE)){
+ char *zKey = (char *)fts3HashKey(pE);
+ int nKey = fts3HashKeysize(pE);
+ if( nTerm==0 || (nKey>=nTerm && 0==memcmp(zKey, zTerm, nTerm)) ){
+ if( nElem==nAlloc ){
+ Fts3HashElem **aElem2;
+ nAlloc += 16;
+ aElem2 = (Fts3HashElem **)sqlite3_realloc(
+ aElem, nAlloc*sizeof(Fts3HashElem *)
+ );
+ if( !aElem2 ){
+ rc = SQLITE_NOMEM;
+ nElem = 0;
+ break;
+ }
+ aElem = aElem2;
+ }
+ aElem[nElem++] = pE;
+ }
+ }
+
+ /* If more than one term matches the prefix, sort the Fts3HashElem
+ ** objects in term order using qsort(). This uses the same comparison
+ ** callback as is used when flushing terms to disk.
+ */
+ if( nElem>1 ){
+ qsort(aElem, nElem, sizeof(Fts3HashElem *), fts3CompareElemByTerm);
+ }
+
+ }else{
+ Fts3HashElem *pE = fts3HashFindElem(&p->pendingTerms, zTerm, nTerm);
+ if( pE ){
+ aElem = &pE;
+ nElem = 1;
+ }
+ }
+
+ if( nElem>0 ){
+ int nByte = sizeof(Fts3SegReader) + (nElem+1)*sizeof(Fts3HashElem *);
+ pReader = (Fts3SegReader *)sqlite3_malloc(nByte);
+ if( !pReader ){
+ rc = SQLITE_NOMEM;
+ }else{
+ memset(pReader, 0, nByte);
+ pReader->iIdx = 0x7FFFFFFF;
+ pReader->ppNextElem = (Fts3HashElem **)&pReader[1];
+ memcpy(pReader->ppNextElem, aElem, nElem*sizeof(Fts3HashElem *));
+ fts3SegReaderNext(pReader);
+ }
+ }
+
+ if( isPrefix ){
+ sqlite3_free(aElem);
+ }
+ *ppReader = pReader;
+ return rc;
+}
+
+
+/*
+** The second argument to this function is expected to be a statement of
+** the form:
+**
+** SELECT
+** idx, -- col 0
+** start_block, -- col 1
+** leaves_end_block, -- col 2
+** end_block, -- col 3
+** root -- col 4
+** FROM %_segdir ...
+**
+** This function allocates and initializes a Fts3SegReader structure to
+** iterate through the terms stored in the segment identified by the
+** current row that pStmt is pointing to.
+**
+** If successful, the Fts3SegReader is left pointing to the first term
+** in the segment and SQLITE_OK is returned. Otherwise, an SQLite error
+** code is returned.
+*/
+static int fts3SegReaderNew(
+ Fts3Table *p, /* Virtual table handle */
+ sqlite3_stmt *pStmt, /* See above */
+ int iAge, /* Segment "age". */
+ Fts3SegReader **ppReader /* OUT: Allocated Fts3SegReader */
+){
+ return sqlite3Fts3SegReaderNew(p, iAge,
+ sqlite3_column_int64(pStmt, 1),
+ sqlite3_column_int64(pStmt, 2),
+ sqlite3_column_int64(pStmt, 3),
+ sqlite3_column_blob(pStmt, 4),
+ sqlite3_column_bytes(pStmt, 4),
+ ppReader
+ );
+}
+
+/*
+** Compare the entries pointed to by two Fts3SegReader structures.
+** Comparison is as follows:
+**
+** 1) EOF is greater than not EOF.
+**
+** 2) The current terms (if any) are compared using memcmp(). If one
+** term is a prefix of another, the longer term is considered the
+** larger.
+**
+** 3) By segment age. An older segment is considered larger.
+*/
+static int fts3SegReaderCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){
+ int rc;
+ if( pLhs->aNode && pRhs->aNode ){
+ int rc2 = pLhs->nTerm - pRhs->nTerm;
+ if( rc2<0 ){
+ rc = memcmp(pLhs->zTerm, pRhs->zTerm, pLhs->nTerm);
+ }else{
+ rc = memcmp(pLhs->zTerm, pRhs->zTerm, pRhs->nTerm);
+ }
+ if( rc==0 ){
+ rc = rc2;
+ }
+ }else{
+ rc = (pLhs->aNode==0) - (pRhs->aNode==0);
+ }
+ if( rc==0 ){
+ rc = pRhs->iIdx - pLhs->iIdx;
+ }
+ assert( rc!=0 );
+ return rc;
+}
+
+/*
+** A different comparison function for SegReader structures. In this
+** version, it is assumed that each SegReader points to an entry in
+** a doclist for identical terms. Comparison is made as follows:
+**
+** 1) EOF (end of doclist in this case) is greater than not EOF.
+**
+** 2) By current docid.
+**
+** 3) By segment age. An older segment is considered larger.
+*/
+static int fts3SegReaderDoclistCmp(Fts3SegReader *pLhs, Fts3SegReader *pRhs){
+ int rc = (pLhs->pOffsetList==0)-(pRhs->pOffsetList==0);
+ if( rc==0 ){
+ if( pLhs->iDocid==pRhs->iDocid ){
+ rc = pRhs->iIdx - pLhs->iIdx;
+ }else{
+ rc = (pLhs->iDocid > pRhs->iDocid) ? 1 : -1;
+ }
+ }
+ assert( pLhs->aNode && pRhs->aNode );
+ return rc;
+}
+
+/*
+** Compare the term that the Fts3SegReader object passed as the first argument
+** points to with the term specified by arguments zTerm and nTerm.
+**
+** If the pSeg iterator is already at EOF, return 0. Otherwise, return
+** -ve if the pSeg term is less than zTerm/nTerm, 0 if the two terms are
+** equal, or +ve if the pSeg term is greater than zTerm/nTerm.
+*/
+static int fts3SegReaderTermCmp(
+ Fts3SegReader *pSeg, /* Segment reader object */
+ const char *zTerm, /* Term to compare to */
+ int nTerm /* Size of term zTerm in bytes */
+){
+ int res = 0;
+ if( pSeg->aNode ){
+ if( pSeg->nTerm>nTerm ){
+ res = memcmp(pSeg->zTerm, zTerm, nTerm);
+ }else{
+ res = memcmp(pSeg->zTerm, zTerm, pSeg->nTerm);
+ }
+ if( res==0 ){
+ res = pSeg->nTerm-nTerm;
+ }
+ }
+ return res;
+}
+
+/*
+** Argument apSegment is an array of nSegment elements. It is known that
+** the final (nSegment-nSuspect) members are already in sorted order
+** (according to the comparison function provided). This function shuffles
+** the array around until all entries are in sorted order.
+*/
+static void fts3SegReaderSort(
+ Fts3SegReader **apSegment, /* Array to sort entries of */
+ int nSegment, /* Size of apSegment array */
+ int nSuspect, /* Unsorted entry count */
+ int (*xCmp)(Fts3SegReader *, Fts3SegReader *) /* Comparison function */
+){
+ int i; /* Iterator variable */
+
+ assert( nSuspect<=nSegment );
+
+ if( nSuspect==nSegment ) nSuspect--;
+ for(i=nSuspect-1; i>=0; i--){
+ int j;
+ for(j=i; j<(nSegment-1); j++){
+ Fts3SegReader *pTmp;
+ if( xCmp(apSegment[j], apSegment[j+1])<0 ) break;
+ pTmp = apSegment[j+1];
+ apSegment[j+1] = apSegment[j];
+ apSegment[j] = pTmp;
+ }
+ }
+
+#ifndef NDEBUG
+ /* Check that the list really is sorted now. */
+ for(i=0; i<(nSuspect-1); i++){
+ assert( xCmp(apSegment[i], apSegment[i+1])<0 );
+ }
+#endif
+}
+
+/*
+** Insert a record into the %_segments table.
+*/
+static int fts3WriteSegment(
+ Fts3Table *p, /* Virtual table handle */
+ sqlite3_int64 iBlock, /* Block id for new block */
+ char *z, /* Pointer to buffer containing block data */
+ int n /* Size of buffer z in bytes */
+){
+ sqlite3_stmt *pStmt;
+ int rc = fts3SqlStmt(p, SQL_INSERT_SEGMENTS, &pStmt, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3_bind_int64(pStmt, 1, iBlock);
+ sqlite3_bind_blob(pStmt, 2, z, n, SQLITE_STATIC);
+ sqlite3_step(pStmt);
+ rc = sqlite3_reset(pStmt);
+ }
+ return rc;
+}
+
+/*
+** Insert a record into the %_segdir table.
+*/
+static int fts3WriteSegdir(
+ Fts3Table *p, /* Virtual table handle */
+ int iLevel, /* Value for "level" field */
+ int iIdx, /* Value for "idx" field */
+ sqlite3_int64 iStartBlock, /* Value for "start_block" field */
+ sqlite3_int64 iLeafEndBlock, /* Value for "leaves_end_block" field */
+ sqlite3_int64 iEndBlock, /* Value for "end_block" field */
+ char *zRoot, /* Blob value for "root" field */
+ int nRoot /* Number of bytes in buffer zRoot */
+){
+ sqlite3_stmt *pStmt;
+ int rc = fts3SqlStmt(p, SQL_INSERT_SEGDIR, &pStmt, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3_bind_int(pStmt, 1, iLevel);
+ sqlite3_bind_int(pStmt, 2, iIdx);
+ sqlite3_bind_int64(pStmt, 3, iStartBlock);
+ sqlite3_bind_int64(pStmt, 4, iLeafEndBlock);
+ sqlite3_bind_int64(pStmt, 5, iEndBlock);
+ sqlite3_bind_blob(pStmt, 6, zRoot, nRoot, SQLITE_STATIC);
+ sqlite3_step(pStmt);
+ rc = sqlite3_reset(pStmt);
+ }
+ return rc;
+}
+
+/*
+** Return the size of the common prefix (if any) shared by zPrev and
+** zNext, in bytes. For example,
+**
+** fts3PrefixCompress("abc", 3, "abcdef", 6) // returns 3
+** fts3PrefixCompress("abX", 3, "abcdef", 6) // returns 2
+** fts3PrefixCompress("abX", 3, "Xbcdef", 6) // returns 0
+*/
+static int fts3PrefixCompress(
+ const char *zPrev, /* Buffer containing previous term */
+ int nPrev, /* Size of buffer zPrev in bytes */
+ const char *zNext, /* Buffer containing next term */
+ int nNext /* Size of buffer zNext in bytes */
+){
+ int n;
+ UNUSED_PARAMETER(nNext);
+ for(n=0; n<nPrev && zPrev[n]==zNext[n]; n++);
+ return n;
+}
+
+/*
+** Add term zTerm to the SegmentNode. It is guaranteed that zTerm is larger
+** (according to memcmp) than the previous term.
+*/
+static int fts3NodeAddTerm(
+ Fts3Table *p, /* Virtual table handle */
+ SegmentNode **ppTree, /* IN/OUT: SegmentNode handle */
+ int isCopyTerm, /* True if zTerm/nTerm is transient */
+ const char *zTerm, /* Pointer to buffer containing term */
+ int nTerm /* Size of term in bytes */
+){
+ SegmentNode *pTree = *ppTree;
+ int rc;
+ SegmentNode *pNew;
+
+ /* First try to append the term to the current node. Return early if
+ ** this is possible.
+ */
+ if( pTree ){
+ int nData = pTree->nData; /* Current size of node in bytes */
+ int nReq = nData; /* Required space after adding zTerm */
+ int nPrefix; /* Number of bytes of prefix compression */
+ int nSuffix; /* Suffix length */
+
+ nPrefix = fts3PrefixCompress(pTree->zTerm, pTree->nTerm, zTerm, nTerm);
+ nSuffix = nTerm-nPrefix;
+
+ nReq += sqlite3Fts3VarintLen(nPrefix)+sqlite3Fts3VarintLen(nSuffix)+nSuffix;
+ if( nReq<=p->nNodeSize || !pTree->zTerm ){
+
+ if( nReq>p->nNodeSize ){
+ /* An unusual case: this is the first term to be added to the node
+ ** and the static node buffer (p->nNodeSize bytes) is not large
+ ** enough. Use a separately malloced buffer instead This wastes
+ ** p->nNodeSize bytes, but since this scenario only comes about when
+ ** the database contain two terms that share a prefix of almost 2KB,
+ ** this is not expected to be a serious problem.
+ */
+ assert( pTree->aData==(char *)&pTree[1] );
+ pTree->aData = (char *)sqlite3_malloc(nReq);
+ if( !pTree->aData ){
+ return SQLITE_NOMEM;
+ }
+ }
+
+ if( pTree->zTerm ){
+ /* There is no prefix-length field for first term in a node */
+ nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nPrefix);
+ }
+
+ nData += sqlite3Fts3PutVarint(&pTree->aData[nData], nSuffix);
+ memcpy(&pTree->aData[nData], &zTerm[nPrefix], nSuffix);
+ pTree->nData = nData + nSuffix;
+ pTree->nEntry++;
+
+ if( isCopyTerm ){
+ if( pTree->nMalloc<nTerm ){
+ char *zNew = sqlite3_realloc(pTree->zMalloc, nTerm*2);
+ if( !zNew ){
+ return SQLITE_NOMEM;
+ }
+ pTree->nMalloc = nTerm*2;
+ pTree->zMalloc = zNew;
+ }
+ pTree->zTerm = pTree->zMalloc;
+ memcpy(pTree->zTerm, zTerm, nTerm);
+ pTree->nTerm = nTerm;
+ }else{
+ pTree->zTerm = (char *)zTerm;
+ pTree->nTerm = nTerm;
+ }
+ return SQLITE_OK;
+ }
+ }
+
+ /* If control flows to here, it was not possible to append zTerm to the
+ ** current node. Create a new node (a right-sibling of the current node).
+ ** If this is the first node in the tree, the term is added to it.
+ **
+ ** Otherwise, the term is not added to the new node, it is left empty for
+ ** now. Instead, the term is inserted into the parent of pTree. If pTree
+ ** has no parent, one is created here.
+ */
+ pNew = (SegmentNode *)sqlite3_malloc(sizeof(SegmentNode) + p->nNodeSize);
+ if( !pNew ){
+ return SQLITE_NOMEM;
+ }
+ memset(pNew, 0, sizeof(SegmentNode));
+ pNew->nData = 1 + FTS3_VARINT_MAX;
+ pNew->aData = (char *)&pNew[1];
+
+ if( pTree ){
+ SegmentNode *pParent = pTree->pParent;
+ rc = fts3NodeAddTerm(p, &pParent, isCopyTerm, zTerm, nTerm);
+ if( pTree->pParent==0 ){
+ pTree->pParent = pParent;
+ }
+ pTree->pRight = pNew;
+ pNew->pLeftmost = pTree->pLeftmost;
+ pNew->pParent = pParent;
+ pNew->zMalloc = pTree->zMalloc;
+ pNew->nMalloc = pTree->nMalloc;
+ pTree->zMalloc = 0;
+ }else{
+ pNew->pLeftmost = pNew;
+ rc = fts3NodeAddTerm(p, &pNew, isCopyTerm, zTerm, nTerm);
+ }
+
+ *ppTree = pNew;
+ return rc;
+}
+
+/*
+** Helper function for fts3NodeWrite().
+*/
+static int fts3TreeFinishNode(
+ SegmentNode *pTree,
+ int iHeight,
+ sqlite3_int64 iLeftChild
+){
+ int nStart;
+ assert( iHeight>=1 && iHeight<128 );
+ nStart = FTS3_VARINT_MAX - sqlite3Fts3VarintLen(iLeftChild);
+ pTree->aData[nStart] = (char)iHeight;
+ sqlite3Fts3PutVarint(&pTree->aData[nStart+1], iLeftChild);
+ return nStart;
+}
+
+/*
+** Write the buffer for the segment node pTree and all of its peers to the
+** database. Then call this function recursively to write the parent of
+** pTree and its peers to the database.
+**
+** Except, if pTree is a root node, do not write it to the database. Instead,
+** set output variables *paRoot and *pnRoot to contain the root node.
+**
+** If successful, SQLITE_OK is returned and output variable *piLast is
+** set to the largest blockid written to the database (or zero if no
+** blocks were written to the db). Otherwise, an SQLite error code is
+** returned.
+*/
+static int fts3NodeWrite(
+ Fts3Table *p, /* Virtual table handle */
+ SegmentNode *pTree, /* SegmentNode handle */
+ int iHeight, /* Height of this node in tree */
+ sqlite3_int64 iLeaf, /* Block id of first leaf node */
+ sqlite3_int64 iFree, /* Block id of next free slot in %_segments */
+ sqlite3_int64 *piLast, /* OUT: Block id of last entry written */
+ char **paRoot, /* OUT: Data for root node */
+ int *pnRoot /* OUT: Size of root node in bytes */
+){
+ int rc = SQLITE_OK;
+
+ if( !pTree->pParent ){
+ /* Root node of the tree. */
+ int nStart = fts3TreeFinishNode(pTree, iHeight, iLeaf);
+ *piLast = iFree-1;
+ *pnRoot = pTree->nData - nStart;
+ *paRoot = &pTree->aData[nStart];
+ }else{
+ SegmentNode *pIter;
+ sqlite3_int64 iNextFree = iFree;
+ sqlite3_int64 iNextLeaf = iLeaf;
+ for(pIter=pTree->pLeftmost; pIter && rc==SQLITE_OK; pIter=pIter->pRight){
+ int nStart = fts3TreeFinishNode(pIter, iHeight, iNextLeaf);
+ int nWrite = pIter->nData - nStart;
+
+ rc = fts3WriteSegment(p, iNextFree, &pIter->aData[nStart], nWrite);
+ iNextFree++;
+ iNextLeaf += (pIter->nEntry+1);
+ }
+ if( rc==SQLITE_OK ){
+ assert( iNextLeaf==iFree );
+ rc = fts3NodeWrite(
+ p, pTree->pParent, iHeight+1, iFree, iNextFree, piLast, paRoot, pnRoot
+ );
+ }
+ }
+
+ return rc;
+}
+
+/*
+** Free all memory allocations associated with the tree pTree.
+*/
+static void fts3NodeFree(SegmentNode *pTree){
+ if( pTree ){
+ SegmentNode *p = pTree->pLeftmost;
+ fts3NodeFree(p->pParent);
+ while( p ){
+ SegmentNode *pRight = p->pRight;
+ if( p->aData!=(char *)&p[1] ){
+ sqlite3_free(p->aData);
+ }
+ assert( pRight==0 || p->zMalloc==0 );
+ sqlite3_free(p->zMalloc);
+ sqlite3_free(p);
+ p = pRight;
+ }
+ }
+}
+
+/*
+** Add a term to the segment being constructed by the SegmentWriter object
+** *ppWriter. When adding the first term to a segment, *ppWriter should
+** be passed NULL. This function will allocate a new SegmentWriter object
+** and return it via the input/output variable *ppWriter in this case.
+**
+** If successful, SQLITE_OK is returned. Otherwise, an SQLite error code.
+*/
+static int fts3SegWriterAdd(
+ Fts3Table *p, /* Virtual table handle */
+ SegmentWriter **ppWriter, /* IN/OUT: SegmentWriter handle */
+ int isCopyTerm, /* True if buffer zTerm must be copied */
+ const char *zTerm, /* Pointer to buffer containing term */
+ int nTerm, /* Size of term in bytes */
+ const char *aDoclist, /* Pointer to buffer containing doclist */
+ int nDoclist /* Size of doclist in bytes */
+){
+ int nPrefix; /* Size of term prefix in bytes */
+ int nSuffix; /* Size of term suffix in bytes */
+ int nReq; /* Number of bytes required on leaf page */
+ int nData;
+ SegmentWriter *pWriter = *ppWriter;
+
+ if( !pWriter ){
+ int rc;
+ sqlite3_stmt *pStmt;
+
+ /* Allocate the SegmentWriter structure */
+ pWriter = (SegmentWriter *)sqlite3_malloc(sizeof(SegmentWriter));
+ if( !pWriter ) return SQLITE_NOMEM;
+ memset(pWriter, 0, sizeof(SegmentWriter));
+ *ppWriter = pWriter;
+
+ /* Allocate a buffer in which to accumulate data */
+ pWriter->aData = (char *)sqlite3_malloc(p->nNodeSize);
+ if( !pWriter->aData ) return SQLITE_NOMEM;
+ pWriter->nSize = p->nNodeSize;
+
+ /* Find the next free blockid in the %_segments table */
+ rc = fts3SqlStmt(p, SQL_NEXT_SEGMENTS_ID, &pStmt, 0);
+ if( rc!=SQLITE_OK ) return rc;
+ if( SQLITE_ROW==sqlite3_step(pStmt) ){
+ pWriter->iFree = sqlite3_column_int64(pStmt, 0);
+ pWriter->iFirst = pWriter->iFree;
+ }
+ rc = sqlite3_reset(pStmt);
+ if( rc!=SQLITE_OK ) return rc;
+ }
+ nData = pWriter->nData;
+
+ nPrefix = fts3PrefixCompress(pWriter->zTerm, pWriter->nTerm, zTerm, nTerm);
+ nSuffix = nTerm-nPrefix;
+
+ /* Figure out how many bytes are required by this new entry */
+ nReq = sqlite3Fts3VarintLen(nPrefix) + /* varint containing prefix size */
+ sqlite3Fts3VarintLen(nSuffix) + /* varint containing suffix size */
+ nSuffix + /* Term suffix */
+ sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */
+ nDoclist; /* Doclist data */
+
+ if( nData>0 && nData+nReq>p->nNodeSize ){
+ int rc;
+
+ /* The current leaf node is full. Write it out to the database. */
+ rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, nData);
+ if( rc!=SQLITE_OK ) return rc;
+
+ /* Add the current term to the interior node tree. The term added to
+ ** the interior tree must:
+ **
+ ** a) be greater than the largest term on the leaf node just written
+ ** to the database (still available in pWriter->zTerm), and
+ **
+ ** b) be less than or equal to the term about to be added to the new
+ ** leaf node (zTerm/nTerm).
+ **
+ ** In other words, it must be the prefix of zTerm 1 byte longer than
+ ** the common prefix (if any) of zTerm and pWriter->zTerm.
+ */
+ assert( nPrefix<nTerm );
+ rc = fts3NodeAddTerm(p, &pWriter->pTree, isCopyTerm, zTerm, nPrefix+1);
+ if( rc!=SQLITE_OK ) return rc;
+
+ nData = 0;
+ pWriter->nTerm = 0;
+
+ nPrefix = 0;
+ nSuffix = nTerm;
+ nReq = 1 + /* varint containing prefix size */
+ sqlite3Fts3VarintLen(nTerm) + /* varint containing suffix size */
+ nTerm + /* Term suffix */
+ sqlite3Fts3VarintLen(nDoclist) + /* Size of doclist */
+ nDoclist; /* Doclist data */
+ }
+
+ /* If the buffer currently allocated is too small for this entry, realloc
+ ** the buffer to make it large enough.
+ */
+ if( nReq>pWriter->nSize ){
+ char *aNew = sqlite3_realloc(pWriter->aData, nReq);
+ if( !aNew ) return SQLITE_NOMEM;
+ pWriter->aData = aNew;
+ pWriter->nSize = nReq;
+ }
+ assert( nData+nReq<=pWriter->nSize );
+
+ /* Append the prefix-compressed term and doclist to the buffer. */
+ nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nPrefix);
+ nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nSuffix);
+ memcpy(&pWriter->aData[nData], &zTerm[nPrefix], nSuffix);
+ nData += nSuffix;
+ nData += sqlite3Fts3PutVarint(&pWriter->aData[nData], nDoclist);
+ memcpy(&pWriter->aData[nData], aDoclist, nDoclist);
+ pWriter->nData = nData + nDoclist;
+
+ /* Save the current term so that it can be used to prefix-compress the next.
+ ** If the isCopyTerm parameter is true, then the buffer pointed to by
+ ** zTerm is transient, so take a copy of the term data. Otherwise, just
+ ** store a copy of the pointer.
+ */
+ if( isCopyTerm ){
+ if( nTerm>pWriter->nMalloc ){
+ char *zNew = sqlite3_realloc(pWriter->zMalloc, nTerm*2);
+ if( !zNew ){
+ return SQLITE_NOMEM;
+ }
+ pWriter->nMalloc = nTerm*2;
+ pWriter->zMalloc = zNew;
+ pWriter->zTerm = zNew;
+ }
+ assert( pWriter->zTerm==pWriter->zMalloc );
+ memcpy(pWriter->zTerm, zTerm, nTerm);
+ }else{
+ pWriter->zTerm = (char *)zTerm;
+ }
+ pWriter->nTerm = nTerm;
+
+ return SQLITE_OK;
+}
+
+/*
+** Flush all data associated with the SegmentWriter object pWriter to the
+** database. This function must be called after all terms have been added
+** to the segment using fts3SegWriterAdd(). If successful, SQLITE_OK is
+** returned. Otherwise, an SQLite error code.
+*/
+static int fts3SegWriterFlush(
+ Fts3Table *p, /* Virtual table handle */
+ SegmentWriter *pWriter, /* SegmentWriter to flush to the db */
+ int iLevel, /* Value for 'level' column of %_segdir */
+ int iIdx /* Value for 'idx' column of %_segdir */
+){
+ int rc; /* Return code */
+ if( pWriter->pTree ){
+ sqlite3_int64 iLast = 0; /* Largest block id written to database */
+ sqlite3_int64 iLastLeaf; /* Largest leaf block id written to db */
+ char *zRoot = NULL; /* Pointer to buffer containing root node */
+ int nRoot = 0; /* Size of buffer zRoot */
+
+ iLastLeaf = pWriter->iFree;
+ rc = fts3WriteSegment(p, pWriter->iFree++, pWriter->aData, pWriter->nData);
+ if( rc==SQLITE_OK ){
+ rc = fts3NodeWrite(p, pWriter->pTree, 1,
+ pWriter->iFirst, pWriter->iFree, &iLast, &zRoot, &nRoot);
+ }
+ if( rc==SQLITE_OK ){
+ rc = fts3WriteSegdir(
+ p, iLevel, iIdx, pWriter->iFirst, iLastLeaf, iLast, zRoot, nRoot);
+ }
+ }else{
+ /* The entire tree fits on the root node. Write it to the segdir table. */
+ rc = fts3WriteSegdir(
+ p, iLevel, iIdx, 0, 0, 0, pWriter->aData, pWriter->nData);
+ }
+ return rc;
+}
+
+/*
+** Release all memory held by the SegmentWriter object passed as the
+** first argument.
+*/
+static void fts3SegWriterFree(SegmentWriter *pWriter){
+ if( pWriter ){
+ sqlite3_free(pWriter->aData);
+ sqlite3_free(pWriter->zMalloc);
+ fts3NodeFree(pWriter->pTree);
+ sqlite3_free(pWriter);
+ }
+}
+
+/*
+** The first value in the apVal[] array is assumed to contain an integer.
+** This function tests if there exist any documents with docid values that
+** are different from that integer. i.e. if deleting the document with docid
+** apVal[0] would mean the FTS3 table were empty.
+**
+** If successful, *pisEmpty is set to true if the table is empty except for
+** document apVal[0], or false otherwise, and SQLITE_OK is returned. If an
+** error occurs, an SQLite error code is returned.
+*/
+static int fts3IsEmpty(Fts3Table *p, sqlite3_value **apVal, int *pisEmpty){
+ sqlite3_stmt *pStmt;
+ int rc;
+ rc = fts3SqlStmt(p, SQL_IS_EMPTY, &pStmt, apVal);
+ if( rc==SQLITE_OK ){
+ if( SQLITE_ROW==sqlite3_step(pStmt) ){
+ *pisEmpty = sqlite3_column_int(pStmt, 0);
+ }
+ rc = sqlite3_reset(pStmt);
+ }
+ return rc;
+}
+
+/*
+** Set *pnSegment to the number of segments of level iLevel in the database.
+**
+** Return SQLITE_OK if successful, or an SQLite error code if not.
+*/
+static int fts3SegmentCount(Fts3Table *p, int iLevel, int *pnSegment){
+ sqlite3_stmt *pStmt;
+ int rc;
+
+ assert( iLevel>=0 );
+ rc = fts3SqlStmt(p, SQL_SELECT_LEVEL_COUNT, &pStmt, 0);
+ if( rc!=SQLITE_OK ) return rc;
+ sqlite3_bind_int(pStmt, 1, iLevel);
+ if( SQLITE_ROW==sqlite3_step(pStmt) ){
+ *pnSegment = sqlite3_column_int(pStmt, 0);
+ }
+ return sqlite3_reset(pStmt);
+}
+
+/*
+** Set *pnSegment to the total number of segments in the database. Set
+** *pnMax to the largest segment level in the database (segment levels
+** are stored in the 'level' column of the %_segdir table).
+**
+** Return SQLITE_OK if successful, or an SQLite error code if not.
+*/
+static int fts3SegmentCountMax(Fts3Table *p, int *pnSegment, int *pnMax){
+ sqlite3_stmt *pStmt;
+ int rc;
+
+ rc = fts3SqlStmt(p, SQL_SELECT_SEGDIR_COUNT_MAX, &pStmt, 0);
+ if( rc!=SQLITE_OK ) return rc;
+ if( SQLITE_ROW==sqlite3_step(pStmt) ){
+ *pnSegment = sqlite3_column_int(pStmt, 0);
+ *pnMax = sqlite3_column_int(pStmt, 1);
+ }
+ return sqlite3_reset(pStmt);
+}
+
+/*
+** This function is used after merging multiple segments into a single large
+** segment to delete the old, now redundant, segment b-trees. Specifically,
+** it:
+**
+** 1) Deletes all %_segments entries for the segments associated with
+** each of the SegReader objects in the array passed as the third
+** argument, and
+**
+** 2) deletes all %_segdir entries with level iLevel, or all %_segdir
+** entries regardless of level if (iLevel<0).
+**
+** SQLITE_OK is returned if successful, otherwise an SQLite error code.
+*/
+static int fts3DeleteSegdir(
+ Fts3Table *p, /* Virtual table handle */
+ int iLevel, /* Level of %_segdir entries to delete */
+ Fts3SegReader **apSegment, /* Array of SegReader objects */
+ int nReader /* Size of array apSegment */
+){
+ int rc; /* Return Code */
+ int i; /* Iterator variable */
+ sqlite3_stmt *pDelete; /* SQL statement to delete rows */
+
+ rc = fts3SqlStmt(p, SQL_DELETE_SEGMENTS_RANGE, &pDelete, 0);
+ for(i=0; rc==SQLITE_OK && i<nReader; i++){
+ Fts3SegReader *pSegment = apSegment[i];
+ if( pSegment->iStartBlock ){
+ sqlite3_bind_int64(pDelete, 1, pSegment->iStartBlock);
+ sqlite3_bind_int64(pDelete, 2, pSegment->iEndBlock);
+ sqlite3_step(pDelete);
+ rc = sqlite3_reset(pDelete);
+ }
+ }
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+
+ if( iLevel>=0 ){
+ rc = fts3SqlStmt(p, SQL_DELETE_SEGDIR_BY_LEVEL, &pDelete, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3_bind_int(pDelete, 1, iLevel);
+ sqlite3_step(pDelete);
+ rc = sqlite3_reset(pDelete);
+ }
+ }else{
+ fts3SqlExec(&rc, p, SQL_DELETE_ALL_SEGDIR, 0);
+ }
+
+ return rc;
+}
+
+/*
+** When this function is called, buffer *ppList (size *pnList bytes) contains
+** a position list that may (or may not) feature multiple columns. This
+** function adjusts the pointer *ppList and the length *pnList so that they
+** identify the subset of the position list that corresponds to column iCol.
+**
+** If there are no entries in the input position list for column iCol, then
+** *pnList is set to zero before returning.
+*/
+static void fts3ColumnFilter(
+ int iCol, /* Column to filter on */
+ char **ppList, /* IN/OUT: Pointer to position list */
+ int *pnList /* IN/OUT: Size of buffer *ppList in bytes */
+){
+ char *pList = *ppList;
+ int nList = *pnList;
+ char *pEnd = &pList[nList];
+ int iCurrent = 0;
+ char *p = pList;
+
+ assert( iCol>=0 );
+ while( 1 ){
+ char c = 0;
+ while( p<pEnd && (c | *p)&0xFE ) c = *p++ & 0x80;
+
+ if( iCol==iCurrent ){
+ nList = (int)(p - pList);
+ break;
+ }
+
+ nList -= (int)(p - pList);
+ pList = p;
+ if( nList==0 ){
+ break;
+ }
+ p = &pList[1];
+ p += sqlite3Fts3GetVarint32(p, &iCurrent);
+ }
+
+ *ppList = pList;
+ *pnList = nList;
+}
+
+/*
+** sqlite3Fts3SegReaderIterate() callback used when merging multiple
+** segments to create a single, larger segment.
+*/
+static int fts3MergeCallback(
+ Fts3Table *p, /* FTS3 Virtual table handle */
+ void *pContext, /* Pointer to SegmentWriter* to write with */
+ char *zTerm, /* Term to write to the db */
+ int nTerm, /* Number of bytes in zTerm */
+ char *aDoclist, /* Doclist associated with zTerm */
+ int nDoclist /* Number of bytes in doclist */
+){
+ SegmentWriter **ppW = (SegmentWriter **)pContext;
+ return fts3SegWriterAdd(p, ppW, 1, zTerm, nTerm, aDoclist, nDoclist);
+}
+
+/*
+** sqlite3Fts3SegReaderIterate() callback used when flushing the contents
+** of the pending-terms hash table to the database.
+*/
+static int fts3FlushCallback(
+ Fts3Table *p, /* FTS3 Virtual table handle */
+ void *pContext, /* Pointer to SegmentWriter* to write with */
+ char *zTerm, /* Term to write to the db */
+ int nTerm, /* Number of bytes in zTerm */
+ char *aDoclist, /* Doclist associated with zTerm */
+ int nDoclist /* Number of bytes in doclist */
+){
+ SegmentWriter **ppW = (SegmentWriter **)pContext;
+ return fts3SegWriterAdd(p, ppW, 0, zTerm, nTerm, aDoclist, nDoclist);
+}
+
+/*
+** This function is used to iterate through a contiguous set of terms
+** stored in the full-text index. It merges data contained in one or
+** more segments to support this.
+**
+** The second argument is passed an array of pointers to SegReader objects
+** allocated with sqlite3Fts3SegReaderNew(). This function merges the range
+** of terms selected by each SegReader. If a single term is present in
+** more than one segment, the associated doclists are merged. For each
+** term and (possibly merged) doclist in the merged range, the callback
+** function xFunc is invoked with its arguments set as follows.
+**
+** arg 0: Copy of 'p' parameter passed to this function
+** arg 1: Copy of 'pContext' parameter passed to this function
+** arg 2: Pointer to buffer containing term
+** arg 3: Size of arg 2 buffer in bytes
+** arg 4: Pointer to buffer containing doclist
+** arg 5: Size of arg 2 buffer in bytes
+**
+** The 4th argument to this function is a pointer to a structure of type
+** Fts3SegFilter, defined in fts3Int.h. The contents of this structure
+** further restrict the range of terms that callbacks are made for and
+** modify the behaviour of this function. See comments above structure
+** definition for details.
+*/
+int sqlite3Fts3SegReaderIterate(
+ Fts3Table *p, /* Virtual table handle */
+ Fts3SegReader **apSegment, /* Array of Fts3SegReader objects */
+ int nSegment, /* Size of apSegment array */
+ Fts3SegFilter *pFilter, /* Restrictions on range of iteration */
+ int (*xFunc)(Fts3Table *, void *, char *, int, char *, int), /* Callback */
+ void *pContext /* Callback context (2nd argument) */
+){
+ int i; /* Iterator variable */
+ char *aBuffer = 0; /* Buffer to merge doclists in */
+ int nAlloc = 0; /* Allocated size of aBuffer buffer */
+ int rc = SQLITE_OK; /* Return code */
+
+ int isIgnoreEmpty = (pFilter->flags & FTS3_SEGMENT_IGNORE_EMPTY);
+ int isRequirePos = (pFilter->flags & FTS3_SEGMENT_REQUIRE_POS);
+ int isColFilter = (pFilter->flags & FTS3_SEGMENT_COLUMN_FILTER);
+ int isPrefix = (pFilter->flags & FTS3_SEGMENT_PREFIX);
+
+ /* If there are zero segments, this function is a no-op. This scenario
+ ** comes about only when reading from an empty database.
+ */
+ if( nSegment==0 ) goto finished;
+
+ /* If the Fts3SegFilter defines a specific term (or term prefix) to search
+ ** for, then advance each segment iterator until it points to a term of
+ ** equal or greater value than the specified term. This prevents many
+ ** unnecessary merge/sort operations for the case where single segment
+ ** b-tree leaf nodes contain more than one term.
+ */
+ if( pFilter->zTerm ){
+ int nTerm = pFilter->nTerm;
+ const char *zTerm = pFilter->zTerm;
+ for(i=0; i<nSegment; i++){
+ Fts3SegReader *pSeg = apSegment[i];
+ while( fts3SegReaderTermCmp(pSeg, zTerm, nTerm)<0 ){
+ rc = fts3SegReaderNext(pSeg);
+ if( rc!=SQLITE_OK ) goto finished; }
+ }
+ }
+
+ fts3SegReaderSort(apSegment, nSegment, nSegment, fts3SegReaderCmp);
+ while( apSegment[0]->aNode ){
+ int nTerm = apSegment[0]->nTerm;
+ char *zTerm = apSegment[0]->zTerm;
+ int nMerge = 1;
+
+ /* If this is a prefix-search, and if the term that apSegment[0] points
+ ** to does not share a suffix with pFilter->zTerm/nTerm, then all
+ ** required callbacks have been made. In this case exit early.
+ **
+ ** Similarly, if this is a search for an exact match, and the first term
+ ** of segment apSegment[0] is not a match, exit early.
+ */
+ if( pFilter->zTerm ){
+ if( nTerm<pFilter->nTerm
+ || (!isPrefix && nTerm>pFilter->nTerm)
+ || memcmp(zTerm, pFilter->zTerm, pFilter->nTerm)
+ ){
+ goto finished;
+ }
+ }
+
+ while( nMerge<nSegment
+ && apSegment[nMerge]->aNode
+ && apSegment[nMerge]->nTerm==nTerm
+ && 0==memcmp(zTerm, apSegment[nMerge]->zTerm, nTerm)
+ ){
+ nMerge++;
+ }
+
+ assert( isIgnoreEmpty || (isRequirePos && !isColFilter) );
+ if( nMerge==1 && !isIgnoreEmpty ){
+ Fts3SegReader *p0 = apSegment[0];
+ rc = xFunc(p, pContext, zTerm, nTerm, p0->aDoclist, p0->nDoclist);
+ if( rc!=SQLITE_OK ) goto finished;
+ }else{
+ int nDoclist = 0; /* Size of doclist */
+ sqlite3_int64 iPrev = 0; /* Previous docid stored in doclist */
+
+ /* The current term of the first nMerge entries in the array
+ ** of Fts3SegReader objects is the same. The doclists must be merged
+ ** and a single term added to the new segment.
+ */
+ for(i=0; i<nMerge; i++){
+ fts3SegReaderFirstDocid(apSegment[i]);
+ }
+ fts3SegReaderSort(apSegment, nMerge, nMerge, fts3SegReaderDoclistCmp);
+ while( apSegment[0]->pOffsetList ){
+ int j; /* Number of segments that share a docid */
+ char *pList;
+ int nList;
+ int nByte;
+ sqlite3_int64 iDocid = apSegment[0]->iDocid;
+ fts3SegReaderNextDocid(apSegment[0], &pList, &nList);
+ j = 1;
+ while( j<nMerge
+ && apSegment[j]->pOffsetList
+ && apSegment[j]->iDocid==iDocid
+ ){
+ fts3SegReaderNextDocid(apSegment[j], 0, 0);
+ j++;
+ }
+
+ if( isColFilter ){
+ fts3ColumnFilter(pFilter->iCol, &pList, &nList);
+ }
+
+ if( !isIgnoreEmpty || nList>0 ){
+ nByte = sqlite3Fts3VarintLen(iDocid-iPrev) + (isRequirePos?nList+1:0);
+ if( nDoclist+nByte>nAlloc ){
+ char *aNew;
+ nAlloc = nDoclist+nByte*2;
+ aNew = sqlite3_realloc(aBuffer, nAlloc);
+ if( !aNew ){
+ rc = SQLITE_NOMEM;
+ goto finished;
+ }
+ aBuffer = aNew;
+ }
+ nDoclist += sqlite3Fts3PutVarint(&aBuffer[nDoclist], iDocid-iPrev);
+ iPrev = iDocid;
+ if( isRequirePos ){
+ memcpy(&aBuffer[nDoclist], pList, nList);
+ nDoclist += nList;
+ aBuffer[nDoclist++] = '\0';
+ }
+ }
+
+ fts3SegReaderSort(apSegment, nMerge, j, fts3SegReaderDoclistCmp);
+ }
+
+ if( nDoclist>0 ){
+ rc = xFunc(p, pContext, zTerm, nTerm, aBuffer, nDoclist);
+ if( rc!=SQLITE_OK ) goto finished;
+ }
+ }
+
+ /* If there is a term specified to filter on, and this is not a prefix
+ ** search, return now. The callback that corresponds to the required
+ ** term (if such a term exists in the index) has already been made.
+ */
+ if( pFilter->zTerm && !isPrefix ){
+ goto finished;
+ }
+
+ for(i=0; i<nMerge; i++){
+ rc = fts3SegReaderNext(apSegment[i]);
+ if( rc!=SQLITE_OK ) goto finished;
+ }
+ fts3SegReaderSort(apSegment, nSegment, nMerge, fts3SegReaderCmp);
+ }
+
+ finished:
+ sqlite3_free(aBuffer);
+ return rc;
+}
+
+/*
+** Merge all level iLevel segments in the database into a single
+** iLevel+1 segment. Or, if iLevel<0, merge all segments into a
+** single segment with a level equal to the numerically largest level
+** currently present in the database.
+**
+** If this function is called with iLevel<0, but there is only one
+** segment in the database, SQLITE_DONE is returned immediately.
+** Otherwise, if successful, SQLITE_OK is returned. If an error occurs,
+** an SQLite error code is returned.
+*/
+static int fts3SegmentMerge(Fts3Table *p, int iLevel){
+ int i; /* Iterator variable */
+ int rc; /* Return code */
+ int iIdx; /* Index of new segment */
+ int iNewLevel; /* Level to create new segment at */
+ sqlite3_stmt *pStmt = 0;
+ SegmentWriter *pWriter = 0;
+ int nSegment = 0; /* Number of segments being merged */
+ Fts3SegReader **apSegment = 0; /* Array of Segment iterators */
+ Fts3SegReader *pPending = 0; /* Iterator for pending-terms */
+ Fts3SegFilter filter; /* Segment term filter condition */
+
+ if( iLevel<0 ){
+ /* This call is to merge all segments in the database to a single
+ ** segment. The level of the new segment is equal to the the numerically
+ ** greatest segment level currently present in the database. The index
+ ** of the new segment is always 0.
+ */
+ iIdx = 0;
+ rc = sqlite3Fts3SegReaderPending(p, 0, 0, 1, &pPending);
+ if( rc!=SQLITE_OK ) goto finished;
+ rc = fts3SegmentCountMax(p, &nSegment, &iNewLevel);
+ if( rc!=SQLITE_OK ) goto finished;
+ nSegment += (pPending!=0);
+ if( nSegment<=1 ){
+ return SQLITE_DONE;
+ }
+ }else{
+ /* This call is to merge all segments at level iLevel. Find the next
+ ** available segment index at level iLevel+1. The call to
+ ** fts3AllocateSegdirIdx() will merge the segments at level iLevel+1 to
+ ** a single iLevel+2 segment if necessary.
+ */
+ iNewLevel = iLevel+1;
+ rc = fts3AllocateSegdirIdx(p, iNewLevel, &iIdx);
+ if( rc!=SQLITE_OK ) goto finished;
+ rc = fts3SegmentCount(p, iLevel, &nSegment);
+ if( rc!=SQLITE_OK ) goto finished;
+ }
+ assert( nSegment>0 );
+ assert( iNewLevel>=0 );
+
+ /* Allocate space for an array of pointers to segment iterators. */
+ apSegment = (Fts3SegReader**)sqlite3_malloc(sizeof(Fts3SegReader *)*nSegment);
+ if( !apSegment ){
+ rc = SQLITE_NOMEM;
+ goto finished;
+ }
+ memset(apSegment, 0, sizeof(Fts3SegReader *)*nSegment);
+
+ /* Allocate a Fts3SegReader structure for each segment being merged. A
+ ** Fts3SegReader stores the state data required to iterate through all
+ ** entries on all leaves of a single segment.
+ */
+ assert( SQL_SELECT_LEVEL+1==SQL_SELECT_ALL_LEVEL);
+ rc = fts3SqlStmt(p, SQL_SELECT_LEVEL+(iLevel<0), &pStmt, 0);
+ if( rc!=SQLITE_OK ) goto finished;
+ sqlite3_bind_int(pStmt, 1, iLevel);
+ for(i=0; SQLITE_ROW==(sqlite3_step(pStmt)); i++){
+ rc = fts3SegReaderNew(p, pStmt, i, &apSegment[i]);
+ if( rc!=SQLITE_OK ){
+ goto finished;
+ }
+ }
+ rc = sqlite3_reset(pStmt);
+ if( pPending ){
+ apSegment[i] = pPending;
+ pPending = 0;
+ }
+ pStmt = 0;
+ if( rc!=SQLITE_OK ) goto finished;
+
+ memset(&filter, 0, sizeof(Fts3SegFilter));
+ filter.flags = FTS3_SEGMENT_REQUIRE_POS;
+ filter.flags |= (iLevel<0 ? FTS3_SEGMENT_IGNORE_EMPTY : 0);
+ rc = sqlite3Fts3SegReaderIterate(p, apSegment, nSegment,
+ &filter, fts3MergeCallback, (void *)&pWriter
+ );
+ if( rc!=SQLITE_OK ) goto finished;
+
+ rc = fts3DeleteSegdir(p, iLevel, apSegment, nSegment);
+ if( rc==SQLITE_OK ){
+ rc = fts3SegWriterFlush(p, pWriter, iNewLevel, iIdx);
+ }
+
+ finished:
+ fts3SegWriterFree(pWriter);
+ if( apSegment ){
+ for(i=0; i<nSegment; i++){
+ sqlite3Fts3SegReaderFree(p, apSegment[i]);
+ }
+ sqlite3_free(apSegment);
+ }
+ sqlite3Fts3SegReaderFree(p, pPending);
+ sqlite3_reset(pStmt);
+ return rc;
+}
+
+
+/*
+** Flush the contents of pendingTerms to a level 0 segment.
+*/
+int sqlite3Fts3PendingTermsFlush(Fts3Table *p){
+ int rc; /* Return Code */
+ int idx; /* Index of new segment created */
+ SegmentWriter *pWriter = 0; /* Used to write the segment */
+ Fts3SegReader *pReader = 0; /* Used to iterate through the hash table */
+
+ /* Allocate a SegReader object to iterate through the contents of the
+ ** pending-terms table. If an error occurs, or if there are no terms
+ ** in the pending-terms table, return immediately.
+ */
+ rc = sqlite3Fts3SegReaderPending(p, 0, 0, 1, &pReader);
+ if( rc!=SQLITE_OK || pReader==0 ){
+ return rc;
+ }
+
+ /* Determine the next index at level 0. If level 0 is already full, this
+ ** call may merge all existing level 0 segments into a single level 1
+ ** segment.
+ */
+ rc = fts3AllocateSegdirIdx(p, 0, &idx);
+
+ /* If no errors have occured, iterate through the contents of the
+ ** pending-terms hash table using the Fts3SegReader iterator. The callback
+ ** writes each term (along with its doclist) to the database via the
+ ** SegmentWriter handle pWriter.
+ */
+ if( rc==SQLITE_OK ){
+ void *c = (void *)&pWriter; /* SegReaderIterate() callback context */
+ Fts3SegFilter f; /* SegReaderIterate() parameters */
+
+ memset(&f, 0, sizeof(Fts3SegFilter));
+ f.flags = FTS3_SEGMENT_REQUIRE_POS;
+ rc = sqlite3Fts3SegReaderIterate(p, &pReader, 1, &f, fts3FlushCallback, c);
+ }
+ assert( pWriter || rc!=SQLITE_OK );
+
+ /* If no errors have occured, flush the SegmentWriter object to the
+ ** database. Then delete the SegmentWriter and Fts3SegReader objects
+ ** allocated by this function.
+ */
+ if( rc==SQLITE_OK ){
+ rc = fts3SegWriterFlush(p, pWriter, 0, idx);
+ }
+ fts3SegWriterFree(pWriter);
+ sqlite3Fts3SegReaderFree(p, pReader);
+
+ if( rc==SQLITE_OK ){
+ sqlite3Fts3PendingTermsClear(p);
+ }
+ return rc;
+}
+
+/*
+** Encode N integers as varints into a blob.
+*/
+static void fts3EncodeIntArray(
+ int N, /* The number of integers to encode */
+ u32 *a, /* The integer values */
+ char *zBuf, /* Write the BLOB here */
+ int *pNBuf /* Write number of bytes if zBuf[] used here */
+){
+ int i, j;
+ for(i=j=0; i<N; i++){
+ j += sqlite3Fts3PutVarint(&zBuf[j], (sqlite3_int64)a[i]);
+ }
+ *pNBuf = j;
+}
+
+/*
+** Decode a blob of varints into N integers
+*/
+static void fts3DecodeIntArray(
+ int N, /* The number of integers to decode */
+ u32 *a, /* Write the integer values */
+ const char *zBuf, /* The BLOB containing the varints */
+ int nBuf /* size of the BLOB */
+){
+ int i, j;
+ UNUSED_PARAMETER(nBuf);
+ for(i=j=0; i<N; i++){
+ sqlite3_int64 x;
+ j += sqlite3Fts3GetVarint(&zBuf[j], &x);
+ assert(j<=nBuf);
+ a[i] = (u32)(x & 0xffffffff);
+ }
+}
+
+/*
+** Fill in the document size auxiliary information for the matchinfo
+** structure. The auxiliary information is:
+**
+** N Total number of documents in the full-text index
+** a0 Average length of column 0 over the whole index
+** n0 Length of column 0 on the matching row
+** ...
+** aM Average length of column M over the whole index
+** nM Length of column M on the matching row
+**
+** The fts3MatchinfoDocsizeLocal() routine fills in the nX values.
+** The fts3MatchinfoDocsizeGlobal() routine fills in N and the aX values.
+*/
+int sqlite3Fts3MatchinfoDocsizeLocal(Fts3Cursor *pCur, u32 *a){
+ const char *pBlob; /* The BLOB holding %_docsize info */
+ int nBlob; /* Size of the BLOB */
+ sqlite3_stmt *pStmt; /* Statement for reading and writing */
+ int i, j; /* Loop counters */
+ sqlite3_int64 x; /* Varint value */
+ int rc; /* Result code from subfunctions */
+ Fts3Table *p; /* The FTS table */
+
+ p = (Fts3Table*)pCur->base.pVtab;
+ rc = fts3SqlStmt(p, SQL_SELECT_DOCSIZE, &pStmt, 0);
+ if( rc ){
+ return rc;
+ }
+ sqlite3_bind_int64(pStmt, 1, pCur->iPrevId);
+ if( sqlite3_step(pStmt)==SQLITE_ROW ){
+ nBlob = sqlite3_column_bytes(pStmt, 0);
+ pBlob = (const char*)sqlite3_column_blob(pStmt, 0);
+ for(i=j=0; i<p->nColumn && j<nBlob; i++){
+ j = sqlite3Fts3GetVarint(&pBlob[j], &x);
+ a[2+i*2] = (u32)(x & 0xffffffff);
+ }
+ }
+ sqlite3_reset(pStmt);
+ return SQLITE_OK;
+}
+int sqlite3Fts3MatchinfoDocsizeGlobal(Fts3Cursor *pCur, u32 *a){
+ const char *pBlob; /* The BLOB holding %_stat info */
+ int nBlob; /* Size of the BLOB */
+ sqlite3_stmt *pStmt; /* Statement for reading and writing */
+ int i, j; /* Loop counters */
+ sqlite3_int64 x; /* Varint value */
+ int nDoc; /* Number of documents */
+ int rc; /* Result code from subfunctions */
+ Fts3Table *p; /* The FTS table */
+
+ p = (Fts3Table*)pCur->base.pVtab;
+ rc = fts3SqlStmt(p, SQL_SELECT_DOCTOTAL, &pStmt, 0);
+ if( rc ){
+ return rc;
+ }
+ if( sqlite3_step(pStmt)==SQLITE_ROW ){
+ nBlob = sqlite3_column_bytes(pStmt, 0);
+ pBlob = (const char*)sqlite3_column_blob(pStmt, 0);
+ j = sqlite3Fts3GetVarint(pBlob, &x);
+ a[0] = nDoc = (u32)(x & 0xffffffff);
+ for(i=0; i<p->nColumn && j<nBlob; i++){
+ j = sqlite3Fts3GetVarint(&pBlob[j], &x);
+ a[1+i*2] = ((u32)(x & 0xffffffff) + nDoc/2)/nDoc;
+ }
+ }
+ sqlite3_reset(pStmt);
+ return SQLITE_OK;
+}
+
+/*
+** Insert the sizes (in tokens) for each column of the document
+** with docid equal to p->iPrevDocid. The sizes are encoded as
+** a blob of varints.
+*/
+static void fts3InsertDocsize(
+ int *pRC, /* Result code */
+ Fts3Table *p, /* Table into which to insert */
+ u32 *aSz /* Sizes of each column */
+){
+ char *pBlob; /* The BLOB encoding of the document size */
+ int nBlob; /* Number of bytes in the BLOB */
+ sqlite3_stmt *pStmt; /* Statement used to insert the encoding */
+ int rc; /* Result code from subfunctions */
+
+ if( *pRC ) return;
+ pBlob = sqlite3_malloc( 10*p->nColumn );
+ if( pBlob==0 ){
+ *pRC = SQLITE_NOMEM;
+ return;
+ }
+ fts3EncodeIntArray(p->nColumn, aSz, pBlob, &nBlob);
+ rc = fts3SqlStmt(p, SQL_REPLACE_DOCSIZE, &pStmt, 0);
+ if( rc ){
+ sqlite3_free(pBlob);
+ *pRC = rc;
+ return;
+ }
+ sqlite3_bind_int64(pStmt, 1, p->iPrevDocid);
+ sqlite3_bind_blob(pStmt, 2, pBlob, nBlob, sqlite3_free);
+ sqlite3_step(pStmt);
+ *pRC = sqlite3_reset(pStmt);
+}
+
+/*
+** Update the 0 record of the %_stat table so that it holds a blob
+** which contains the document count followed by the cumulative
+** document sizes for all columns.
+*/
+static void fts3UpdateDocTotals(
+ int *pRC, /* The result code */
+ Fts3Table *p, /* Table being updated */
+ u32 *aSzIns, /* Size increases */
+ u32 *aSzDel, /* Size decreases */
+ int nChng /* Change in the number of documents */
+){
+ char *pBlob; /* Storage for BLOB written into %_stat */
+ int nBlob; /* Size of BLOB written into %_stat */
+ u32 *a; /* Array of integers that becomes the BLOB */
+ sqlite3_stmt *pStmt; /* Statement for reading and writing */
+ int i; /* Loop counter */
+ int rc; /* Result code from subfunctions */
+
+ if( *pRC ) return;
+ a = sqlite3_malloc( (sizeof(u32)+10)*(p->nColumn+1) );
+ if( a==0 ){
+ *pRC = SQLITE_NOMEM;
+ return;
+ }
+ pBlob = (char*)&a[p->nColumn+1];
+ rc = fts3SqlStmt(p, SQL_SELECT_DOCTOTAL, &pStmt, 0);
+ if( rc ){
+ sqlite3_free(a);
+ *pRC = rc;
+ return;
+ }
+ if( sqlite3_step(pStmt)==SQLITE_ROW ){
+ fts3DecodeIntArray(p->nColumn+1, a,
+ sqlite3_column_blob(pStmt, 0),
+ sqlite3_column_bytes(pStmt, 0));
+ }else{
+ memset(a, 0, sizeof(u32)*(p->nColumn+1) );
+ }
+ sqlite3_reset(pStmt);
+ if( nChng<0 && a[0]<(u32)(-nChng) ){
+ a[0] = 0;
+ }else{
+ a[0] += nChng;
+ }
+ for(i=0; i<p->nColumn; i++){
+ u32 x = a[i+1];
+ if( x+aSzIns[i] < aSzDel[i] ){
+ x = 0;
+ }else{
+ x = x + aSzIns[i] - aSzDel[i];
+ }
+ a[i+1] = x;
+ }
+ fts3EncodeIntArray(p->nColumn+1, a, pBlob, &nBlob);
+ rc = fts3SqlStmt(p, SQL_REPLACE_DOCTOTAL, &pStmt, 0);
+ if( rc ){
+ sqlite3_free(a);
+ *pRC = rc;
+ return;
+ }
+ sqlite3_bind_blob(pStmt, 1, pBlob, nBlob, SQLITE_STATIC);
+ sqlite3_step(pStmt);
+ *pRC = sqlite3_reset(pStmt);
+ sqlite3_free(a);
+}
+
+/*
+** Handle a 'special' INSERT of the form:
+**
+** "INSERT INTO tbl(tbl) VALUES(<expr>)"
+**
+** Argument pVal contains the result of <expr>. Currently the only
+** meaningful value to insert is the text 'optimize'.
+*/
+static int fts3SpecialInsert(Fts3Table *p, sqlite3_value *pVal){
+ int rc; /* Return Code */
+ const char *zVal = (const char *)sqlite3_value_text(pVal);
+ int nVal = sqlite3_value_bytes(pVal);
+
+ if( !zVal ){
+ return SQLITE_NOMEM;
+ }else if( nVal==8 && 0==sqlite3_strnicmp(zVal, "optimize", 8) ){
+ rc = fts3SegmentMerge(p, -1);
+ if( rc==SQLITE_DONE ){
+ rc = SQLITE_OK;
+ }else{
+ sqlite3Fts3PendingTermsClear(p);
+ }
+#ifdef SQLITE_TEST
+ }else if( nVal>9 && 0==sqlite3_strnicmp(zVal, "nodesize=", 9) ){
+ p->nNodeSize = atoi(&zVal[9]);
+ rc = SQLITE_OK;
+ }else if( nVal>11 && 0==sqlite3_strnicmp(zVal, "maxpending=", 9) ){
+ p->nMaxPendingData = atoi(&zVal[11]);
+ rc = SQLITE_OK;
+#endif
+ }else{
+ rc = SQLITE_ERROR;
+ }
+
+ return rc;
+}
+
+/*
+** This function does the work for the xUpdate method of FTS3 virtual
+** tables.
+*/
+int sqlite3Fts3UpdateMethod(
+ sqlite3_vtab *pVtab, /* FTS3 vtab object */
+ int nArg, /* Size of argument array */
+ sqlite3_value **apVal, /* Array of arguments */
+ sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */
+){
+ Fts3Table *p = (Fts3Table *)pVtab;
+ int rc = SQLITE_OK; /* Return Code */
+ int isRemove = 0; /* True for an UPDATE or DELETE */
+ sqlite3_int64 iRemove = 0; /* Rowid removed by UPDATE or DELETE */
+ u32 *aSzIns; /* Sizes of inserted documents */
+ u32 *aSzDel; /* Sizes of deleted documents */
+ int nChng = 0; /* Net change in number of documents */
+
+
+ /* Allocate space to hold the change in document sizes */
+ aSzIns = sqlite3_malloc( sizeof(aSzIns[0])*p->nColumn*2 );
+ if( aSzIns==0 ) return SQLITE_NOMEM;
+ aSzDel = &aSzIns[p->nColumn];
+ memset(aSzIns, 0, sizeof(aSzIns[0])*p->nColumn*2);
+
+ /* If this is a DELETE or UPDATE operation, remove the old record. */
+ if( sqlite3_value_type(apVal[0])!=SQLITE_NULL ){
+ int isEmpty;
+ rc = fts3IsEmpty(p, apVal, &isEmpty);
+ if( rc==SQLITE_OK ){
+ if( isEmpty ){
+ /* Deleting this row means the whole table is empty. In this case
+ ** delete the contents of all three tables and throw away any
+ ** data in the pendingTerms hash table.
+ */
+ rc = fts3DeleteAll(p);
+ }else{
+ isRemove = 1;
+ iRemove = sqlite3_value_int64(apVal[0]);
+ rc = fts3PendingTermsDocid(p, iRemove);
+ fts3DeleteTerms(&rc, p, apVal, aSzDel);
+ fts3SqlExec(&rc, p, SQL_DELETE_CONTENT, apVal);
+ if( p->bHasDocsize ){
+ fts3SqlExec(&rc, p, SQL_DELETE_DOCSIZE, apVal);
+ nChng--;
+ }
+ }
+ }
+ }else if( sqlite3_value_type(apVal[p->nColumn+2])!=SQLITE_NULL ){
+ sqlite3_free(aSzIns);
+ return fts3SpecialInsert(p, apVal[p->nColumn+2]);
+ }
+
+ /* If this is an INSERT or UPDATE operation, insert the new record. */
+ if( nArg>1 && rc==SQLITE_OK ){
+ rc = fts3InsertData(p, apVal, pRowid);
+ if( rc==SQLITE_OK && (!isRemove || *pRowid!=iRemove) ){
+ rc = fts3PendingTermsDocid(p, *pRowid);
+ }
+ if( rc==SQLITE_OK ){
+ rc = fts3InsertTerms(p, apVal, aSzIns);
+ }
+ if( p->bHasDocsize ){
+ nChng++;
+ fts3InsertDocsize(&rc, p, aSzIns);
+ }
+ }
+
+ if( p->bHasDocsize ){
+ fts3UpdateDocTotals(&rc, p, aSzIns, aSzDel, nChng);
+ }
+
+ sqlite3_free(aSzIns);
+ return rc;
+}
+
+/*
+** Flush any data in the pending-terms hash table to disk. If successful,
+** merge all segments in the database (including the new segment, if
+** there was any data to flush) into a single segment.
+*/
+int sqlite3Fts3Optimize(Fts3Table *p){
+ int rc;
+ rc = sqlite3_exec(p->db, "SAVEPOINT fts3", 0, 0, 0);
+ if( rc==SQLITE_OK ){
+ rc = fts3SegmentMerge(p, -1);
+ if( rc==SQLITE_OK ){
+ rc = sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0);
+ if( rc==SQLITE_OK ){
+ sqlite3Fts3PendingTermsClear(p);
+ }
+ }else{
+ sqlite3_exec(p->db, "ROLLBACK TO fts3", 0, 0, 0);
+ sqlite3_exec(p->db, "RELEASE fts3", 0, 0, 0);
+ }
+ }
+ return rc;
+}
+
+#endif
diff --git a/third_party/sqlite/src/ext/fts3/mkfts3amal.tcl b/third_party/sqlite/src/ext/fts3/mkfts3amal.tcl
index 0590487..0590487 100755..100644
--- a/third_party/sqlite/src/ext/fts3/mkfts3amal.tcl
+++ b/third_party/sqlite/src/ext/fts3/mkfts3amal.tcl
diff --git a/third_party/sqlite/src/ext/icu/README.txt b/third_party/sqlite/src/ext/icu/README.txt
index 5c995cc..c5cadb5 100644
--- a/third_party/sqlite/src/ext/icu/README.txt
+++ b/third_party/sqlite/src/ext/icu/README.txt
@@ -139,7 +139,7 @@ SQLite. Documentation follows.
3.2 The SQLITE_MAX_LIKE_PATTERN_LENGTH Macro
Passing very long patterns to the built-in SQLite LIKE operator can
- cause a stack overflow. To curb this problem, SQLite defines the
+ cause excessive CPU usage. To curb this problem, SQLite defines the
SQLITE_MAX_LIKE_PATTERN_LENGTH macro as the maximum length of a
pattern in bytes (irrespective of encoding). The default value is
defined in internal header file "limits.h".
@@ -167,4 +167,3 @@ SQLite. Documentation follows.
malicious users may execute arbitrary SQL (i.e. gears), they
should be prevented from invoking the icu_load_collation() function,
possibly using the authorisation callback.
-
diff --git a/third_party/sqlite/src/ext/rtree/README b/third_party/sqlite/src/ext/rtree/README
index 3736f45..3736f45 100755..100644
--- a/third_party/sqlite/src/ext/rtree/README
+++ b/third_party/sqlite/src/ext/rtree/README
diff --git a/third_party/sqlite/src/ext/rtree/rtree.c b/third_party/sqlite/src/ext/rtree/rtree.c
index 5a4f570..b805676 100644
--- a/third_party/sqlite/src/ext/rtree/rtree.c
+++ b/third_party/sqlite/src/ext/rtree/rtree.c
@@ -11,8 +11,45 @@
*************************************************************************
** This file contains code for implementations of the r-tree and r*-tree
** algorithms packaged as an SQLite virtual table module.
+*/
+
+/*
+** Database Format of R-Tree Tables
+** --------------------------------
+**
+** The data structure for a single virtual r-tree table is stored in three
+** native SQLite tables declared as follows. In each case, the '%' character
+** in the table name is replaced with the user-supplied name of the r-tree
+** table.
+**
+** CREATE TABLE %_node(nodeno INTEGER PRIMARY KEY, data BLOB)
+** CREATE TABLE %_parent(nodeno INTEGER PRIMARY KEY, parentnode INTEGER)
+** CREATE TABLE %_rowid(rowid INTEGER PRIMARY KEY, nodeno INTEGER)
+**
+** The data for each node of the r-tree structure is stored in the %_node
+** table. For each node that is not the root node of the r-tree, there is
+** an entry in the %_parent table associating the node with its parent.
+** And for each row of data in the table, there is an entry in the %_rowid
+** table that maps from the entries rowid to the id of the node that it
+** is stored on.
+**
+** The root node of an r-tree always exists, even if the r-tree table is
+** empty. The nodeno of the root node is always 1. All other nodes in the
+** table must be the same size as the root node. The content of each node
+** is formatted as follows:
+**
+** 1. If the node is the root node (node 1), then the first 2 bytes
+** of the node contain the tree depth as a big-endian integer.
+** For non-root nodes, the first 2 bytes are left unused.
+**
+** 2. The next 2 bytes contain the number of entries currently
+** stored in the node.
**
-** $Id: rtree.c,v 1.14 2009/08/06 18:36:47 danielk1977 Exp $
+** 3. The remainder of the node contains the node entries. Each entry
+** consists of a single 8-byte integer followed by an even number
+** of 4-byte coordinates. For leaf nodes the integer is the rowid
+** of a record. For internal nodes it is the node number of a
+** child page.
*/
#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_RTREE)
@@ -55,6 +92,9 @@
#define AssignCells splitNodeStartree
#endif
+#if !defined(NDEBUG) && !defined(SQLITE_DEBUG)
+# define NDEBUG 1
+#endif
#ifndef SQLITE_CORE
#include "sqlite3ext.h"
@@ -67,6 +107,7 @@
#include <assert.h>
#ifndef SQLITE_AMALGAMATION
+#include "sqlite3rtree.h"
typedef sqlite3_int64 i64;
typedef unsigned char u8;
typedef unsigned int u32;
@@ -77,6 +118,8 @@ typedef struct RtreeCursor RtreeCursor;
typedef struct RtreeNode RtreeNode;
typedef struct RtreeCell RtreeCell;
typedef struct RtreeConstraint RtreeConstraint;
+typedef struct RtreeMatchArg RtreeMatchArg;
+typedef struct RtreeGeomCallback RtreeGeomCallback;
typedef union RtreeCoord RtreeCoord;
/* The rtree may have between 1 and RTREE_MAX_DIMENSIONS dimensions. */
@@ -146,6 +189,15 @@ struct Rtree {
#define RTREE_REINSERT(p) RTREE_MINCELLS(p)
#define RTREE_MAXCELLS 51
+/*
+** The smallest possible node-size is (512-64)==448 bytes. And the largest
+** supported cell size is 48 bytes (8 byte rowid + ten 4 byte coordinates).
+** Therefore all non-root nodes must contain at least 3 entries. Since
+** 2^40 is greater than 2^64, an r-tree structure always has a depth of
+** 40 or less.
+*/
+#define RTREE_MAX_DEPTH 40
+
/*
** An rtree cursor object.
*/
@@ -178,35 +230,23 @@ union RtreeCoord {
** A search constraint.
*/
struct RtreeConstraint {
- int iCoord; /* Index of constrained coordinate */
- int op; /* Constraining operation */
- double rValue; /* Constraint value. */
+ int iCoord; /* Index of constrained coordinate */
+ int op; /* Constraining operation */
+ double rValue; /* Constraint value. */
+ int (*xGeom)(sqlite3_rtree_geometry *, int, double *, int *);
+ sqlite3_rtree_geometry *pGeom; /* Constraint callback argument for a MATCH */
};
/* Possible values for RtreeConstraint.op */
-#define RTREE_EQ 0x41
-#define RTREE_LE 0x42
-#define RTREE_LT 0x43
-#define RTREE_GE 0x44
-#define RTREE_GT 0x45
+#define RTREE_EQ 0x41
+#define RTREE_LE 0x42
+#define RTREE_LT 0x43
+#define RTREE_GE 0x44
+#define RTREE_GT 0x45
+#define RTREE_MATCH 0x46
/*
** An rtree structure node.
-**
-** Data format (RtreeNode.zData):
-**
-** 1. If the node is the root node (node 1), then the first 2 bytes
-** of the node contain the tree depth as a big-endian integer.
-** For non-root nodes, the first 2 bytes are left unused.
-**
-** 2. The next 2 bytes contain the number of entries currently
-** stored in the node.
-**
-** 3. The remainder of the node contains the node entries. Each entry
-** consists of a single 8-byte integer followed by an even number
-** of 4-byte coordinates. For leaf nodes the integer is the rowid
-** of a record. For internal nodes it is the node number of a
-** child page.
*/
struct RtreeNode {
RtreeNode *pParent; /* Parent node */
@@ -226,6 +266,40 @@ struct RtreeCell {
RtreeCoord aCoord[RTREE_MAX_DIMENSIONS*2];
};
+
+/*
+** Value for the first field of every RtreeMatchArg object. The MATCH
+** operator tests that the first field of a blob operand matches this
+** value to avoid operating on invalid blobs (which could cause a segfault).
+*/
+#define RTREE_GEOMETRY_MAGIC 0x891245AB
+
+/*
+** An instance of this structure must be supplied as a blob argument to
+** the right-hand-side of an SQL MATCH operator used to constrain an
+** r-tree query.
+*/
+struct RtreeMatchArg {
+ u32 magic; /* Always RTREE_GEOMETRY_MAGIC */
+ int (*xGeom)(sqlite3_rtree_geometry *, int, double *, int *);
+ void *pContext;
+ int nParam;
+ double aParam[1];
+};
+
+/*
+** When a geometry callback is created (see sqlite3_rtree_geometry_callback),
+** a single instance of the following structure is allocated. It is used
+** as the context for the user-function created by by s_r_g_c(). The object
+** is eventually deleted by the destructor mechanism provided by
+** sqlite3_create_function_v2() (which is called by s_r_g_c() to create
+** the geometry callback function).
+*/
+struct RtreeGeomCallback {
+ int (*xGeom)(sqlite3_rtree_geometry *, int, double *, int *);
+ void *pContext;
+};
+
#ifndef MAX
# define MAX(x,y) ((x) < (y) ? (y) : (x))
#endif
@@ -308,10 +382,8 @@ static void nodeReference(RtreeNode *p){
** Clear the content of node p (set all bytes to 0x00).
*/
static void nodeZero(Rtree *pRtree, RtreeNode *p){
- if( p ){
- memset(&p->zData[2], 0, pRtree->iNodeSize-2);
- p->isDirty = 1;
- }
+ memset(&p->zData[2], 0, pRtree->iNodeSize-2);
+ p->isDirty = 1;
}
/*
@@ -331,7 +403,6 @@ static int nodeHash(i64 iNode){
*/
static RtreeNode *nodeHashLookup(Rtree *pRtree, i64 iNode){
RtreeNode *p;
- assert( iNode!=0 );
for(p=pRtree->aHash[nodeHash(iNode)]; p && p->iNode!=iNode; p=p->pNext);
return p;
}
@@ -340,13 +411,11 @@ static RtreeNode *nodeHashLookup(Rtree *pRtree, i64 iNode){
** Add node pNode to the node hash table.
*/
static void nodeHashInsert(Rtree *pRtree, RtreeNode *pNode){
- if( pNode ){
- int iHash;
- assert( pNode->pNext==0 );
- iHash = nodeHash(pNode->iNode);
- pNode->pNext = pRtree->aHash[iHash];
- pRtree->aHash[iHash] = pNode;
- }
+ int iHash;
+ assert( pNode->pNext==0 );
+ iHash = nodeHash(pNode->iNode);
+ pNode->pNext = pRtree->aHash[iHash];
+ pRtree->aHash[iHash] = pNode;
}
/*
@@ -368,11 +437,11 @@ static void nodeHashDelete(Rtree *pRtree, RtreeNode *pNode){
** assigned a node number when nodeWrite() is called to write the
** node contents out to the database.
*/
-static RtreeNode *nodeNew(Rtree *pRtree, RtreeNode *pParent, int zero){
+static RtreeNode *nodeNew(Rtree *pRtree, RtreeNode *pParent){
RtreeNode *pNode;
pNode = (RtreeNode *)sqlite3_malloc(sizeof(RtreeNode) + pRtree->iNodeSize);
if( pNode ){
- memset(pNode, 0, sizeof(RtreeNode) + (zero?pRtree->iNodeSize:0));
+ memset(pNode, 0, sizeof(RtreeNode) + pRtree->iNodeSize);
pNode->zData = (u8 *)&pNode[1];
pNode->nRef = 1;
pNode->pParent = pParent;
@@ -393,6 +462,7 @@ nodeAcquire(
RtreeNode **ppNode /* OUT: Acquired node */
){
int rc;
+ int rc2 = SQLITE_OK;
RtreeNode *pNode;
/* Check if the requested node is already in the hash table. If so,
@@ -409,38 +479,63 @@ nodeAcquire(
return SQLITE_OK;
}
- pNode = (RtreeNode *)sqlite3_malloc(sizeof(RtreeNode) + pRtree->iNodeSize);
- if( !pNode ){
- *ppNode = 0;
- return SQLITE_NOMEM;
- }
- pNode->pParent = pParent;
- pNode->zData = (u8 *)&pNode[1];
- pNode->nRef = 1;
- pNode->iNode = iNode;
- pNode->isDirty = 0;
- pNode->pNext = 0;
-
sqlite3_bind_int64(pRtree->pReadNode, 1, iNode);
rc = sqlite3_step(pRtree->pReadNode);
if( rc==SQLITE_ROW ){
const u8 *zBlob = sqlite3_column_blob(pRtree->pReadNode, 0);
- memcpy(pNode->zData, zBlob, pRtree->iNodeSize);
- nodeReference(pParent);
- }else{
- sqlite3_free(pNode);
- pNode = 0;
+ if( pRtree->iNodeSize==sqlite3_column_bytes(pRtree->pReadNode, 0) ){
+ pNode = (RtreeNode *)sqlite3_malloc(sizeof(RtreeNode)+pRtree->iNodeSize);
+ if( !pNode ){
+ rc2 = SQLITE_NOMEM;
+ }else{
+ pNode->pParent = pParent;
+ pNode->zData = (u8 *)&pNode[1];
+ pNode->nRef = 1;
+ pNode->iNode = iNode;
+ pNode->isDirty = 0;
+ pNode->pNext = 0;
+ memcpy(pNode->zData, zBlob, pRtree->iNodeSize);
+ nodeReference(pParent);
+ }
+ }
}
-
- *ppNode = pNode;
rc = sqlite3_reset(pRtree->pReadNode);
+ if( rc==SQLITE_OK ) rc = rc2;
- if( rc==SQLITE_OK && iNode==1 ){
+ /* If the root node was just loaded, set pRtree->iDepth to the height
+ ** of the r-tree structure. A height of zero means all data is stored on
+ ** the root node. A height of one means the children of the root node
+ ** are the leaves, and so on. If the depth as specified on the root node
+ ** is greater than RTREE_MAX_DEPTH, the r-tree structure must be corrupt.
+ */
+ if( pNode && iNode==1 ){
pRtree->iDepth = readInt16(pNode->zData);
+ if( pRtree->iDepth>RTREE_MAX_DEPTH ){
+ rc = SQLITE_CORRUPT;
+ }
}
- assert( (rc==SQLITE_OK && pNode) || (pNode==0 && rc!=SQLITE_OK) );
- nodeHashInsert(pRtree, pNode);
+ /* If no error has occurred so far, check if the "number of entries"
+ ** field on the node is too large. If so, set the return code to
+ ** SQLITE_CORRUPT.
+ */
+ if( pNode && rc==SQLITE_OK ){
+ if( NCELL(pNode)>((pRtree->iNodeSize-4)/pRtree->nBytesPerCell) ){
+ rc = SQLITE_CORRUPT;
+ }
+ }
+
+ if( rc==SQLITE_OK ){
+ if( pNode!=0 ){
+ nodeHashInsert(pRtree, pNode);
+ }else{
+ rc = SQLITE_CORRUPT;
+ }
+ *ppNode = pNode;
+ }else{
+ sqlite3_free(pNode);
+ *ppNode = 0;
+ }
return rc;
}
@@ -493,8 +588,7 @@ nodeInsertCell(
nMaxCell = (pRtree->iNodeSize-4)/pRtree->nBytesPerCell;
nCell = NCELL(pNode);
- assert(nCell<=nMaxCell);
-
+ assert( nCell<=nMaxCell );
if( nCell<nMaxCell ){
nodeOverwriteCell(pRtree, pNode, pCell, nCell);
writeInt16(&pNode->zData[2], nCell+1);
@@ -714,6 +808,25 @@ static int rtreeOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
return rc;
}
+
+/*
+** Free the RtreeCursor.aConstraint[] array and its contents.
+*/
+static void freeCursorConstraints(RtreeCursor *pCsr){
+ if( pCsr->aConstraint ){
+ int i; /* Used to iterate through constraint array */
+ for(i=0; i<pCsr->nConstraint; i++){
+ sqlite3_rtree_geometry *pGeom = pCsr->aConstraint[i].pGeom;
+ if( pGeom ){
+ if( pGeom->xDelUser ) pGeom->xDelUser(pGeom->pUser);
+ sqlite3_free(pGeom);
+ }
+ }
+ sqlite3_free(pCsr->aConstraint);
+ pCsr->aConstraint = 0;
+ }
+}
+
/*
** Rtree virtual table module xClose method.
*/
@@ -721,7 +834,7 @@ static int rtreeClose(sqlite3_vtab_cursor *cur){
Rtree *pRtree = (Rtree *)(cur->pVtab);
int rc;
RtreeCursor *pCsr = (RtreeCursor *)cur;
- sqlite3_free(pCsr->aConstraint);
+ freeCursorConstraints(pCsr);
rc = nodeRelease(pRtree, pCsr->pNode);
sqlite3_free(pCsr);
return rc;
@@ -738,13 +851,39 @@ static int rtreeEof(sqlite3_vtab_cursor *cur){
return (pCsr->pNode==0);
}
+/*
+** The r-tree constraint passed as the second argument to this function is
+** guaranteed to be a MATCH constraint.
+*/
+static int testRtreeGeom(
+ Rtree *pRtree, /* R-Tree object */
+ RtreeConstraint *pConstraint, /* MATCH constraint to test */
+ RtreeCell *pCell, /* Cell to test */
+ int *pbRes /* OUT: Test result */
+){
+ int i;
+ double aCoord[RTREE_MAX_DIMENSIONS*2];
+ int nCoord = pRtree->nDim*2;
+
+ assert( pConstraint->op==RTREE_MATCH );
+ assert( pConstraint->pGeom );
+
+ for(i=0; i<nCoord; i++){
+ aCoord[i] = DCOORD(pCell->aCoord[i]);
+ }
+ return pConstraint->xGeom(pConstraint->pGeom, nCoord, aCoord, pbRes);
+}
+
/*
** Cursor pCursor currently points to a cell in a non-leaf page.
-** Return true if the sub-tree headed by the cell is filtered
+** Set *pbEof to true if the sub-tree headed by the cell is filtered
** (excluded) by the constraints in the pCursor->aConstraint[]
** array, or false otherwise.
+**
+** Return SQLITE_OK if successful or an SQLite error code if an error
+** occurs within a geometry callback.
*/
-static int testRtreeCell(Rtree *pRtree, RtreeCursor *pCursor){
+static int testRtreeCell(Rtree *pRtree, RtreeCursor *pCursor, int *pbEof){
RtreeCell cell;
int ii;
int bRes = 0;
@@ -756,31 +895,55 @@ static int testRtreeCell(Rtree *pRtree, RtreeCursor *pCursor){
double cell_max = DCOORD(cell.aCoord[(p->iCoord>>1)*2+1]);
assert(p->op==RTREE_LE || p->op==RTREE_LT || p->op==RTREE_GE
- || p->op==RTREE_GT || p->op==RTREE_EQ
+ || p->op==RTREE_GT || p->op==RTREE_EQ || p->op==RTREE_MATCH
);
switch( p->op ){
- case RTREE_LE: case RTREE_LT: bRes = p->rValue<cell_min; break;
- case RTREE_GE: case RTREE_GT: bRes = p->rValue>cell_max; break;
- case RTREE_EQ:
+ case RTREE_LE: case RTREE_LT:
+ bRes = p->rValue<cell_min;
+ break;
+
+ case RTREE_GE: case RTREE_GT:
+ bRes = p->rValue>cell_max;
+ break;
+
+ case RTREE_EQ:
bRes = (p->rValue>cell_max || p->rValue<cell_min);
break;
+
+ default: {
+ int rc;
+ assert( p->op==RTREE_MATCH );
+ rc = testRtreeGeom(pRtree, p, &cell, &bRes);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ bRes = !bRes;
+ break;
+ }
}
}
- return bRes;
+ *pbEof = bRes;
+ return SQLITE_OK;
}
/*
-** Return true if the cell that cursor pCursor currently points to
+** Test if the cell that cursor pCursor currently points to
** would be filtered (excluded) by the constraints in the
-** pCursor->aConstraint[] array, or false otherwise.
+** pCursor->aConstraint[] array. If so, set *pbEof to true before
+** returning. If the cell is not filtered (excluded) by the constraints,
+** set pbEof to zero.
+**
+** Return SQLITE_OK if successful or an SQLite error code if an error
+** occurs within a geometry callback.
**
** This function assumes that the cell is part of a leaf node.
*/
-static int testRtreeEntry(Rtree *pRtree, RtreeCursor *pCursor){
+static int testRtreeEntry(Rtree *pRtree, RtreeCursor *pCursor, int *pbEof){
RtreeCell cell;
int ii;
+ *pbEof = 0;
nodeGetCell(pRtree, pCursor->pNode, pCursor->iCell, &cell);
for(ii=0; ii<pCursor->nConstraint; ii++){
@@ -788,7 +951,7 @@ static int testRtreeEntry(Rtree *pRtree, RtreeCursor *pCursor){
double coord = DCOORD(cell.aCoord[p->iCoord]);
int res;
assert(p->op==RTREE_LE || p->op==RTREE_LT || p->op==RTREE_GE
- || p->op==RTREE_GT || p->op==RTREE_EQ
+ || p->op==RTREE_GT || p->op==RTREE_EQ || p->op==RTREE_MATCH
);
switch( p->op ){
case RTREE_LE: res = (coord<=p->rValue); break;
@@ -796,12 +959,24 @@ static int testRtreeEntry(Rtree *pRtree, RtreeCursor *pCursor){
case RTREE_GE: res = (coord>=p->rValue); break;
case RTREE_GT: res = (coord>p->rValue); break;
case RTREE_EQ: res = (coord==p->rValue); break;
+ default: {
+ int rc;
+ assert( p->op==RTREE_MATCH );
+ rc = testRtreeGeom(pRtree, p, &cell, &res);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
+ break;
+ }
}
- if( !res ) return 1;
+ if( !res ){
+ *pbEof = 1;
+ return SQLITE_OK;
+ }
}
- return 0;
+ return SQLITE_OK;
}
/*
@@ -828,13 +1003,13 @@ static int descendToCell(
assert( iHeight>=0 );
if( iHeight==0 ){
- isEof = testRtreeEntry(pRtree, pCursor);
+ rc = testRtreeEntry(pRtree, pCursor, &isEof);
}else{
- isEof = testRtreeCell(pRtree, pCursor);
+ rc = testRtreeCell(pRtree, pCursor, &isEof);
}
- if( isEof || iHeight==0 ){
+ if( rc!=SQLITE_OK || isEof || iHeight==0 ){
*pEof = isEof;
- return SQLITE_OK;
+ return rc;
}
iRowid = nodeGetRowid(pRtree, pCursor->pNode, pCursor->iCell);
@@ -870,24 +1045,34 @@ static int descendToCell(
** One of the cells in node pNode is guaranteed to have a 64-bit
** integer value equal to iRowid. Return the index of this cell.
*/
-static int nodeRowidIndex(Rtree *pRtree, RtreeNode *pNode, i64 iRowid){
+static int nodeRowidIndex(
+ Rtree *pRtree,
+ RtreeNode *pNode,
+ i64 iRowid,
+ int *piIndex
+){
int ii;
- for(ii=0; nodeGetRowid(pRtree, pNode, ii)!=iRowid; ii++){
- assert( ii<(NCELL(pNode)-1) );
+ int nCell = NCELL(pNode);
+ for(ii=0; ii<nCell; ii++){
+ if( nodeGetRowid(pRtree, pNode, ii)==iRowid ){
+ *piIndex = ii;
+ return SQLITE_OK;
+ }
}
- return ii;
+ return SQLITE_CORRUPT;
}
/*
** Return the index of the cell containing a pointer to node pNode
** in its parent. If pNode is the root node, return -1.
*/
-static int nodeParentIndex(Rtree *pRtree, RtreeNode *pNode){
+static int nodeParentIndex(Rtree *pRtree, RtreeNode *pNode, int *piIndex){
RtreeNode *pParent = pNode->pParent;
if( pParent ){
- return nodeRowidIndex(pRtree, pParent, pNode->iNode);
+ return nodeRowidIndex(pRtree, pParent, pNode->iNode, piIndex);
}
- return -1;
+ *piIndex = -1;
+ return SQLITE_OK;
}
/*
@@ -898,13 +1083,17 @@ static int rtreeNext(sqlite3_vtab_cursor *pVtabCursor){
RtreeCursor *pCsr = (RtreeCursor *)pVtabCursor;
int rc = SQLITE_OK;
+ /* RtreeCursor.pNode must not be NULL. If is is NULL, then this cursor is
+ ** already at EOF. It is against the rules to call the xNext() method of
+ ** a cursor that has already reached EOF.
+ */
+ assert( pCsr->pNode );
+
if( pCsr->iStrategy==1 ){
/* This "scan" is a direct lookup by rowid. There is no next entry. */
nodeRelease(pRtree, pCsr->pNode);
pCsr->pNode = 0;
- }
-
- else if( pCsr->pNode ){
+ }else{
/* Move to the next entry that matches the configured constraints. */
int iHeight = 0;
while( pCsr->pNode ){
@@ -918,7 +1107,10 @@ static int rtreeNext(sqlite3_vtab_cursor *pVtabCursor){
}
}
pCsr->pNode = pNode->pParent;
- pCsr->iCell = nodeParentIndex(pRtree, pNode);
+ rc = nodeParentIndex(pRtree, pNode, &pCsr->iCell);
+ if( rc!=SQLITE_OK ){
+ return rc;
+ }
nodeReference(pCsr->pNode);
nodeRelease(pRtree, pNode);
iHeight++;
@@ -986,6 +1178,51 @@ static int findLeafNode(Rtree *pRtree, i64 iRowid, RtreeNode **ppLeaf){
return rc;
}
+/*
+** This function is called to configure the RtreeConstraint object passed
+** as the second argument for a MATCH constraint. The value passed as the
+** first argument to this function is the right-hand operand to the MATCH
+** operator.
+*/
+static int deserializeGeometry(sqlite3_value *pValue, RtreeConstraint *pCons){
+ RtreeMatchArg *p;
+ sqlite3_rtree_geometry *pGeom;
+ int nBlob;
+
+ /* Check that value is actually a blob. */
+ if( !sqlite3_value_type(pValue)==SQLITE_BLOB ) return SQLITE_ERROR;
+
+ /* Check that the blob is roughly the right size. */
+ nBlob = sqlite3_value_bytes(pValue);
+ if( nBlob<sizeof(RtreeMatchArg)
+ || ((nBlob-sizeof(RtreeMatchArg))%sizeof(double))!=0
+ ){
+ return SQLITE_ERROR;
+ }
+
+ pGeom = (sqlite3_rtree_geometry *)sqlite3_malloc(
+ sizeof(sqlite3_rtree_geometry) + nBlob
+ );
+ if( !pGeom ) return SQLITE_NOMEM;
+ memset(pGeom, 0, sizeof(sqlite3_rtree_geometry));
+ p = (RtreeMatchArg *)&pGeom[1];
+
+ memcpy(p, sqlite3_value_blob(pValue), nBlob);
+ if( p->magic!=RTREE_GEOMETRY_MAGIC
+ || nBlob!=(sizeof(RtreeMatchArg) + (p->nParam-1)*sizeof(double))
+ ){
+ sqlite3_free(pGeom);
+ return SQLITE_ERROR;
+ }
+
+ pGeom->pContext = p->pContext;
+ pGeom->nParam = p->nParam;
+ pGeom->aParam = p->aParam;
+
+ pCons->xGeom = p->xGeom;
+ pCons->pGeom = pGeom;
+ return SQLITE_OK;
+}
/*
** Rtree virtual table module xFilter method.
@@ -1004,8 +1241,7 @@ static int rtreeFilter(
rtreeReference(pRtree);
- sqlite3_free(pCsr->aConstraint);
- pCsr->aConstraint = 0;
+ freeCursorConstraints(pCsr);
pCsr->iStrategy = idxNum;
if( idxNum==1 ){
@@ -1014,8 +1250,9 @@ static int rtreeFilter(
i64 iRowid = sqlite3_value_int64(argv[0]);
rc = findLeafNode(pRtree, iRowid, &pLeaf);
pCsr->pNode = pLeaf;
- if( pLeaf && rc==SQLITE_OK ){
- pCsr->iCell = nodeRowidIndex(pRtree, pLeaf, iRowid);
+ if( pLeaf ){
+ assert( rc==SQLITE_OK );
+ rc = nodeRowidIndex(pRtree, pLeaf, iRowid, &pCsr->iCell);
}
}else{
/* Normal case - r-tree scan. Set up the RtreeCursor.aConstraint array
@@ -1027,12 +1264,24 @@ static int rtreeFilter(
if( !pCsr->aConstraint ){
rc = SQLITE_NOMEM;
}else{
+ memset(pCsr->aConstraint, 0, sizeof(RtreeConstraint)*argc);
assert( (idxStr==0 && argc==0) || strlen(idxStr)==argc*2 );
for(ii=0; ii<argc; ii++){
RtreeConstraint *p = &pCsr->aConstraint[ii];
p->op = idxStr[ii*2];
p->iCoord = idxStr[ii*2+1]-'a';
- p->rValue = sqlite3_value_double(argv[ii]);
+ if( p->op==RTREE_MATCH ){
+ /* A MATCH operator. The right-hand-side must be a blob that
+ ** can be cast into an RtreeMatchArg object. One created using
+ ** an sqlite3_rtree_geometry_callback() SQL user function.
+ */
+ rc = deserializeGeometry(argv[ii], p);
+ if( rc!=SQLITE_OK ){
+ break;
+ }
+ }else{
+ p->rValue = sqlite3_value_double(argv[ii]);
+ }
}
}
}
@@ -1073,11 +1322,10 @@ static int rtreeFilter(
** idxNum idxStr Strategy
** ------------------------------------------------
** 1 Unused Direct lookup by rowid.
-** 2 See below R-tree query.
-** 3 Unused Full table scan.
+** 2 See below R-tree query or full-table scan.
** ------------------------------------------------
**
-** If strategy 1 or 3 is used, then idxStr is not meaningful. If strategy
+** If strategy 1 is used, then idxStr is not meaningful. If strategy
** 2 is used, idxStr is formatted to contain 2 bytes for each
** constraint used. The first two bytes of idxStr correspond to
** the constraint in sqlite3_index_info.aConstraintUsage[] with
@@ -1093,6 +1341,7 @@ static int rtreeFilter(
** < 0x43 ('C')
** >= 0x44 ('D')
** > 0x45 ('E')
+** MATCH 0x46 ('F')
** ----------------------
**
** The second of each pair of bytes identifies the coordinate column
@@ -1131,7 +1380,9 @@ static int rtreeBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){
return SQLITE_OK;
}
- if( p->usable && p->iColumn>0 ){
+ if( p->usable && (p->iColumn>0 || p->op==SQLITE_INDEX_CONSTRAINT_MATCH) ){
+ int j, opmsk;
+ static const unsigned char compatible[] = { 0, 0, 1, 1, 2, 2 };
u8 op = 0;
switch( p->op ){
case SQLITE_INDEX_CONSTRAINT_EQ: op = RTREE_EQ; break;
@@ -1139,31 +1390,33 @@ static int rtreeBestIndex(sqlite3_vtab *tab, sqlite3_index_info *pIdxInfo){
case SQLITE_INDEX_CONSTRAINT_LE: op = RTREE_LE; break;
case SQLITE_INDEX_CONSTRAINT_LT: op = RTREE_LT; break;
case SQLITE_INDEX_CONSTRAINT_GE: op = RTREE_GE; break;
+ default:
+ assert( p->op==SQLITE_INDEX_CONSTRAINT_MATCH );
+ op = RTREE_MATCH;
+ break;
}
- if( op ){
- /* Make sure this particular constraint has not been used before.
- ** If it has been used before, ignore it.
- **
- ** A <= or < can be used if there is a prior >= or >.
- ** A >= or > can be used if there is a prior < or <=.
- ** A <= or < is disqualified if there is a prior <=, <, or ==.
- ** A >= or > is disqualified if there is a prior >=, >, or ==.
- ** A == is disqualifed if there is any prior constraint.
- */
- int j, opmsk;
- static const unsigned char compatible[] = { 0, 0, 1, 1, 2, 2 };
- assert( compatible[RTREE_EQ & 7]==0 );
- assert( compatible[RTREE_LT & 7]==1 );
- assert( compatible[RTREE_LE & 7]==1 );
- assert( compatible[RTREE_GT & 7]==2 );
- assert( compatible[RTREE_GE & 7]==2 );
- cCol = p->iColumn - 1 + 'a';
- opmsk = compatible[op & 7];
- for(j=0; j<iIdx; j+=2){
- if( zIdxStr[j+1]==cCol && (compatible[zIdxStr[j] & 7] & opmsk)!=0 ){
- op = 0;
- break;
- }
+ assert( op!=0 );
+
+ /* Make sure this particular constraint has not been used before.
+ ** If it has been used before, ignore it.
+ **
+ ** A <= or < can be used if there is a prior >= or >.
+ ** A >= or > can be used if there is a prior < or <=.
+ ** A <= or < is disqualified if there is a prior <=, <, or ==.
+ ** A >= or > is disqualified if there is a prior >=, >, or ==.
+ ** A == is disqualifed if there is any prior constraint.
+ */
+ assert( compatible[RTREE_EQ & 7]==0 );
+ assert( compatible[RTREE_LT & 7]==1 );
+ assert( compatible[RTREE_LE & 7]==1 );
+ assert( compatible[RTREE_GT & 7]==2 );
+ assert( compatible[RTREE_GE & 7]==2 );
+ cCol = p->iColumn - 1 + 'a';
+ opmsk = compatible[op & 7];
+ for(j=0; j<iIdx; j+=2){
+ if( zIdxStr[j+1]==cCol && (compatible[zIdxStr[j] & 7] & opmsk)!=0 ){
+ op = 0;
+ break;
}
}
if( op ){
@@ -1271,7 +1524,12 @@ static float cellOverlap(
int ii;
float overlap = 0.0;
for(ii=0; ii<nCell; ii++){
- if( ii!=iExclude ){
+#if VARIANT_RSTARTREE_CHOOSESUBTREE
+ if( ii!=iExclude )
+#else
+ assert( iExclude==-1 );
+#endif
+ {
int jj;
float o = 1.0;
for(jj=0; jj<(pRtree->nDim*2); jj+=2){
@@ -1364,22 +1622,31 @@ static int ChooseLeaf(
** the smallest area.
*/
for(iCell=0; iCell<nCell; iCell++){
+ int bBest = 0;
float growth;
float area;
float overlap = 0.0;
nodeGetCell(pRtree, pNode, iCell, &cell);
growth = cellGrowth(pRtree, &cell, pCell);
area = cellArea(pRtree, &cell);
+
#if VARIANT_RSTARTREE_CHOOSESUBTREE
if( ii==(pRtree->iDepth-1) ){
overlap = cellOverlapEnlargement(pRtree,&cell,pCell,aCell,nCell,iCell);
}
-#endif
if( (iCell==0)
|| (overlap<fMinOverlap)
|| (overlap==fMinOverlap && growth<fMinGrowth)
|| (overlap==fMinOverlap && growth==fMinGrowth && area<fMinArea)
){
+ bBest = 1;
+ }
+#else
+ if( iCell==0||growth<fMinGrowth||(growth==fMinGrowth && area<fMinArea) ){
+ bBest = 1;
+ }
+#endif
+ if( bBest ){
fMinOverlap = overlap;
fMinGrowth = growth;
fMinArea = area;
@@ -1402,16 +1669,20 @@ static int ChooseLeaf(
** the node pNode. This function updates the bounding box cells in
** all ancestor elements.
*/
-static void AdjustTree(
+static int AdjustTree(
Rtree *pRtree, /* Rtree table */
RtreeNode *pNode, /* Adjust ancestry of this node. */
RtreeCell *pCell /* This cell was just inserted */
){
RtreeNode *p = pNode;
while( p->pParent ){
- RtreeCell cell;
RtreeNode *pParent = p->pParent;
- int iCell = nodeParentIndex(pRtree, p);
+ RtreeCell cell;
+ int iCell;
+
+ if( nodeParentIndex(pRtree, p, &iCell) ){
+ return SQLITE_CORRUPT;
+ }
nodeGetCell(pRtree, pParent, iCell, &cell);
if( !cellContains(pRtree, &cell, pCell) ){
@@ -1421,6 +1692,7 @@ static void AdjustTree(
p = pParent;
}
+ return SQLITE_OK;
}
/*
@@ -1486,8 +1758,8 @@ static void LinearPickSeeds(
** variables iLeftSeek and iRightSeed.
*/
for(i=0; i<pRtree->nDim; i++){
- float x1 = aCell[0].aCoord[i*2];
- float x2 = aCell[0].aCoord[i*2+1];
+ float x1 = DCOORD(aCell[0].aCoord[i*2]);
+ float x2 = DCOORD(aCell[0].aCoord[i*2+1]);
float x3 = x1;
float x4 = x2;
int jj;
@@ -1496,8 +1768,8 @@ static void LinearPickSeeds(
int iCellRight = 0;
for(jj=1; jj<nCell; jj++){
- float left = aCell[jj].aCoord[i*2];
- float right = aCell[jj].aCoord[i*2+1];
+ float left = DCOORD(aCell[jj].aCoord[i*2]);
+ float right = DCOORD(aCell[jj].aCoord[i*2+1]);
if( left<x1 ) x1 = left;
if( right>x4 ) x4 = right;
@@ -1855,6 +2127,9 @@ static int splitNodeGuttman(
int i;
aiUsed = sqlite3_malloc(sizeof(int)*nCell);
+ if( !aiUsed ){
+ return SQLITE_NOMEM;
+ }
memset(aiUsed, 0, sizeof(int)*nCell);
PickSeeds(pRtree, aCell, nCell, &iLeftSeed, &iRightSeed);
@@ -1946,14 +2221,14 @@ static int SplitNode(
nCell++;
if( pNode->iNode==1 ){
- pRight = nodeNew(pRtree, pNode, 1);
- pLeft = nodeNew(pRtree, pNode, 1);
+ pRight = nodeNew(pRtree, pNode);
+ pLeft = nodeNew(pRtree, pNode);
pRtree->iDepth++;
pNode->isDirty = 1;
writeInt16(pNode->zData, pRtree->iDepth);
}else{
pLeft = pNode;
- pRight = nodeNew(pRtree, pLeft->pParent, 1);
+ pRight = nodeNew(pRtree, pLeft->pParent);
nodeReference(pLeft);
}
@@ -1970,8 +2245,12 @@ static int SplitNode(
goto splitnode_out;
}
- /* Ensure both child nodes have node numbers assigned to them. */
- if( (0==pRight->iNode && SQLITE_OK!=(rc = nodeWrite(pRtree, pRight)))
+ /* Ensure both child nodes have node numbers assigned to them by calling
+ ** nodeWrite(). Node pRight always needs a node number, as it was created
+ ** by nodeNew() above. But node pLeft sometimes already has a node number.
+ ** In this case avoid the all to nodeWrite().
+ */
+ if( SQLITE_OK!=(rc = nodeWrite(pRtree, pRight))
|| (0==pLeft->iNode && SQLITE_OK!=(rc = nodeWrite(pRtree, pLeft)))
){
goto splitnode_out;
@@ -1987,9 +2266,15 @@ static int SplitNode(
}
}else{
RtreeNode *pParent = pLeft->pParent;
- int iCell = nodeParentIndex(pRtree, pLeft);
- nodeOverwriteCell(pRtree, pParent, &leftbbox, iCell);
- AdjustTree(pRtree, pParent, &leftbbox);
+ int iCell;
+ rc = nodeParentIndex(pRtree, pLeft, &iCell);
+ if( rc==SQLITE_OK ){
+ nodeOverwriteCell(pRtree, pParent, &leftbbox, iCell);
+ rc = AdjustTree(pRtree, pParent, &leftbbox);
+ }
+ if( rc!=SQLITE_OK ){
+ goto splitnode_out;
+ }
}
if( (rc = rtreeInsertCell(pRtree, pRight->pParent, &rightbbox, iHeight+1)) ){
goto splitnode_out;
@@ -2033,20 +2318,43 @@ splitnode_out:
return rc;
}
+/*
+** If node pLeaf is not the root of the r-tree and its pParent pointer is
+** still NULL, load all ancestor nodes of pLeaf into memory and populate
+** the pLeaf->pParent chain all the way up to the root node.
+**
+** This operation is required when a row is deleted (or updated - an update
+** is implemented as a delete followed by an insert). SQLite provides the
+** rowid of the row to delete, which can be used to find the leaf on which
+** the entry resides (argument pLeaf). Once the leaf is located, this
+** function is called to determine its ancestry.
+*/
static int fixLeafParent(Rtree *pRtree, RtreeNode *pLeaf){
int rc = SQLITE_OK;
- if( pLeaf->iNode!=1 && pLeaf->pParent==0 ){
- sqlite3_bind_int64(pRtree->pReadParent, 1, pLeaf->iNode);
- if( sqlite3_step(pRtree->pReadParent)==SQLITE_ROW ){
- i64 iNode = sqlite3_column_int64(pRtree->pReadParent, 0);
- rc = nodeAcquire(pRtree, iNode, 0, &pLeaf->pParent);
- }else{
- rc = SQLITE_ERROR;
- }
- sqlite3_reset(pRtree->pReadParent);
- if( rc==SQLITE_OK ){
- rc = fixLeafParent(pRtree, pLeaf->pParent);
+ RtreeNode *pChild = pLeaf;
+ while( rc==SQLITE_OK && pChild->iNode!=1 && pChild->pParent==0 ){
+ int rc2 = SQLITE_OK; /* sqlite3_reset() return code */
+ sqlite3_bind_int64(pRtree->pReadParent, 1, pChild->iNode);
+ rc = sqlite3_step(pRtree->pReadParent);
+ if( rc==SQLITE_ROW ){
+ RtreeNode *pTest; /* Used to test for reference loops */
+ i64 iNode; /* Node number of parent node */
+
+ /* Before setting pChild->pParent, test that we are not creating a
+ ** loop of references (as we would if, say, pChild==pParent). We don't
+ ** want to do this as it leads to a memory leak when trying to delete
+ ** the referenced counted node structures.
+ */
+ iNode = sqlite3_column_int64(pRtree->pReadParent, 0);
+ for(pTest=pLeaf; pTest && pTest->iNode!=iNode; pTest=pTest->pParent);
+ if( !pTest ){
+ rc2 = nodeAcquire(pRtree, iNode, 0, &pChild->pParent);
+ }
}
+ rc = sqlite3_reset(pRtree->pReadParent);
+ if( rc==SQLITE_OK ) rc = rc2;
+ if( rc==SQLITE_OK && !pChild->pParent ) rc = SQLITE_CORRUPT;
+ pChild = pChild->pParent;
}
return rc;
}
@@ -2055,18 +2363,24 @@ static int deleteCell(Rtree *, RtreeNode *, int, int);
static int removeNode(Rtree *pRtree, RtreeNode *pNode, int iHeight){
int rc;
+ int rc2;
RtreeNode *pParent;
int iCell;
assert( pNode->nRef==1 );
/* Remove the entry in the parent cell. */
- iCell = nodeParentIndex(pRtree, pNode);
- pParent = pNode->pParent;
- pNode->pParent = 0;
- if( SQLITE_OK!=(rc = deleteCell(pRtree, pParent, iCell, iHeight+1))
- || SQLITE_OK!=(rc = nodeRelease(pRtree, pParent))
- ){
+ rc = nodeParentIndex(pRtree, pNode, &iCell);
+ if( rc==SQLITE_OK ){
+ pParent = pNode->pParent;
+ pNode->pParent = 0;
+ rc = deleteCell(pRtree, pParent, iCell, iHeight+1);
+ }
+ rc2 = nodeRelease(pRtree, pParent);
+ if( rc==SQLITE_OK ){
+ rc = rc2;
+ }
+ if( rc!=SQLITE_OK ){
return rc;
}
@@ -2096,8 +2410,9 @@ static int removeNode(Rtree *pRtree, RtreeNode *pNode, int iHeight){
return SQLITE_OK;
}
-static void fixBoundingBox(Rtree *pRtree, RtreeNode *pNode){
+static int fixBoundingBox(Rtree *pRtree, RtreeNode *pNode){
RtreeNode *pParent = pNode->pParent;
+ int rc = SQLITE_OK;
if( pParent ){
int ii;
int nCell = NCELL(pNode);
@@ -2109,10 +2424,13 @@ static void fixBoundingBox(Rtree *pRtree, RtreeNode *pNode){
cellUnion(pRtree, &box, &cell);
}
box.iRowid = pNode->iNode;
- ii = nodeParentIndex(pRtree, pNode);
- nodeOverwriteCell(pRtree, pParent, &box, ii);
- fixBoundingBox(pRtree, pParent);
+ rc = nodeParentIndex(pRtree, pNode, &ii);
+ if( rc==SQLITE_OK ){
+ nodeOverwriteCell(pRtree, pParent, &box, ii);
+ rc = fixBoundingBox(pRtree, pParent);
+ }
}
+ return rc;
}
/*
@@ -2120,6 +2438,7 @@ static void fixBoundingBox(Rtree *pRtree, RtreeNode *pNode){
** cell, adjust the r-tree data structure if required.
*/
static int deleteCell(Rtree *pRtree, RtreeNode *pNode, int iCell, int iHeight){
+ RtreeNode *pParent;
int rc;
if( SQLITE_OK!=(rc = fixLeafParent(pRtree, pNode)) ){
@@ -2136,14 +2455,13 @@ static int deleteCell(Rtree *pRtree, RtreeNode *pNode, int iCell, int iHeight){
** cell in the parent node so that it tightly contains the updated
** node.
*/
- if( pNode->iNode!=1 ){
- RtreeNode *pParent = pNode->pParent;
- if( (pParent->iNode!=1 || NCELL(pParent)!=1)
- && (NCELL(pNode)<RTREE_MINCELLS(pRtree))
- ){
+ pParent = pNode->pParent;
+ assert( pParent || pNode->iNode==1 );
+ if( pParent ){
+ if( NCELL(pNode)<RTREE_MINCELLS(pRtree) ){
rc = removeNode(pRtree, pNode, iHeight);
}else{
- fixBoundingBox(pRtree, pNode);
+ rc = fixBoundingBox(pRtree, pNode);
}
}
@@ -2226,7 +2544,7 @@ static int Reinsert(
}
}
if( rc==SQLITE_OK ){
- fixBoundingBox(pRtree, pNode);
+ rc = fixBoundingBox(pRtree, pNode);
}
for(; rc==SQLITE_OK && ii<nCell; ii++){
/* Find a node to store this cell in. pNode->iNode currently contains
@@ -2280,11 +2598,13 @@ static int rtreeInsertCell(
rc = SplitNode(pRtree, pNode, pCell, iHeight);
#endif
}else{
- AdjustTree(pRtree, pNode, pCell);
- if( iHeight==0 ){
- rc = rowidWrite(pRtree, pCell->iRowid, pNode->iNode);
- }else{
- rc = parentWrite(pRtree, pCell->iRowid, pNode->iNode);
+ rc = AdjustTree(pRtree, pNode, pCell);
+ if( rc==SQLITE_OK ){
+ if( iHeight==0 ){
+ rc = rowidWrite(pRtree, pCell->iRowid, pNode->iNode);
+ }else{
+ rc = parentWrite(pRtree, pCell->iRowid, pNode->iNode);
+ }
}
}
return rc;
@@ -2354,7 +2674,6 @@ static int rtreeUpdate(
rtreeReference(pRtree);
assert(nData>=1);
- assert(hashIsEmpty(pRtree));
/* If azData[0] is not an SQL NULL value, it is the rowid of a
** record to delete from the r-tree table. The following block does
@@ -2380,8 +2699,10 @@ static int rtreeUpdate(
/* Delete the cell in question from the leaf node. */
if( rc==SQLITE_OK ){
int rc2;
- iCell = nodeRowidIndex(pRtree, pLeaf, iDelete);
- rc = deleteCell(pRtree, pLeaf, iCell, 0);
+ rc = nodeRowidIndex(pRtree, pLeaf, iDelete, &iCell);
+ if( rc==SQLITE_OK ){
+ rc = deleteCell(pRtree, pLeaf, iCell, 0);
+ }
rc2 = nodeRelease(pRtree, pLeaf);
if( rc==SQLITE_OK ){
rc = rc2;
@@ -2403,19 +2724,20 @@ static int rtreeUpdate(
** the root node (the operation that Gutman's paper says to perform
** in this scenario).
*/
- if( rc==SQLITE_OK && pRtree->iDepth>0 ){
- if( rc==SQLITE_OK && NCELL(pRoot)==1 ){
- RtreeNode *pChild;
- i64 iChild = nodeGetRowid(pRtree, pRoot, 0);
- rc = nodeAcquire(pRtree, iChild, pRoot, &pChild);
- if( rc==SQLITE_OK ){
- rc = removeNode(pRtree, pChild, pRtree->iDepth-1);
- }
- if( rc==SQLITE_OK ){
- pRtree->iDepth--;
- writeInt16(pRoot->zData, pRtree->iDepth);
- pRoot->isDirty = 1;
- }
+ if( rc==SQLITE_OK && pRtree->iDepth>0 && NCELL(pRoot)==1 ){
+ int rc2;
+ RtreeNode *pChild;
+ i64 iChild = nodeGetRowid(pRtree, pRoot, 0);
+ rc = nodeAcquire(pRtree, iChild, pRoot, &pChild);
+ if( rc==SQLITE_OK ){
+ rc = removeNode(pRtree, pChild, pRtree->iDepth-1);
+ }
+ rc2 = nodeRelease(pRtree, pChild);
+ if( rc==SQLITE_OK ) rc = rc2;
+ if( rc==SQLITE_OK ){
+ pRtree->iDepth--;
+ writeInt16(pRoot->zData, pRtree->iDepth);
+ pRoot->isDirty = 1;
}
}
@@ -2481,6 +2803,7 @@ static int rtreeUpdate(
}
rc = sqlite3_reset(pRtree->pReadRowid);
}
+ *pRowid = cell.iRowid;
if( rc==SQLITE_OK ){
rc = ChooseLeaf(pRtree, &cell, 0, &pLeaf);
@@ -2618,31 +2941,69 @@ static int rtreeSqlInit(
}
/*
-** This routine queries database handle db for the page-size used by
-** database zDb. If successful, the page-size in bytes is written to
-** *piPageSize and SQLITE_OK returned. Otherwise, and an SQLite error
-** code is returned.
+** The second argument to this function contains the text of an SQL statement
+** that returns a single integer value. The statement is compiled and executed
+** using database connection db. If successful, the integer value returned
+** is written to *piVal and SQLITE_OK returned. Otherwise, an SQLite error
+** code is returned and the value of *piVal after returning is not defined.
*/
-static int getPageSize(sqlite3 *db, const char *zDb, int *piPageSize){
+static int getIntFromStmt(sqlite3 *db, const char *zSql, int *piVal){
int rc = SQLITE_NOMEM;
- char *zSql;
- sqlite3_stmt *pStmt = 0;
-
- zSql = sqlite3_mprintf("PRAGMA %Q.page_size", zDb);
- if( !zSql ){
- return SQLITE_NOMEM;
+ if( zSql ){
+ sqlite3_stmt *pStmt = 0;
+ rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
+ if( rc==SQLITE_OK ){
+ if( SQLITE_ROW==sqlite3_step(pStmt) ){
+ *piVal = sqlite3_column_int(pStmt, 0);
+ }
+ rc = sqlite3_finalize(pStmt);
+ }
}
+ return rc;
+}
- rc = sqlite3_prepare_v2(db, zSql, -1, &pStmt, 0);
- sqlite3_free(zSql);
- if( rc!=SQLITE_OK ){
- return rc;
+/*
+** This function is called from within the xConnect() or xCreate() method to
+** determine the node-size used by the rtree table being created or connected
+** to. If successful, pRtree->iNodeSize is populated and SQLITE_OK returned.
+** Otherwise, an SQLite error code is returned.
+**
+** If this function is being called as part of an xConnect(), then the rtree
+** table already exists. In this case the node-size is determined by inspecting
+** the root node of the tree.
+**
+** Otherwise, for an xCreate(), use 64 bytes less than the database page-size.
+** This ensures that each node is stored on a single database page. If the
+** database page-size is so large that more than RTREE_MAXCELLS entries
+** would fit in a single node, use a smaller node-size.
+*/
+static int getNodeSize(
+ sqlite3 *db, /* Database handle */
+ Rtree *pRtree, /* Rtree handle */
+ int isCreate /* True for xCreate, false for xConnect */
+){
+ int rc;
+ char *zSql;
+ if( isCreate ){
+ int iPageSize;
+ zSql = sqlite3_mprintf("PRAGMA %Q.page_size", pRtree->zDb);
+ rc = getIntFromStmt(db, zSql, &iPageSize);
+ if( rc==SQLITE_OK ){
+ pRtree->iNodeSize = iPageSize-64;
+ if( (4+pRtree->nBytesPerCell*RTREE_MAXCELLS)<pRtree->iNodeSize ){
+ pRtree->iNodeSize = 4+pRtree->nBytesPerCell*RTREE_MAXCELLS;
+ }
+ }
+ }else{
+ zSql = sqlite3_mprintf(
+ "SELECT length(data) FROM '%q'.'%q_node' WHERE nodeno = 1",
+ pRtree->zDb, pRtree->zName
+ );
+ rc = getIntFromStmt(db, zSql, &pRtree->iNodeSize);
}
- if( SQLITE_ROW==sqlite3_step(pStmt) ){
- *piPageSize = sqlite3_column_int(pStmt, 0);
- }
- return sqlite3_finalize(pStmt);
+ sqlite3_free(zSql);
+ return rc;
}
/*
@@ -2663,11 +3024,10 @@ static int rtreeInit(
int isCreate /* True for xCreate, false for xConnect */
){
int rc = SQLITE_OK;
- int iPageSize = 0;
Rtree *pRtree;
int nDb; /* Length of string argv[1] */
int nName; /* Length of string argv[2] */
- int eCoordType = (int)pAux;
+ int eCoordType = (pAux ? RTREE_COORD_INT32 : RTREE_COORD_REAL32);
const char *aErrMsg[] = {
0, /* 0 */
@@ -2682,11 +3042,6 @@ static int rtreeInit(
return SQLITE_ERROR;
}
- rc = getPageSize(db, argv[1], &iPageSize);
- if( rc!=SQLITE_OK ){
- return rc;
- }
-
/* Allocate the sqlite3_vtab structure */
nDb = strlen(argv[1]);
nName = strlen(argv[2]);
@@ -2705,44 +3060,37 @@ static int rtreeInit(
memcpy(pRtree->zDb, argv[1], nDb);
memcpy(pRtree->zName, argv[2], nName);
- /* Figure out the node size to use. By default, use 64 bytes less than
- ** the database page-size. This ensures that each node is stored on
- ** a single database page.
- **
- ** If the databasd page-size is so large that more than RTREE_MAXCELLS
- ** entries would fit in a single node, use a smaller node-size.
- */
- pRtree->iNodeSize = iPageSize-64;
- if( (4+pRtree->nBytesPerCell*RTREE_MAXCELLS)<pRtree->iNodeSize ){
- pRtree->iNodeSize = 4+pRtree->nBytesPerCell*RTREE_MAXCELLS;
- }
+ /* Figure out the node size to use. */
+ rc = getNodeSize(db, pRtree, isCreate);
/* Create/Connect to the underlying relational database schema. If
** that is successful, call sqlite3_declare_vtab() to configure
** the r-tree table schema.
*/
- if( (rc = rtreeSqlInit(pRtree, db, argv[1], argv[2], isCreate)) ){
- *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db));
- }else{
- char *zSql = sqlite3_mprintf("CREATE TABLE x(%s", argv[3]);
- char *zTmp;
- int ii;
- for(ii=4; zSql && ii<argc; ii++){
- zTmp = zSql;
- zSql = sqlite3_mprintf("%s, %s", zTmp, argv[ii]);
- sqlite3_free(zTmp);
- }
- if( zSql ){
- zTmp = zSql;
- zSql = sqlite3_mprintf("%s);", zTmp);
- sqlite3_free(zTmp);
- }
- if( !zSql ){
- rc = SQLITE_NOMEM;
- }else if( SQLITE_OK!=(rc = sqlite3_declare_vtab(db, zSql)) ){
+ if( rc==SQLITE_OK ){
+ if( (rc = rtreeSqlInit(pRtree, db, argv[1], argv[2], isCreate)) ){
*pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db));
+ }else{
+ char *zSql = sqlite3_mprintf("CREATE TABLE x(%s", argv[3]);
+ char *zTmp;
+ int ii;
+ for(ii=4; zSql && ii<argc; ii++){
+ zTmp = zSql;
+ zSql = sqlite3_mprintf("%s, %s", zTmp, argv[ii]);
+ sqlite3_free(zTmp);
+ }
+ if( zSql ){
+ zTmp = zSql;
+ zSql = sqlite3_mprintf("%s);", zTmp);
+ sqlite3_free(zTmp);
+ }
+ if( !zSql ){
+ rc = SQLITE_NOMEM;
+ }else if( SQLITE_OK!=(rc = sqlite3_declare_vtab(db, zSql)) ){
+ *pzErr = sqlite3_mprintf("%s", sqlite3_errmsg(db));
+ }
+ sqlite3_free(zSql);
}
- sqlite3_free(zSql);
}
if( rc==SQLITE_OK ){
@@ -2825,12 +3173,10 @@ static void rtreedepth(sqlite3_context *ctx, int nArg, sqlite3_value **apArg){
** function "rtreenode".
*/
int sqlite3RtreeInit(sqlite3 *db){
- int rc = SQLITE_OK;
+ const int utf8 = SQLITE_UTF8;
+ int rc;
- if( rc==SQLITE_OK ){
- int utf8 = SQLITE_UTF8;
- rc = sqlite3_create_function(db, "rtreenode", 2, utf8, 0, rtreenode, 0, 0);
- }
+ rc = sqlite3_create_function(db, "rtreenode", 2, utf8, 0, rtreenode, 0, 0);
if( rc==SQLITE_OK ){
int utf8 = SQLITE_UTF8;
rc = sqlite3_create_function(db, "rtreedepth", 1, utf8, 0,rtreedepth, 0, 0);
@@ -2847,6 +3193,70 @@ int sqlite3RtreeInit(sqlite3 *db){
return rc;
}
+/*
+** A version of sqlite3_free() that can be used as a callback. This is used
+** in two places - as the destructor for the blob value returned by the
+** invocation of a geometry function, and as the destructor for the geometry
+** functions themselves.
+*/
+static void doSqlite3Free(void *p){
+ sqlite3_free(p);
+}
+
+/*
+** Each call to sqlite3_rtree_geometry_callback() creates an ordinary SQLite
+** scalar user function. This C function is the callback used for all such
+** registered SQL functions.
+**
+** The scalar user functions return a blob that is interpreted by r-tree
+** table MATCH operators.
+*/
+static void geomCallback(sqlite3_context *ctx, int nArg, sqlite3_value **aArg){
+ RtreeGeomCallback *pGeomCtx = (RtreeGeomCallback *)sqlite3_user_data(ctx);
+ RtreeMatchArg *pBlob;
+ int nBlob;
+
+ nBlob = sizeof(RtreeMatchArg) + (nArg-1)*sizeof(double);
+ pBlob = (RtreeMatchArg *)sqlite3_malloc(nBlob);
+ if( !pBlob ){
+ sqlite3_result_error_nomem(ctx);
+ }else{
+ int i;
+ pBlob->magic = RTREE_GEOMETRY_MAGIC;
+ pBlob->xGeom = pGeomCtx->xGeom;
+ pBlob->pContext = pGeomCtx->pContext;
+ pBlob->nParam = nArg;
+ for(i=0; i<nArg; i++){
+ pBlob->aParam[i] = sqlite3_value_double(aArg[i]);
+ }
+ sqlite3_result_blob(ctx, pBlob, nBlob, doSqlite3Free);
+ }
+}
+
+/*
+** Register a new geometry function for use with the r-tree MATCH operator.
+*/
+int sqlite3_rtree_geometry_callback(
+ sqlite3 *db,
+ const char *zGeom,
+ int (*xGeom)(sqlite3_rtree_geometry *, int, double *, int *),
+ void *pContext
+){
+ RtreeGeomCallback *pGeomCtx; /* Context object for new user-function */
+
+ /* Allocate and populate the context object. */
+ pGeomCtx = (RtreeGeomCallback *)sqlite3_malloc(sizeof(RtreeGeomCallback));
+ if( !pGeomCtx ) return SQLITE_NOMEM;
+ pGeomCtx->xGeom = xGeom;
+ pGeomCtx->pContext = pContext;
+
+ /* Create the new user-function. Register a destructor function to delete
+ ** the context object when it is no longer required. */
+ return sqlite3_create_function_v2(db, zGeom, -1, SQLITE_ANY,
+ (void *)pGeomCtx, geomCallback, 0, 0, doSqlite3Free
+ );
+}
+
#if !SQLITE_CORE
int sqlite3_extension_init(
sqlite3 *db,
diff --git a/third_party/sqlite/src/ext/rtree/rtree1.test b/third_party/sqlite/src/ext/rtree/rtree1.test
index 8c1675b..fe5fa0a 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree1.test
+++ b/third_party/sqlite/src/ext/rtree/rtree1.test
@@ -11,11 +11,9 @@
#
# The focus of this file is testing the r-tree extension.
#
-# $Id: rtree1.test,v 1.7 2009/07/17 16:54:48 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source [file join [file dirname [info script]] rtree_util.tcl]
source $testdir/tester.tcl
@@ -401,4 +399,21 @@ do_test rtree-10.1 {
catchsql { CREATE VIRTUAL TABLE t7 USING rtree(index, x1, y1, x2, y2) }
} {1 {near "index": syntax error}}
+#-------------------------------------------------------------------------
+# Test last_insert_rowid().
+#
+do_test rtree-11.1 {
+ execsql {
+ CREATE VIRTUAL TABLE t8 USING rtree(idx, x1, x2, y1, y2);
+ INSERT INTO t8 VALUES(1, 1.0, 1.0, 2.0, 2.0);
+ SELECT last_insert_rowid();
+ }
+} {1}
+do_test rtree-11.2 {
+ execsql {
+ INSERT INTO t8 VALUES(NULL, 1.0, 1.0, 2.0, 2.0);
+ SELECT last_insert_rowid();
+ }
+} {2}
+
finish_test
diff --git a/third_party/sqlite/src/ext/rtree/rtree2.test b/third_party/sqlite/src/ext/rtree/rtree2.test
index 7e38c8f..f5d15cc 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree2.test
+++ b/third_party/sqlite/src/ext/rtree/rtree2.test
@@ -11,11 +11,9 @@
#
# The focus of this file is testing the r-tree extension.
#
-# $Id: rtree2.test,v 1.4 2008/07/14 15:37:01 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source [file join [file dirname [info script]] rtree_util.tcl]
source $testdir/tester.tcl
@@ -29,7 +27,7 @@ set ::NROW 1000
set ::NDEL 10
set ::NSELECT 100
-if {[info exists ISQUICK] && $ISQUICK} {
+if {[info exists G(isquick)] && $G(isquick)} {
set ::NROW 100
set ::NSELECT 10
}
diff --git a/third_party/sqlite/src/ext/rtree/rtree3.test b/third_party/sqlite/src/ext/rtree/rtree3.test
index b83ceeb4..fea5513 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree3.test
+++ b/third_party/sqlite/src/ext/rtree/rtree3.test
@@ -12,56 +12,95 @@
# The focus of this file is testing that the r-tree correctly handles
# out-of-memory conditions.
#
-# $Id: rtree3.test,v 1.2 2008/06/23 15:55:52 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source $testdir/tester.tcl
-
+source $testdir/malloc_common.tcl
ifcapable !rtree {
finish_test
return
}
-# Only run these tests if memory debugging is turned on.
+# Test summary:
#
-source $testdir/malloc_common.tcl
-if {!$MEMDEBUG} {
- puts "Skipping malloc tests: not compiled with -DSQLITE_MEMDEBUG..."
- finish_test
- return
+# rtree3-1: Test OOM in simple CREATE TABLE, INSERT, DELETE and SELECT
+# commands on an almost empty table.
+#
+# rtree3-2: Test OOM in a DROP TABLE command.
+#
+# rtree3-3a: Test OOM during a transaction to insert 100 pseudo-random rows.
+#
+# rtree3-3b: Test OOM during a transaction deleting all entries in the
+# database constructed in [rtree3-3a] in pseudo-random order.
+#
+# rtree3-4a: OOM during "SELECT count(*) FROM ..." on a big table.
+#
+# rtree3-4b: OOM while deleting rows from a big table.
+#
+# rtree3-5: Test OOM while inserting rows into a big table.
+#
+# rtree3-6: Test OOM while deleting all rows of a table, one at a time.
+#
+# rtree3-7: OOM during an ALTER TABLE RENAME TABLE command.
+#
+# rtree3-8: Test OOM while registering the r-tree module with sqlite.
+#
+
+do_faultsim_test rtree3-1 -faults oom* -prep {
+ faultsim_delete_and_reopen
+} -body {
+ execsql {
+ BEGIN TRANSACTION;
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
+ INSERT INTO rt VALUES(NULL, 13, 15, 17, 19);
+ DELETE FROM rt WHERE ii = 1;
+ SELECT * FROM rt;
+ SELECT ii FROM rt WHERE ii = 2;
+ COMMIT;
+ }
}
-do_malloc_test rtree3-1 -sqlbody {
- BEGIN TRANSACTION;
- CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
- INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
- INSERT INTO rt VALUES(NULL, 13, 15, 17, 19);
- DELETE FROM rt WHERE ii = 1;
- SELECT * FROM rt;
- SELECT ii FROM rt WHERE ii = 2;
- COMMIT;
-}
-do_malloc_test rtree3-2 -sqlprep {
- CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
- INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
-} -sqlbody {
- DROP TABLE rt;
-}
+do_test rtree3-2.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
+ }
+ faultsim_save_and_close
+} {}
+do_faultsim_test rtree3-2 -faults oom* -prep {
+ faultsim_restore_and_reopen
+} -body {
+ execsql { DROP TABLE rt }
+}
+do_malloc_test rtree3-3.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
+ }
+ faultsim_save_and_close
+} {}
-do_malloc_test rtree3-3 -sqlprep {
- CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
- INSERT INTO rt VALUES(NULL, 3, 5, 7, 9);
-} -tclbody {
+do_faultsim_test rtree3-3a -faults oom* -prep {
+ faultsim_restore_and_reopen
+} -body {
db eval BEGIN
for {set ii 0} {$ii < 100} {incr ii} {
set f [expr rand()]
db eval {INSERT INTO rt VALUES(NULL, $f*10.0, $f*10.0, $f*15.0, $f*15.0)}
}
db eval COMMIT
+}
+faultsim_save_and_close
+
+do_faultsim_test rtree3-3b -faults oom* -prep {
+ faultsim_restore_and_reopen
+} -body {
db eval BEGIN
for {set ii 0} {$ii < 100} {incr ii} {
set f [expr rand()]
@@ -70,5 +109,129 @@ do_malloc_test rtree3-3 -sqlprep {
db eval COMMIT
}
-finish_test
+do_test rtree3-4.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ BEGIN;
+ PRAGMA page_size = 512;
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ }
+ for {set i 0} {$i < 1500} {incr i} {
+ execsql { INSERT INTO rt VALUES($i, $i, $i+1, $i, $i+1) }
+ }
+ execsql { COMMIT }
+ faultsim_save_and_close
+} {}
+
+do_faultsim_test rtree3-4a -faults oom-* -prep {
+ faultsim_restore_and_reopen
+} -body {
+ db eval { SELECT count(*) FROM rt }
+} -test {
+ faultsim_test_result {0 1500}
+}
+do_faultsim_test rtree3-4b -faults oom-transient -prep {
+ faultsim_restore_and_reopen
+} -body {
+ db eval { DELETE FROM rt WHERE ii BETWEEN 1 AND 100 }
+} -test {
+ faultsim_test_result {0 {}}
+}
+
+do_test rtree3-5.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ BEGIN;
+ PRAGMA page_size = 512;
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ }
+ for {set i 0} {$i < 100} {incr i} {
+ execsql { INSERT INTO rt VALUES($i, $i, $i+1, $i, $i+1) }
+ }
+ execsql { COMMIT }
+ faultsim_save_and_close
+} {}
+do_faultsim_test rtree3-5 -faults oom-* -prep {
+ faultsim_restore_and_reopen
+} -body {
+ for {set i 100} {$i < 110} {incr i} {
+ execsql { INSERT INTO rt VALUES($i, $i, $i+1, $i, $i+1) }
+ }
+} -test {
+ faultsim_test_result {0 {}}
+}
+
+do_test rtree3-6.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ BEGIN;
+ PRAGMA page_size = 512;
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2);
+ }
+ for {set i 0} {$i < 50} {incr i} {
+ execsql { INSERT INTO rt VALUES($i, $i, $i+1, $i, $i+1) }
+ }
+ execsql { COMMIT }
+ faultsim_save_and_close
+} {}
+do_faultsim_test rtree3-6 -faults oom-* -prep {
+ faultsim_restore_and_reopen
+} -body {
+ execsql BEGIN
+ for {set i 0} {$i < 50} {incr i} {
+ execsql { DELETE FROM rt WHERE ii=$i }
+ }
+ execsql COMMIT
+} -test {
+ faultsim_test_result {0 {}}
+}
+
+do_test rtree3-7.prep {
+ faultsim_delete_and_reopen
+ execsql { CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2) }
+ faultsim_save_and_close
+} {}
+do_faultsim_test rtree3-7 -faults oom-* -prep {
+ faultsim_restore_and_reopen
+} -body {
+ execsql { ALTER TABLE rt RENAME TO rt2 }
+} -test {
+ faultsim_test_result {0 {}}
+}
+
+do_faultsim_test rtree3-8 -faults oom-* -prep {
+ catch { db close }
+} -body {
+ sqlite3 db test.db
+}
+
+do_faultsim_test rtree3-9 -faults oom-* -prep {
+ sqlite3 db :memory:
+} -body {
+ set rc [register_cube_geom db]
+ if {$rc != "SQLITE_OK"} { error $rc }
+} -test {
+ faultsim_test_result {0 {}} {1 SQLITE_NOMEM}
+}
+
+do_test rtree3-10.prep {
+ faultsim_delete_and_reopen
+ execsql {
+ CREATE VIRTUAL TABLE rt USING rtree(ii, x1, x2, y1, y2, z1, z2);
+ INSERT INTO rt VALUES(1, 10, 10, 10, 11, 11, 11);
+ INSERT INTO rt VALUES(2, 5, 6, 6, 7, 7, 8);
+ }
+ faultsim_save_and_close
+} {}
+do_faultsim_test rtree3-10 -faults oom-* -prep {
+ faultsim_restore_and_reopen
+ register_cube_geom db
+ execsql { SELECT * FROM rt }
+} -body {
+ execsql { SELECT ii FROM rt WHERE ii MATCH cube(4.5, 5.5, 6.5, 1, 1, 1) }
+} -test {
+ faultsim_test_result {0 2}
+}
+
+finish_test
diff --git a/third_party/sqlite/src/ext/rtree/rtree4.test b/third_party/sqlite/src/ext/rtree/rtree4.test
index d73e7a6..708d335 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree4.test
+++ b/third_party/sqlite/src/ext/rtree/rtree4.test
@@ -11,11 +11,9 @@
#
# Randomized test cases for the rtree extension.
#
-# $Id: rtree4.test,v 1.3 2008/06/23 15:55:52 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source $testdir/tester.tcl
@@ -25,7 +23,7 @@ ifcapable !rtree {
}
set ::NROW 2500
-if {[info exists ISQUICK] && $ISQUICK} {
+if {[info exists G(isquick)] && $G(isquick)} {
set ::NROW 250
}
@@ -95,7 +93,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
for {set i 1} {$i<$::NROW} {incr i} {
# Do a random insert
#
- do_test rtree-$nDim.2.$i.1 {
+ do_test rtree4-$nDim.2.$i.1 {
set vlist {}
for {set j 0} {$j<$nDim} {incr j} {
set mn [rand 10000]
@@ -115,7 +113,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mn$j>=$mn mx$j<=$mx
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.2 {
+ do_test rtree4-$nDim.2.$i.2 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -128,7 +126,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mx$j>=$mn mn$j<=$mx
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.3 {
+ do_test rtree4-$nDim.2.$i.3 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -145,7 +143,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mn$j>=$mn mx$j<=$mx
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.3 {
+ do_test rtree4-$nDim.2.$i.3 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -162,7 +160,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mx$j>$mn mn$j<$mx
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.4 {
+ do_test rtree4-$nDim.2.$i.4 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -178,7 +176,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mn$j>=-10000 mx$j<10000
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.5 {
+ do_test rtree4-$nDim.2.$i.5 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -194,7 +192,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mx$j>-10000 mn$j<=10000
}
set where "WHERE [join $where { AND }]"
- do_test rtree-$nDim.2.$i.6 {
+ do_test rtree4-$nDim.2.$i.6 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -210,7 +208,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mn$j>=$mn1 mn$j>$mn2 mx$j<$mx1 mx$j<=$mx2
}
set where "WHERE [join [scramble $where] { AND }]"
- do_test rtree-$nDim.2.$i.7 {
+ do_test rtree4-$nDim.2.$i.7 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
@@ -226,7 +224,7 @@ for {set nDim 1} {$nDim<=5} {incr nDim} {
lappend where mx$j>=$mn1 mx$j>$mn2 mn$j<$mx1 mn$j<=$mx2
}
set where "WHERE [join [scramble $where] { AND }]"
- do_test rtree-$nDim.2.$i.8 {
+ do_test rtree4-$nDim.2.$i.8 {
list $where [db eval "SELECT id FROM rx $where ORDER BY id"]
} [list $where [db eval "SELECT id FROM bx $where ORDER BY id"]]
}
diff --git a/third_party/sqlite/src/ext/rtree/rtree5.test b/third_party/sqlite/src/ext/rtree/rtree5.test
index 4fa007f..ea2946f 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree5.test
+++ b/third_party/sqlite/src/ext/rtree/rtree5.test
@@ -12,11 +12,9 @@
# The focus of this file is testing the r-tree extension when it is
# configured to store values as 32 bit integers.
#
-# $Id: rtree5.test,v 1.1 2008/07/14 15:37:01 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source $testdir/tester.tcl
diff --git a/third_party/sqlite/src/ext/rtree/rtree6.test b/third_party/sqlite/src/ext/rtree/rtree6.test
index affa8fe..0a65c15 100644
--- a/third_party/sqlite/src/ext/rtree/rtree6.test
+++ b/third_party/sqlite/src/ext/rtree/rtree6.test
@@ -8,12 +8,11 @@
# May you share freely, never taking more than you give.
#
#***********************************************************************
-#
-# $Id: rtree6.test,v 1.1 2008/09/01 12:47:00 danielk1977 Exp $
+#
#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source $testdir/tester.tcl
@@ -89,8 +88,8 @@ do_test rtree6.2.2 {
do_test rtree6.2.3 {
query_plan {SELECT * FROM t1,t2 WHERE k=ii}
} [list \
- {TABLE t2} \
- {TABLE t1 VIRTUAL TABLE INDEX 1:} \
+ {TABLE t1 VIRTUAL TABLE INDEX 2:} \
+ {TABLE t2 USING PRIMARY KEY} \
]
do_test rtree6.2.4 {
@@ -103,9 +102,8 @@ do_test rtree6.2.4 {
do_test rtree6.2.5 {
query_plan {SELECT * FROM t1,t2 WHERE k=ii AND x1<v}
} [list \
- {TABLE t2} \
- {TABLE t1 VIRTUAL TABLE INDEX 1:} \
+ {TABLE t1 VIRTUAL TABLE INDEX 2:} \
+ {TABLE t2 USING PRIMARY KEY} \
]
finish_test
-
diff --git a/third_party/sqlite/src/ext/rtree/rtree7.test b/third_party/sqlite/src/ext/rtree/rtree7.test
new file mode 100644
index 0000000..31dae0c
--- /dev/null
+++ b/third_party/sqlite/src/ext/rtree/rtree7.test
@@ -0,0 +1,58 @@
+# 2010 February 16
+#
+# The author disclaims copyright to this source code. In place of
+# a legal notice, here is a blessing:
+#
+# May you do good and not evil.
+# May you find forgiveness for yourself and forgive others.
+# May you share freely, never taking more than you give.
+#
+#***********************************************************************
+#
+# Test that nothing goes wrong if an rtree table is created, then the
+# database page-size is modified. At one point (3.6.22), this was causing
+# malfunctions.
+#
+
+if {![info exists testdir]} {
+ set testdir [file join [file dirname [info script]] .. .. test]
+}
+source $testdir/tester.tcl
+
+ifcapable !rtree||!vacuum {
+ finish_test
+ return
+}
+
+do_test rtree7-1.1 {
+ execsql {
+ PRAGMA page_size = 1024;
+ CREATE VIRTUAL TABLE rt USING rtree(id, x1, x2, y1, y2);
+ INSERT INTO rt VALUES(1, 1, 2, 3, 4);
+ }
+} {}
+do_test rtree7-1.2 {
+ execsql { SELECT * FROM rt }
+} {1 1.0 2.0 3.0 4.0}
+do_test rtree7-1.3 {
+ execsql {
+ PRAGMA page_size = 2048;
+ VACUUM;
+ SELECT * FROM rt;
+ }
+} {1 1.0 2.0 3.0 4.0}
+do_test rtree7-1.4 {
+ for {set i 2} {$i <= 51} {incr i} {
+ execsql { INSERT INTO rt VALUES($i, 1, 2, 3, 4) }
+ }
+ execsql { SELECT sum(x1), sum(x2), sum(y1), sum(y2) FROM rt }
+} {51.0 102.0 153.0 204.0}
+do_test rtree7-1.5 {
+ execsql {
+ PRAGMA page_size = 512;
+ VACUUM;
+ SELECT sum(x1), sum(x2), sum(y1), sum(y2) FROM rt
+ }
+} {51.0 102.0 153.0 204.0}
+
+finish_test
diff --git a/third_party/sqlite/src/ext/rtree/rtree8.test b/third_party/sqlite/src/ext/rtree/rtree8.test
new file mode 100644
index 0000000..bf22cbf
--- /dev/null
+++ b/third_party/sqlite/src/ext/rtree/rtree8.test
@@ -0,0 +1,171 @@
+# 2010 February 16
+#
+# The author disclaims copyright to this source code. In place of
+# a legal notice, here is a blessing:
+#
+# May you do good and not evil.
+# May you find forgiveness for yourself and forgive others.
+# May you share freely, never taking more than you give.
+#
+#***********************************************************************
+#
+#
+
+if {![info exists testdir]} {
+ set testdir [file join [file dirname [info script]] .. .. test]
+}
+source $testdir/tester.tcl
+ifcapable !rtree { finish_test ; return }
+
+#-------------------------------------------------------------------------
+# The following block of tests - rtree8-1.* - feature reading and writing
+# an r-tree table while there exist open cursors on it.
+#
+proc populate_t1 {n} {
+ execsql { DELETE FROM t1 }
+ for {set i 1} {$i <= $n} {incr i} {
+ execsql { INSERT INTO t1 VALUES($i, $i, $i+2) }
+ }
+}
+
+# A DELETE while a cursor is reading the table.
+#
+do_test rtree8-1.1.1 {
+ execsql { PRAGMA page_size = 512 }
+ execsql { CREATE VIRTUAL TABLE t1 USING rtree_i32(id, x1, x2) }
+ populate_t1 5
+} {}
+do_test rtree8-1.1.2 {
+ set res [list]
+ db eval { SELECT * FROM t1 } {
+ lappend res $x1 $x2
+ if {$id==3} { db eval { DELETE FROM t1 WHERE id>3 } }
+ }
+ set res
+} {1 3 2 4 3 5}
+do_test rtree8-1.1.3 {
+ execsql { SELECT * FROM t1 }
+} {1 1 3 2 2 4 3 3 5}
+
+# Many SELECTs on the same small table.
+#
+proc nested_select {n} {
+ set ::max $n
+ db eval { SELECT * FROM t1 } {
+ if {$id == $n} { nested_select [expr $n+1] }
+ }
+ return $::max
+}
+do_test rtree8-1.2.1 { populate_t1 50 } {}
+do_test rtree8-1.2.2 { nested_select 1 } {51}
+
+# This test runs many SELECT queries simultaneously against a large
+# table, causing a collision in the hash-table used to store r-tree
+# nodes internally.
+#
+populate_t1 1500
+do_execsql_test rtree8-1.3.1 { SELECT max(nodeno) FROM t1_node } {164}
+do_test rtree8-1.3.2 {
+ set rowids [execsql {SELECT min(rowid) FROM t1_rowid GROUP BY nodeno}]
+ set stmt_list [list]
+ foreach row $rowids {
+ set stmt [sqlite3_prepare db "SELECT * FROM t1 WHERE id = $row" -1 tail]
+ sqlite3_step $stmt
+ lappend res_list [sqlite3_column_int $stmt 0]
+ lappend stmt_list $stmt
+ }
+} {}
+do_test rtree8-1.3.3 { set res_list } $rowids
+do_execsql_test rtree8-1.3.4 { SELECT count(*) FROM t1 } {1500}
+do_test rtree8-1.3.5 {
+ foreach stmt $stmt_list { sqlite3_finalize $stmt }
+} {}
+
+
+#-------------------------------------------------------------------------
+# The following block of tests - rtree8-2.* - test a couple of database
+# corruption cases. In this case things are not corrupted at the b-tree
+# level, but the contents of the various tables used internally by an
+# r-tree table are inconsistent.
+#
+populate_t1 50
+do_execsql_test rtree8-2.1.1 { SELECT max(nodeno) FROM t1_node } {5}
+do_execsql_test rtree8-2.1.2 { DELETE FROM t1_node } {}
+for {set i 1} {$i <= 50} {incr i} {
+ do_catchsql_test rtree8-2.1.3.$i {
+ SELECT * FROM t1 WHERE id = $i
+ } {1 {database disk image is malformed}}
+}
+do_catchsql_test rtree8-2.1.4 {
+ SELECT * FROM t1
+} {1 {database disk image is malformed}}
+do_catchsql_test rtree8-2.1.5 {
+ DELETE FROM t1
+} {1 {database disk image is malformed}}
+
+do_execsql_test rtree8-2.1.6 {
+ DROP TABLE t1;
+ CREATE VIRTUAL TABLE t1 USING rtree_i32(id, x1, x2);
+} {}
+
+
+populate_t1 50
+do_execsql_test rtree8-2.2.1 {
+ DELETE FROM t1_parent
+} {}
+do_catchsql_test rtree8-2.2.2 {
+ DELETE FROM t1 WHERE id=25
+} {1 {database disk image is malformed}}
+do_execsql_test rtree8-2.2.3 {
+ DROP TABLE t1;
+ CREATE VIRTUAL TABLE t1 USING rtree_i32(id, x1, x2);
+} {}
+
+
+#-------------------------------------------------------------------------
+# Test that trying to use the MATCH operator with the r-tree module does
+# not confuse it.
+#
+populate_t1 10
+do_catchsql_test rtree8-3.1 {
+ SELECT * FROM t1 WHERE x1 MATCH '1234'
+} {1 {SQL logic error or missing database}}
+
+#-------------------------------------------------------------------------
+# Test a couple of invalid arguments to rtreedepth().
+#
+do_catchsql_test rtree8-4.1 {
+ SELECT rtreedepth('hello world')
+} {1 {Invalid argument to rtreedepth()}}
+do_catchsql_test rtree8-4.2 {
+ SELECT rtreedepth(X'00')
+} {1 {Invalid argument to rtreedepth()}}
+
+
+#-------------------------------------------------------------------------
+# Delete half of a lopsided tree.
+#
+do_execsql_test rtree8-5.1 {
+ CREATE VIRTUAL TABLE t2 USING rtree_i32(id, x1, x2)
+} {}
+do_test rtree8-5.2 {
+ execsql BEGIN
+ for {set i 0} {$i < 100} {incr i} {
+ execsql { INSERT INTO t2 VALUES($i, 100, 101) }
+ }
+ for {set i 100} {$i < 200} {incr i} {
+ execsql { INSERT INTO t2 VALUES($i, 1000, 1001) }
+ }
+ execsql COMMIT
+} {}
+do_test rtree8-5.3 {
+ execsql BEGIN
+ for {set i 0} {$i < 200} {incr i} {
+ execsql { DELETE FROM t2 WHERE id = $i }
+ }
+ execsql COMMIT
+} {}
+
+
+finish_test
+
diff --git a/third_party/sqlite/src/ext/rtree/rtree9.test b/third_party/sqlite/src/ext/rtree/rtree9.test
new file mode 100644
index 0000000..ddee277
--- /dev/null
+++ b/third_party/sqlite/src/ext/rtree/rtree9.test
@@ -0,0 +1,125 @@
+# 2010 August 28
+#
+# The author disclaims copyright to this source code. In place of
+# a legal notice, here is a blessing:
+#
+# May you do good and not evil.
+# May you find forgiveness for yourself and forgive others.
+# May you share freely, never taking more than you give.
+#
+#***********************************************************************
+# This file contains tests for the r-tree module. Specifically, it tests
+# that custom r-tree queries (geometry callbacks) work.
+#
+
+if {![info exists testdir]} {
+ set testdir [file join [file dirname [info script]] .. .. test]
+}
+source $testdir/tester.tcl
+ifcapable !rtree { finish_test ; return }
+
+register_cube_geom db
+
+do_execsql_test rtree9-1.1 {
+ CREATE VIRTUAL TABLE rt USING rtree(id, x1, x2, y1, y2, z1, z2);
+ INSERT INTO rt VALUES(1, 1, 2, 1, 2, 1, 2);
+} {}
+do_execsql_test rtree9-1.2 {
+ SELECT * FROM rt WHERE id MATCH cube(0, 0, 0, 2, 2, 2);
+} {1 1.0 2.0 1.0 2.0 1.0 2.0}
+do_execsql_test rtree9-1.3 {
+ SELECT * FROM rt WHERE id MATCH cube(3, 3, 3, 2, 2, 2);
+} {}
+do_execsql_test rtree9-1.4 {
+ DELETE FROM rt;
+} {}
+
+
+for {set i 0} {$i < 1000} {incr i} {
+ set x [expr $i%10]
+ set y [expr ($i/10)%10]
+ set z [expr ($i/100)%10]
+ execsql { INSERT INTO rt VALUES($i, $x, $x+1, $y, $y+1, $z, $z+1) }
+}
+do_execsql_test rtree9-2.1 {
+ SELECT id FROM rt WHERE id MATCH cube(2.5, 2.5, 2.5, 1, 1, 1) ORDER BY id;
+} {222 223 232 233 322 323 332 333}
+do_execsql_test rtree9-2.2 {
+ SELECT id FROM rt WHERE id MATCH cube(5.5, 5.5, 5.5, 1, 1, 1) ORDER BY id;
+} {555 556 565 566 655 656 665 666}
+
+
+do_execsql_test rtree9-3.1 {
+ CREATE VIRTUAL TABLE rt32 USING rtree_i32(id, x1, x2, y1, y2, z1, z2);
+} {}
+for {set i 0} {$i < 1000} {incr i} {
+ set x [expr $i%10]
+ set y [expr ($i/10)%10]
+ set z [expr ($i/100)%10]
+ execsql { INSERT INTO rt32 VALUES($i, $x, $x+1, $y, $y+1, $z, $z+1) }
+}
+do_execsql_test rtree9-3.2 {
+ SELECT id FROM rt32 WHERE id MATCH cube(3, 3, 3, 1, 1, 1) ORDER BY id;
+} {222 223 224 232 233 234 242 243 244 322 323 324 332 333 334 342 343 344 422 423 424 432 433 434 442 443 444}
+do_execsql_test rtree9-3.3 {
+ SELECT id FROM rt32 WHERE id MATCH cube(5.5, 5.5, 5.5, 1, 1, 1) ORDER BY id;
+} {555 556 565 566 655 656 665 666}
+
+
+do_catchsql_test rtree9-4.1 {
+ SELECT id FROM rt32 WHERE id MATCH cube(5.5, 5.5, 1, 1, 1) ORDER BY id;
+} {1 {SQL logic error or missing database}}
+for {set x 2} {$x<200} {incr x 2} {
+ do_catchsql_test rtree9-4.2.[expr $x/2] {
+ SELECT id FROM rt WHERE id MATCH randomblob($x)
+ } {1 {SQL logic error or missing database}}
+}
+do_catchsql_test rtree9-4.3 {
+ SELECT id FROM rt WHERE id MATCH CAST(
+ (cube(5.5, 5.5, 5.5, 1, 1, 1) || X'1234567812345678') AS blob
+ )
+} {1 {SQL logic error or missing database}}
+
+
+#-------------------------------------------------------------------------
+# Test the example 2d "circle" geometry callback.
+#
+register_circle_geom db
+
+breakpoint
+do_execsql_test rtree9-5.1 {
+ CREATE VIRTUAL TABLE rt2 USING rtree(id, xmin, xmax, ymin, ymax);
+
+ INSERT INTO rt2 VALUES(1, 1, 2, 1, 2);
+ INSERT INTO rt2 VALUES(2, 1, 2, -2, -1);
+ INSERT INTO rt2 VALUES(3, -2, -1, -2, -1);
+ INSERT INTO rt2 VALUES(4, -2, -1, 1, 2);
+
+ INSERT INTO rt2 VALUES(5, 2, 3, 2, 3);
+ INSERT INTO rt2 VALUES(6, 2, 3, -3, -2);
+ INSERT INTO rt2 VALUES(7, -3, -2, -3, -2);
+ INSERT INTO rt2 VALUES(8, -3, -2, 2, 3);
+
+ INSERT INTO rt2 VALUES(9, 1.8, 3, 1.8, 3);
+ INSERT INTO rt2 VALUES(10, 1.8, 3, -3, -1.8);
+ INSERT INTO rt2 VALUES(11, -3, -1.8, -3, -1.8);
+ INSERT INTO rt2 VALUES(12, -3, -1.8, 1.8, 3);
+
+ INSERT INTO rt2 VALUES(13, -15, 15, 1.8, 2.2);
+ INSERT INTO rt2 VALUES(14, -15, 15, -2.2, -1.8);
+ INSERT INTO rt2 VALUES(15, 1.8, 2.2, -15, 15);
+ INSERT INTO rt2 VALUES(16, -2.2, -1.8, -15, 15);
+
+ INSERT INTO rt2 VALUES(17, -100, 100, -100, 100);
+} {}
+
+do_execsql_test rtree9-5.2 {
+ SELECT id FROM rt2 WHERE id MATCH circle(0.0, 0.0, 2.0);
+} {1 2 3 4 13 14 15 16 17}
+
+do_execsql_test rtree9-5.3 {
+ UPDATE rt2 SET xmin=xmin+5, ymin=ymin+5, xmax=xmax+5, ymax=ymax+5;
+ SELECT id FROM rt2 WHERE id MATCH circle(5.0, 5.0, 2.0);
+} {1 2 3 4 13 14 15 16 17}
+
+finish_test
diff --git a/third_party/sqlite/src/ext/rtree/rtreeA.test b/third_party/sqlite/src/ext/rtree/rtreeA.test
new file mode 100644
index 0000000..e377b01
--- /dev/null
+++ b/third_party/sqlite/src/ext/rtree/rtreeA.test
@@ -0,0 +1,220 @@
+# 2010 September 22
+#
+# The author disclaims copyright to this source code. In place of
+# a legal notice, here is a blessing:
+#
+# May you do good and not evil.
+# May you find forgiveness for yourself and forgive others.
+# May you share freely, never taking more than you give.
+#
+#***********************************************************************
+# This file contains tests for the r-tree module. Specifically, it tests
+# that corrupt or inconsistent databases do not cause crashes in the r-tree
+# module.
+#
+
+if {![info exists testdir]} {
+ set testdir [file join [file dirname [info script]] .. .. test]
+}
+source $testdir/tester.tcl
+ifcapable !rtree { finish_test ; return }
+
+proc create_t1 {} {
+ db close
+ forcedelete test.db
+ sqlite3 db test.db
+ execsql {
+ PRAGMA page_size = 1024;
+ CREATE VIRTUAL TABLE t1 USING rtree(id, x1, x2, y1, y2);
+ }
+}
+proc populate_t1 {} {
+ execsql BEGIN
+ for {set i 0} {$i < 500} {incr i} {
+ set x2 [expr $i+5]
+ set y2 [expr $i+5]
+ execsql { INSERT INTO t1 VALUES($i, $i, $x2, $i, $y2) }
+ }
+ execsql COMMIT
+}
+
+proc truncate_node {nodeno nTrunc} {
+ set blob [db one {SELECT data FROM t1_node WHERE nodeno=$nodeno}]
+ if {$nTrunc<0} {set nTrunc "end-$nTrunc"}
+ set blob [string range $blob 0 $nTrunc]
+ db eval { UPDATE t1_node SET data = $blob WHERE nodeno=$nodeno }
+}
+
+proc set_tree_depth {tbl {newvalue ""}} {
+ set blob [db one "SELECT data FROM ${tbl}_node WHERE nodeno=1"]
+
+ if {$newvalue == ""} {
+ binary scan $blob Su oldvalue
+ return $oldvalue
+ }
+
+ set blob [binary format Sua* $newvalue [string range $blob 2 end]]
+ db eval "UPDATE ${tbl}_node SET data = \$blob WHERE nodeno=1"
+ return [set_tree_depth $tbl]
+}
+
+proc set_entry_count {tbl nodeno {newvalue ""}} {
+ set blob [db one "SELECT data FROM ${tbl}_node WHERE nodeno=$nodeno"]
+
+ if {$newvalue == ""} {
+ binary scan [string range $blob 2 end] Su oldvalue
+ return $oldvalue
+ }
+
+ set blob [binary format a*Sua* \
+ [string range $blob 0 1] $newvalue [string range $blob 4 end]
+ ]
+ db eval "UPDATE ${tbl}_node SET data = \$blob WHERE nodeno=$nodeno"
+ return [set_entry_count $tbl $nodeno]
+}
+
+
+proc do_corruption_tests {prefix args} {
+ set testarray [lindex $args end]
+ set errormsg {database disk image is malformed}
+
+ foreach {z value} [lrange $args 0 end-1] {
+ set n [string length $z]
+ if {$n>=2 && [string equal -length $n $z "-error"]} {
+ set errormsg $value
+ }
+ }
+
+ foreach {tn sql} $testarray {
+ do_catchsql_test $prefix.$tn $sql [list 1 $errormsg]
+ }
+}
+
+#-------------------------------------------------------------------------
+# Test the libraries response if the %_node table is completely empty
+# (i.e. the root node is missing), or has been removed from the database
+# entirely.
+#
+create_t1
+populate_t1
+do_execsql_test rtreeA-1.0 {
+ DELETE FROM t1_node;
+} {}
+
+do_corruption_tests rtreeA-1.1 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+ 4 "SELECT * FROM t1 WHERE x1<10 AND x2>12"
+}
+
+do_execsql_test rtreeA-1.2.0 { DROP TABLE t1_node } {}
+do_corruption_tests rtreeA-1.2 -error "SQL logic error or missing database" {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+ 4 "SELECT * FROM t1 WHERE x1<10 AND x2>12"
+}
+
+#-------------------------------------------------------------------------
+# Test the libraries response if some of the entries in the %_node table
+# are the wrong size.
+#
+create_t1
+populate_t1
+do_test rtreeA-2.1.0 {
+ set nodes [db eval {select nodeno FROM t1_node}]
+ foreach {a b c} $nodes { truncate_node $c 200 }
+} {}
+do_corruption_tests rtreeA-2.1 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+ 4 "SELECT * FROM t1 WHERE x1<10 AND x2>12"
+}
+
+create_t1
+populate_t1
+do_test rtreeA-2.2.0 { truncate_node 1 200 } {}
+do_corruption_tests rtreeA-2.2 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+ 4 "SELECT * FROM t1 WHERE x1<10 AND x2>12"
+}
+
+#-------------------------------------------------------------------------
+# Set the "depth" of the tree stored on the root node incorrectly. Test
+# that this does not cause any problems.
+#
+create_t1
+populate_t1
+do_test rtreeA-3.1.0.1 { set_tree_depth t1 } {1}
+do_test rtreeA-3.1.0.2 { set_tree_depth t1 3 } {3}
+do_corruption_tests rtreeA-3.1 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+}
+
+do_test rtreeA-3.2.0 { set_tree_depth t1 1000 } {1000}
+do_corruption_tests rtreeA-3.2 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+}
+
+create_t1
+populate_t1
+do_test rtreeA-3.3.0 {
+ execsql { DELETE FROM t1 WHERE rowid = 0 }
+ set_tree_depth t1 65535
+} {65535}
+do_corruption_tests rtreeA-3.3 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+}
+
+#-------------------------------------------------------------------------
+# Set the "number of entries" field on some nodes incorrectly.
+#
+create_t1
+populate_t1
+do_test rtreeA-4.1.0 {
+ set_entry_count t1 1 4000
+} {4000}
+do_corruption_tests rtreeA-4.1 {
+ 1 "SELECT * FROM t1"
+ 2 "SELECT * FROM t1 WHERE rowid=5"
+ 3 "INSERT INTO t1 VALUES(1000, 1, 2, 3, 4)"
+ 4 "SELECT * FROM t1 WHERE x1<10 AND x2>12"
+}
+
+#-------------------------------------------------------------------------
+# Remove entries from the %_parent table and check that this does not
+# cause a crash.
+#
+create_t1
+populate_t1
+do_execsql_test rtreeA-5.1.0 { DELETE FROM t1_parent } {}
+do_corruption_tests rtreeA-5.1 {
+ 1 "DELETE FROM t1 WHERE rowid = 5"
+ 2 "DELETE FROM t1"
+}
+
+#-------------------------------------------------------------------------
+# Add some bad entries to the %_parent table.
+#
+create_t1
+populate_t1
+do_execsql_test rtreeA-6.1.0 {
+ UPDATE t1_parent set parentnode = parentnode+1
+} {}
+do_corruption_tests rtreeA-6.1 {
+ 1 "DELETE FROM t1 WHERE rowid = 5"
+ 2 "UPDATE t1 SET x1=x1+1, x2=x2+1"
+}
+
+
+finish_test
diff --git a/third_party/sqlite/src/ext/rtree/rtree_perf.tcl b/third_party/sqlite/src/ext/rtree/rtree_perf.tcl
index fa3a4d3..e42e685 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree_perf.tcl
+++ b/third_party/sqlite/src/ext/rtree/rtree_perf.tcl
@@ -72,5 +72,3 @@ set rtree_select_time [time {
}
}]
puts "$rtree_select_time"
-
-
diff --git a/third_party/sqlite/src/ext/rtree/rtree_util.tcl b/third_party/sqlite/src/ext/rtree/rtree_util.tcl
index 55482e4..50a1b58 100755..100644
--- a/third_party/sqlite/src/ext/rtree/rtree_util.tcl
+++ b/third_party/sqlite/src/ext/rtree/rtree_util.tcl
@@ -13,8 +13,6 @@
# analyzing r-tree structures created with this module. It is
# used by both test procedures and the r-tree viewer application.
#
-# $Id: rtree_util.tcl,v 1.1 2008/05/26 18:41:54 danielk1977 Exp $
-#
#--------------------------------------------------------------------------
@@ -192,4 +190,3 @@ proc rtree_treedump {db zTab} {
set d [rtree_depth $db $zTab]
rtree_nodetreedump $db $zTab "" $d 1
}
-
diff --git a/third_party/sqlite/src/ext/rtree/sqlite3rtree.h b/third_party/sqlite/src/ext/rtree/sqlite3rtree.h
new file mode 100644
index 0000000..cffb300
--- /dev/null
+++ b/third_party/sqlite/src/ext/rtree/sqlite3rtree.h
@@ -0,0 +1,56 @@
+/*
+** 2010 August 30
+**
+** The author disclaims copyright to this source code. In place of
+** a legal notice, here is a blessing:
+**
+** May you do good and not evil.
+** May you find forgiveness for yourself and forgive others.
+** May you share freely, never taking more than you give.
+**
+*************************************************************************
+*/
+
+#ifndef _SQLITE3RTREE_H_
+#define _SQLITE3RTREE_H_
+
+#include <sqlite3.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct sqlite3_rtree_geometry sqlite3_rtree_geometry;
+
+/*
+** Register a geometry callback named zGeom that can be used as part of an
+** R-Tree geometry query as follows:
+**
+** SELECT ... FROM <rtree> WHERE <rtree col> MATCH $zGeom(... params ...)
+*/
+int sqlite3_rtree_geometry_callback(
+ sqlite3 *db,
+ const char *zGeom,
+ int (*xGeom)(sqlite3_rtree_geometry *, int nCoord, double *aCoord, int *pRes),
+ void *pContext
+);
+
+
+/*
+** A pointer to a structure of the following type is passed as the first
+** argument to callbacks registered using rtree_geometry_callback().
+*/
+struct sqlite3_rtree_geometry {
+ void *pContext; /* Copy of pContext passed to s_r_g_c() */
+ int nParam; /* Size of array aParam[] */
+ double *aParam; /* Parameters passed to SQL geom function */
+ void *pUser; /* Callback implementation user data */
+ void (*xDelUser)(void *); /* Called by SQLite to clean up pUser */
+};
+
+
+#ifdef __cplusplus
+} /* end of the 'extern "C"' block */
+#endif
+
+#endif /* ifndef _SQLITE3RTREE_H_ */
diff --git a/third_party/sqlite/src/ext/rtree/tkt3363.test b/third_party/sqlite/src/ext/rtree/tkt3363.test
index 11c5192..db05ed5 100644
--- a/third_party/sqlite/src/ext/rtree/tkt3363.test
+++ b/third_party/sqlite/src/ext/rtree/tkt3363.test
@@ -11,11 +11,9 @@
#
# The focus of this file is testing that ticket #3363 is fixed.
#
-# $Id: tkt3363.test,v 1.1 2008/09/08 11:07:03 danielk1977 Exp $
-#
if {![info exists testdir]} {
- set testdir [file join [file dirname $argv0] .. .. test]
+ set testdir [file join [file dirname [info script]] .. .. test]
}
source [file join [file dirname [info script]] rtree_util.tcl]
source $testdir/tester.tcl
@@ -50,5 +48,3 @@ do_test tkt3363.1.4 {
} {7}
finish_test
-
-
diff --git a/third_party/sqlite/src/ext/rtree/viewrtree.tcl b/third_party/sqlite/src/ext/rtree/viewrtree.tcl
index 2b4dd1b..794677f 100755..100644
--- a/third_party/sqlite/src/ext/rtree/viewrtree.tcl
+++ b/third_party/sqlite/src/ext/rtree/viewrtree.tcl
@@ -186,4 +186,3 @@ proc cell_report {db zTab iParent iCell} {
view_node
bind .c <Configure> view_node
-