diff options
Diffstat (limited to 'third_party/sqlite/src/ext/fts3/fts3.c')
| -rw-r--r-- | third_party/sqlite/src/ext/fts3/fts3.c | 8537 |
1 files changed, 1945 insertions, 6592 deletions
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 |