path: root/gettext-tools/src/write-java.c

/* Writing Java ResourceBundles.
   Copyright (C) 2001-2003, 2005-2010, 2015-2016 Free Software Foundation, Inc.
   Written by Bruno Haible <haible@clisp.cons.org>, 2001.

   This program is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */

#ifdef HAVE_CONFIG_H
# include <config.h>
#endif
#include <alloca.h>

/* Specification.  */
#include "write-java.h"

#include <errno.h>
#include <limits.h>
#include <stdbool.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#include <sys/stat.h>
#if !defined S_ISDIR && defined S_IFDIR
# define S_ISDIR(mode) (((mode) & S_IFMT) == S_IFDIR)
#endif
#if !S_IRUSR && S_IREAD
# define S_IRUSR S_IREAD
#endif
#if !S_IRUSR
# define S_IRUSR 00400
#endif
#if !S_IWUSR && S_IWRITE
# define S_IWUSR S_IWRITE
#endif
#if !S_IWUSR
# define S_IWUSR 00200
#endif
#if !S_IXUSR && S_IEXEC
# define S_IXUSR S_IEXEC
#endif
#if !S_IXUSR
# define S_IXUSR 00100
#endif

#include "c-ctype.h"
#include "error.h"
#include "xerror.h"
#include "xvasprintf.h"
#include "javacomp.h"
#include "message.h"
#include "msgfmt.h"
#include "msgl-iconv.h"
#include "plural-exp.h"
#include "po-charset.h"
#include "xalloc.h"
#include "xmalloca.h"
#include "minmax.h"
#include "concat-filename.h"
#include "fwriteerror.h"
#include "clean-temp.h"
#include "unistr.h"
#include "gettext.h"

#define _(str) gettext (str)


/* Check that the resource name is a valid Java class name.  To simplify
   things, we allow only ASCII characters in the class name.
   Return the number of dots in the class name, or -1 if not OK.  */
static int
check_resource_name (const char *name)
{
  int ndots = 0;
  const char *p = name;

  for (;;)
    {
      /* First character, see Character.isJavaIdentifierStart.  */
      if (!(c_isalpha (*p) || (*p == '$') || (*p == '_')))
        return -1;
      /* Following characters, see Character.isJavaIdentifierPart.  */
      do
        p++;
      while (c_isalpha (*p) || (*p == '$') || (*p == '_') || c_isdigit (*p));
      if (*p == '\0')
        break;
      if (*p != '.')
        return -1;
      p++;
      ndots++;
    }
  return ndots;
}


/* Return the Java hash code of a string mod 2^31.
   The Java String.hashCode() function returns the same values across
   Java implementations.
   (See http://www.javasoft.com/docs/books/jls/clarify.html)
   It returns a signed 32-bit integer.  We add a mod 2^31 afterwards;
   this removes one bit but greatly simplifies the following "mod hash_size"
   and "mod (hash_size - 2)" operations.  */
static unsigned int
string_hashcode (const char *str)
{
  const char *str_limit = str + strlen (str);
  int hash = 0;
  while (str < str_limit)
    {
      ucs4_t uc;
      str += u8_mbtouc (&uc, (const unsigned char *) str, str_limit - str);
      if (uc < 0x10000)
        /* Single UCS-2 'char'.  */
        hash = 31 * hash + uc;
      else
        {
          /* UTF-16 surrogate: two 'char's.  */
          ucs4_t uc1 = 0xd800 + ((uc - 0x10000) >> 10);
          ucs4_t uc2 = 0xdc00 + ((uc - 0x10000) & 0x3ff);
          hash = 31 * hash + uc1;
          hash = 31 * hash + uc2;
        }
    }
  return hash & 0x7fffffff;
}


/* Return the Java hash code of a (msgctxt, msgid) pair mod 2^31.  */
static unsigned int
msgid_hashcode (const char *msgctxt, const char *msgid)
{
  if (msgctxt == NULL)
    return string_hashcode (msgid);
  else
    {
      size_t msgctxt_len = strlen (msgctxt);
      size_t msgid_len = strlen (msgid);
      size_t combined_len = msgctxt_len + 1 + msgid_len;
      char *combined;
      unsigned int result;

      combined = (char *) xmalloca (combined_len);
      memcpy (combined, msgctxt, msgctxt_len);
      combined[msgctxt_len] = MSGCTXT_SEPARATOR;
      memcpy (combined + msgctxt_len + 1, msgid, msgid_len + 1);

      result = string_hashcode (combined);

      freea (combined);

      return result;
    }
}


/* Compute a good hash table size for the given set of msgids.  */
static unsigned int
compute_hashsize (message_list_ty *mlp, bool *collisionp)
{
  /* This is an O(n^2) algorithm, but should be sufficient because few
     programs have more than 1000 messages in a single domain.  */
#define XXN 3  /* can be tweaked */
#define XXS 3  /* can be tweaked */
  unsigned int n = mlp->nitems;
  unsigned int *hashcodes =
    (unsigned int *) xmalloca (n * sizeof (unsigned int));
  unsigned int hashsize;
  unsigned int best_hashsize;
  unsigned int best_score;
  size_t j;

  for (j = 0; j < n; j++)
    hashcodes[j] = msgid_hashcode (mlp->item[j]->msgctxt, mlp->item[j]->msgid);

  /* Try all numbers between n and 3*n.  The score depends on the size of the
     table -- the smaller the better -- and the number of collision lookups,
     i.e. total number of times that 1 + (hashcode % (hashsize - 2))
     is added to the index during lookup.  If there are collisions, only odd
     hashsize values are allowed.  */
  best_hashsize = 0;
  best_score = UINT_MAX;
  for (hashsize = n; hashsize <= XXN * n; hashsize++)
    {
      char *bitmap;
      unsigned int score;

      /* Premature end of the loop if all future scores are known to be
         larger than the already reached best_score.  This relies on the
         ascending loop and on the fact that score >= hashsize.  */
      if (hashsize >= best_score)
        break;

      bitmap = XNMALLOC (hashsize, char);
      memset (bitmap, 0, hashsize);

      score = 0;
      for (j = 0; j < n; j++)
        {
          unsigned int idx = hashcodes[j] % hashsize;

          if (bitmap[idx] != 0)
            {
              /* Collision.  Cannot deal with it if hashsize is even.  */
              if ((hashsize % 2) == 0)
                /* Try next hashsize.  */
                goto bad_hashsize;
              else
                {
                  unsigned int idx0 = idx;
                  unsigned int incr = 1 + (hashcodes[j] % (hashsize - 2));
                  score += 2;   /* Big penalty for the additional division */
                  do
                    {
                      score++;  /* Small penalty for each loop round */
                      idx += incr;
                      if (idx >= hashsize)
                        idx -= hashsize;
                      if (idx == idx0)
                        /* Searching for a hole, we performed a whole round
                           across the table.  This happens particularly
                           frequently if gcd(hashsize,incr) > 1.  Try next
                           hashsize.  */
                        goto bad_hashsize;
                    }
                  while (bitmap[idx] != 0);
                }
            }
          bitmap[idx] = 1;
        }

      /* Big hashsize also gives a penalty.  */
      score = XXS * score + hashsize;

      /* If for any incr between 1 and hashsize - 2, an whole round
         (idx0, idx0 + incr, ...) is occupied, and the lookup function
         must deal with collisions, then some inputs would lead to
         an endless loop in the lookup function.  */
      if (score > hashsize)
        {
          unsigned int incr;

          /* Since the set { idx0, idx0 + incr, ... } depends only on idx0
             and gcd(hashsize,incr), we only need to conside incr that
             divides hashsize.  */
          for (incr = 1; incr <= hashsize / 2; incr++)
            if ((hashsize % incr) == 0)
              {
                unsigned int idx0;

                for (idx0 = 0; idx0 < incr; idx0++)
                  {
                    bool full = true;
                    unsigned int idx;

                    for (idx = idx0; idx < hashsize; idx += incr)
                      if (bitmap[idx] == 0)
                        {
                          full = false;
                          break;
                        }
                    if (full)
                      /* A whole round is occupied.  */
                      goto bad_hashsize;
                  }
              }
        }

      if (false)
        bad_hashsize:
        score = UINT_MAX;

      free (bitmap);

      if (score < best_score)
        {
          best_score = score;
          best_hashsize = hashsize;
        }
    }
  if (best_hashsize == 0 || best_score < best_hashsize)
    abort ();

  freea (hashcodes);

  /* There are collisions if and only if best_score > best_hashsize.  */
  *collisionp = (best_score > best_hashsize);
  return best_hashsize;
}


struct table_item { unsigned int index; message_ty *mp; };

static int
compare_index (const void *pval1, const void *pval2)
{
  return (int)((const struct table_item *) pval1)->index
         - (int)((const struct table_item *) pval2)->index;
}

/* Compute the list of messages and table indices, sorted according to the
   indices.  */
static struct table_item *
compute_table_items (message_list_ty *mlp, unsigned int hashsize)
{
  unsigned int n = mlp->nitems;
  struct table_item *arr = XNMALLOC (n, struct table_item);
  char *bitmap;
  size_t j;

  bitmap = XNMALLOC (hashsize, char);
  memset (bitmap, 0, hashsize);

  for (j = 0; j < n; j++)
    {
      unsigned int hashcode =
        msgid_hashcode (mlp->item[j]->msgctxt, mlp->item[j]->msgid);
      unsigned int idx = hashcode % hashsize;

      if (bitmap[idx] != 0)
        {
          unsigned int incr = 1 + (hashcode % (hashsize - 2));
          do
            {
              idx += incr;
              if (idx >= hashsize)
                idx -= hashsize;
            }
          while (bitmap[idx] != 0);
        }
      bitmap[idx] = 1;

      arr[j].index = idx;
      arr[j].mp = mlp->item[j];
    }

  free (bitmap);

  qsort (arr, n, sizeof (arr[0]), compare_index);

  return arr;
}


/* Write a string in Java Unicode notation to the given stream.  */
static void
write_java_string (FILE *stream, const char *str)
{
  static const char hexdigit[] = "0123456789abcdef";
  const char *str_limit = str + strlen (str);

  fprintf (stream, "\"");
  while (str < str_limit)
    {
      ucs4_t uc;
      str += u8_mbtouc (&uc, (const unsigned char *) str, str_limit - str);
      if (uc < 0x10000)
        {
          /* Single UCS-2 'char'.  */
          if (uc == 0x000a)
            fprintf (stream, "\\n");
          else if (uc == 0x000d)
            fprintf (stream, "\\r");
          else if (uc == 0x0022)
            fprintf (stream, "\\\"");
          else if (uc == 0x005c)
            fprintf (stream, "\\\\");
          else if (uc >= 0x0020 && uc < 0x007f)
            fprintf (stream, "%c", (int) uc);
          else
            fprintf (stream, "\\u%c%c%c%c",
                     hexdigit[(uc >> 12) & 0x0f], hexdigit[(uc >> 8) & 0x0f],
                     hexdigit[(uc >> 4) & 0x0f], hexdigit[uc & 0x0f]);
        }
      else
        {
          /* UTF-16 surrogate: two 'char's.  */
          ucs4_t uc1 = 0xd800 + ((uc - 0x10000) >> 10);
          ucs4_t uc2 = 0xdc00 + ((uc - 0x10000) & 0x3ff);
          fprintf (stream, "\\u%c%c%c%c",
                   hexdigit[(uc1 >> 12) & 0x0f], hexdigit[(uc1 >> 8) & 0x0f],
                   hexdigit[(uc1 >> 4) & 0x0f], hexdigit[uc1 & 0x0f]);
          fprintf (stream, "\\u%c%c%c%c",
                   hexdigit[(uc2 >> 12) & 0x0f], hexdigit[(uc2 >> 8) & 0x0f],
                   hexdigit[(uc2 >> 4) & 0x0f], hexdigit[uc2 & 0x0f]);
        }
    }
  fprintf (stream, "\"");
}


/* Write a (msgctxt, msgid) pair as a string in Java Unicode notation to the
   given stream.  */
static void
write_java_msgid (FILE *stream, message_ty *mp)
{
  const char *msgctxt = mp->msgctxt;
  const char *msgid = mp->msgid;

  if (msgctxt == NULL)
    write_java_string (stream, msgid);
  else
    {
      size_t msgctxt_len = strlen (msgctxt);
      size_t msgid_len = strlen (msgid);
      size_t combined_len = msgctxt_len + 1 + msgid_len;
      char *combined;

      combined = (char *) xmalloca (combined_len);
      memcpy (combined, msgctxt, msgctxt_len);
      combined[msgctxt_len] = MSGCTXT_SEPARATOR;
      memcpy (combined + msgctxt_len + 1, msgid, msgid_len + 1);

      write_java_string (stream, combined);

      freea (combined);
    }
}


/* Write Java code that returns the value for a message.  If the message
   has plural forms, it is an expression of type String[], otherwise it is
   an expression of type String.  */
static void
write_java_msgstr (FILE *stream, message_ty *mp)
{
  if (mp->msgid_plural != NULL)
    {
      bool first;
      const char *p;

      fprintf (stream, "new java.lang.String[] { ");
      for (p = mp->msgstr, first = true;
           p < mp->msgstr + mp->msgstr_len;
           p += strlen (p) + 1, first = false)
        {
          if (!first)
            fprintf (stream, ", ");
          write_java_string (stream, p);
        }
      fprintf (stream, " }");
    }
  else
    {
      if (mp->msgstr_len != strlen (mp->msgstr) + 1)
        abort ();

      write_java_string (stream, mp->msgstr);
    }
}


/* Writes the body of the function which returns the local value for a key
   named 'msgid'.  */
static void
write_lookup_code (FILE *stream, unsigned int hashsize, bool collisions)
{
  fprintf (stream, "    int hash_val = msgid.hashCode() & 0x7fffffff;\n");
  fprintf (stream, "    int idx = (hash_val %% %d) << 1;\n", hashsize);
  if (collisions)
    {
      fprintf (stream, "    {\n");
      fprintf (stream, "      java.lang.Object found = table[idx];\n");
      fprintf (stream, "      if (found == null)\n");
      fprintf (stream, "        return null;\n");
      fprintf (stream, "      if (msgid.equals(found))\n");
      fprintf (stream, "        return table[idx + 1];\n");
      fprintf (stream, "    }\n");
      fprintf (stream, "    int incr = ((hash_val %% %d) + 1) << 1;\n",
               hashsize - 2);
      fprintf (stream, "    for (;;) {\n");
      fprintf (stream, "      idx += incr;\n");
      fprintf (stream, "      if (idx >= %d)\n", 2 * hashsize);
      fprintf (stream, "        idx -= %d;\n", 2 * hashsize);
      fprintf (stream, "      java.lang.Object found = table[idx];\n");
      fprintf (stream, "      if (found == null)\n");
      fprintf (stream, "        return null;\n");
      fprintf (stream, "      if (msgid.equals(found))\n");
      fprintf (stream, "        return table[idx + 1];\n");
      fprintf (stream, "    }\n");
    }
  else
    {
      fprintf (stream, "    java.lang.Object found = table[idx];\n");
      fprintf (stream, "    if (found != null && msgid.equals(found))\n");
      fprintf (stream, "      return table[idx + 1];\n");
      fprintf (stream, "    return null;\n");
    }
}


/* Tests whether a plural expression, evaluated according to the C rules,
   can only produce the values 0 and 1.  */
static bool
is_expression_boolean (struct expression *exp)
{
  switch (exp->operation)
    {
    case var:
    case mult:
    case divide:
    case module:
    case plus:
    case minus:
      return false;
    case lnot:
    case less_than:
    case greater_than:
    case less_or_equal:
    case greater_or_equal:
    case equal:
    case not_equal:
    case land:
    case lor:
      return true;
    case num:
      return (exp->val.num == 0 || exp->val.num == 1);
    case qmop:
      return is_expression_boolean (exp->val.args[1])
             && is_expression_boolean (exp->val.args[2]);
    default:
      abort ();
    }
}


/* Write Java code that evaluates a plural expression according to the C rules.
   The variable is called 'n'.  */
static void
write_java_expression (FILE *stream, const struct expression *exp, bool as_boolean)
{
  /* We use parentheses everywhere.  This frees us from tracking the priority
     of arithmetic operators.  */
  if (as_boolean)
    {
      /* Emit a Java expression of type 'boolean'.  */
      switch (exp->operation)
        {
        case num:
          fprintf (stream, "%s", exp->val.num ? "true" : "false");
          return;
        case lnot:
          fprintf (stream, "(!");
          write_java_expression (stream, exp->val.args[0], true);
          fprintf (stream, ")");
          return;
        case less_than:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " < ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case greater_than:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " > ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case less_or_equal:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " <= ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case greater_or_equal:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " >= ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case equal:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " == ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case not_equal:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " != ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case land:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], true);
          fprintf (stream, " && ");
          write_java_expression (stream, exp->val.args[1], true);
          fprintf (stream, ")");
          return;
        case lor:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], true);
          fprintf (stream, " || ");
          write_java_expression (stream, exp->val.args[1], true);
          fprintf (stream, ")");
          return;
        case qmop:
          if (is_expression_boolean (exp->val.args[1])
              && is_expression_boolean (exp->val.args[2]))
            {
              fprintf (stream, "(");
              write_java_expression (stream, exp->val.args[0], true);
              fprintf (stream, " ? ");
              write_java_expression (stream, exp->val.args[1], true);
              fprintf (stream, " : ");
              write_java_expression (stream, exp->val.args[2], true);
              fprintf (stream, ")");
              return;
            }
          /*FALLTHROUGH*/
        case var:
        case mult:
        case divide:
        case module:
        case plus:
        case minus:
          fprintf (stream, "(");
          write_java_expression (stream, exp, false);
          fprintf (stream, " != 0)");
          return;
        default:
          abort ();
        }
    }
  else
    {
      /* Emit a Java expression of type 'long'.  */
      switch (exp->operation)
        {
        case var:
          fprintf (stream, "n");
          return;
        case num:
          fprintf (stream, "%lu", exp->val.num);
          return;
        case mult:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " * ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case divide:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " / ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case module:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " %% ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case plus:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " + ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case minus:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], false);
          fprintf (stream, " - ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, ")");
          return;
        case qmop:
          fprintf (stream, "(");
          write_java_expression (stream, exp->val.args[0], true);
          fprintf (stream, " ? ");
          write_java_expression (stream, exp->val.args[1], false);
          fprintf (stream, " : ");
          write_java_expression (stream, exp->val.args[2], false);
          fprintf (stream, ")");
          return;
        case lnot:
        case less_than:
        case greater_than:
        case less_or_equal:
        case greater_or_equal:
        case equal:
        case not_equal:
        case land:
        case lor:
          fprintf (stream, "(");
          write_java_expression (stream, exp, true);
          fprintf (stream, " ? 1 : 0)");
          return;
        default:
          abort ();
        }
    }
}


/* Write the Java initialization statements for the Java 1.1.x case,
   for items j, start_index <= j < end_index.  */
static void
write_java1_init_statements (FILE *stream, message_list_ty *mlp,
                             size_t start_index, size_t end_index)
{
  size_t j;

  for (j = start_index; j < end_index; j++)
    {
      fprintf (stream, "    t.put(");
      write_java_msgid (stream, mlp->item[j]);
      fprintf (stream, ",");
      write_java_msgstr (stream, mlp->item[j]);
      fprintf (stream, ");\n");
    }
}


/* Write the Java initialization statements for the Java 2 case,
   for items j, start_index <= j < end_index.  */
static void
write_java2_init_statements (FILE *stream, message_list_ty *mlp,
                             const struct table_item *table_items,
                             size_t start_index, size_t end_index)
{
  size_t j;

  for (j = start_index; j < end_index; j++)
    {
      const struct table_item *ti = &table_items[j];

      fprintf (stream, "    t[%d] = ", 2 * ti->index);
      write_java_msgid (stream, ti->mp);
      fprintf (stream, ";\n");
      fprintf (stream, "    t[%d] = ", 2 * ti->index + 1);
      write_java_msgstr (stream, ti->mp);
      fprintf (stream, ";\n");
    }
}


/* Write the Java code for the ResourceBundle subclass to the given stream.
   Note that we use fully qualified class names and no "import" statements,
   because applications can have their own classes called X.Y.ResourceBundle
   or X.Y.String.  */
static void
write_java_code (FILE *stream, const char *class_name, message_list_ty *mlp,
                 bool assume_java2)
{
  const char *last_dot;
  unsigned int plurals;
  size_t j;

  fprintf (stream,
           "/* Automatically generated by GNU msgfmt.  Do not modify!  */\n");
  last_dot = strrchr (class_name, '.');
  if (last_dot != NULL)
    {
      fprintf (stream, "package ");
      fwrite (class_name, 1, last_dot - class_name, stream);
      fprintf (stream, ";\npublic class %s", last_dot + 1);
    }
  else
    fprintf (stream, "public class %s", class_name);
  fprintf (stream, " extends java.util.ResourceBundle {\n");

  /* Determine whether there are plural messages.  */
  plurals = 0;
  for (j = 0; j < mlp->nitems; j++)
    if (mlp->item[j]->msgid_plural != NULL)
      plurals++;

  if (assume_java2)
    {
      unsigned int hashsize;
      bool collisions;
      struct table_item *table_items;
      const char *table_eltype;

      /* Determine the hash table size and whether it leads to collisions.  */
      hashsize = compute_hashsize (mlp, &collisions);

      /* Determines which indices in the table contain a message.  The others
         are null.  */
      table_items = compute_table_items (mlp, hashsize);

      /* Emit the table of pairs (msgid, msgstr).  If there are plurals,
         it is of type Object[], otherwise of type String[].  We use a static
         code block because that makes less code:  The Java compilers also
         generate code for the 'null' entries, which is dumb.  */
      table_eltype = (plurals ? "java.lang.Object" : "java.lang.String");
      fprintf (stream, "  private static final %s[] table;\n", table_eltype);
      {
        /* With the Sun javac compiler, each assignment takes 5 to 8 bytes
           of bytecode, therefore for each message, up to 16 bytes are needed.
           Since the bytecode of every method, including the <clinit> method
           that contains the static initializers, is limited to 64 KB, only ca,
           65536 / 16 = 4096 messages can be initialized in a single method.
           Account for other Java compilers and for plurals by limiting it to
           1000.  */
        const size_t max_items_per_method = 1000;

        if (mlp->nitems > max_items_per_method)
          {
            unsigned int k;
            size_t start_j;
            size_t end_j;

            for (k = 0, start_j = 0, end_j = start_j + max_items_per_method;
                 start_j < mlp->nitems;
                 k++, start_j = end_j, end_j = start_j + max_items_per_method)
              {
                fprintf (stream, "  static void clinit_part_%u (%s[] t) {\n",
                         k, table_eltype);
                write_java2_init_statements (stream, mlp, table_items,
                                             start_j, MIN (end_j, mlp->nitems));
                fprintf (stream, "  }\n");
              }
          }
        fprintf (stream, "  static {\n");
        fprintf (stream, "    %s[] t = new %s[%d];\n", table_eltype,
                 table_eltype, 2 * hashsize);
        if (mlp->nitems > max_items_per_method)
          {
            unsigned int k;
            size_t start_j;

            for (k = 0, start_j = 0;
                 start_j < mlp->nitems;
                 k++, start_j += max_items_per_method)
              fprintf (stream, "    clinit_part_%u(t);\n", k);
          }
        else
          write_java2_init_statements (stream, mlp, table_items,
                                       0, mlp->nitems);
        fprintf (stream, "    table = t;\n");
        fprintf (stream, "  }\n");
      }

      /* Emit the msgid_plural strings.  Only used by msgunfmt.  */
      if (plurals)
        {
          bool first;
          fprintf (stream, "  public static final java.lang.String[] get_msgid_plural_table () {\n");
          fprintf (stream, "    return new java.lang.String[] { ");
          first = true;
          for (j = 0; j < mlp->nitems; j++)
            {
              struct table_item *ti = &table_items[j];
              if (ti->mp->msgid_plural != NULL)
                {
                  if (!first)
                    fprintf (stream, ", ");
                  write_java_string (stream, ti->mp->msgid_plural);
                  first = false;
                }
            }
          fprintf (stream, " };\n");
          fprintf (stream, "  }\n");
        }

      if (plurals)
        {
          /* Emit the lookup function.  It is a common subroutine for
             handleGetObject and ngettext.  */
          fprintf (stream, "  public java.lang.Object lookup (java.lang.String msgid) {\n");
          write_lookup_code (stream, hashsize, collisions);
          fprintf (stream, "  }\n");
        }

      /* Emit the handleGetObject function.  It is declared abstract in
         ResourceBundle.  It implements a local version of gettext.  */
      fprintf (stream, "  public java.lang.Object handleGetObject (java.lang.String msgid) throws java.util.MissingResourceException {\n");
      if (plurals)
        {
          fprintf (stream, "    java.lang.Object value = lookup(msgid);\n");
          fprintf (stream, "    return (value instanceof java.lang.String[] ? ((java.lang.String[])value)[0] : value);\n");
        }
      else
        write_lookup_code (stream, hashsize, collisions);
      fprintf (stream, "  }\n");

      /* Emit the getKeys function.  It is declared abstract in ResourceBundle.
         The inner class is not avoidable.  */
      fprintf (stream, "  public java.util.Enumeration getKeys () {\n");
      fprintf (stream, "    return\n");
      fprintf (stream, "      new java.util.Enumeration() {\n");
      fprintf (stream, "        private int idx = 0;\n");
      fprintf (stream, "        { while (idx < %d && table[idx] == null) idx += 2; }\n",
               2 * hashsize);
      fprintf (stream, "        public boolean hasMoreElements () {\n");
      fprintf (stream, "          return (idx < %d);\n", 2 * hashsize);
      fprintf (stream, "        }\n");
      fprintf (stream, "        public java.lang.Object nextElement () {\n");
      fprintf (stream, "          java.lang.Object key = table[idx];\n");
      fprintf (stream, "          do idx += 2; while (idx < %d && table[idx] == null);\n",
               2 * hashsize);
      fprintf (stream, "          return key;\n");
      fprintf (stream, "        }\n");
      fprintf (stream, "      };\n");
      fprintf (stream, "  }\n");
    }
  else
    {
      /* Java 1.1.x uses a different hash function.  If compatibility with
         this Java version is required, the hash table must be built at run time,
         not at compile time.  */
      fprintf (stream, "  private static final java.util.Hashtable table;\n");
      {
        /* With the Sun javac compiler, each 'put' call takes 9 to 11 bytes
           of bytecode, therefore for each message, up to 11 bytes are needed.
           Since the bytecode of every method, including the <clinit> method
           that contains the static initializers, is limited to 64 KB, only ca,
           65536 / 11 = 5958 messages can be initialized in a single method.
           Account for other Java compilers and for plurals by limiting it to
           1500.  */
        const size_t max_items_per_method = 1500;

        if (mlp->nitems > max_items_per_method)
          {
            unsigned int k;
            size_t start_j;
            size_t end_j;

            for (k = 0, start_j = 0, end_j = start_j + max_items_per_method;
                 start_j < mlp->nitems;
                 k++, start_j = end_j, end_j = start_j + max_items_per_method)
              {
                fprintf (stream, "  static void clinit_part_%u (java.util.Hashtable t) {\n",
                         k);
                write_java1_init_statements (stream, mlp,
                                             start_j, MIN (end_j, mlp->nitems));
                fprintf (stream, "  }\n");
              }
          }
        fprintf (stream, "  static {\n");
        fprintf (stream, "    java.util.Hashtable t = new java.util.Hashtable();\n");
        if (mlp->nitems > max_items_per_method)
          {
            unsigned int k;
            size_t start_j;

            for (k = 0, start_j = 0;
                 start_j < mlp->nitems;
                 k++, start_j += max_items_per_method)
              fprintf (stream, "    clinit_part_%u(t);\n", k);
          }
        else
          write_java1_init_statements (stream, mlp, 0, mlp->nitems);
        fprintf (stream, "    table = t;\n");
        fprintf (stream, "  }\n");
      }

      /* Emit the msgid_plural strings.  Only used by msgunfmt.  */
      if (plurals)
        {
          fprintf (stream, "  public static final java.util.Hashtable get_msgid_plural_table () {\n");
          fprintf (stream, "    java.util.Hashtable p = new java.util.Hashtable();\n");
          for (j = 0; j < mlp->nitems; j++)
            if (mlp->item[j]->msgid_plural != NULL)
              {
                fprintf (stream, "    p.put(");
                write_java_msgid (stream, mlp->item[j]);
                fprintf (stream, ",");
                write_java_string (stream, mlp->item[j]->msgid_plural);
                fprintf (stream, ");\n");
              }
          fprintf (stream, "    return p;\n");
          fprintf (stream, "  }\n");
        }

      if (plurals)
        {
          /* Emit the lookup function.  It is a common subroutine for
             handleGetObject and ngettext.  */
          fprintf (stream, "  public java.lang.Object lookup (java.lang.String msgid) {\n");
          fprintf (stream, "    return table.get(msgid);\n");
          fprintf (stream, "  }\n");
        }

      /* Emit the handleGetObject function.  It is declared abstract in
         ResourceBundle.  It implements a local version of gettext.  */
      fprintf (stream, "  public java.lang.Object handleGetObject (java.lang.String msgid) throws java.util.MissingResourceException {\n");
      if (plurals)
        {
          fprintf (stream, "    java.lang.Object value = table.get(msgid);\n");
          fprintf (stream, "    return (value instanceof java.lang.String[] ? ((java.lang.String[])value)[0] : value);\n");
        }
      else
        fprintf (stream, "    return table.get(msgid);\n");
      fprintf (stream, "  }\n");

      /* Emit the getKeys function.  It is declared abstract in
         ResourceBundle.  */
      fprintf (stream, "  public java.util.Enumeration getKeys () {\n");
      fprintf (stream, "    return table.keys();\n");
      fprintf (stream, "  }\n");
    }

  /* Emit the pluralEval function.  It is a subroutine for ngettext.  */
  if (plurals)
    {
      message_ty *header_entry;
      const struct expression *plural;
      unsigned long int nplurals;

      header_entry = message_list_search (mlp, NULL, "");
      extract_plural_expression (header_entry ? header_entry->msgstr : NULL,
                                 &plural, &nplurals);

      fprintf (stream, "  public static long pluralEval (long n) {\n");
      fprintf (stream, "    return ");
      write_java_expression (stream, plural, false);
      fprintf (stream, ";\n");
      fprintf (stream, "  }\n");
    }

  /* Emit the getParent function.  It is a subroutine for ngettext.  */
  fprintf (stream, "  public java.util.ResourceBundle getParent () {\n");
  fprintf (stream, "    return parent;\n");
  fprintf (stream, "  }\n");

  fprintf (stream, "}\n");
}


int
msgdomain_write_java (message_list_ty *mlp, const char *canon_encoding,
                      const char *resource_name, const char *locale_name,
                      const char *directory,
                      bool assume_java2,
                      bool output_source)
{
  int retval;
  struct temp_dir *tmpdir;
  int ndots;
  char *class_name;
  char **subdirs;
  char *java_file_name;
  FILE *java_file;
  const char *java_sources[1];
  const char *source_dir_name;

  /* If no entry for this resource/domain, don't even create the file.  */
  if (mlp->nitems == 0)
    return 0;

  retval = 1;

  /* Convert the messages to Unicode.  */
  iconv_message_list (mlp, canon_encoding, po_charset_utf8, NULL);

  if (output_source)
    {
      tmpdir = NULL;
      source_dir_name = directory;
    }
  else
    {
      /* Create a temporary directory where we can put the Java file.  */
      tmpdir = create_temp_dir ("msg", NULL, false);
      if (tmpdir == NULL)
        goto quit1;
      source_dir_name = tmpdir->dir_name;
    }

  /* Assign a default value to the resource name.  */
  if (resource_name == NULL)
    resource_name = "Messages";

  /* Prepare the list of subdirectories.  */
  ndots = check_resource_name (resource_name);
  if (ndots < 0)
    {
      error (0, 0, _("not a valid Java class name: %s"), resource_name);
      goto quit2;
    }

  if (locale_name != NULL)
    class_name = xasprintf ("%s_%s", resource_name, locale_name);
  else
    class_name = xstrdup (resource_name);

  subdirs = (ndots > 0 ? (char **) xmalloca (ndots * sizeof (char *)) : NULL);
  {
    const char *p;
    const char *last_dir;
    int i;

    last_dir = source_dir_name;
    p = resource_name;
    for (i = 0; i < ndots; i++)
      {
        const char *q = strchr (p, '.');
        size_t n = q - p;
        char *part = (char *) xmalloca (n + 1);
        memcpy (part, p, n);
        part[n] = '\0';
        subdirs[i] = xconcatenated_filename (last_dir, part, NULL);
        freea (part);
        last_dir = subdirs[i];
        p = q + 1;
      }

    if (locale_name != NULL)
      {
        char *suffix = xasprintf ("_%s.java", locale_name);
        java_file_name = xconcatenated_filename (last_dir, p, suffix);
        free (suffix);
      }
    else
      java_file_name = xconcatenated_filename (last_dir, p, ".java");
  }

  /* If OUTPUT_SOURCE, write the Java file in DIRECTORY and return.  */
  if (output_source)
    {
      int i;

      for (i = 0; i < ndots; i++)
        {
          if (mkdir (subdirs[i], S_IRUSR | S_IWUSR | S_IXUSR) < 0)
            {
              error (0, errno, _("failed to create \"%s\""), subdirs[i]);
              goto quit3;
            }
        }

      java_file = fopen (java_file_name, "w");
      if (java_file == NULL)
        {
          error (0, errno, _("failed to create \"%s\""), java_file_name);
          goto quit3;
        }

      write_java_code (java_file, class_name, mlp, assume_java2);

      if (fwriteerror (java_file))
        {
          error (0, errno, _("error while writing \"%s\" file"),
                 java_file_name);
          goto quit3;
        }

      retval = 0;
      goto quit3;
    }

  /* Create the subdirectories.  This is needed because some older Java
     compilers verify that the source of class A.B.C really sits in a
     directory whose name ends in /A/B.  */
  {
    int i;

    for (i = 0; i < ndots; i++)
      {
        register_temp_subdir (tmpdir, subdirs[i]);
        if (mkdir (subdirs[i], S_IRUSR | S_IWUSR | S_IXUSR) < 0)
          {
            error (0, errno, _("failed to create \"%s\""), subdirs[i]);
            unregister_temp_subdir (tmpdir, subdirs[i]);
            goto quit3;
          }
      }
  }

  /* Create the Java file.  */
  register_temp_file (tmpdir, java_file_name);
  java_file = fopen_temp (java_file_name, "w");
  if (java_file == NULL)
    {
      error (0, errno, _("failed to create \"%s\""), java_file_name);
      unregister_temp_file (tmpdir, java_file_name);
      goto quit3;
    }

  write_java_code (java_file, class_name, mlp, assume_java2);

  if (fwriteerror_temp (java_file))
    {
      error (0, errno, _("error while writing \"%s\" file"), java_file_name);
      goto quit3;
    }

  /* Compile the Java file to a .class file.
     directory must be non-NULL, because when the -d option is omitted, the
     Java compilers create the class files in the source file's directory -
     which is in a temporary directory in our case.  */
  java_sources[0] = java_file_name;
  if (compile_java_class (java_sources, 1, NULL, 0, "1.3", "1.1", directory,
                          true, false, true, verbose > 0))
    {
      if (!verbose)
        error (0, 0, _("\
compilation of Java class failed, please try --verbose or set $JAVAC"));
      else
        error (0, 0, _("\
compilation of Java class failed, please try to set $JAVAC"));
      goto quit3;
    }

  retval = 0;

 quit3:
  {
    int i;
    free (java_file_name);
    for (i = 0; i < ndots; i++)
      free (subdirs[i]);
  }
  freea (subdirs);
  free (class_name);
 quit2:
  if (tmpdir != NULL)
    cleanup_temp_dir (tmpdir);
 quit1:
  return retval;
}