iconvconfig.c

Newlib 嵌入式 C库 标准实现代码

  new_module (from, to - from, to, module - to, directory, module, wp - module,
	      cost, need_ext);
}


/* Read the config file and add the data for this directory to that.  */
static int
handle_dir (const char *dir)
{
  char *infile;
  FILE *fp;
  char *line = NULL;
  size_t linelen = 0;
  size_t dirlen = strlen (dir);
  char *tmp;

  if (dir[dirlen - 1] != '/')
    {
      char *newp = (char *) xmalloc (dirlen + 2);
      dir = memcpy (newp, dir, dirlen);
      newp[dirlen++] = '/';
      newp[dirlen] = '\0';
    }

  infile = (char *) alloca (dirlen + sizeof "gconv-modules");
  tmp = mempcpy (infile, dir, dirlen);
  tmp += dirlen;
  strcpy (tmp, "gconv-modules");

  fp = fopen (infile, "r");
  if (fp == NULL)
    {
      error (0, errno, "cannot open `%s'", infile);
      return 1;
    }

  /* No threads present.  */
  __fsetlocking (fp, FSETLOCKING_BYCALLER);

  while (!feof_unlocked (fp))
    {
      char *rp, *endp, *word;
      ssize_t n = __getdelim (&line, &linelen, '\n', fp);

      if (n < 0)
	/* An error occurred.  */
	break;

      rp = line;
      /* Terminate the line (excluding comments or newline) with a NUL
	 byte to simplify the following code.  */
      endp = strchr (rp, '#');
      if (endp != NULL)
	*endp = '\0';
      else
	if (rp[n - 1] == '\n')
	  rp[n - 1] = '\0';

      while (isspace (*rp))
	++rp;

      /* If this is an empty line go on with the next one.  */
      if (rp == endp)
	continue;

      word = rp;
      while (*rp != '\0' && !isspace (*rp))
	++rp;

      if (rp - word == sizeof ("alias") - 1
	  && memcmp (word, "alias", sizeof ("alias") - 1) == 0)
	add_alias (rp);
      else if (rp - word == sizeof ("module") - 1
	       && memcmp (word, "module", sizeof ("module") - 1) == 0)
	add_module (rp, dir);
      /* else */
	/* Otherwise ignore the line.  */
    }

  free (line);

  fclose (fp);

  return 0;
}


static void
append_module (const void *nodep, VISIT value, int level)
{
  struct module *mo;

  if (value != leaf && value != postorder)
    return;

  mo = *(struct module **) nodep;

  if (nmodule_list > 0
      && strcmp (module_list[nmodule_list - 1]->fromname, mo->fromname) == 0)
    {
      /* Same name.  */
      mo->next = module_list[nmodule_list - 1];
      module_list[nmodule_list - 1] = mo;

      return;
    }

  if (nmodule_list_max == nmodule_list)
    {
      nmodule_list_max += 50;
      module_list = (struct module **) xrealloc (module_list,
						 (nmodule_list_max
						  * sizeof (struct module *)));
    }

  module_list[nmodule_list++] = mo;
}


static void
get_modules (void)
{
  twalk (modules, append_module);
}


static void
add_builtins (void)
{
  size_t cnt;

  /* Add all aliases.  */
  for (cnt = 0; cnt < nbuiltin_alias; ++cnt)
    new_alias (builtin_alias[cnt].from,
	       strlen (builtin_alias[cnt].from) + 1,
	       builtin_alias[cnt].to,
	       strlen (builtin_alias[cnt].to) + 1);

  /* add the builtin transformations.  */
  for (cnt = 0; cnt < nbuiltin_trans; ++cnt)
    new_module (builtin_trans[cnt].from,
		strlen (builtin_trans[cnt].from) + 1,
		builtin_trans[cnt].to,
		strlen (builtin_trans[cnt].to) + 1,
		"", builtin_trans[cnt].module,
		strlen (builtin_trans[cnt].module) + 1,
		builtin_trans[cnt].cost, 0);
}


static int
name_compare (const void *p1, const void *p2)
{
  const struct name *n1 = (const struct name *) p1;
  const struct name *n2 = (const struct name *) p2;

  return strcmp (n1->name, n2->name);
}


static struct name *
new_name (const char *str, struct Strent *strent)
{
  struct name *newp = (struct name *) xmalloc (sizeof (struct name));

  newp->name = str;
  newp->strent = strent;
  newp->module_idx = -1;
  newp->hashval = hash_string (str);

  ++nnames;

  return newp;
}


static void
generate_name_list (void)
{
  size_t i;

  /* A name we always need.  */
  tsearch (new_name ("INTERNAL", strtabadd (strtab, "INTERNAL",
					    sizeof ("INTERNAL"))),
	   &names, name_compare);

  for (i = 0; i < nmodule_list; ++i)
    {
      struct module *runp;

      if (strcmp (module_list[i]->fromname, "INTERNAL") != 0)
	tsearch (new_name (module_list[i]->fromname,
			   module_list[i]->fromname_strent),
		 &names, name_compare);

      for (runp = module_list[i]; runp != NULL; runp = runp->next)
	if (strcmp (runp->toname, "INTERNAL") != 0)
	  tsearch (new_name (runp->toname, runp->toname_strent),
		   &names, name_compare);
    }
}


static int
name_to_module_idx (const char *name, int add)
{
  struct name **res;
  struct name fake_name = { .name = name };
  int idx;

  res = (struct name **) tfind (&fake_name, &names, name_compare);
  if (res == NULL)
    abort ();

  idx = (*res)->module_idx;
  if (idx == -1 && add)
    /* No module index assigned yet.  */
    idx = (*res)->module_idx = nname_info++;

  return idx;
}


static void
generate_name_info (void)
{
  size_t i;
  int idx;

  name_info = (struct name_info *) xcalloc (nmodule_list + 1,
					    sizeof (struct name_info));

  /* First add a special entry for the INTERNAL name.  This must have
     index zero.  */
  idx = name_to_module_idx ("INTERNAL", 1);
  name_info[0].canonical_name = "INTERNAL";
  name_info[0].canonical_strent = strtabadd (strtab, "INTERNAL",
					     sizeof ("INTERNAL"));
  assert (nname_info == 1);

  for (i = 0; i < nmodule_list; ++i)
    {
      struct module *runp;

      for (runp = module_list[i]; runp != NULL; runp = runp->next)
	if (strcmp (runp->fromname, "INTERNAL") == 0)
	  {
	    idx = name_to_module_idx (runp->toname, 1);
	    name_info[idx].from_internal = runp;
	    assert (name_info[idx].canonical_name == NULL
		    || strcmp (name_info[idx].canonical_name,
			       runp->toname) == 0);
	    name_info[idx].canonical_name = runp->toname;
	    name_info[idx].canonical_strent = runp->toname_strent;
	  }
	else if (strcmp (runp->toname, "INTERNAL") == 0)
	  {
	    idx = name_to_module_idx (runp->fromname, 1);
	    name_info[idx].to_internal = runp;
	    assert (name_info[idx].canonical_name == NULL
		    || strcmp (name_info[idx].canonical_name,
			       runp->fromname) == 0);
	    name_info[idx].canonical_name = runp->fromname;
	    name_info[idx].canonical_strent = runp->fromname_strent;
	  }
	else
	  {
	    /* This is a transformation not to or from the INTERNAL
	       encoding.  */
	    int from_idx = name_to_module_idx (runp->fromname, 1);
	    int to_idx = name_to_module_idx (runp->toname, 1);
	    struct other_conv_list *newp;

	    newp = (struct other_conv_list *)
	      xmalloc (sizeof (struct other_conv_list));
	    newp->other_conv.module_idx = to_idx;
	    newp->other_conv.module = runp;
	    newp->other_conv.next = NULL; /* XXX Allow multiple module sequence */
	    newp->dest_idx = to_idx;
	    newp->next = name_info[from_idx].other_conv_list;
	    name_info[from_idx].other_conv_list = newp;
	    assert (name_info[from_idx].canonical_name == NULL
		    || strcmp (name_info[from_idx].canonical_name,
			       runp->fromname) == 0);
	    name_info[from_idx].canonical_name = runp->fromname;
	    name_info[from_idx].canonical_strent = runp->fromname_strent;

	    ++nextra_modules;
	  }
    }

  /* Now add the module index information for all the aliases.  */
  for (i = 0; i < nalias_list; ++i)
    {
      struct name fake_name = { .name = alias_list[i]->toname };
      struct name **tonamep;

      tonamep = (struct name **) tfind (&fake_name, &names, name_compare);
      if (tonamep != NULL)
	{
	  struct name *newp = new_name (alias_list[i]->fromname,
					alias_list[i]->froment);
	  newp->module_idx = (*tonamep)->module_idx;
	  tsearch (newp, &names, name_compare);
	}
    }
}


static int
is_prime (unsigned long int candidate)
{
  /* No even number and none less than 10 will be passed here.  */
  unsigned long int divn = 3;
  unsigned long int sq = divn * divn;

  while (sq < candidate && candidate % divn != 0)
    {
      ++divn;
      sq += 4 * divn;
      ++divn;
    }

  return candidate % divn != 0;
}


static uint32_t
next_prime (uint32_t seed)
{
  /* Make it definitely odd.  */
  seed |= 1;

  while (!is_prime (seed))
    seed += 2;

  return seed;
}


/* Format of the output file.

   Offset   Length       Description
   0000     4            Magic header bytes
   0004     4            Offset of string table (stoff)
   0008     4            Offset of name hashing table (hoff)
   000C     4            Hashing table size (hsize)
   0010     4            Offset of module table (moff)
   0014     4            Offset of other conversion module table (ooff)

   stoff    ???          String table

   hoff     8*hsize      Array of tuples
                            string table offset
                            module index

   moff     ???          Array of tuples
                            canonical name offset
                            from-internal module dir name offset
                            from-internal module name off
                            to-internal module dir name offset
                            to-internal module name offset
                            offset into other conversion table

   ooff     ???          One or more of
                            number of steps/modules
                            one or more of tuple
                              canonical name offset for output
                              module dir name offset
                              module name offset
                         (following last entry with step count 0)
*/
static int
write_output (void)
{
  int fd;
  char *string_table;
  size_t string_table_size;
  struct gconvcache_header header;
  struct hash_entry *hash_table;
  size_t hash_size;
  struct module_entry *module_table;
  char *extra_table;
  char *cur_extra_table;
  size_t n;
  int idx;
  struct iovec iov[6];
  static const gidx_t null_word;
  size_t total;
  char tmpfname[sizeof (GCONV_MODULES_CACHE) + strlen (".XXXXXX")];

  /* Function to insert the names.  */
  static void name_insert (const void *nodep, VISIT value, int level)
    {
      struct name *name;
      unsigned int idx;
      unsigned int hval2;

      if (value != leaf && value != postorder)
	return;

      name = *(struct name **) nodep;
      idx = name->hashval % hash_size;
      hval2 = 1 + name->hashval % (hash_size - 2);

      while (hash_table[idx].string_offset != 0)
	if ((idx += hval2) >= hash_size)
	  idx -= hash_size;

      hash_table[idx].string_offset = strtaboffset (name->strent);

      assert (name->module_idx != -1);
      hash_table[idx].module_idx = name->module_idx;
    }

  /* Open the output file.  */
  strcpy (stpcpy (tmpfname, GCONV_MODULES_CACHE), ".XXXXXX");
  fd = mkstemp (tmpfname);
  if (fd == -1)
    return 1;

  /* Create the string table.  */
  string_table = strtabfinalize (strtab, &string_table_size);

  /* Create the hashing table.  We know how many strings we have.
     Creating a perfect hash table is not reasonable here.  Therefore
     we use open hashing and a table size which is the next prime 40%
     larger than the number of strings.  */
  hash_size = next_prime (nnames * 1.4);
  hash_table = (struct hash_entry *) xcalloc (hash_size,
					      sizeof (struct hash_entry));
  /* Fill the hash table.  */
  twalk (names, name_insert);

  /* Create the section for the module list.  */
  module_table = (struct module_entry *) xcalloc (sizeof (struct module_entry),
						  nname_info);

  /* Allocate memory for the non-INTERNAL conversions.  The allocated
     memory can be more than is actually needed.  */
  extra_table = (char *) xcalloc (sizeof (struct extra_entry)
				  + sizeof (gidx_t)
				  + sizeof (struct extra_entry_module),
				  nextra_modules);
  cur_extra_table = extra_table;

  /* Fill in the module information.  */
  for (n = 0; n < nname_info; ++n)
    {
      module_table[n].canonname_offset =
	strtaboffset (name_info[n].canonical_strent);

      if (name_info[n].from_internal == NULL)
	{
	  module_table[n].fromdir_offset = 0;
	  module_table[n].fromname_offset = 0;
	}
      else
	{
	  module_table[n].fromdir_offset =
	    strtaboffset (name_info[n].from_internal->directory_strent);
	  module_table[n].fromname_offset =
	    strtaboffset (name_info[n].from_internal->filename_strent);
	}

      if (name_info[n].to_internal == NULL)
	{
	  module_table[n].todir_offset = 0;
	  module_table[n].toname_offset = 0;
	}
      else
	{
	  module_table[n].todir_offset =
	    strtaboffset (name_info[n].to_internal->directory_strent);
	  module_table[n].toname_offset =
	    strtaboffset (name_info[n].to_internal->filename_strent);
	}

      if (name_info[n].other_conv_list != NULL)
	{
	  struct other_conv_list *other = name_info[n].other_conv_list;

	  /* Store the reference.  We add 1 to distinguish the entry
	     at offset zero from the case where no extra modules are
	     available.  The file reader has to account for the
	     offset.  */
	  module_table[n].extra_offset = 1 + cur_extra_table - extra_table;

	  do
	    {
	      struct other_conv *runp;
	      struct extra_entry *extra;

	      /* Allocate new entry.  */
	      extra = (struct extra_entry *) cur_extra_table;
	      cur_extra_table += sizeof (struct extra_entry);
	      extra->module_cnt = 0;

	      runp = &other->other_conv;
	      do
		{
		  cur_extra_table += sizeof (struct extra_entry_module);
		  extra->module[extra->module_cnt].outname_offset =
		    runp->next == NULL
		    ? other->dest_idx : runp->next->module_idx;
		  extra->module[extra->module_cnt].dir_offset =
		    strtaboffset (runp->module->directory_strent);
		  extra->module[extra->module_cnt].name_offset =
		    strtaboffset (runp->module->filename_strent);
		  ++extra->module_cnt;

		  runp = runp->next;
		}
	      while (runp != NULL);

	      other = other->next;
	    }
	  while (other != NULL);

	  /* Final module_cnt is zero.  */
	  *((gidx_t *) cur_extra_table) = 0;
	  cur_extra_table += sizeof (gidx_t);
	}
    }

  header.magic = GCONVCACHE_MAGIC;

  iov[0].iov_base = &header;
  iov[0].iov_len = sizeof (struct gconvcache_header);
  total = iov[0].iov_len;

  header.string_offset = total;
  iov[1].iov_base = string_table;
  iov[1].iov_len = string_table_size;
  total += iov[1].iov_len;

  idx = 2;
  if ((string_table_size & (sizeof (gidx_t) - 1)) != 0)
    {
      iov[2].iov_base = (void *) &null_word;
      iov[2].iov_len = (sizeof (gidx_t)
			- (string_table_size & (sizeof (gidx_t) - 1)));
      total += iov[2].iov_len;
      ++idx;
    }

  header.hash_offset = total;
  header.hash_size = hash_size;
  iov[idx].iov_base = hash_table;
  iov[idx].iov_len = hash_size * sizeof (struct hash_entry);
  total += iov[idx].iov_len;
  ++idx;

  header.module_offset = total;
  iov[idx].iov_base = module_table;
  iov[idx].iov_len = nname_info * sizeof (struct module_entry);
  total += iov[idx].iov_len;
  ++idx;

  assert (cur_extra_table - extra_table
	  <= ((sizeof (struct extra_entry) + sizeof (gidx_t)
	       + sizeof (struct extra_entry_module))
	      * nextra_modules));
  header.otherconv_offset = total;
  iov[idx].iov_base = extra_table;
  iov[idx].iov_len = cur_extra_table - extra_table;
  total += iov[idx].iov_len;
  ++idx;

  if (TEMP_FAILURE_RETRY (writev (fd, iov, idx)) != total
      /* The file was created with mode 0600.  Make it world-readable.  */
      || fchmod (fd, 0644) != 0
      /* Rename the file, possibly replacing an old one.  */
      || rename (tmpfname, GCONV_MODULES_CACHE) != 0)
    {
      int save_errno = errno;
      close (fd);
      unlink (tmpfname);
      error (EXIT_FAILURE, save_errno,
	     gettext ("cannot generate output file"));
    }

  close (fd);

  return 0;
}