summary refs log tree commit diff
path: root/intl
diff options
context:
space:
mode:
authorUlrich Drepper <drepper@redhat.com>2005-04-04 22:37:38 +0000
committerUlrich Drepper <drepper@redhat.com>2005-04-04 22:37:38 +0000
commit4e82c610255f6a186c20c73e74f8e71dcda98efc (patch)
treeb55a3e6692a0261c576330950793bde88524e0e1 /intl
parent7a50b1f6d1bd959ae5dfb5539d9cd8935eb8d926 (diff)
downloadglibc-4e82c610255f6a186c20c73e74f8e71dcda98efc.tar.gz
glibc-4e82c610255f6a186c20c73e74f8e71dcda98efc.tar.xz
glibc-4e82c610255f6a186c20c73e74f8e71dcda98efc.zip
* intl/tst-gettext3.c: New file.
	* intl/tst-gettext3.sh: New file.
	* intl/Makefile (distribute): Add tst-gettext3.sh.
	(test-srcs): Add tst-gettext3.
	(tests): Depend on tst-gettext3.out.
	(tst-gettext3.out): New rule.
	(CFLAGS-tst-gettext3.c): New variable.

	Fix bug exposed by tst-gettext3.
	* intl/gettextP.h (struct converted_domain): New type.
	(struct loaded_domain): Remove the conv, conv_tab fields. Add
	conversions, nconversions fields.
	(_nl_init_domain_conv): Remove declaration.
	(_nl_free_domain_conv): Remove declaration.
	(_nl_find_msg): Add convert argument.
	* intl/dcigettext.c (DCIGETTEXT): Call _nl_find_msg with convert=1.
	(_nl_find_msg): Add convert argument. When a conversion to a different
	charset is needed, create a new converted_domain element, instead of
	throwing away the old converted translations.
	(get_output_charset): New function.
	* intl/loadmsgcat.c (_nl_init_domain_conv): Remove function.
	(_nl_free_domain_conv): Remove function.
	(_nl_load_domain): Initialize the conversions array to empty. Use
	_nl_find_msg instead of _nl_init_domain_conv to retrieve the header
	entry.
	(_nl_unload_domain): Free the conversions array and its contents.

	* intl/gettextP.h (struct loaded_domain): Remove codeset_cntr field.
	(struct binding): Likewise.
	* intl/bindtextdom.c (set_binding_values): Drop codeset_cntr
	modifications.
Diffstat (limited to 'intl')
-rw-r--r--intl/bindtextdom.c4
-rw-r--r--intl/dcigettext.c485
-rw-r--r--intl/gettextP.h45
-rw-r--r--intl/loadmsgcat.c172
-rw-r--r--intl/tst-gettext3.c60
-rw-r--r--intl/tst-gettext3.sh45
6 files changed, 489 insertions, 322 deletions
diff --git a/intl/bindtextdom.c b/intl/bindtextdom.c
index 546da8e29d..fd527a180a 100644
--- a/intl/bindtextdom.c
+++ b/intl/bindtextdom.c
@@ -207,7 +207,6 @@ set_binding_values (domainname, dirnamep, codesetp)
 			free (binding->codeset);
 
 		      binding->codeset = result;
-		      ++binding->codeset_cntr;
 		      modified = 1;
 		    }
 		}
@@ -271,8 +270,6 @@ set_binding_values (domainname, dirnamep, codesetp)
 	/* The default value.  */
 	new_binding->dirname = (char *) _nl_default_dirname;
 
-      new_binding->codeset_cntr = 0;
-
       if (codesetp)
 	{
 	  const char *codeset = *codesetp;
@@ -293,7 +290,6 @@ set_binding_values (domainname, dirnamep, codesetp)
 	      memcpy (result, codeset, len);
 #endif
 	      codeset = result;
-	      ++new_binding->codeset_cntr;
 	    }
 	  *codesetp = codeset;
 	  new_binding->codeset = (char *) codeset;
diff --git a/intl/dcigettext.c b/intl/dcigettext.c
index 8163064edc..c73c719d94 100644
--- a/intl/dcigettext.c
+++ b/intl/dcigettext.c
@@ -326,6 +326,10 @@ static struct transmem_list *transmem_list;
 #else
 typedef unsigned char transmem_block_t;
 #endif
+#if defined _LIBC || HAVE_ICONV
+static const char *get_output_charset PARAMS ((struct binding *domainbinding))
+     internal_function;
+#endif
 
 
 /* Names for the libintl functions are a problem.  They must not clash
@@ -597,7 +601,7 @@ DCIGETTEXT (domainname, msgid1, msgid2, plural, n, category)
 
       if (domain != NULL)
 	{
-	  retval = _nl_find_msg (domain, binding, msgid1, &retlen);
+	  retval = _nl_find_msg (domain, binding, msgid1, 1, &retlen);
 
 	  if (retval == NULL)
 	    {
@@ -606,7 +610,7 @@ DCIGETTEXT (domainname, msgid1, msgid2, plural, n, category)
 	      for (cnt = 0; domain->successor[cnt] != NULL; ++cnt)
 		{
 		  retval = _nl_find_msg (domain->successor[cnt], binding,
-					 msgid1, &retlen);
+					 msgid1, 1, &retlen);
 
 		  if (retval != NULL)
 		    {
@@ -683,10 +687,11 @@ DCIGETTEXT (domainname, msgid1, msgid2, plural, n, category)
 
 char *
 internal_function
-_nl_find_msg (domain_file, domainbinding, msgid, lengthp)
+_nl_find_msg (domain_file, domainbinding, msgid, convert, lengthp)
      struct loaded_l10nfile *domain_file;
      struct binding *domainbinding;
      const char *msgid;
+     int convert;
      size_t *lengthp;
 {
   struct loaded_domain *domain;
@@ -793,192 +798,317 @@ _nl_find_msg (domain_file, domainbinding, msgid, lengthp)
     }
 
 #if defined _LIBC || HAVE_ICONV
-  if (domain->codeset_cntr
-      != (domainbinding != NULL ? domainbinding->codeset_cntr : 0))
+  if (convert)
     {
-      /* The domain's codeset has changed through bind_textdomain_codeset()
-	 since the message catalog was initialized or last accessed.  We
-	 have to reinitialize the converter.  */
-      _nl_free_domain_conv (domain);
-      _nl_init_domain_conv (domain_file, domain, domainbinding);
-    }
+      /* We are supposed to do a conversion.  */
+      const char *encoding = get_output_charset (domainbinding);
+
+      /* Search whether a table with converted translations for this
+	 encoding has already been allocated.  */
+      size_t nconversions = domain->nconversions;
+      struct converted_domain *convd = NULL;
+      size_t i;
+
+      for (i = nconversions; i > 0; )
+	{
+	  i--;
+	  if (strcmp (domain->conversions[i].encoding, encoding) == 0)
+	    {
+	      convd = &domain->conversions[i];
+	      break;
+	    }
+	}
 
-  if (
+      if (convd == NULL)
+	{
+	  /* Allocate a table for the converted translations for this
+	     encoding.  */
+	  struct converted_domain *new_conversions =
+	    (struct converted_domain *)
+	    (domain->conversions != NULL
+	     ? realloc (domain->conversions,
+			(nconversions + 1) * sizeof (struct converted_domain))
+	     : malloc ((nconversions + 1) * sizeof (struct converted_domain)));
+
+	  if (__builtin_expect (new_conversions == NULL, 0))
+	    /* Nothing we can do, no more memory.  */
+	    goto converted;
+	  domain->conversions = new_conversions;
+
+	  /* Copy the 'encoding' string to permanent storage.  */
+	  encoding = strdup (encoding);
+	  if (__builtin_expect (encoding == NULL, 0))
+	    /* Nothing we can do, no more memory.  */
+	    goto converted;
+
+	  convd = &new_conversions[nconversions];
+	  convd->encoding = encoding;
+
+	  /* Find out about the character set the file is encoded with.
+	     This can be found (in textual form) in the entry "".  If this
+	     entry does not exist or if this does not contain the 'charset='
+	     information, we will assume the charset matches the one the
+	     current locale and we don't have to perform any conversion.  */
 # ifdef _LIBC
-      domain->conv != (__gconv_t) -1
+	  convd->conv = (__gconv_t) -1;
 # else
 #  if HAVE_ICONV
-      domain->conv != (iconv_t) -1
+	  convd->conv = (iconv_t) -1;
 #  endif
 # endif
-      )
-    {
-      /* We are supposed to do a conversion.  First allocate an
-	 appropriate table with the same structure as the table
-	 of translations in the file, where we can put the pointers
-	 to the converted strings in.
-	 There is a slight complication with plural entries.  They
-	 are represented by consecutive NUL terminated strings.  We
-	 handle this case by converting RESULTLEN bytes, including
-	 NULs.  */
-
-      if (domain->conv_tab == NULL
-	  && ((domain->conv_tab =
-		 (char **) calloc (nstrings + domain->n_sysdep_strings,
-				   sizeof (char *)))
-	      == NULL))
-	/* Mark that we didn't succeed allocating a table.  */
-	domain->conv_tab = (char **) -1;
-
-      if (__builtin_expect (domain->conv_tab == (char **) -1, 0))
-	/* Nothing we can do, no more memory.  */
-	goto converted;
-
-      if (domain->conv_tab[act] == NULL)
+	  {
+	    char *nullentry;
+	    size_t nullentrylen;
+
+	    /* Get the header entry.  This is a recursion, but it doesn't
+	       reallocate domain->conversions because we pass convert = 0.  */
+	    nullentry =
+	      _nl_find_msg (domain_file, domainbinding, "", 0, &nullentrylen);
+
+	    if (nullentry != NULL)
+	      {
+		const char *charsetstr;
+
+		charsetstr = strstr (nullentry, "charset=");
+		if (charsetstr != NULL)
+		  {
+		    size_t len;
+		    char *charset;
+		    const char *outcharset;
+
+		    charsetstr += strlen ("charset=");
+		    len = strcspn (charsetstr, " \t\n");
+
+		    charset = (char *) alloca (len + 1);
+# if defined _LIBC || HAVE_MEMPCPY
+		    *((char *) mempcpy (charset, charsetstr, len)) = '\0';
+# else
+		    memcpy (charset, charsetstr, len);
+		    charset[len] = '\0';
+# endif
+
+		    outcharset = encoding;
+
+# ifdef _LIBC
+		    /* We always want to use transliteration.  */
+		    outcharset = norm_add_slashes (outcharset, "TRANSLIT");
+		    charset = norm_add_slashes (charset, "");
+		    if (__gconv_open (outcharset, charset, &convd->conv,
+				      GCONV_AVOID_NOCONV)
+			!= __GCONV_OK)
+		      convd->conv = (__gconv_t) -1;
+# else
+#  if HAVE_ICONV
+		    /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
+		       we want to use transliteration.  */
+#   if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
+       || _LIBICONV_VERSION >= 0x0105
+		    if (strchr (outcharset, '/') == NULL)
+		      {
+			char *tmp;
+
+			len = strlen (outcharset);
+			tmp = (char *) alloca (len + 10 + 1);
+			memcpy (tmp, outcharset, len);
+			memcpy (tmp + len, "//TRANSLIT", 10 + 1);
+			outcharset = tmp;
+
+			convd->conv = iconv_open (outcharset, charset);
+
+			freea (outcharset);
+		      }
+		    else
+#   endif
+		      convd->conv = iconv_open (outcharset, charset);
+#  endif
+# endif
+
+		    freea (charset);
+		  }
+	      }
+	  }
+	  convd->conv_tab = NULL;
+	  /* Here domain->conversions is still == new_conversions.  */
+	  domain->nconversions++;
+	}
+
+      if (
+# ifdef _LIBC
+	  convd->conv != (__gconv_t) -1
+# else
+#  if HAVE_ICONV
+	  convd->conv != (iconv_t) -1
+#  endif
+# endif
+	  )
 	{
-	  /* We haven't used this string so far, so it is not
-	     translated yet.  Do this now.  */
-	  /* We use a bit more efficient memory handling.
-	     We allocate always larger blocks which get used over
-	     time.  This is faster than many small allocations.   */
-	  __libc_lock_define_initialized (static, lock)
+	  /* We are supposed to do a conversion.  First allocate an
+	     appropriate table with the same structure as the table
+	     of translations in the file, where we can put the pointers
+	     to the converted strings in.
+	     There is a slight complication with plural entries.  They
+	     are represented by consecutive NUL terminated strings.  We
+	     handle this case by converting RESULTLEN bytes, including
+	     NULs.  */
+
+	  if (convd->conv_tab == NULL
+	      && ((convd->conv_tab =
+		    (char **) calloc (nstrings + domain->n_sysdep_strings,
+				      sizeof (char *)))
+		  == NULL))
+	    /* Mark that we didn't succeed allocating a table.  */
+	    convd->conv_tab = (char **) -1;
+
+	  if (__builtin_expect (convd->conv_tab == (char **) -1, 0))
+	    /* Nothing we can do, no more memory.  */
+	    goto converted;
+
+	  if (convd->conv_tab[act] == NULL)
+	    {
+	      /* We haven't used this string so far, so it is not
+		 translated yet.  Do this now.  */
+	      /* We use a bit more efficient memory handling.
+		 We allocate always larger blocks which get used over
+		 time.  This is faster than many small allocations.   */
+	      __libc_lock_define_initialized (static, lock)
 # define INITIAL_BLOCK_SIZE	4080
-	  static unsigned char *freemem;
-	  static size_t freemem_size;
+	      static unsigned char *freemem;
+	      static size_t freemem_size;
 
-	  const unsigned char *inbuf;
-	  unsigned char *outbuf;
-	  int malloc_count;
+	      const unsigned char *inbuf;
+	      unsigned char *outbuf;
+	      int malloc_count;
 # ifndef _LIBC
-	  transmem_block_t *transmem_list = NULL;
+	      transmem_block_t *transmem_list = NULL;
 # endif
 
-	  __libc_lock_lock (lock);
+	      __libc_lock_lock (lock);
 
-	  inbuf = (const unsigned char *) result;
-	  outbuf = freemem + sizeof (size_t);
+	      inbuf = (const unsigned char *) result;
+	      outbuf = freemem + sizeof (size_t);
 
-	  malloc_count = 0;
-	  while (1)
-	    {
-	      transmem_block_t *newmem;
+	      malloc_count = 0;
+	      while (1)
+		{
+		  transmem_block_t *newmem;
 # ifdef _LIBC
-	      size_t non_reversible;
-	      int res;
+		  size_t non_reversible;
+		  int res;
 
-	      if (freemem_size < sizeof (size_t))
-		goto resize_freemem;
+		  if (freemem_size < sizeof (size_t))
+		    goto resize_freemem;
 
-	      res = __gconv (domain->conv,
-			     &inbuf, inbuf + resultlen,
-			     &outbuf,
-			     outbuf + freemem_size - sizeof (size_t),
-			     &non_reversible);
+		  res = __gconv (convd->conv,
+				 &inbuf, inbuf + resultlen,
+				 &outbuf,
+				 outbuf + freemem_size - sizeof (size_t),
+				 &non_reversible);
 
-	      if (res == __GCONV_OK || res == __GCONV_EMPTY_INPUT)
-		break;
+		  if (res == __GCONV_OK || res == __GCONV_EMPTY_INPUT)
+		    break;
 
-	      if (res != __GCONV_FULL_OUTPUT)
-		{
-		  __libc_lock_unlock (lock);
-		  goto converted;
-		}
+		  if (res != __GCONV_FULL_OUTPUT)
+		    {
+		      __libc_lock_unlock (lock);
+		      goto converted;
+		    }
 
-	      inbuf = (const unsigned char *) result;
+		  inbuf = (const unsigned char *) result;
 # else
 #  if HAVE_ICONV
-	      const char *inptr = (const char *) inbuf;
-	      size_t inleft = resultlen;
-	      char *outptr = (char *) outbuf;
-	      size_t outleft;
-
-	      if (freemem_size < sizeof (size_t))
-		goto resize_freemem;
-
-	      outleft = freemem_size - sizeof (size_t);
-	      if (iconv (domain->conv,
-			 (ICONV_CONST char **) &inptr, &inleft,
-			 &outptr, &outleft)
-		  != (size_t) (-1))
-		{
-		  outbuf = (unsigned char *) outptr;
-		  break;
-		}
-	      if (errno != E2BIG)
-		{
-		  __libc_lock_unlock (lock);
-		  goto converted;
-		}
+		  const char *inptr = (const char *) inbuf;
+		  size_t inleft = resultlen;
+		  char *outptr = (char *) outbuf;
+		  size_t outleft;
+
+		  if (freemem_size < sizeof (size_t))
+		    goto resize_freemem;
+
+		  outleft = freemem_size - sizeof (size_t);
+		  if (iconv (convd->conv,
+			     (ICONV_CONST char **) &inptr, &inleft,
+			     &outptr, &outleft)
+		      != (size_t) (-1))
+		    {
+		      outbuf = (unsigned char *) outptr;
+		      break;
+		    }
+		  if (errno != E2BIG)
+		    {
+		      __libc_lock_unlock (lock);
+		      goto converted;
+		    }
 #  endif
 # endif
 
-	    resize_freemem:
-	      /* We must allocate a new buffer or resize the old one.  */
-	      if (malloc_count > 0)
-		{
-		  ++malloc_count;
-		  freemem_size = malloc_count * INITIAL_BLOCK_SIZE;
-		  newmem = (transmem_block_t *) realloc (transmem_list,
-							 freemem_size);
+		resize_freemem:
+		  /* We must allocate a new buffer or resize the old one.  */
+		  if (malloc_count > 0)
+		    {
+		      ++malloc_count;
+		      freemem_size = malloc_count * INITIAL_BLOCK_SIZE;
+		      newmem = (transmem_block_t *) realloc (transmem_list,
+							     freemem_size);
 # ifdef _LIBC
-		  if (newmem != NULL)
-		    transmem_list = transmem_list->next;
+		      if (newmem != NULL)
+			transmem_list = transmem_list->next;
+		      else
+			{
+			  struct transmem_list *old = transmem_list;
+
+			  transmem_list = transmem_list->next;
+			  free (old);
+			}
+# endif
+		    }
 		  else
 		    {
-		      struct transmem_list *old = transmem_list;
-
-		      transmem_list = transmem_list->next;
-		      free (old);
+		      malloc_count = 1;
+		      freemem_size = INITIAL_BLOCK_SIZE;
+		      newmem = (transmem_block_t *) malloc (freemem_size);
+		    }
+		  if (__builtin_expect (newmem == NULL, 0))
+		    {
+		      freemem = NULL;
+		      freemem_size = 0;
+		      __libc_lock_unlock (lock);
+		      goto converted;
 		    }
-# endif
-		}
-	      else
-		{
-		  malloc_count = 1;
-		  freemem_size = INITIAL_BLOCK_SIZE;
-		  newmem = (transmem_block_t *) malloc (freemem_size);
-		}
-	      if (__builtin_expect (newmem == NULL, 0))
-		{
-		  freemem = NULL;
-		  freemem_size = 0;
-		  __libc_lock_unlock (lock);
-		  goto converted;
-		}
 
 # ifdef _LIBC
-	      /* Add the block to the list of blocks we have to free
-                 at some point.  */
-	      newmem->next = transmem_list;
-	      transmem_list = newmem;
+		  /* Add the block to the list of blocks we have to free
+		     at some point.  */
+		  newmem->next = transmem_list;
+		  transmem_list = newmem;
 
-	      freemem = (unsigned char *) newmem->data;
-	      freemem_size -= offsetof (struct transmem_list, data);
+		  freemem = (unsigned char *) newmem->data;
+		  freemem_size -= offsetof (struct transmem_list, data);
 # else
-	      transmem_list = newmem;
-	      freemem = newmem;
+		  transmem_list = newmem;
+		  freemem = newmem;
 # endif
 
-	      outbuf = freemem + sizeof (size_t);
+		  outbuf = freemem + sizeof (size_t);
+		}
+
+	      /* We have now in our buffer a converted string.  Put this
+		 into the table of conversions.  */
+	      *(size_t *) freemem = outbuf - freemem - sizeof (size_t);
+	      convd->conv_tab[act] = (char *) freemem;
+	      /* Shrink freemem, but keep it aligned.  */
+	      freemem_size -= outbuf - freemem;
+	      freemem = outbuf;
+	      freemem += freemem_size & (alignof (size_t) - 1);
+	      freemem_size = freemem_size & ~ (alignof (size_t) - 1);
+
+	      __libc_lock_unlock (lock);
 	    }
 
-	  /* We have now in our buffer a converted string.  Put this
-	     into the table of conversions.  */
-	  *(size_t *) freemem = outbuf - freemem - sizeof (size_t);
-	  domain->conv_tab[act] = (char *) freemem;
-	  /* Shrink freemem, but keep it aligned.  */
-	  freemem_size -= outbuf - freemem;
-	  freemem = outbuf;
-	  freemem += freemem_size & (alignof (size_t) - 1);
-	  freemem_size = freemem_size & ~ (alignof (size_t) - 1);
-
-	  __libc_lock_unlock (lock);
+	  /* Now convd->conv_tab[act] contains the translation of all
+	     the plural variants.  */
+	  result = convd->conv_tab[act] + sizeof (size_t);
+	  resultlen = *(size_t *) convd->conv_tab[act];
 	}
-
-      /* Now domain->conv_tab[act] contains the translation of all
-	 the plural variants.  */
-      result = domain->conv_tab[act] + sizeof (size_t);
-      resultlen = *(size_t *) domain->conv_tab[act];
     }
 
  converted:
@@ -1122,6 +1252,61 @@ guess_category_value (category, categoryname)
   return language != NULL && strcmp (retval, "C") != 0 ? language : retval;
 }
 
+#if defined _LIBC || HAVE_ICONV
+/* Returns the output charset.  */
+static const char *
+internal_function
+get_output_charset (domainbinding)
+     struct binding *domainbinding;
+{
+  /* The output charset should normally be determined by the locale.  But
+     sometimes the locale is not used or not correctly set up, so we provide
+     a possibility for the user to override this: the OUTPUT_CHARSET
+     environment variable.  Moreover, the value specified through
+     bind_textdomain_codeset overrides both.  */
+  if (domainbinding != NULL && domainbinding->codeset != NULL)
+    return domainbinding->codeset;
+  else
+    {
+      /* For speed reasons, we look at the value of OUTPUT_CHARSET only
+	 once.  This is a user variable that is not supposed to change
+	 during a program run.  */
+      static char *output_charset_cache;
+      static int output_charset_cached;
+
+      if (!output_charset_cached)
+	{
+	  const char *value = getenv ("OUTPUT_CHARSET");
+
+	  if (value != NULL && value[0] != '\0')
+	    {
+	      size_t len = strlen (value) + 1;
+	      char *value_copy = (char *) malloc (len);
+
+	      if (value_copy != NULL)
+		memcpy (value_copy, value, len);
+	      output_charset_cache = value_copy;
+	    }
+	  output_charset_cached = 1;
+	}
+
+      if (output_charset_cache != NULL)
+	return output_charset_cache;
+      else
+	{
+# ifdef _LIBC
+	  return _NL_CURRENT (LC_CTYPE, CODESET);
+# else
+#  if HAVE_ICONV
+	  extern const char *locale_charset PARAMS ((void);
+	  return locale_charset ();
+#  endif
+# endif
+	}
+    }
+}
+#endif
+
 /* @@ begin of epilog @@ */
 
 /* We don't want libintl.a to depend on any other library.  So we
diff --git a/intl/gettextP.h b/intl/gettextP.h
index 46b51e1008..f18535a5b3 100644
--- a/intl/gettextP.h
+++ b/intl/gettextP.h
@@ -1,5 +1,5 @@
 /* Header describing internals of libintl library.
-   Copyright (C) 1995-1999, 2000, 2001, 2004 Free Software Foundation, Inc.
+   Copyright (C) 1995-1999, 2000, 2001, 2004-2005 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Written by Ulrich Drepper <drepper@cygnus.com>, 1995.
 
@@ -88,6 +88,26 @@ struct sysdep_string_desc
   const char *pointer;
 };
 
+/* Cache of translated strings after charset conversion.
+   Note: The strings are converted to the target encoding only on an as-needed
+   basis.  */
+struct converted_domain
+{
+  /* The target encoding name.  */
+  const char *encoding;
+  /* The descriptor for conversion from the message catalog's encoding to
+     this target encoding.  */
+#ifdef _LIBC
+  __gconv_t conv;
+#else
+# if HAVE_ICONV
+  iconv_t conv;
+# endif
+#endif
+  /* The table of translated strings after charset conversion.  */
+  char **conv_tab;
+};
+
 /* The representation of an opened message catalog.  */
 struct loaded_domain
 {
@@ -123,15 +143,9 @@ struct loaded_domain
   /* 1 if the hash table uses a different endianness than this machine.  */
   int must_swap_hash_tab;
 
-  int codeset_cntr;
-#ifdef _LIBC
-  __gconv_t conv;
-#else
-# if HAVE_ICONV
-  iconv_t conv;
-# endif
-#endif
-  char **conv_tab;
+  /* Cache of charset conversions of the translated strings.  */
+  struct converted_domain *conversions;
+  size_t nconversions;
 
   struct expression *plural;
   unsigned long int nplurals;
@@ -151,7 +165,6 @@ struct binding
 {
   struct binding *next;
   char *dirname;
-  int codeset_cntr;	/* Incremented each time codeset changes.  */
   char *codeset;
   char domainname[ZERO];
 };
@@ -173,16 +186,10 @@ struct loaded_l10nfile *_nl_find_domain PARAMS ((const char *__dirname,
 void _nl_load_domain PARAMS ((struct loaded_l10nfile *__domain,
 			      struct binding *__domainbinding))
      internal_function;
-const char *_nl_init_domain_conv PARAMS ((struct loaded_l10nfile *__domain_file,
-					  struct loaded_domain *__domain,
-					  struct binding *__domainbinding))
-     internal_function;
-void _nl_free_domain_conv PARAMS ((struct loaded_domain *__domain))
-     internal_function;
 
 char *_nl_find_msg PARAMS ((struct loaded_l10nfile *domain_file,
-			    struct binding *domainbinding,
-			    const char *msgid, size_t *lengthp))
+			    struct binding *domainbinding, const char *msgid,
+			    int convert, size_t *lengthp))
      internal_function;
 
 #ifdef _LIBC
diff --git a/intl/loadmsgcat.c b/intl/loadmsgcat.c
index efefc69a43..1f55531097 100644
--- a/intl/loadmsgcat.c
+++ b/intl/loadmsgcat.c
@@ -1,5 +1,5 @@
 /* Load needed message catalogs.
-   Copyright (C) 1995-2004 Free Software Foundation, Inc.
+   Copyright (C) 1995-2005 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -752,146 +752,6 @@ get_sysdep_segment_value (name)
   return NULL;
 }
 
-/* Initialize the codeset dependent parts of an opened message catalog.
-   Return the header entry.  */
-const char *
-internal_function
-_nl_init_domain_conv (domain_file, domain, domainbinding)
-     struct loaded_l10nfile *domain_file;
-     struct loaded_domain *domain;
-     struct binding *domainbinding;
-{
-  /* Find out about the character set the file is encoded with.
-     This can be found (in textual form) in the entry "".  If this
-     entry does not exist or if this does not contain the `charset='
-     information, we will assume the charset matches the one the
-     current locale and we don't have to perform any conversion.  */
-  char *nullentry;
-  size_t nullentrylen;
-
-  /* Preinitialize fields, to avoid recursion during _nl_find_msg.  */
-  domain->codeset_cntr =
-    (domainbinding != NULL ? domainbinding->codeset_cntr : 0);
-#ifdef _LIBC
-  domain->conv = (__gconv_t) -1;
-#else
-# if HAVE_ICONV
-  domain->conv = (iconv_t) -1;
-# endif
-#endif
-  domain->conv_tab = NULL;
-
-  /* Get the header entry.  */
-  nullentry = _nl_find_msg (domain_file, domainbinding, "", &nullentrylen);
-
-  if (nullentry != NULL)
-    {
-#if defined _LIBC || HAVE_ICONV
-      const char *charsetstr;
-
-      charsetstr = strstr (nullentry, "charset=");
-      if (charsetstr != NULL)
-	{
-	  size_t len;
-	  char *charset;
-	  const char *outcharset;
-
-	  charsetstr += strlen ("charset=");
-	  len = strcspn (charsetstr, " \t\n");
-
-	  charset = (char *) alloca (len + 1);
-# if defined _LIBC || HAVE_MEMPCPY
-	  *((char *) mempcpy (charset, charsetstr, len)) = '\0';
-# else
-	  memcpy (charset, charsetstr, len);
-	  charset[len] = '\0';
-# endif
-
-	  /* The output charset should normally be determined by the
-	     locale.  But sometimes the locale is not used or not correctly
-	     set up, so we provide a possibility for the user to override
-	     this.  Moreover, the value specified through
-	     bind_textdomain_codeset overrides both.  */
-	  if (domainbinding != NULL && domainbinding->codeset != NULL)
-	    outcharset = domainbinding->codeset;
-	  else
-	    {
-	      outcharset = getenv ("OUTPUT_CHARSET");
-	      if (outcharset == NULL || outcharset[0] == '\0')
-		{
-# ifdef _LIBC
-		  outcharset = _NL_CURRENT (LC_CTYPE, CODESET);
-# else
-#  if HAVE_ICONV
-		  extern const char *locale_charset PARAMS ((void));
-		  outcharset = locale_charset ();
-#  endif
-# endif
-		}
-	    }
-
-# ifdef _LIBC
-	  /* We always want to use transliteration.  */
-	  outcharset = norm_add_slashes (outcharset, "TRANSLIT");
-	  charset = norm_add_slashes (charset, "");
-	  if (__gconv_open (outcharset, charset, &domain->conv,
-			    GCONV_AVOID_NOCONV)
-	      != __GCONV_OK)
-	    domain->conv = (__gconv_t) -1;
-# else
-#  if HAVE_ICONV
-	  /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
-	     we want to use transliteration.  */
-#   if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
-       || _LIBICONV_VERSION >= 0x0105
-	  if (strchr (outcharset, '/') == NULL)
-	    {
-	      char *tmp;
-
-	      len = strlen (outcharset);
-	      tmp = (char *) alloca (len + 10 + 1);
-	      memcpy (tmp, outcharset, len);
-	      memcpy (tmp + len, "//TRANSLIT", 10 + 1);
-	      outcharset = tmp;
-
-	      domain->conv = iconv_open (outcharset, charset);
-
-	      freea (outcharset);
-	    }
-	  else
-#   endif
-	    domain->conv = iconv_open (outcharset, charset);
-#  endif
-# endif
-
-	  freea (charset);
-	}
-#endif /* _LIBC || HAVE_ICONV */
-    }
-
-  return nullentry;
-}
-
-/* Frees the codeset dependent parts of an opened message catalog.  */
-void
-internal_function
-_nl_free_domain_conv (domain)
-     struct loaded_domain *domain;
-{
-  if (domain->conv_tab != NULL && domain->conv_tab != (char **) -1)
-    free (domain->conv_tab);
-
-#ifdef _LIBC
-  if (domain->conv != (__gconv_t) -1)
-    __gconv_close (domain->conv);
-#else
-# if HAVE_ICONV
-  if (domain->conv != (iconv_t) -1)
-    iconv_close (domain->conv);
-# endif
-#endif
-}
-
 /* Load the message catalogs specified by FILENAME.  If it is no valid
    message catalog do nothing.  */
 void
@@ -913,6 +773,7 @@ _nl_load_domain (domain_file, domainbinding)
   struct loaded_domain *domain;
   int revision;
   const char *nullentry;
+  size_t nullentrylen;
 
   __libc_lock_lock_recursive (lock);
   if (domain_file->decided != 0)
@@ -920,8 +781,8 @@ _nl_load_domain (domain_file, domainbinding)
       /* There are two possibilities:
 
          + is is the same thread calling again during this
-	   initialization via _nl_init_domain_conv and _nl_find_msg.  We
-	   have initialized everything this call needs.
+	   initialization via _nl_find_msg.  We have initialized
+	   everything this call needs.
 
 	 + this is another thread which tried to initialize this object.
            Not necessary anymore since if the lock is available this
@@ -1388,12 +1249,12 @@ _nl_load_domain (domain_file, domainbinding)
       goto out;
     }
 
-  /* Now initialize the character set converter from the character set
-     the file is encoded with (found in the header entry) to the domain's
-     specified character set or the locale's character set.  */
-  nullentry = _nl_init_domain_conv (domain_file, domain, domainbinding);
+  /* No caches of converted translations so far.  */
+  domain->conversions = NULL;
+  domain->nconversions = 0;
 
-  /* Also look for a plural specification.  */
+  /* Get the header entry and look for a plural specification.  */
+  nullentry = _nl_find_msg (domain_file, domainbinding, "", 0, &nullentrylen);
   EXTRACT_PLURAL_EXPRESSION (nullentry, &domain->plural, &domain->nplurals);
 
  out:
@@ -1412,10 +1273,23 @@ internal_function __libc_freeres_fn_section
 _nl_unload_domain (domain)
      struct loaded_domain *domain;
 {
+  size_t i;
+
   if (domain->plural != &__gettext_germanic_plural)
     __gettext_free_exp (domain->plural);
 
-  _nl_free_domain_conv (domain);
+  for (i = 0; i < domain->nconversions; i++)
+    {
+      struct converted_domain *convd = &domain->conversions[i];
+
+      free ((char *) convd->encoding);
+      if (convd->conv_tab != NULL && convd->conv_tab != (char **) -1)
+	free (convd->conv_tab);
+      if (convd->conv != (__gconv_t) -1)
+	__gconv_close (convd->conv);
+    }
+  if (domain->conversions != NULL)
+    free (domain->conversions);
 
   if (domain->malloced)
     free (domain->malloced);
diff --git a/intl/tst-gettext3.c b/intl/tst-gettext3.c
new file mode 100644
index 0000000000..917967b383
--- /dev/null
+++ b/intl/tst-gettext3.c
@@ -0,0 +1,60 @@
+/* Test that the gettext() results come out in the correct encoding for
+   locales that differ only in their encoding.
+   Copyright (C) 2001, 2005 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+   Contributed by Bruno Haible <bruno@clisp.org>, 2001, 2005.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library 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
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <libintl.h>
+#include <locale.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+int
+main (void)
+{
+  char *s;
+  int result = 0;
+
+  unsetenv ("LANGUAGE");
+  unsetenv ("OUTPUT_CHARSET");
+  textdomain ("codeset");
+  bindtextdomain ("codeset", OBJPFX "domaindir");
+
+  setlocale (LC_ALL, "de_DE.ISO-8859-1");
+
+  /* Here we expect output in ISO-8859-1.  */
+  s = gettext ("cheese");
+  if (strcmp (s, "K\344se"))
+    {
+      printf ("call 1 returned: %s\n", s);
+      result = 1;
+    }
+
+  setlocale (LC_ALL, "de_DE.UTF-8");
+
+  /* Here we expect output in UTF-8.  */
+  s = gettext ("cheese");
+  if (strcmp (s, "K\303\244se"))
+    {
+      printf ("call 2 returned: %s\n", s);
+      result = 1;
+    }
+
+  return result;
+}
diff --git a/intl/tst-gettext3.sh b/intl/tst-gettext3.sh
new file mode 100644
index 0000000000..7bd977cf2d
--- /dev/null
+++ b/intl/tst-gettext3.sh
@@ -0,0 +1,45 @@
+#! /bin/sh
+# Test that the gettext() results come out in the correct encoding for
+# locales that differ only in their encoding.
+# Copyright (C) 2001, 2002, 2005 Free Software Foundation, Inc.
+# This file is part of the GNU C Library.
+#
+
+# The GNU C Library is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+
+# The GNU C Library 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
+# Lesser General Public License for more details.
+
+# You should have received a copy of the GNU Lesser General Public
+# License along with the GNU C Library; if not, write to the Free
+# Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+# 02111-1307 USA.
+
+common_objpfx=$1
+objpfx=$2
+
+LC_ALL=C
+export LC_ALL
+
+# Generate the test data.
+test -d ${objpfx}domaindir || mkdir ${objpfx}domaindir
+# Create the domain directories.
+test -d ${objpfx}domaindir/de_DE || mkdir ${objpfx}domaindir/de_DE
+test -d ${objpfx}domaindir/de_DE/LC_MESSAGES || mkdir ${objpfx}domaindir/de_DE/LC_MESSAGES
+# Populate them.
+msgfmt -o ${objpfx}domaindir/de_DE/LC_MESSAGES/codeset.mo tstcodeset.po
+
+GCONV_PATH=${common_objpfx}iconvdata
+export GCONV_PATH
+LOCPATH=${common_objpfx}localedata
+export LOCPATH
+
+${common_objpfx}elf/ld.so --library-path $common_objpfx \
+${objpfx}tst-gettext3 > ${objpfx}tst-gettext3.out
+
+exit $?