Use O_CLOEXEC in tzfile handling
[glibc.git] / catgets / gencat.c
blob2a48122fe8deb27ff7f942f5bb8953a70fa0ccc7
1 /* Copyright (C) 1996-2008, 2009, 2010, 2011 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
3 Contributed by Ulrich Drepper <drepper@redhat.com>, 1996.
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published
7 by the Free Software Foundation; version 2 of the License, or
8 (at your option) any later version.
10 This program is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 GNU General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software Foundation,
17 Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
19 #ifdef HAVE_CONFIG_H
20 # include "config.h"
21 #endif
23 #include <argp.h>
24 #include <assert.h>
25 #include <ctype.h>
26 #include <endian.h>
27 #include <errno.h>
28 #include <error.h>
29 #include <fcntl.h>
30 #include <iconv.h>
31 #include <langinfo.h>
32 #include <locale.h>
33 #include <libintl.h>
34 #include <limits.h>
35 #include <nl_types.h>
36 #include <obstack.h>
37 #include <stdint.h>
38 #include <stdio.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <unistd.h>
42 #include <wchar.h>
44 #include "version.h"
46 #include "catgetsinfo.h"
49 #define SWAPU32(w) \
50 (((w) << 24) | (((w) & 0xff00) << 8) | (((w) >> 8) & 0xff00) | ((w) >> 24))
52 struct message_list
54 int number;
55 const char *message;
57 const char *fname;
58 size_t line;
59 const char *symbol;
61 struct message_list *next;
65 struct set_list
67 int number;
68 int deleted;
69 struct message_list *messages;
70 int last_message;
72 const char *fname;
73 size_t line;
74 const char *symbol;
76 struct set_list *next;
80 struct catalog
82 struct set_list *all_sets;
83 struct set_list *current_set;
84 size_t total_messages;
85 wint_t quote_char;
86 int last_set;
88 struct obstack mem_pool;
92 /* If non-zero force creation of new file, not using existing one. */
93 static int force_new;
95 /* Name of output file. */
96 static const char *output_name;
98 /* Name of generated C header file. */
99 static const char *header_name;
101 /* Name and version of program. */
102 static void print_version (FILE *stream, struct argp_state *state);
103 void (*argp_program_version_hook) (FILE *, struct argp_state *) = print_version;
105 #define OPT_NEW 1
107 /* Definitions of arguments for argp functions. */
108 static const struct argp_option options[] =
110 { "header", 'H', N_("NAME"), 0,
111 N_("Create C header file NAME containing symbol definitions") },
112 { "new", OPT_NEW, NULL, 0,
113 N_("Do not use existing catalog, force new output file") },
114 { "output", 'o', N_("NAME"), 0, N_("Write output to file NAME") },
115 { NULL, 0, NULL, 0, NULL }
118 /* Short description of program. */
119 static const char doc[] = N_("Generate message catalog.\
120 \vIf INPUT-FILE is -, input is read from standard input. If OUTPUT-FILE\n\
121 is -, output is written to standard output.\n");
123 /* Strings for arguments in help texts. */
124 static const char args_doc[] = N_("\
125 -o OUTPUT-FILE [INPUT-FILE]...\n[OUTPUT-FILE [INPUT-FILE]...]");
127 /* Prototype for option handler. */
128 static error_t parse_opt (int key, char *arg, struct argp_state *state);
130 /* Function to print some extra text in the help message. */
131 static char *more_help (int key, const char *text, void *input);
133 /* Data structure to communicate with argp functions. */
134 static struct argp argp =
136 options, parse_opt, args_doc, doc, NULL, more_help
140 /* Wrapper functions with error checking for standard functions. */
141 extern void *xmalloc (size_t n);
142 extern void *xcalloc (size_t n, size_t s);
143 extern void *xrealloc (void *o, size_t n);
144 extern char *xstrdup (const char *);
146 /* Prototypes for local functions. */
147 static void error_print (void);
148 static struct catalog *read_input_file (struct catalog *current,
149 const char *fname);
150 static void write_out (struct catalog *result, const char *output_name,
151 const char *header_name);
152 static struct set_list *find_set (struct catalog *current, int number);
153 static void normalize_line (const char *fname, size_t line, iconv_t cd,
154 wchar_t *string, wchar_t quote_char,
155 wchar_t escape_char);
156 static void read_old (struct catalog *catalog, const char *file_name);
157 static int open_conversion (const char *codesetp, iconv_t *cd_towcp,
158 iconv_t *cd_tombp, wchar_t *escape_charp);
162 main (int argc, char *argv[])
164 struct catalog *result;
165 int remaining;
167 /* Set program name for messages. */
168 error_print_progname = error_print;
170 /* Set locale via LC_ALL. */
171 setlocale (LC_ALL, "");
173 /* Set the text message domain. */
174 textdomain (PACKAGE);
176 /* Initialize local variables. */
177 result = NULL;
179 /* Parse and process arguments. */
180 argp_parse (&argp, argc, argv, 0, &remaining, NULL);
182 /* Determine output file. */
183 if (output_name == NULL)
184 output_name = remaining < argc ? argv[remaining++] : "-";
186 /* Process all input files. */
187 setlocale (LC_CTYPE, "C");
188 if (remaining < argc)
190 result = read_input_file (result, argv[remaining]);
191 while (++remaining < argc);
192 else
193 result = read_input_file (NULL, "-");
195 /* Write out the result. */
196 if (result != NULL)
197 write_out (result, output_name, header_name);
199 return error_message_count != 0;
203 /* Handle program arguments. */
204 static error_t
205 parse_opt (int key, char *arg, struct argp_state *state)
207 switch (key)
209 case 'H':
210 header_name = arg;
211 break;
212 case OPT_NEW:
213 force_new = 1;
214 break;
215 case 'o':
216 output_name = arg;
217 break;
218 default:
219 return ARGP_ERR_UNKNOWN;
221 return 0;
225 static char *
226 more_help (int key, const char *text, void *input)
228 switch (key)
230 case ARGP_KEY_HELP_EXTRA:
231 /* We print some extra information. */
232 return strdup (gettext ("\
233 For bug reporting instructions, please see:\n\
234 <http://www.gnu.org/software/libc/bugs.html>.\n"));
235 default:
236 break;
238 return (char *) text;
241 /* Print the version information. */
242 static void
243 print_version (FILE *stream, struct argp_state *state)
245 fprintf (stream, "gencat (GNU %s) %s\n", PACKAGE, VERSION);
246 fprintf (stream, gettext ("\
247 Copyright (C) %s Free Software Foundation, Inc.\n\
248 This is free software; see the source for copying conditions. There is NO\n\
249 warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.\n\
250 "), "2011");
251 fprintf (stream, gettext ("Written by %s.\n"), "Ulrich Drepper");
255 /* The address of this function will be assigned to the hook in the
256 error functions. */
257 static void
258 error_print ()
260 /* We don't want the program name to be printed in messages. Emacs'
261 compile.el does not like this. */
265 static struct catalog *
266 read_input_file (struct catalog *current, const char *fname)
268 FILE *fp;
269 char *buf;
270 size_t len;
271 size_t line_number;
272 wchar_t *wbuf;
273 size_t wbufsize;
274 iconv_t cd_towc = (iconv_t) -1;
275 iconv_t cd_tomb = (iconv_t) -1;
276 wchar_t escape_char = L'\\';
277 char *codeset = NULL;
279 if (strcmp (fname, "-") == 0 || strcmp (fname, "/dev/stdin") == 0)
281 fp = stdin;
282 fname = gettext ("*standard input*");
284 else
285 fp = fopen (fname, "r");
286 if (fp == NULL)
288 error (0, errno, gettext ("cannot open input file `%s'"), fname);
289 return current;
292 /* If we haven't seen anything yet, allocate result structure. */
293 if (current == NULL)
295 current = (struct catalog *) xcalloc (1, sizeof (*current));
297 #define obstack_chunk_alloc malloc
298 #define obstack_chunk_free free
299 obstack_init (&current->mem_pool);
301 current->current_set = find_set (current, NL_SETD);
304 buf = NULL;
305 len = 0;
306 line_number = 0;
308 wbufsize = 1024;
309 wbuf = (wchar_t *) xmalloc (wbufsize);
311 while (!feof (fp))
313 int continued;
314 int used;
315 size_t start_line = line_number + 1;
316 char *this_line;
320 int act_len;
322 act_len = getline (&buf, &len, fp);
323 if (act_len <= 0)
324 break;
325 ++line_number;
327 /* It the line continued? */
328 continued = 0;
329 if (buf[act_len - 1] == '\n')
331 --act_len;
333 /* There might be more than one backslash at the end of
334 the line. Only if there is an odd number of them is
335 the line continued. */
336 if (act_len > 0 && buf[act_len - 1] == '\\')
338 int temp_act_len = act_len;
342 --temp_act_len;
343 continued = !continued;
345 while (temp_act_len > 0 && buf[temp_act_len - 1] == '\\');
347 if (continued)
348 --act_len;
352 /* Append to currently selected line. */
353 obstack_grow (&current->mem_pool, buf, act_len);
355 while (continued);
357 obstack_1grow (&current->mem_pool, '\0');
358 this_line = (char *) obstack_finish (&current->mem_pool);
360 used = 0;
361 if (this_line[0] == '$')
363 if (isblank (this_line[1]))
365 int cnt = 1;
366 while (isblank (this_line[cnt]))
367 ++cnt;
368 if (strncmp (&this_line[cnt], "codeset=", 8) != 0)
369 /* This is a comment line. Do nothing. */;
370 else if (codeset != NULL)
371 /* Ignore multiple codeset. */;
372 else
374 int start = cnt + 8;
375 cnt = start;
376 while (this_line[cnt] != '\0' && !isspace (this_line[cnt]))
377 ++cnt;
378 if (cnt != start)
380 int len = cnt - start;
381 codeset = xmalloc (len + 1);
382 *((char *) mempcpy (codeset, &this_line[start], len))
383 = '\0';
387 else if (strncmp (&this_line[1], "set", 3) == 0)
389 int cnt = sizeof ("set");
390 int set_number;
391 const char *symbol = NULL;
392 while (isspace (this_line[cnt]))
393 ++cnt;
395 if (isdigit (this_line[cnt]))
397 set_number = atol (&this_line[cnt]);
399 /* If the given number for the character set is
400 higher than any we used for symbolic set names
401 avoid clashing by using only higher numbers for
402 the following symbolic definitions. */
403 if (set_number > current->last_set)
404 current->last_set = set_number;
406 else
408 /* See whether it is a reasonable identifier. */
409 int start = cnt;
410 while (isalnum (this_line[cnt]) || this_line[cnt] == '_')
411 ++cnt;
413 if (cnt == start)
415 /* No correct character found. */
416 error_at_line (0, 0, fname, start_line,
417 gettext ("illegal set number"));
418 set_number = 0;
420 else
422 /* We have found seomthing that looks like a
423 correct identifier. */
424 struct set_list *runp;
426 this_line[cnt] = '\0';
427 used = 1;
428 symbol = &this_line[start];
430 /* Test whether the identifier was already used. */
431 runp = current->all_sets;
432 while (runp != 0)
433 if (runp->symbol != NULL
434 && strcmp (runp->symbol, symbol) == 0)
435 break;
436 else
437 runp = runp->next;
439 if (runp != NULL)
441 /* We cannot allow duplicate identifiers for
442 message sets. */
443 error_at_line (0, 0, fname, start_line,
444 gettext ("duplicate set definition"));
445 error_at_line (0, 0, runp->fname, runp->line,
446 gettext ("\
447 this is the first definition"));
448 set_number = 0;
450 else
451 /* Allocate next free message set for identifier. */
452 set_number = ++current->last_set;
456 if (set_number != 0)
458 /* We found a legal set number. */
459 current->current_set = find_set (current, set_number);
460 if (symbol != NULL)
461 used = 1;
462 current->current_set->symbol = symbol;
463 current->current_set->fname = fname;
464 current->current_set->line = start_line;
467 else if (strncmp (&this_line[1], "delset", 6) == 0)
469 int cnt = sizeof ("delset");
470 while (isspace (this_line[cnt]))
471 ++cnt;
473 if (isdigit (this_line[cnt]))
475 size_t set_number = atol (&this_line[cnt]);
476 struct set_list *set;
478 /* Mark the message set with the given number as
479 deleted. */
480 set = find_set (current, set_number);
481 set->deleted = 1;
483 else
485 /* See whether it is a reasonable identifier. */
486 int start = cnt;
487 while (isalnum (this_line[cnt]) || this_line[cnt] == '_')
488 ++cnt;
490 if (cnt == start)
491 error_at_line (0, 0, fname, start_line,
492 gettext ("illegal set number"));
493 else
495 const char *symbol;
496 struct set_list *runp;
498 this_line[cnt] = '\0';
499 used = 1;
500 symbol = &this_line[start];
502 /* We have a symbolic set name. This name must
503 appear somewhere else in the catalogs read so
504 far. */
505 for (runp = current->all_sets; runp != NULL;
506 runp = runp->next)
508 if (strcmp (runp->symbol, symbol) == 0)
510 runp->deleted = 1;
511 break;
514 if (runp == NULL)
515 /* Name does not exist before. */
516 error_at_line (0, 0, fname, start_line,
517 gettext ("unknown set `%s'"), symbol);
521 else if (strncmp (&this_line[1], "quote", 5) == 0)
523 char buf[2];
524 char *bufptr;
525 size_t buflen;
526 char *wbufptr;
527 size_t wbuflen;
528 int cnt;
530 cnt = sizeof ("quote");
531 while (isspace (this_line[cnt]))
532 ++cnt;
534 /* We need the conversion. */
535 if (cd_towc == (iconv_t) -1
536 && open_conversion (codeset, &cd_towc, &cd_tomb,
537 &escape_char) != 0)
538 /* Something is wrong. */
539 goto out;
541 /* Yes, the quote char can be '\0'; this means no quote
542 char. The function using the information works on
543 wide characters so we have to convert it here. */
544 buf[0] = this_line[cnt];
545 buf[1] = '\0';
546 bufptr = buf;
547 buflen = 2;
549 wbufptr = (char *) wbuf;
550 wbuflen = wbufsize;
552 /* Flush the state. */
553 iconv (cd_towc, NULL, NULL, NULL, NULL);
555 iconv (cd_towc, &bufptr, &buflen, &wbufptr, &wbuflen);
556 if (buflen != 0 || (wchar_t *) wbufptr != &wbuf[2])
557 error_at_line (0, 0, fname, start_line,
558 gettext ("invalid quote character"));
559 else
560 /* Use the converted wide character. */
561 current->quote_char = wbuf[0];
563 else
565 int cnt;
566 cnt = 2;
567 while (this_line[cnt] != '\0' && !isspace (this_line[cnt]))
568 ++cnt;
569 this_line[cnt] = '\0';
570 error_at_line (0, 0, fname, start_line,
571 gettext ("unknown directive `%s': line ignored"),
572 &this_line[1]);
575 else if (isalnum (this_line[0]) || this_line[0] == '_')
577 const char *ident = this_line;
578 char *line = this_line;
579 int message_number;
582 ++line;
583 while (line[0] != '\0' && !isspace (line[0]));
584 if (line[0] != '\0')
585 *line++ = '\0'; /* Terminate the identifier. */
587 /* Now we found the beginning of the message itself. */
589 if (isdigit (ident[0]))
591 struct message_list *runp;
592 struct message_list *lastp;
594 message_number = atoi (ident);
596 /* Find location to insert the new message. */
597 runp = current->current_set->messages;
598 lastp = NULL;
599 while (runp != NULL)
600 if (runp->number == message_number)
601 break;
602 else
604 lastp = runp;
605 runp = runp->next;
607 if (runp != NULL)
609 /* Oh, oh. There is already a message with this
610 number in the message set. */
611 if (runp->symbol == NULL)
613 /* The existing message had its number specified
614 by the user. Fatal collision type uh, oh. */
615 error_at_line (0, 0, fname, start_line,
616 gettext ("duplicated message number"));
617 error_at_line (0, 0, runp->fname, runp->line,
618 gettext ("this is the first definition"));
619 message_number = 0;
621 else
623 /* Collision was with number auto-assigned to a
624 symbolic. Change existing symbolic number
625 and move to end the list (if not already there). */
626 runp->number = ++current->current_set->last_message;
628 if (runp->next != NULL)
630 struct message_list *endp;
632 if (lastp == NULL)
633 current->current_set->messages=runp->next;
634 else
635 lastp->next=runp->next;
637 endp = runp->next;
638 while (endp->next != NULL)
639 endp = endp->next;
641 endp->next = runp;
642 runp->next = NULL;
646 ident = NULL; /* We don't have a symbol. */
648 if (message_number != 0
649 && message_number > current->current_set->last_message)
650 current->current_set->last_message = message_number;
652 else if (ident[0] != '\0')
654 struct message_list *runp;
656 /* Test whether the symbolic name was not used for
657 another message in this message set. */
658 runp = current->current_set->messages;
659 while (runp != NULL)
660 if (runp->symbol != NULL && strcmp (ident, runp->symbol) == 0)
661 break;
662 else
663 runp = runp->next;
664 if (runp != NULL)
666 /* The name is already used. */
667 error_at_line (0, 0, fname, start_line, gettext ("\
668 duplicated message identifier"));
669 error_at_line (0, 0, runp->fname, runp->line,
670 gettext ("this is the first definition"));
671 message_number = 0;
673 else
674 /* Give the message the next unused number. */
675 message_number = ++current->current_set->last_message;
677 else
678 message_number = 0;
680 if (message_number != 0)
682 char *inbuf;
683 size_t inlen;
684 char *outbuf;
685 size_t outlen;
686 struct message_list *newp;
687 size_t line_len = strlen (line) + 1;
688 size_t ident_len = 0;
690 /* We need the conversion. */
691 if (cd_towc == (iconv_t) -1
692 && open_conversion (codeset, &cd_towc, &cd_tomb,
693 &escape_char) != 0)
694 /* Something is wrong. */
695 goto out;
697 /* Convert to a wide character string. We have to
698 interpret escape sequences which will be impossible
699 without doing the conversion if the codeset of the
700 message is stateful. */
701 while (1)
703 inbuf = line;
704 inlen = line_len;
705 outbuf = (char *) wbuf;
706 outlen = wbufsize;
708 /* Flush the state. */
709 iconv (cd_towc, NULL, NULL, NULL, NULL);
711 iconv (cd_towc, &inbuf, &inlen, &outbuf, &outlen);
712 if (inlen == 0)
714 /* The string is converted. */
715 assert (outlen < wbufsize);
716 assert (wbuf[(wbufsize - outlen) / sizeof (wchar_t) - 1]
717 == L'\0');
718 break;
721 if (outlen != 0)
723 /* Something is wrong with this string, we ignore it. */
724 error_at_line (0, 0, fname, start_line, gettext ("\
725 invalid character: message ignored"));
726 goto ignore;
729 /* The output buffer is too small. */
730 wbufsize *= 2;
731 wbuf = (wchar_t *) xrealloc (wbuf, wbufsize);
734 /* Strip quote characters, change escape sequences into
735 correct characters etc. */
736 normalize_line (fname, start_line, cd_towc, wbuf,
737 current->quote_char, escape_char);
739 if (ident)
740 ident_len = line - this_line;
742 /* Now the string is free of escape sequences. Convert it
743 back into a multibyte character string. First free the
744 memory allocated for the original string. */
745 obstack_free (&current->mem_pool, this_line);
747 used = 1; /* Yes, we use the line. */
749 /* Now fill in the new string. It should never happen that
750 the replaced string is longer than the original. */
751 inbuf = (char *) wbuf;
752 inlen = (wcslen (wbuf) + 1) * sizeof (wchar_t);
754 outlen = obstack_room (&current->mem_pool);
755 obstack_blank (&current->mem_pool, outlen);
756 this_line = (char *) obstack_base (&current->mem_pool);
757 outbuf = this_line + ident_len;
758 outlen -= ident_len;
760 /* Flush the state. */
761 iconv (cd_tomb, NULL, NULL, NULL, NULL);
763 iconv (cd_tomb, &inbuf, &inlen, &outbuf, &outlen);
764 if (inlen != 0)
766 error_at_line (0, 0, fname, start_line,
767 gettext ("invalid line"));
768 goto ignore;
770 assert (outbuf[-1] == '\0');
772 /* Free the memory in the obstack we don't use. */
773 obstack_blank (&current->mem_pool, -(int) outlen);
774 line = obstack_finish (&current->mem_pool);
776 newp = (struct message_list *) xmalloc (sizeof (*newp));
777 newp->number = message_number;
778 newp->message = line + ident_len;
779 /* Remember symbolic name; is NULL if no is given. */
780 newp->symbol = ident ? line : NULL;
781 /* Remember where we found the character. */
782 newp->fname = fname;
783 newp->line = start_line;
785 /* Find place to insert to message. We keep them in a
786 sorted single linked list. */
787 if (current->current_set->messages == NULL
788 || current->current_set->messages->number > message_number)
790 newp->next = current->current_set->messages;
791 current->current_set->messages = newp;
793 else
795 struct message_list *runp;
796 runp = current->current_set->messages;
797 while (runp->next != NULL)
798 if (runp->next->number > message_number)
799 break;
800 else
801 runp = runp->next;
802 newp->next = runp->next;
803 runp->next = newp;
806 ++current->total_messages;
808 else
810 size_t cnt;
812 cnt = 0;
813 /* See whether we have any non-white space character in this
814 line. */
815 while (this_line[cnt] != '\0' && isspace (this_line[cnt]))
816 ++cnt;
818 if (this_line[cnt] != '\0')
819 /* Yes, some unknown characters found. */
820 error_at_line (0, 0, fname, start_line,
821 gettext ("malformed line ignored"));
824 ignore:
825 /* We can save the memory for the line if it was not used. */
826 if (!used)
827 obstack_free (&current->mem_pool, this_line);
830 /* Close the conversion modules. */
831 iconv_close (cd_towc);
832 iconv_close (cd_tomb);
833 free (codeset);
835 out:
836 free (wbuf);
838 if (fp != stdin)
839 fclose (fp);
840 return current;
844 static void
845 write_out (struct catalog *catalog, const char *output_name,
846 const char *header_name)
848 /* Computing the "optimal" size. */
849 struct set_list *set_run;
850 size_t best_total, best_size, best_depth;
851 size_t act_size, act_depth;
852 struct catalog_obj obj;
853 struct obstack string_pool;
854 const char *strings;
855 size_t strings_size;
856 uint32_t *array1, *array2;
857 size_t cnt;
858 int fd;
860 /* If not otherwise told try to read file with existing
861 translations. */
862 if (!force_new)
863 read_old (catalog, output_name);
865 /* Initialize best_size with a very high value. */
866 best_total = best_size = best_depth = UINT_MAX;
868 /* We need some start size for testing. Let's start with
869 TOTAL_MESSAGES / 5, which theoretically provides a mean depth of
870 5. */
871 act_size = 1 + catalog->total_messages / 5;
873 /* We determine the size of a hash table here. Because the message
874 numbers can be chosen arbitrary by the programmer we cannot use
875 the simple method of accessing the array using the message
876 number. The algorithm is based on the trivial hash function
877 NUMBER % TABLE_SIZE, where collisions are stored in a second
878 dimension up to TABLE_DEPTH. We here compute TABLE_SIZE so that
879 the needed space (= TABLE_SIZE * TABLE_DEPTH) is minimal. */
880 while (act_size <= best_total)
882 size_t deep[act_size];
884 act_depth = 1;
885 memset (deep, '\0', act_size * sizeof (size_t));
886 set_run = catalog->all_sets;
887 while (set_run != NULL)
889 struct message_list *message_run;
891 message_run = set_run->messages;
892 while (message_run != NULL)
894 size_t idx = (message_run->number * set_run->number) % act_size;
896 ++deep[idx];
897 if (deep[idx] > act_depth)
899 act_depth = deep[idx];
900 if (act_depth * act_size > best_total)
901 break;
903 message_run = message_run->next;
905 set_run = set_run->next;
908 if (act_depth * act_size <= best_total)
910 /* We have found a better solution. */
911 best_total = act_depth * act_size;
912 best_size = act_size;
913 best_depth = act_depth;
916 ++act_size;
919 /* let's be prepared for an empty message file. */
920 if (best_size == UINT_MAX)
922 best_size = 1;
923 best_depth = 1;
926 /* OK, now we have the size we will use. Fill in the header, build
927 the table and the second one with swapped byte order. */
928 obj.magic = CATGETS_MAGIC;
929 obj.plane_size = best_size;
930 obj.plane_depth = best_depth;
932 /* Allocate room for all needed arrays. */
933 array1 =
934 (uint32_t *) alloca (best_size * best_depth * sizeof (uint32_t) * 3);
935 memset (array1, '\0', best_size * best_depth * sizeof (uint32_t) * 3);
936 array2
937 = (uint32_t *) alloca (best_size * best_depth * sizeof (uint32_t) * 3);
938 obstack_init (&string_pool);
940 set_run = catalog->all_sets;
941 while (set_run != NULL)
943 struct message_list *message_run;
945 message_run = set_run->messages;
946 while (message_run != NULL)
948 size_t idx = (((message_run->number * set_run->number) % best_size)
949 * 3);
950 /* Determine collision depth. */
951 while (array1[idx] != 0)
952 idx += best_size * 3;
954 /* Store set number, message number and pointer into string
955 space, relative to the first string. */
956 array1[idx + 0] = set_run->number;
957 array1[idx + 1] = message_run->number;
958 array1[idx + 2] = obstack_object_size (&string_pool);
960 /* Add current string to the continuous space containing all
961 strings. */
962 obstack_grow0 (&string_pool, message_run->message,
963 strlen (message_run->message));
965 message_run = message_run->next;
968 set_run = set_run->next;
970 strings_size = obstack_object_size (&string_pool);
971 strings = obstack_finish (&string_pool);
973 /* Compute ARRAY2 by changing the byte order. */
974 for (cnt = 0; cnt < best_size * best_depth * 3; ++cnt)
975 array2[cnt] = SWAPU32 (array1[cnt]);
977 /* Now we can write out the whole data. */
978 if (strcmp (output_name, "-") == 0
979 || strcmp (output_name, "/dev/stdout") == 0)
980 fd = STDOUT_FILENO;
981 else
983 fd = creat (output_name, 0666);
984 if (fd < 0)
985 error (EXIT_FAILURE, errno, gettext ("cannot open output file `%s'"),
986 output_name);
989 /* Write out header. */
990 write (fd, &obj, sizeof (obj));
992 /* We always write out the little endian version of the index
993 arrays. */
994 #if __BYTE_ORDER == __LITTLE_ENDIAN
995 write (fd, array1, best_size * best_depth * sizeof (uint32_t) * 3);
996 write (fd, array2, best_size * best_depth * sizeof (uint32_t) * 3);
997 #elif __BYTE_ORDER == __BIG_ENDIAN
998 write (fd, array2, best_size * best_depth * sizeof (uint32_t) * 3);
999 write (fd, array1, best_size * best_depth * sizeof (uint32_t) * 3);
1000 #else
1001 # error Cannot handle __BYTE_ORDER byte order
1002 #endif
1004 /* Finally write the strings. */
1005 write (fd, strings, strings_size);
1007 if (fd != STDOUT_FILENO)
1008 close (fd);
1010 /* If requested now write out the header file. */
1011 if (header_name != NULL)
1013 int first = 1;
1014 FILE *fp;
1016 /* Open output file. "-" or "/dev/stdout" means write to
1017 standard output. */
1018 if (strcmp (header_name, "-") == 0
1019 || strcmp (header_name, "/dev/stdout") == 0)
1020 fp = stdout;
1021 else
1023 fp = fopen (header_name, "w");
1024 if (fp == NULL)
1025 error (EXIT_FAILURE, errno,
1026 gettext ("cannot open output file `%s'"), header_name);
1029 /* Iterate over all sets and all messages. */
1030 set_run = catalog->all_sets;
1031 while (set_run != NULL)
1033 struct message_list *message_run;
1035 /* If the current message set has a symbolic name write this
1036 out first. */
1037 if (set_run->symbol != NULL)
1038 fprintf (fp, "%s#define %sSet %#x\t/* %s:%Zu */\n",
1039 first ? "" : "\n", set_run->symbol, set_run->number - 1,
1040 set_run->fname, set_run->line);
1041 first = 0;
1043 message_run = set_run->messages;
1044 while (message_run != NULL)
1046 /* If the current message has a symbolic name write
1047 #define out. But we have to take care for the set
1048 not having a symbolic name. */
1049 if (message_run->symbol != NULL)
1051 if (set_run->symbol == NULL)
1052 fprintf (fp, "#define AutomaticSet%d%s %#x\t/* %s:%Zu */\n",
1053 set_run->number, message_run->symbol,
1054 message_run->number, message_run->fname,
1055 message_run->line);
1056 else
1057 fprintf (fp, "#define %s%s %#x\t/* %s:%Zu */\n",
1058 set_run->symbol, message_run->symbol,
1059 message_run->number, message_run->fname,
1060 message_run->line);
1063 message_run = message_run->next;
1066 set_run = set_run->next;
1069 if (fp != stdout)
1070 fclose (fp);
1075 static struct set_list *
1076 find_set (struct catalog *current, int number)
1078 struct set_list *result = current->all_sets;
1080 /* We must avoid set number 0 because a set of this number signals
1081 in the tables that the entry is not occupied. */
1082 ++number;
1084 while (result != NULL)
1085 if (result->number == number)
1086 return result;
1087 else
1088 result = result->next;
1090 /* Prepare new message set. */
1091 result = (struct set_list *) xcalloc (1, sizeof (*result));
1092 result->number = number;
1093 result->next = current->all_sets;
1094 current->all_sets = result;
1096 return result;
1100 /* Normalize given string *in*place* by processing escape sequences
1101 and quote characters. */
1102 static void
1103 normalize_line (const char *fname, size_t line, iconv_t cd, wchar_t *string,
1104 wchar_t quote_char, wchar_t escape_char)
1106 int is_quoted;
1107 wchar_t *rp = string;
1108 wchar_t *wp = string;
1110 if (quote_char != L'\0' && *rp == quote_char)
1112 is_quoted = 1;
1113 ++rp;
1115 else
1116 is_quoted = 0;
1118 while (*rp != L'\0')
1119 if (*rp == quote_char)
1120 /* We simply end the string when we find the first time an
1121 not-escaped quote character. */
1122 break;
1123 else if (*rp == escape_char)
1125 ++rp;
1126 if (quote_char != L'\0' && *rp == quote_char)
1127 /* This is an extension to XPG. */
1128 *wp++ = *rp++;
1129 else
1130 /* Recognize escape sequences. */
1131 switch (*rp)
1133 case L'n':
1134 *wp++ = L'\n';
1135 ++rp;
1136 break;
1137 case L't':
1138 *wp++ = L'\t';
1139 ++rp;
1140 break;
1141 case L'v':
1142 *wp++ = L'\v';
1143 ++rp;
1144 break;
1145 case L'b':
1146 *wp++ = L'\b';
1147 ++rp;
1148 break;
1149 case L'r':
1150 *wp++ = L'\r';
1151 ++rp;
1152 break;
1153 case L'f':
1154 *wp++ = L'\f';
1155 ++rp;
1156 break;
1157 case L'0' ... L'7':
1159 int number;
1160 char cbuf[2];
1161 char *cbufptr;
1162 size_t cbufin;
1163 wchar_t wcbuf[2];
1164 char *wcbufptr;
1165 size_t wcbufin;
1167 number = *rp++ - L'0';
1168 while (number <= (255 / 8) && *rp >= L'0' && *rp <= L'7')
1170 number *= 8;
1171 number += *rp++ - L'0';
1174 cbuf[0] = (char) number;
1175 cbuf[1] = '\0';
1176 cbufptr = cbuf;
1177 cbufin = 2;
1179 wcbufptr = (char *) wcbuf;
1180 wcbufin = sizeof (wcbuf);
1182 /* Flush the state. */
1183 iconv (cd, NULL, NULL, NULL, NULL);
1185 iconv (cd, &cbufptr, &cbufin, &wcbufptr, &wcbufin);
1186 if (cbufptr != &cbuf[2] || (wchar_t *) wcbufptr != &wcbuf[2])
1187 error_at_line (0, 0, fname, line,
1188 gettext ("invalid escape sequence"));
1189 else
1190 *wp++ = wcbuf[0];
1192 break;
1193 default:
1194 if (*rp == escape_char)
1196 *wp++ = escape_char;
1197 ++rp;
1199 else
1200 /* Simply ignore the backslash character. */;
1201 break;
1204 else
1205 *wp++ = *rp++;
1207 /* If we saw a quote character at the beginning we expect another
1208 one at the end. */
1209 if (is_quoted && *rp != quote_char)
1210 error_at_line (0, 0, fname, line, gettext ("unterminated message"));
1212 /* Terminate string. */
1213 *wp = L'\0';
1214 return;
1218 static void
1219 read_old (struct catalog *catalog, const char *file_name)
1221 struct catalog_info old_cat_obj;
1222 struct set_list *set = NULL;
1223 int last_set = -1;
1224 size_t cnt;
1226 /* Try to open catalog, but don't look through the NLSPATH. */
1227 if (__open_catalog (file_name, NULL, NULL, &old_cat_obj) != 0)
1229 if (errno == ENOENT)
1230 /* No problem, the catalog simply does not exist. */
1231 return;
1232 else
1233 error (EXIT_FAILURE, errno,
1234 gettext ("while opening old catalog file"));
1237 /* OK, we have the catalog loaded. Now read all messages and merge
1238 them. When set and message number clash for any message the new
1239 one is used. If the new one is empty it indicates that the
1240 message should be deleted. */
1241 for (cnt = 0; cnt < old_cat_obj.plane_size * old_cat_obj.plane_depth; ++cnt)
1243 struct message_list *message, *last;
1245 if (old_cat_obj.name_ptr[cnt * 3 + 0] == 0)
1246 /* No message in this slot. */
1247 continue;
1249 if (old_cat_obj.name_ptr[cnt * 3 + 0] - 1 != (uint32_t) last_set)
1251 last_set = old_cat_obj.name_ptr[cnt * 3 + 0] - 1;
1252 set = find_set (catalog, old_cat_obj.name_ptr[cnt * 3 + 0] - 1);
1255 last = NULL;
1256 message = set->messages;
1257 while (message != NULL)
1259 if ((uint32_t) message->number >= old_cat_obj.name_ptr[cnt * 3 + 1])
1260 break;
1261 last = message;
1262 message = message->next;
1265 if (message == NULL
1266 || (uint32_t) message->number > old_cat_obj.name_ptr[cnt * 3 + 1])
1268 /* We have found a message which is not yet in the catalog.
1269 Insert it at the right position. */
1270 struct message_list *newp;
1272 newp = (struct message_list *) xmalloc (sizeof(*newp));
1273 newp->number = old_cat_obj.name_ptr[cnt * 3 + 1];
1274 newp->message =
1275 &old_cat_obj.strings[old_cat_obj.name_ptr[cnt * 3 + 2]];
1276 newp->fname = NULL;
1277 newp->line = 0;
1278 newp->symbol = NULL;
1279 newp->next = message;
1281 if (last == NULL)
1282 set->messages = newp;
1283 else
1284 last->next = newp;
1286 ++catalog->total_messages;
1288 else if (*message->message == '\0')
1290 /* The new empty message has overridden the old one thus
1291 "deleting" it as required. Now remove the empty remains. */
1292 if (last == NULL)
1293 set->messages = message->next;
1294 else
1295 last->next = message->next;
1301 static int
1302 open_conversion (const char *codeset, iconv_t *cd_towcp, iconv_t *cd_tombp,
1303 wchar_t *escape_charp)
1305 char buf[2];
1306 char *bufptr;
1307 size_t bufsize;
1308 wchar_t wbuf[2];
1309 char *wbufptr;
1310 size_t wbufsize;
1312 /* If the input file does not specify the codeset use the locale's. */
1313 if (codeset == NULL)
1315 setlocale (LC_ALL, "");
1316 codeset = nl_langinfo (CODESET);
1317 setlocale (LC_ALL, "C");
1320 /* Get the conversion modules. */
1321 *cd_towcp = iconv_open ("WCHAR_T", codeset);
1322 *cd_tombp = iconv_open (codeset, "WCHAR_T");
1323 if (*cd_towcp == (iconv_t) -1 || *cd_tombp == (iconv_t) -1)
1325 error (0, 0, gettext ("conversion modules not available"));
1326 if (*cd_towcp != (iconv_t) -1)
1327 iconv_close (*cd_towcp);
1329 return 1;
1332 /* One special case for historical reasons is the backslash
1333 character. In some codesets the byte value 0x5c is not mapped to
1334 U005c in Unicode. These charsets then don't have a backslash
1335 character at all. Therefore we have to live with whatever the
1336 codeset provides and recognize, instead of the U005c, the character
1337 the byte value 0x5c is mapped to. */
1338 buf[0] = '\\';
1339 buf[1] = '\0';
1340 bufptr = buf;
1341 bufsize = 2;
1343 wbufptr = (char *) wbuf;
1344 wbufsize = sizeof (wbuf);
1346 iconv (*cd_towcp, &bufptr, &bufsize, &wbufptr, &wbufsize);
1347 if (bufsize != 0 || wbufsize != 0)
1349 /* Something went wrong, we couldn't convert the byte 0x5c. Go
1350 on with using U005c. */
1351 error (0, 0, gettext ("cannot determine escape character"));
1352 *escape_charp = L'\\';
1354 else
1355 *escape_charp = wbuf[0];
1357 return 0;