CodingGuidelines: our documents are in AsciiDoc
[git] / convert.c
1 #include "cache.h"
2 #include "attr.h"
3 #include "run-command.h"
4 #include "quote.h"
5 #include "sigchain.h"
6
7 /*
8  * convert.c - convert a file when checking it out and checking it in.
9  *
10  * This should use the pathname to decide on whether it wants to do some
11  * more interesting conversions (automatic gzip/unzip, general format
12  * conversions etc etc), but by default it just does automatic CRLF<->LF
13  * translation when the "text" attribute or "auto_crlf" option is set.
14  */
15
16 enum crlf_action {
17         CRLF_GUESS = -1,
18         CRLF_BINARY = 0,
19         CRLF_TEXT,
20         CRLF_INPUT,
21         CRLF_CRLF,
22         CRLF_AUTO
23 };
24
25 struct text_stat {
26         /* NUL, CR, LF and CRLF counts */
27         unsigned nul, cr, lf, crlf;
28
29         /* These are just approximations! */
30         unsigned printable, nonprintable;
31 };
32
33 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
34 {
35         unsigned long i;
36
37         memset(stats, 0, sizeof(*stats));
38
39         for (i = 0; i < size; i++) {
40                 unsigned char c = buf[i];
41                 if (c == '\r') {
42                         stats->cr++;
43                         if (i+1 < size && buf[i+1] == '\n')
44                                 stats->crlf++;
45                         continue;
46                 }
47                 if (c == '\n') {
48                         stats->lf++;
49                         continue;
50                 }
51                 if (c == 127)
52                         /* DEL */
53                         stats->nonprintable++;
54                 else if (c < 32) {
55                         switch (c) {
56                                 /* BS, HT, ESC and FF */
57                         case '\b': case '\t': case '\033': case '\014':
58                                 stats->printable++;
59                                 break;
60                         case 0:
61                                 stats->nul++;
62                                 /* fall through */
63                         default:
64                                 stats->nonprintable++;
65                         }
66                 }
67                 else
68                         stats->printable++;
69         }
70
71         /* If file ends with EOF then don't count this EOF as non-printable. */
72         if (size >= 1 && buf[size-1] == '\032')
73                 stats->nonprintable--;
74 }
75
76 /*
77  * The same heuristics as diff.c::mmfile_is_binary()
78  */
79 static int is_binary(unsigned long size, struct text_stat *stats)
80 {
81
82         if (stats->nul)
83                 return 1;
84         if ((stats->printable >> 7) < stats->nonprintable)
85                 return 1;
86         /*
87          * Other heuristics? Average line length might be relevant,
88          * as might LF vs CR vs CRLF counts..
89          *
90          * NOTE! It might be normal to have a low ratio of CRLF to LF
91          * (somebody starts with a LF-only file and edits it with an editor
92          * that adds CRLF only to lines that are added..). But do  we
93          * want to support CR-only? Probably not.
94          */
95         return 0;
96 }
97
98 static enum eol output_eol(enum crlf_action crlf_action)
99 {
100         switch (crlf_action) {
101         case CRLF_BINARY:
102                 return EOL_UNSET;
103         case CRLF_CRLF:
104                 return EOL_CRLF;
105         case CRLF_INPUT:
106                 return EOL_LF;
107         case CRLF_GUESS:
108                 if (!auto_crlf)
109                         return EOL_UNSET;
110                 /* fall through */
111         case CRLF_TEXT:
112         case CRLF_AUTO:
113                 if (auto_crlf == AUTO_CRLF_TRUE)
114                         return EOL_CRLF;
115                 else if (auto_crlf == AUTO_CRLF_INPUT)
116                         return EOL_LF;
117                 else if (core_eol == EOL_UNSET)
118                         return EOL_NATIVE;
119         }
120         return core_eol;
121 }
122
123 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
124                             struct text_stat *stats, enum safe_crlf checksafe)
125 {
126         if (!checksafe)
127                 return;
128
129         if (output_eol(crlf_action) == EOL_LF) {
130                 /*
131                  * CRLFs would not be restored by checkout:
132                  * check if we'd remove CRLFs
133                  */
134                 if (stats->crlf) {
135                         if (checksafe == SAFE_CRLF_WARN)
136                                 warning("CRLF will be replaced by LF in %s.\nThe file will have its original line endings in your working directory.", path);
137                         else /* i.e. SAFE_CRLF_FAIL */
138                                 die("CRLF would be replaced by LF in %s.", path);
139                 }
140         } else if (output_eol(crlf_action) == EOL_CRLF) {
141                 /*
142                  * CRLFs would be added by checkout:
143                  * check if we have "naked" LFs
144                  */
145                 if (stats->lf != stats->crlf) {
146                         if (checksafe == SAFE_CRLF_WARN)
147                                 warning("LF will be replaced by CRLF in %s.\nThe file will have its original line endings in your working directory.", path);
148                         else /* i.e. SAFE_CRLF_FAIL */
149                                 die("LF would be replaced by CRLF in %s", path);
150                 }
151         }
152 }
153
154 static int has_cr_in_index(const char *path)
155 {
156         int pos, len;
157         unsigned long sz;
158         enum object_type type;
159         void *data;
160         int has_cr;
161         struct index_state *istate = &the_index;
162
163         len = strlen(path);
164         pos = index_name_pos(istate, path, len);
165         if (pos < 0) {
166                 /*
167                  * We might be in the middle of a merge, in which
168                  * case we would read stage #2 (ours).
169                  */
170                 int i;
171                 for (i = -pos - 1;
172                      (pos < 0 && i < istate->cache_nr &&
173                       !strcmp(istate->cache[i]->name, path));
174                      i++)
175                         if (ce_stage(istate->cache[i]) == 2)
176                                 pos = i;
177         }
178         if (pos < 0)
179                 return 0;
180         data = read_sha1_file(istate->cache[pos]->sha1, &type, &sz);
181         if (!data || type != OBJ_BLOB) {
182                 free(data);
183                 return 0;
184         }
185
186         has_cr = memchr(data, '\r', sz) != NULL;
187         free(data);
188         return has_cr;
189 }
190
191 static int crlf_to_git(const char *path, const char *src, size_t len,
192                        struct strbuf *buf,
193                        enum crlf_action crlf_action, enum safe_crlf checksafe)
194 {
195         struct text_stat stats;
196         char *dst;
197
198         if (crlf_action == CRLF_BINARY ||
199             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) ||
200             (src && !len))
201                 return 0;
202
203         /*
204          * If we are doing a dry-run and have no source buffer, there is
205          * nothing to analyze; we must assume we would convert.
206          */
207         if (!buf && !src)
208                 return 1;
209
210         gather_stats(src, len, &stats);
211
212         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
213                 /*
214                  * We're currently not going to even try to convert stuff
215                  * that has bare CR characters. Does anybody do that crazy
216                  * stuff?
217                  */
218                 if (stats.cr != stats.crlf)
219                         return 0;
220
221                 /*
222                  * And add some heuristics for binary vs text, of course...
223                  */
224                 if (is_binary(len, &stats))
225                         return 0;
226
227                 if (crlf_action == CRLF_GUESS) {
228                         /*
229                          * If the file in the index has any CR in it, do not convert.
230                          * This is the new safer autocrlf handling.
231                          */
232                         if (has_cr_in_index(path))
233                                 return 0;
234                 }
235         }
236
237         check_safe_crlf(path, crlf_action, &stats, checksafe);
238
239         /* Optimization: No CR? Nothing to convert, regardless. */
240         if (!stats.cr)
241                 return 0;
242
243         /*
244          * At this point all of our source analysis is done, and we are sure we
245          * would convert. If we are in dry-run mode, we can give an answer.
246          */
247         if (!buf)
248                 return 1;
249
250         /* only grow if not in place */
251         if (strbuf_avail(buf) + buf->len < len)
252                 strbuf_grow(buf, len - buf->len);
253         dst = buf->buf;
254         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
255                 /*
256                  * If we guessed, we already know we rejected a file with
257                  * lone CR, and we can strip a CR without looking at what
258                  * follow it.
259                  */
260                 do {
261                         unsigned char c = *src++;
262                         if (c != '\r')
263                                 *dst++ = c;
264                 } while (--len);
265         } else {
266                 do {
267                         unsigned char c = *src++;
268                         if (! (c == '\r' && (1 < len && *src == '\n')))
269                                 *dst++ = c;
270                 } while (--len);
271         }
272         strbuf_setlen(buf, dst - buf->buf);
273         return 1;
274 }
275
276 static int crlf_to_worktree(const char *path, const char *src, size_t len,
277                             struct strbuf *buf, enum crlf_action crlf_action)
278 {
279         char *to_free = NULL;
280         struct text_stat stats;
281
282         if (!len || output_eol(crlf_action) != EOL_CRLF)
283                 return 0;
284
285         gather_stats(src, len, &stats);
286
287         /* No LF? Nothing to convert, regardless. */
288         if (!stats.lf)
289                 return 0;
290
291         /* Was it already in CRLF format? */
292         if (stats.lf == stats.crlf)
293                 return 0;
294
295         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
296                 if (crlf_action == CRLF_GUESS) {
297                         /* If we have any CR or CRLF line endings, we do not touch it */
298                         /* This is the new safer autocrlf-handling */
299                         if (stats.cr > 0 || stats.crlf > 0)
300                                 return 0;
301                 }
302
303                 /* If we have any bare CR characters, we're not going to touch it */
304                 if (stats.cr != stats.crlf)
305                         return 0;
306
307                 if (is_binary(len, &stats))
308                         return 0;
309         }
310
311         /* are we "faking" in place editing ? */
312         if (src == buf->buf)
313                 to_free = strbuf_detach(buf, NULL);
314
315         strbuf_grow(buf, len + stats.lf - stats.crlf);
316         for (;;) {
317                 const char *nl = memchr(src, '\n', len);
318                 if (!nl)
319                         break;
320                 if (nl > src && nl[-1] == '\r') {
321                         strbuf_add(buf, src, nl + 1 - src);
322                 } else {
323                         strbuf_add(buf, src, nl - src);
324                         strbuf_addstr(buf, "\r\n");
325                 }
326                 len -= nl + 1 - src;
327                 src  = nl + 1;
328         }
329         strbuf_add(buf, src, len);
330
331         free(to_free);
332         return 1;
333 }
334
335 struct filter_params {
336         const char *src;
337         unsigned long size;
338         const char *cmd;
339         const char *path;
340 };
341
342 static int filter_buffer(int in, int out, void *data)
343 {
344         /*
345          * Spawn cmd and feed the buffer contents through its stdin.
346          */
347         struct child_process child_process;
348         struct filter_params *params = (struct filter_params *)data;
349         int write_err, status;
350         const char *argv[] = { NULL, NULL };
351
352         /* apply % substitution to cmd */
353         struct strbuf cmd = STRBUF_INIT;
354         struct strbuf path = STRBUF_INIT;
355         struct strbuf_expand_dict_entry dict[] = {
356                 { "f", NULL, },
357                 { NULL, NULL, },
358         };
359
360         /* quote the path to preserve spaces, etc. */
361         sq_quote_buf(&path, params->path);
362         dict[0].value = path.buf;
363
364         /* expand all %f with the quoted path */
365         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
366         strbuf_release(&path);
367
368         argv[0] = cmd.buf;
369
370         memset(&child_process, 0, sizeof(child_process));
371         child_process.argv = argv;
372         child_process.use_shell = 1;
373         child_process.in = -1;
374         child_process.out = out;
375
376         if (start_command(&child_process))
377                 return error("cannot fork to run external filter %s", params->cmd);
378
379         sigchain_push(SIGPIPE, SIG_IGN);
380
381         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
382         if (close(child_process.in))
383                 write_err = 1;
384         if (write_err)
385                 error("cannot feed the input to external filter %s", params->cmd);
386
387         sigchain_pop(SIGPIPE);
388
389         status = finish_command(&child_process);
390         if (status)
391                 error("external filter %s failed %d", params->cmd, status);
392
393         strbuf_release(&cmd);
394         return (write_err || status);
395 }
396
397 static int apply_filter(const char *path, const char *src, size_t len,
398                         struct strbuf *dst, const char *cmd)
399 {
400         /*
401          * Create a pipeline to have the command filter the buffer's
402          * contents.
403          *
404          * (child --> cmd) --> us
405          */
406         int ret = 1;
407         struct strbuf nbuf = STRBUF_INIT;
408         struct async async;
409         struct filter_params params;
410
411         if (!cmd)
412                 return 0;
413
414         if (!dst)
415                 return 1;
416
417         memset(&async, 0, sizeof(async));
418         async.proc = filter_buffer;
419         async.data = &params;
420         async.out = -1;
421         params.src = src;
422         params.size = len;
423         params.cmd = cmd;
424         params.path = path;
425
426         fflush(NULL);
427         if (start_async(&async))
428                 return 0;       /* error was already reported */
429
430         if (strbuf_read(&nbuf, async.out, len) < 0) {
431                 error("read from external filter %s failed", cmd);
432                 ret = 0;
433         }
434         if (close(async.out)) {
435                 error("read from external filter %s failed", cmd);
436                 ret = 0;
437         }
438         if (finish_async(&async)) {
439                 error("external filter %s failed", cmd);
440                 ret = 0;
441         }
442
443         if (ret) {
444                 strbuf_swap(dst, &nbuf);
445         }
446         strbuf_release(&nbuf);
447         return ret;
448 }
449
450 static struct convert_driver {
451         const char *name;
452         struct convert_driver *next;
453         const char *smudge;
454         const char *clean;
455         int required;
456 } *user_convert, **user_convert_tail;
457
458 static int read_convert_config(const char *var, const char *value, void *cb)
459 {
460         const char *key, *name;
461         int namelen;
462         struct convert_driver *drv;
463
464         /*
465          * External conversion drivers are configured using
466          * "filter.<name>.variable".
467          */
468         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
469                 return 0;
470         for (drv = user_convert; drv; drv = drv->next)
471                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
472                         break;
473         if (!drv) {
474                 drv = xcalloc(1, sizeof(struct convert_driver));
475                 drv->name = xmemdupz(name, namelen);
476                 *user_convert_tail = drv;
477                 user_convert_tail = &(drv->next);
478         }
479
480         /*
481          * filter.<name>.smudge and filter.<name>.clean specifies
482          * the command line:
483          *
484          *      command-line
485          *
486          * The command-line will not be interpolated in any way.
487          */
488
489         if (!strcmp("smudge", key))
490                 return git_config_string(&drv->smudge, var, value);
491
492         if (!strcmp("clean", key))
493                 return git_config_string(&drv->clean, var, value);
494
495         if (!strcmp("required", key)) {
496                 drv->required = git_config_bool(var, value);
497                 return 0;
498         }
499
500         return 0;
501 }
502
503 static int count_ident(const char *cp, unsigned long size)
504 {
505         /*
506          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
507          */
508         int cnt = 0;
509         char ch;
510
511         while (size) {
512                 ch = *cp++;
513                 size--;
514                 if (ch != '$')
515                         continue;
516                 if (size < 3)
517                         break;
518                 if (memcmp("Id", cp, 2))
519                         continue;
520                 ch = cp[2];
521                 cp += 3;
522                 size -= 3;
523                 if (ch == '$')
524                         cnt++; /* $Id$ */
525                 if (ch != ':')
526                         continue;
527
528                 /*
529                  * "$Id: ... "; scan up to the closing dollar sign and discard.
530                  */
531                 while (size) {
532                         ch = *cp++;
533                         size--;
534                         if (ch == '$') {
535                                 cnt++;
536                                 break;
537                         }
538                         if (ch == '\n')
539                                 break;
540                 }
541         }
542         return cnt;
543 }
544
545 static int ident_to_git(const char *path, const char *src, size_t len,
546                         struct strbuf *buf, int ident)
547 {
548         char *dst, *dollar;
549
550         if (!ident || (src && !count_ident(src, len)))
551                 return 0;
552
553         if (!buf)
554                 return 1;
555
556         /* only grow if not in place */
557         if (strbuf_avail(buf) + buf->len < len)
558                 strbuf_grow(buf, len - buf->len);
559         dst = buf->buf;
560         for (;;) {
561                 dollar = memchr(src, '$', len);
562                 if (!dollar)
563                         break;
564                 memmove(dst, src, dollar + 1 - src);
565                 dst += dollar + 1 - src;
566                 len -= dollar + 1 - src;
567                 src  = dollar + 1;
568
569                 if (len > 3 && !memcmp(src, "Id:", 3)) {
570                         dollar = memchr(src + 3, '$', len - 3);
571                         if (!dollar)
572                                 break;
573                         if (memchr(src + 3, '\n', dollar - src - 3)) {
574                                 /* Line break before the next dollar. */
575                                 continue;
576                         }
577
578                         memcpy(dst, "Id$", 3);
579                         dst += 3;
580                         len -= dollar + 1 - src;
581                         src  = dollar + 1;
582                 }
583         }
584         memmove(dst, src, len);
585         strbuf_setlen(buf, dst + len - buf->buf);
586         return 1;
587 }
588
589 static int ident_to_worktree(const char *path, const char *src, size_t len,
590                              struct strbuf *buf, int ident)
591 {
592         unsigned char sha1[20];
593         char *to_free = NULL, *dollar, *spc;
594         int cnt;
595
596         if (!ident)
597                 return 0;
598
599         cnt = count_ident(src, len);
600         if (!cnt)
601                 return 0;
602
603         /* are we "faking" in place editing ? */
604         if (src == buf->buf)
605                 to_free = strbuf_detach(buf, NULL);
606         hash_sha1_file(src, len, "blob", sha1);
607
608         strbuf_grow(buf, len + cnt * 43);
609         for (;;) {
610                 /* step 1: run to the next '$' */
611                 dollar = memchr(src, '$', len);
612                 if (!dollar)
613                         break;
614                 strbuf_add(buf, src, dollar + 1 - src);
615                 len -= dollar + 1 - src;
616                 src  = dollar + 1;
617
618                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
619                 if (len < 3 || memcmp("Id", src, 2))
620                         continue;
621
622                 /* step 3: skip over Id$ or Id:xxxxx$ */
623                 if (src[2] == '$') {
624                         src += 3;
625                         len -= 3;
626                 } else if (src[2] == ':') {
627                         /*
628                          * It's possible that an expanded Id has crept its way into the
629                          * repository, we cope with that by stripping the expansion out.
630                          * This is probably not a good idea, since it will cause changes
631                          * on checkout, which won't go away by stash, but let's keep it
632                          * for git-style ids.
633                          */
634                         dollar = memchr(src + 3, '$', len - 3);
635                         if (!dollar) {
636                                 /* incomplete keyword, no more '$', so just quit the loop */
637                                 break;
638                         }
639
640                         if (memchr(src + 3, '\n', dollar - src - 3)) {
641                                 /* Line break before the next dollar. */
642                                 continue;
643                         }
644
645                         spc = memchr(src + 4, ' ', dollar - src - 4);
646                         if (spc && spc < dollar-1) {
647                                 /* There are spaces in unexpected places.
648                                  * This is probably an id from some other
649                                  * versioning system. Keep it for now.
650                                  */
651                                 continue;
652                         }
653
654                         len -= dollar + 1 - src;
655                         src  = dollar + 1;
656                 } else {
657                         /* it wasn't a "Id$" or "Id:xxxx$" */
658                         continue;
659                 }
660
661                 /* step 4: substitute */
662                 strbuf_addstr(buf, "Id: ");
663                 strbuf_add(buf, sha1_to_hex(sha1), 40);
664                 strbuf_addstr(buf, " $");
665         }
666         strbuf_add(buf, src, len);
667
668         free(to_free);
669         return 1;
670 }
671
672 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
673 {
674         const char *value = check->value;
675
676         if (ATTR_TRUE(value))
677                 return CRLF_TEXT;
678         else if (ATTR_FALSE(value))
679                 return CRLF_BINARY;
680         else if (ATTR_UNSET(value))
681                 ;
682         else if (!strcmp(value, "input"))
683                 return CRLF_INPUT;
684         else if (!strcmp(value, "auto"))
685                 return CRLF_AUTO;
686         return CRLF_GUESS;
687 }
688
689 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
690 {
691         const char *value = check->value;
692
693         if (ATTR_UNSET(value))
694                 ;
695         else if (!strcmp(value, "lf"))
696                 return EOL_LF;
697         else if (!strcmp(value, "crlf"))
698                 return EOL_CRLF;
699         return EOL_UNSET;
700 }
701
702 static struct convert_driver *git_path_check_convert(const char *path,
703                                              struct git_attr_check *check)
704 {
705         const char *value = check->value;
706         struct convert_driver *drv;
707
708         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
709                 return NULL;
710         for (drv = user_convert; drv; drv = drv->next)
711                 if (!strcmp(value, drv->name))
712                         return drv;
713         return NULL;
714 }
715
716 static int git_path_check_ident(const char *path, struct git_attr_check *check)
717 {
718         const char *value = check->value;
719
720         return !!ATTR_TRUE(value);
721 }
722
723 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
724 {
725         if (text_attr == CRLF_BINARY)
726                 return CRLF_BINARY;
727         if (eol_attr == EOL_LF)
728                 return CRLF_INPUT;
729         if (eol_attr == EOL_CRLF)
730                 return CRLF_CRLF;
731         return text_attr;
732 }
733
734 struct conv_attrs {
735         struct convert_driver *drv;
736         enum crlf_action crlf_action;
737         enum eol eol_attr;
738         int ident;
739 };
740
741 static const char *conv_attr_name[] = {
742         "crlf", "ident", "filter", "eol", "text",
743 };
744 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
745
746 static void convert_attrs(struct conv_attrs *ca, const char *path)
747 {
748         int i;
749         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
750
751         if (!ccheck[0].attr) {
752                 for (i = 0; i < NUM_CONV_ATTRS; i++)
753                         ccheck[i].attr = git_attr(conv_attr_name[i]);
754                 user_convert_tail = &user_convert;
755                 git_config(read_convert_config, NULL);
756         }
757
758         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
759                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
760                 if (ca->crlf_action == CRLF_GUESS)
761                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
762                 ca->ident = git_path_check_ident(path, ccheck + 1);
763                 ca->drv = git_path_check_convert(path, ccheck + 2);
764                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
765         } else {
766                 ca->drv = NULL;
767                 ca->crlf_action = CRLF_GUESS;
768                 ca->eol_attr = EOL_UNSET;
769                 ca->ident = 0;
770         }
771 }
772
773 int convert_to_git(const char *path, const char *src, size_t len,
774                    struct strbuf *dst, enum safe_crlf checksafe)
775 {
776         int ret = 0;
777         const char *filter = NULL;
778         int required = 0;
779         struct conv_attrs ca;
780
781         convert_attrs(&ca, path);
782         if (ca.drv) {
783                 filter = ca.drv->clean;
784                 required = ca.drv->required;
785         }
786
787         ret |= apply_filter(path, src, len, dst, filter);
788         if (!ret && required)
789                 die("%s: clean filter '%s' failed", path, ca.drv->name);
790
791         if (ret && dst) {
792                 src = dst->buf;
793                 len = dst->len;
794         }
795         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
796         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
797         if (ret && dst) {
798                 src = dst->buf;
799                 len = dst->len;
800         }
801         return ret | ident_to_git(path, src, len, dst, ca.ident);
802 }
803
804 static int convert_to_working_tree_internal(const char *path, const char *src,
805                                             size_t len, struct strbuf *dst,
806                                             int normalizing)
807 {
808         int ret = 0, ret_filter = 0;
809         const char *filter = NULL;
810         int required = 0;
811         struct conv_attrs ca;
812
813         convert_attrs(&ca, path);
814         if (ca.drv) {
815                 filter = ca.drv->smudge;
816                 required = ca.drv->required;
817         }
818
819         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
820         if (ret) {
821                 src = dst->buf;
822                 len = dst->len;
823         }
824         /*
825          * CRLF conversion can be skipped if normalizing, unless there
826          * is a smudge filter.  The filter might expect CRLFs.
827          */
828         if (filter || !normalizing) {
829                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
830                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
831                 if (ret) {
832                         src = dst->buf;
833                         len = dst->len;
834                 }
835         }
836
837         ret_filter = apply_filter(path, src, len, dst, filter);
838         if (!ret_filter && required)
839                 die("%s: smudge filter %s failed", path, ca.drv->name);
840
841         return ret | ret_filter;
842 }
843
844 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
845 {
846         return convert_to_working_tree_internal(path, src, len, dst, 0);
847 }
848
849 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
850 {
851         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
852         if (ret) {
853                 src = dst->buf;
854                 len = dst->len;
855         }
856         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
857 }
858
859 /*****************************************************************
860  *
861  * Streaming converison support
862  *
863  *****************************************************************/
864
865 typedef int (*filter_fn)(struct stream_filter *,
866                          const char *input, size_t *isize_p,
867                          char *output, size_t *osize_p);
868 typedef void (*free_fn)(struct stream_filter *);
869
870 struct stream_filter_vtbl {
871         filter_fn filter;
872         free_fn free;
873 };
874
875 struct stream_filter {
876         struct stream_filter_vtbl *vtbl;
877 };
878
879 static int null_filter_fn(struct stream_filter *filter,
880                           const char *input, size_t *isize_p,
881                           char *output, size_t *osize_p)
882 {
883         size_t count;
884
885         if (!input)
886                 return 0; /* we do not keep any states */
887         count = *isize_p;
888         if (*osize_p < count)
889                 count = *osize_p;
890         if (count) {
891                 memmove(output, input, count);
892                 *isize_p -= count;
893                 *osize_p -= count;
894         }
895         return 0;
896 }
897
898 static void null_free_fn(struct stream_filter *filter)
899 {
900         ; /* nothing -- null instances are shared */
901 }
902
903 static struct stream_filter_vtbl null_vtbl = {
904         null_filter_fn,
905         null_free_fn,
906 };
907
908 static struct stream_filter null_filter_singleton = {
909         &null_vtbl,
910 };
911
912 int is_null_stream_filter(struct stream_filter *filter)
913 {
914         return filter == &null_filter_singleton;
915 }
916
917
918 /*
919  * LF-to-CRLF filter
920  */
921
922 struct lf_to_crlf_filter {
923         struct stream_filter filter;
924         unsigned has_held:1;
925         char held;
926 };
927
928 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
929                                 const char *input, size_t *isize_p,
930                                 char *output, size_t *osize_p)
931 {
932         size_t count, o = 0;
933         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
934
935         /*
936          * We may be holding onto the CR to see if it is followed by a
937          * LF, in which case we would need to go to the main loop.
938          * Otherwise, just emit it to the output stream.
939          */
940         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
941                 output[o++] = lf_to_crlf->held;
942                 lf_to_crlf->has_held = 0;
943         }
944
945         /* We are told to drain */
946         if (!input) {
947                 *osize_p -= o;
948                 return 0;
949         }
950
951         count = *isize_p;
952         if (count || lf_to_crlf->has_held) {
953                 size_t i;
954                 int was_cr = 0;
955
956                 if (lf_to_crlf->has_held) {
957                         was_cr = 1;
958                         lf_to_crlf->has_held = 0;
959                 }
960
961                 for (i = 0; o < *osize_p && i < count; i++) {
962                         char ch = input[i];
963
964                         if (ch == '\n') {
965                                 output[o++] = '\r';
966                         } else if (was_cr) {
967                                 /*
968                                  * Previous round saw CR and it is not followed
969                                  * by a LF; emit the CR before processing the
970                                  * current character.
971                                  */
972                                 output[o++] = '\r';
973                         }
974
975                         /*
976                          * We may have consumed the last output slot,
977                          * in which case we need to break out of this
978                          * loop; hold the current character before
979                          * returning.
980                          */
981                         if (*osize_p <= o) {
982                                 lf_to_crlf->has_held = 1;
983                                 lf_to_crlf->held = ch;
984                                 continue; /* break but increment i */
985                         }
986
987                         if (ch == '\r') {
988                                 was_cr = 1;
989                                 continue;
990                         }
991
992                         was_cr = 0;
993                         output[o++] = ch;
994                 }
995
996                 *osize_p -= o;
997                 *isize_p -= i;
998
999                 if (!lf_to_crlf->has_held && was_cr) {
1000                         lf_to_crlf->has_held = 1;
1001                         lf_to_crlf->held = '\r';
1002                 }
1003         }
1004         return 0;
1005 }
1006
1007 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1008 {
1009         free(filter);
1010 }
1011
1012 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1013         lf_to_crlf_filter_fn,
1014         lf_to_crlf_free_fn,
1015 };
1016
1017 static struct stream_filter *lf_to_crlf_filter(void)
1018 {
1019         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1020
1021         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1022         return (struct stream_filter *)lf_to_crlf;
1023 }
1024
1025 /*
1026  * Cascade filter
1027  */
1028 #define FILTER_BUFFER 1024
1029 struct cascade_filter {
1030         struct stream_filter filter;
1031         struct stream_filter *one;
1032         struct stream_filter *two;
1033         char buf[FILTER_BUFFER];
1034         int end, ptr;
1035 };
1036
1037 static int cascade_filter_fn(struct stream_filter *filter,
1038                              const char *input, size_t *isize_p,
1039                              char *output, size_t *osize_p)
1040 {
1041         struct cascade_filter *cas = (struct cascade_filter *) filter;
1042         size_t filled = 0;
1043         size_t sz = *osize_p;
1044         size_t to_feed, remaining;
1045
1046         /*
1047          * input -- (one) --> buf -- (two) --> output
1048          */
1049         while (filled < sz) {
1050                 remaining = sz - filled;
1051
1052                 /* do we already have something to feed two with? */
1053                 if (cas->ptr < cas->end) {
1054                         to_feed = cas->end - cas->ptr;
1055                         if (stream_filter(cas->two,
1056                                           cas->buf + cas->ptr, &to_feed,
1057                                           output + filled, &remaining))
1058                                 return -1;
1059                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1060                         filled = sz - remaining;
1061                         continue;
1062                 }
1063
1064                 /* feed one from upstream and have it emit into our buffer */
1065                 to_feed = input ? *isize_p : 0;
1066                 if (input && !to_feed)
1067                         break;
1068                 remaining = sizeof(cas->buf);
1069                 if (stream_filter(cas->one,
1070                                   input, &to_feed,
1071                                   cas->buf, &remaining))
1072                         return -1;
1073                 cas->end = sizeof(cas->buf) - remaining;
1074                 cas->ptr = 0;
1075                 if (input) {
1076                         size_t fed = *isize_p - to_feed;
1077                         *isize_p -= fed;
1078                         input += fed;
1079                 }
1080
1081                 /* do we know that we drained one completely? */
1082                 if (input || cas->end)
1083                         continue;
1084
1085                 /* tell two to drain; we have nothing more to give it */
1086                 to_feed = 0;
1087                 remaining = sz - filled;
1088                 if (stream_filter(cas->two,
1089                                   NULL, &to_feed,
1090                                   output + filled, &remaining))
1091                         return -1;
1092                 if (remaining == (sz - filled))
1093                         break; /* completely drained two */
1094                 filled = sz - remaining;
1095         }
1096         *osize_p -= filled;
1097         return 0;
1098 }
1099
1100 static void cascade_free_fn(struct stream_filter *filter)
1101 {
1102         struct cascade_filter *cas = (struct cascade_filter *)filter;
1103         free_stream_filter(cas->one);
1104         free_stream_filter(cas->two);
1105         free(filter);
1106 }
1107
1108 static struct stream_filter_vtbl cascade_vtbl = {
1109         cascade_filter_fn,
1110         cascade_free_fn,
1111 };
1112
1113 static struct stream_filter *cascade_filter(struct stream_filter *one,
1114                                             struct stream_filter *two)
1115 {
1116         struct cascade_filter *cascade;
1117
1118         if (!one || is_null_stream_filter(one))
1119                 return two;
1120         if (!two || is_null_stream_filter(two))
1121                 return one;
1122
1123         cascade = xmalloc(sizeof(*cascade));
1124         cascade->one = one;
1125         cascade->two = two;
1126         cascade->end = cascade->ptr = 0;
1127         cascade->filter.vtbl = &cascade_vtbl;
1128         return (struct stream_filter *)cascade;
1129 }
1130
1131 /*
1132  * ident filter
1133  */
1134 #define IDENT_DRAINING (-1)
1135 #define IDENT_SKIPPING (-2)
1136 struct ident_filter {
1137         struct stream_filter filter;
1138         struct strbuf left;
1139         int state;
1140         char ident[45]; /* ": x40 $" */
1141 };
1142
1143 static int is_foreign_ident(const char *str)
1144 {
1145         int i;
1146
1147         if (prefixcmp(str, "$Id: "))
1148                 return 0;
1149         for (i = 5; str[i]; i++) {
1150                 if (isspace(str[i]) && str[i+1] != '$')
1151                         return 1;
1152         }
1153         return 0;
1154 }
1155
1156 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1157 {
1158         size_t to_drain = ident->left.len;
1159
1160         if (*osize_p < to_drain)
1161                 to_drain = *osize_p;
1162         if (to_drain) {
1163                 memcpy(*output_p, ident->left.buf, to_drain);
1164                 strbuf_remove(&ident->left, 0, to_drain);
1165                 *output_p += to_drain;
1166                 *osize_p -= to_drain;
1167         }
1168         if (!ident->left.len)
1169                 ident->state = 0;
1170 }
1171
1172 static int ident_filter_fn(struct stream_filter *filter,
1173                            const char *input, size_t *isize_p,
1174                            char *output, size_t *osize_p)
1175 {
1176         struct ident_filter *ident = (struct ident_filter *)filter;
1177         static const char head[] = "$Id";
1178
1179         if (!input) {
1180                 /* drain upon eof */
1181                 switch (ident->state) {
1182                 default:
1183                         strbuf_add(&ident->left, head, ident->state);
1184                 case IDENT_SKIPPING:
1185                         /* fallthru */
1186                 case IDENT_DRAINING:
1187                         ident_drain(ident, &output, osize_p);
1188                 }
1189                 return 0;
1190         }
1191
1192         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1193                 int ch;
1194
1195                 if (ident->state == IDENT_DRAINING) {
1196                         ident_drain(ident, &output, osize_p);
1197                         if (!*osize_p)
1198                                 break;
1199                         continue;
1200                 }
1201
1202                 ch = *(input++);
1203                 (*isize_p)--;
1204
1205                 if (ident->state == IDENT_SKIPPING) {
1206                         /*
1207                          * Skipping until '$' or LF, but keeping them
1208                          * in case it is a foreign ident.
1209                          */
1210                         strbuf_addch(&ident->left, ch);
1211                         if (ch != '\n' && ch != '$')
1212                                 continue;
1213                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1214                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1215                                 strbuf_addstr(&ident->left, ident->ident);
1216                         }
1217                         ident->state = IDENT_DRAINING;
1218                         continue;
1219                 }
1220
1221                 if (ident->state < sizeof(head) &&
1222                     head[ident->state] == ch) {
1223                         ident->state++;
1224                         continue;
1225                 }
1226
1227                 if (ident->state)
1228                         strbuf_add(&ident->left, head, ident->state);
1229                 if (ident->state == sizeof(head) - 1) {
1230                         if (ch != ':' && ch != '$') {
1231                                 strbuf_addch(&ident->left, ch);
1232                                 ident->state = 0;
1233                                 continue;
1234                         }
1235
1236                         if (ch == ':') {
1237                                 strbuf_addch(&ident->left, ch);
1238                                 ident->state = IDENT_SKIPPING;
1239                         } else {
1240                                 strbuf_addstr(&ident->left, ident->ident);
1241                                 ident->state = IDENT_DRAINING;
1242                         }
1243                         continue;
1244                 }
1245
1246                 strbuf_addch(&ident->left, ch);
1247                 ident->state = IDENT_DRAINING;
1248         }
1249         return 0;
1250 }
1251
1252 static void ident_free_fn(struct stream_filter *filter)
1253 {
1254         struct ident_filter *ident = (struct ident_filter *)filter;
1255         strbuf_release(&ident->left);
1256         free(filter);
1257 }
1258
1259 static struct stream_filter_vtbl ident_vtbl = {
1260         ident_filter_fn,
1261         ident_free_fn,
1262 };
1263
1264 static struct stream_filter *ident_filter(const unsigned char *sha1)
1265 {
1266         struct ident_filter *ident = xmalloc(sizeof(*ident));
1267
1268         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1269         strbuf_init(&ident->left, 0);
1270         ident->filter.vtbl = &ident_vtbl;
1271         ident->state = 0;
1272         return (struct stream_filter *)ident;
1273 }
1274
1275 /*
1276  * Return an appropriately constructed filter for the path, or NULL if
1277  * the contents cannot be filtered without reading the whole thing
1278  * in-core.
1279  *
1280  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1281  * large binary blob you would want us not to slurp into the memory!
1282  */
1283 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1284 {
1285         struct conv_attrs ca;
1286         enum crlf_action crlf_action;
1287         struct stream_filter *filter = NULL;
1288
1289         convert_attrs(&ca, path);
1290
1291         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1292                 return filter;
1293
1294         if (ca.ident)
1295                 filter = ident_filter(sha1);
1296
1297         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1298
1299         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1300             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1301                 filter = cascade_filter(filter, &null_filter_singleton);
1302
1303         else if (output_eol(crlf_action) == EOL_CRLF &&
1304                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1305                 filter = cascade_filter(filter, lf_to_crlf_filter());
1306
1307         return filter;
1308 }
1309
1310 void free_stream_filter(struct stream_filter *filter)
1311 {
1312         filter->vtbl->free(filter);
1313 }
1314
1315 int stream_filter(struct stream_filter *filter,
1316                   const char *input, size_t *isize_p,
1317                   char *output, size_t *osize_p)
1318 {
1319         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1320 }