Sync with 2.1.2
[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         unsigned long sz;
157         void *data;
158         int has_cr;
159
160         data = read_blob_data_from_cache(path, &sz);
161         if (!data)
162                 return 0;
163         has_cr = memchr(data, '\r', sz) != NULL;
164         free(data);
165         return has_cr;
166 }
167
168 static int crlf_to_git(const char *path, const char *src, size_t len,
169                        struct strbuf *buf,
170                        enum crlf_action crlf_action, enum safe_crlf checksafe)
171 {
172         struct text_stat stats;
173         char *dst;
174
175         if (crlf_action == CRLF_BINARY ||
176             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE) ||
177             (src && !len))
178                 return 0;
179
180         /*
181          * If we are doing a dry-run and have no source buffer, there is
182          * nothing to analyze; we must assume we would convert.
183          */
184         if (!buf && !src)
185                 return 1;
186
187         gather_stats(src, len, &stats);
188
189         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
190                 /*
191                  * We're currently not going to even try to convert stuff
192                  * that has bare CR characters. Does anybody do that crazy
193                  * stuff?
194                  */
195                 if (stats.cr != stats.crlf)
196                         return 0;
197
198                 /*
199                  * And add some heuristics for binary vs text, of course...
200                  */
201                 if (is_binary(len, &stats))
202                         return 0;
203
204                 if (crlf_action == CRLF_GUESS) {
205                         /*
206                          * If the file in the index has any CR in it, do not convert.
207                          * This is the new safer autocrlf handling.
208                          */
209                         if (has_cr_in_index(path))
210                                 return 0;
211                 }
212         }
213
214         check_safe_crlf(path, crlf_action, &stats, checksafe);
215
216         /* Optimization: No CR? Nothing to convert, regardless. */
217         if (!stats.cr)
218                 return 0;
219
220         /*
221          * At this point all of our source analysis is done, and we are sure we
222          * would convert. If we are in dry-run mode, we can give an answer.
223          */
224         if (!buf)
225                 return 1;
226
227         /* only grow if not in place */
228         if (strbuf_avail(buf) + buf->len < len)
229                 strbuf_grow(buf, len - buf->len);
230         dst = buf->buf;
231         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
232                 /*
233                  * If we guessed, we already know we rejected a file with
234                  * lone CR, and we can strip a CR without looking at what
235                  * follow it.
236                  */
237                 do {
238                         unsigned char c = *src++;
239                         if (c != '\r')
240                                 *dst++ = c;
241                 } while (--len);
242         } else {
243                 do {
244                         unsigned char c = *src++;
245                         if (! (c == '\r' && (1 < len && *src == '\n')))
246                                 *dst++ = c;
247                 } while (--len);
248         }
249         strbuf_setlen(buf, dst - buf->buf);
250         return 1;
251 }
252
253 static int crlf_to_worktree(const char *path, const char *src, size_t len,
254                             struct strbuf *buf, enum crlf_action crlf_action)
255 {
256         char *to_free = NULL;
257         struct text_stat stats;
258
259         if (!len || output_eol(crlf_action) != EOL_CRLF)
260                 return 0;
261
262         gather_stats(src, len, &stats);
263
264         /* No LF? Nothing to convert, regardless. */
265         if (!stats.lf)
266                 return 0;
267
268         /* Was it already in CRLF format? */
269         if (stats.lf == stats.crlf)
270                 return 0;
271
272         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS) {
273                 if (crlf_action == CRLF_GUESS) {
274                         /* If we have any CR or CRLF line endings, we do not touch it */
275                         /* This is the new safer autocrlf-handling */
276                         if (stats.cr > 0 || stats.crlf > 0)
277                                 return 0;
278                 }
279
280                 /* If we have any bare CR characters, we're not going to touch it */
281                 if (stats.cr != stats.crlf)
282                         return 0;
283
284                 if (is_binary(len, &stats))
285                         return 0;
286         }
287
288         /* are we "faking" in place editing ? */
289         if (src == buf->buf)
290                 to_free = strbuf_detach(buf, NULL);
291
292         strbuf_grow(buf, len + stats.lf - stats.crlf);
293         for (;;) {
294                 const char *nl = memchr(src, '\n', len);
295                 if (!nl)
296                         break;
297                 if (nl > src && nl[-1] == '\r') {
298                         strbuf_add(buf, src, nl + 1 - src);
299                 } else {
300                         strbuf_add(buf, src, nl - src);
301                         strbuf_addstr(buf, "\r\n");
302                 }
303                 len -= nl + 1 - src;
304                 src  = nl + 1;
305         }
306         strbuf_add(buf, src, len);
307
308         free(to_free);
309         return 1;
310 }
311
312 struct filter_params {
313         const char *src;
314         unsigned long size;
315         const char *cmd;
316         const char *path;
317 };
318
319 static int filter_buffer(int in, int out, void *data)
320 {
321         /*
322          * Spawn cmd and feed the buffer contents through its stdin.
323          */
324         struct child_process child_process = CHILD_PROCESS_INIT;
325         struct filter_params *params = (struct filter_params *)data;
326         int write_err, status;
327         const char *argv[] = { NULL, NULL };
328
329         /* apply % substitution to cmd */
330         struct strbuf cmd = STRBUF_INIT;
331         struct strbuf path = STRBUF_INIT;
332         struct strbuf_expand_dict_entry dict[] = {
333                 { "f", NULL, },
334                 { NULL, NULL, },
335         };
336
337         /* quote the path to preserve spaces, etc. */
338         sq_quote_buf(&path, params->path);
339         dict[0].value = path.buf;
340
341         /* expand all %f with the quoted path */
342         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
343         strbuf_release(&path);
344
345         argv[0] = cmd.buf;
346
347         child_process.argv = argv;
348         child_process.use_shell = 1;
349         child_process.in = -1;
350         child_process.out = out;
351
352         if (start_command(&child_process))
353                 return error("cannot fork to run external filter %s", params->cmd);
354
355         sigchain_push(SIGPIPE, SIG_IGN);
356
357         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
358         if (close(child_process.in))
359                 write_err = 1;
360         if (write_err)
361                 error("cannot feed the input to external filter %s", params->cmd);
362
363         sigchain_pop(SIGPIPE);
364
365         status = finish_command(&child_process);
366         if (status)
367                 error("external filter %s failed %d", params->cmd, status);
368
369         strbuf_release(&cmd);
370         return (write_err || status);
371 }
372
373 static int apply_filter(const char *path, const char *src, size_t len,
374                         struct strbuf *dst, const char *cmd)
375 {
376         /*
377          * Create a pipeline to have the command filter the buffer's
378          * contents.
379          *
380          * (child --> cmd) --> us
381          */
382         int ret = 1;
383         struct strbuf nbuf = STRBUF_INIT;
384         struct async async;
385         struct filter_params params;
386
387         if (!cmd)
388                 return 0;
389
390         if (!dst)
391                 return 1;
392
393         memset(&async, 0, sizeof(async));
394         async.proc = filter_buffer;
395         async.data = &params;
396         async.out = -1;
397         params.src = src;
398         params.size = len;
399         params.cmd = cmd;
400         params.path = path;
401
402         fflush(NULL);
403         if (start_async(&async))
404                 return 0;       /* error was already reported */
405
406         if (strbuf_read(&nbuf, async.out, len) < 0) {
407                 error("read from external filter %s failed", cmd);
408                 ret = 0;
409         }
410         if (close(async.out)) {
411                 error("read from external filter %s failed", cmd);
412                 ret = 0;
413         }
414         if (finish_async(&async)) {
415                 error("external filter %s failed", cmd);
416                 ret = 0;
417         }
418
419         if (ret) {
420                 strbuf_swap(dst, &nbuf);
421         }
422         strbuf_release(&nbuf);
423         return ret;
424 }
425
426 static struct convert_driver {
427         const char *name;
428         struct convert_driver *next;
429         const char *smudge;
430         const char *clean;
431         int required;
432 } *user_convert, **user_convert_tail;
433
434 static int read_convert_config(const char *var, const char *value, void *cb)
435 {
436         const char *key, *name;
437         int namelen;
438         struct convert_driver *drv;
439
440         /*
441          * External conversion drivers are configured using
442          * "filter.<name>.variable".
443          */
444         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
445                 return 0;
446         for (drv = user_convert; drv; drv = drv->next)
447                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
448                         break;
449         if (!drv) {
450                 drv = xcalloc(1, sizeof(struct convert_driver));
451                 drv->name = xmemdupz(name, namelen);
452                 *user_convert_tail = drv;
453                 user_convert_tail = &(drv->next);
454         }
455
456         /*
457          * filter.<name>.smudge and filter.<name>.clean specifies
458          * the command line:
459          *
460          *      command-line
461          *
462          * The command-line will not be interpolated in any way.
463          */
464
465         if (!strcmp("smudge", key))
466                 return git_config_string(&drv->smudge, var, value);
467
468         if (!strcmp("clean", key))
469                 return git_config_string(&drv->clean, var, value);
470
471         if (!strcmp("required", key)) {
472                 drv->required = git_config_bool(var, value);
473                 return 0;
474         }
475
476         return 0;
477 }
478
479 static int count_ident(const char *cp, unsigned long size)
480 {
481         /*
482          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
483          */
484         int cnt = 0;
485         char ch;
486
487         while (size) {
488                 ch = *cp++;
489                 size--;
490                 if (ch != '$')
491                         continue;
492                 if (size < 3)
493                         break;
494                 if (memcmp("Id", cp, 2))
495                         continue;
496                 ch = cp[2];
497                 cp += 3;
498                 size -= 3;
499                 if (ch == '$')
500                         cnt++; /* $Id$ */
501                 if (ch != ':')
502                         continue;
503
504                 /*
505                  * "$Id: ... "; scan up to the closing dollar sign and discard.
506                  */
507                 while (size) {
508                         ch = *cp++;
509                         size--;
510                         if (ch == '$') {
511                                 cnt++;
512                                 break;
513                         }
514                         if (ch == '\n')
515                                 break;
516                 }
517         }
518         return cnt;
519 }
520
521 static int ident_to_git(const char *path, const char *src, size_t len,
522                         struct strbuf *buf, int ident)
523 {
524         char *dst, *dollar;
525
526         if (!ident || (src && !count_ident(src, len)))
527                 return 0;
528
529         if (!buf)
530                 return 1;
531
532         /* only grow if not in place */
533         if (strbuf_avail(buf) + buf->len < len)
534                 strbuf_grow(buf, len - buf->len);
535         dst = buf->buf;
536         for (;;) {
537                 dollar = memchr(src, '$', len);
538                 if (!dollar)
539                         break;
540                 memmove(dst, src, dollar + 1 - src);
541                 dst += dollar + 1 - src;
542                 len -= dollar + 1 - src;
543                 src  = dollar + 1;
544
545                 if (len > 3 && !memcmp(src, "Id:", 3)) {
546                         dollar = memchr(src + 3, '$', len - 3);
547                         if (!dollar)
548                                 break;
549                         if (memchr(src + 3, '\n', dollar - src - 3)) {
550                                 /* Line break before the next dollar. */
551                                 continue;
552                         }
553
554                         memcpy(dst, "Id$", 3);
555                         dst += 3;
556                         len -= dollar + 1 - src;
557                         src  = dollar + 1;
558                 }
559         }
560         memmove(dst, src, len);
561         strbuf_setlen(buf, dst + len - buf->buf);
562         return 1;
563 }
564
565 static int ident_to_worktree(const char *path, const char *src, size_t len,
566                              struct strbuf *buf, int ident)
567 {
568         unsigned char sha1[20];
569         char *to_free = NULL, *dollar, *spc;
570         int cnt;
571
572         if (!ident)
573                 return 0;
574
575         cnt = count_ident(src, len);
576         if (!cnt)
577                 return 0;
578
579         /* are we "faking" in place editing ? */
580         if (src == buf->buf)
581                 to_free = strbuf_detach(buf, NULL);
582         hash_sha1_file(src, len, "blob", sha1);
583
584         strbuf_grow(buf, len + cnt * 43);
585         for (;;) {
586                 /* step 1: run to the next '$' */
587                 dollar = memchr(src, '$', len);
588                 if (!dollar)
589                         break;
590                 strbuf_add(buf, src, dollar + 1 - src);
591                 len -= dollar + 1 - src;
592                 src  = dollar + 1;
593
594                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
595                 if (len < 3 || memcmp("Id", src, 2))
596                         continue;
597
598                 /* step 3: skip over Id$ or Id:xxxxx$ */
599                 if (src[2] == '$') {
600                         src += 3;
601                         len -= 3;
602                 } else if (src[2] == ':') {
603                         /*
604                          * It's possible that an expanded Id has crept its way into the
605                          * repository, we cope with that by stripping the expansion out.
606                          * This is probably not a good idea, since it will cause changes
607                          * on checkout, which won't go away by stash, but let's keep it
608                          * for git-style ids.
609                          */
610                         dollar = memchr(src + 3, '$', len - 3);
611                         if (!dollar) {
612                                 /* incomplete keyword, no more '$', so just quit the loop */
613                                 break;
614                         }
615
616                         if (memchr(src + 3, '\n', dollar - src - 3)) {
617                                 /* Line break before the next dollar. */
618                                 continue;
619                         }
620
621                         spc = memchr(src + 4, ' ', dollar - src - 4);
622                         if (spc && spc < dollar-1) {
623                                 /* There are spaces in unexpected places.
624                                  * This is probably an id from some other
625                                  * versioning system. Keep it for now.
626                                  */
627                                 continue;
628                         }
629
630                         len -= dollar + 1 - src;
631                         src  = dollar + 1;
632                 } else {
633                         /* it wasn't a "Id$" or "Id:xxxx$" */
634                         continue;
635                 }
636
637                 /* step 4: substitute */
638                 strbuf_addstr(buf, "Id: ");
639                 strbuf_add(buf, sha1_to_hex(sha1), 40);
640                 strbuf_addstr(buf, " $");
641         }
642         strbuf_add(buf, src, len);
643
644         free(to_free);
645         return 1;
646 }
647
648 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
649 {
650         const char *value = check->value;
651
652         if (ATTR_TRUE(value))
653                 return CRLF_TEXT;
654         else if (ATTR_FALSE(value))
655                 return CRLF_BINARY;
656         else if (ATTR_UNSET(value))
657                 ;
658         else if (!strcmp(value, "input"))
659                 return CRLF_INPUT;
660         else if (!strcmp(value, "auto"))
661                 return CRLF_AUTO;
662         return CRLF_GUESS;
663 }
664
665 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
666 {
667         const char *value = check->value;
668
669         if (ATTR_UNSET(value))
670                 ;
671         else if (!strcmp(value, "lf"))
672                 return EOL_LF;
673         else if (!strcmp(value, "crlf"))
674                 return EOL_CRLF;
675         return EOL_UNSET;
676 }
677
678 static struct convert_driver *git_path_check_convert(const char *path,
679                                              struct git_attr_check *check)
680 {
681         const char *value = check->value;
682         struct convert_driver *drv;
683
684         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
685                 return NULL;
686         for (drv = user_convert; drv; drv = drv->next)
687                 if (!strcmp(value, drv->name))
688                         return drv;
689         return NULL;
690 }
691
692 static int git_path_check_ident(const char *path, struct git_attr_check *check)
693 {
694         const char *value = check->value;
695
696         return !!ATTR_TRUE(value);
697 }
698
699 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
700 {
701         if (text_attr == CRLF_BINARY)
702                 return CRLF_BINARY;
703         if (eol_attr == EOL_LF)
704                 return CRLF_INPUT;
705         if (eol_attr == EOL_CRLF)
706                 return CRLF_CRLF;
707         return text_attr;
708 }
709
710 struct conv_attrs {
711         struct convert_driver *drv;
712         enum crlf_action crlf_action;
713         enum eol eol_attr;
714         int ident;
715 };
716
717 static const char *conv_attr_name[] = {
718         "crlf", "ident", "filter", "eol", "text",
719 };
720 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
721
722 static void convert_attrs(struct conv_attrs *ca, const char *path)
723 {
724         int i;
725         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
726
727         if (!ccheck[0].attr) {
728                 for (i = 0; i < NUM_CONV_ATTRS; i++)
729                         ccheck[i].attr = git_attr(conv_attr_name[i]);
730                 user_convert_tail = &user_convert;
731                 git_config(read_convert_config, NULL);
732         }
733
734         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
735                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
736                 if (ca->crlf_action == CRLF_GUESS)
737                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
738                 ca->ident = git_path_check_ident(path, ccheck + 1);
739                 ca->drv = git_path_check_convert(path, ccheck + 2);
740                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
741         } else {
742                 ca->drv = NULL;
743                 ca->crlf_action = CRLF_GUESS;
744                 ca->eol_attr = EOL_UNSET;
745                 ca->ident = 0;
746         }
747 }
748
749 int convert_to_git(const char *path, const char *src, size_t len,
750                    struct strbuf *dst, enum safe_crlf checksafe)
751 {
752         int ret = 0;
753         const char *filter = NULL;
754         int required = 0;
755         struct conv_attrs ca;
756
757         convert_attrs(&ca, path);
758         if (ca.drv) {
759                 filter = ca.drv->clean;
760                 required = ca.drv->required;
761         }
762
763         ret |= apply_filter(path, src, len, dst, filter);
764         if (!ret && required)
765                 die("%s: clean filter '%s' failed", path, ca.drv->name);
766
767         if (ret && dst) {
768                 src = dst->buf;
769                 len = dst->len;
770         }
771         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
772         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
773         if (ret && dst) {
774                 src = dst->buf;
775                 len = dst->len;
776         }
777         return ret | ident_to_git(path, src, len, dst, ca.ident);
778 }
779
780 static int convert_to_working_tree_internal(const char *path, const char *src,
781                                             size_t len, struct strbuf *dst,
782                                             int normalizing)
783 {
784         int ret = 0, ret_filter = 0;
785         const char *filter = NULL;
786         int required = 0;
787         struct conv_attrs ca;
788
789         convert_attrs(&ca, path);
790         if (ca.drv) {
791                 filter = ca.drv->smudge;
792                 required = ca.drv->required;
793         }
794
795         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
796         if (ret) {
797                 src = dst->buf;
798                 len = dst->len;
799         }
800         /*
801          * CRLF conversion can be skipped if normalizing, unless there
802          * is a smudge filter.  The filter might expect CRLFs.
803          */
804         if (filter || !normalizing) {
805                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
806                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
807                 if (ret) {
808                         src = dst->buf;
809                         len = dst->len;
810                 }
811         }
812
813         ret_filter = apply_filter(path, src, len, dst, filter);
814         if (!ret_filter && required)
815                 die("%s: smudge filter %s failed", path, ca.drv->name);
816
817         return ret | ret_filter;
818 }
819
820 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
821 {
822         return convert_to_working_tree_internal(path, src, len, dst, 0);
823 }
824
825 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
826 {
827         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
828         if (ret) {
829                 src = dst->buf;
830                 len = dst->len;
831         }
832         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
833 }
834
835 /*****************************************************************
836  *
837  * Streaming conversion support
838  *
839  *****************************************************************/
840
841 typedef int (*filter_fn)(struct stream_filter *,
842                          const char *input, size_t *isize_p,
843                          char *output, size_t *osize_p);
844 typedef void (*free_fn)(struct stream_filter *);
845
846 struct stream_filter_vtbl {
847         filter_fn filter;
848         free_fn free;
849 };
850
851 struct stream_filter {
852         struct stream_filter_vtbl *vtbl;
853 };
854
855 static int null_filter_fn(struct stream_filter *filter,
856                           const char *input, size_t *isize_p,
857                           char *output, size_t *osize_p)
858 {
859         size_t count;
860
861         if (!input)
862                 return 0; /* we do not keep any states */
863         count = *isize_p;
864         if (*osize_p < count)
865                 count = *osize_p;
866         if (count) {
867                 memmove(output, input, count);
868                 *isize_p -= count;
869                 *osize_p -= count;
870         }
871         return 0;
872 }
873
874 static void null_free_fn(struct stream_filter *filter)
875 {
876         ; /* nothing -- null instances are shared */
877 }
878
879 static struct stream_filter_vtbl null_vtbl = {
880         null_filter_fn,
881         null_free_fn,
882 };
883
884 static struct stream_filter null_filter_singleton = {
885         &null_vtbl,
886 };
887
888 int is_null_stream_filter(struct stream_filter *filter)
889 {
890         return filter == &null_filter_singleton;
891 }
892
893
894 /*
895  * LF-to-CRLF filter
896  */
897
898 struct lf_to_crlf_filter {
899         struct stream_filter filter;
900         unsigned has_held:1;
901         char held;
902 };
903
904 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
905                                 const char *input, size_t *isize_p,
906                                 char *output, size_t *osize_p)
907 {
908         size_t count, o = 0;
909         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
910
911         /*
912          * We may be holding onto the CR to see if it is followed by a
913          * LF, in which case we would need to go to the main loop.
914          * Otherwise, just emit it to the output stream.
915          */
916         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
917                 output[o++] = lf_to_crlf->held;
918                 lf_to_crlf->has_held = 0;
919         }
920
921         /* We are told to drain */
922         if (!input) {
923                 *osize_p -= o;
924                 return 0;
925         }
926
927         count = *isize_p;
928         if (count || lf_to_crlf->has_held) {
929                 size_t i;
930                 int was_cr = 0;
931
932                 if (lf_to_crlf->has_held) {
933                         was_cr = 1;
934                         lf_to_crlf->has_held = 0;
935                 }
936
937                 for (i = 0; o < *osize_p && i < count; i++) {
938                         char ch = input[i];
939
940                         if (ch == '\n') {
941                                 output[o++] = '\r';
942                         } else if (was_cr) {
943                                 /*
944                                  * Previous round saw CR and it is not followed
945                                  * by a LF; emit the CR before processing the
946                                  * current character.
947                                  */
948                                 output[o++] = '\r';
949                         }
950
951                         /*
952                          * We may have consumed the last output slot,
953                          * in which case we need to break out of this
954                          * loop; hold the current character before
955                          * returning.
956                          */
957                         if (*osize_p <= o) {
958                                 lf_to_crlf->has_held = 1;
959                                 lf_to_crlf->held = ch;
960                                 continue; /* break but increment i */
961                         }
962
963                         if (ch == '\r') {
964                                 was_cr = 1;
965                                 continue;
966                         }
967
968                         was_cr = 0;
969                         output[o++] = ch;
970                 }
971
972                 *osize_p -= o;
973                 *isize_p -= i;
974
975                 if (!lf_to_crlf->has_held && was_cr) {
976                         lf_to_crlf->has_held = 1;
977                         lf_to_crlf->held = '\r';
978                 }
979         }
980         return 0;
981 }
982
983 static void lf_to_crlf_free_fn(struct stream_filter *filter)
984 {
985         free(filter);
986 }
987
988 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
989         lf_to_crlf_filter_fn,
990         lf_to_crlf_free_fn,
991 };
992
993 static struct stream_filter *lf_to_crlf_filter(void)
994 {
995         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
996
997         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
998         return (struct stream_filter *)lf_to_crlf;
999 }
1000
1001 /*
1002  * Cascade filter
1003  */
1004 #define FILTER_BUFFER 1024
1005 struct cascade_filter {
1006         struct stream_filter filter;
1007         struct stream_filter *one;
1008         struct stream_filter *two;
1009         char buf[FILTER_BUFFER];
1010         int end, ptr;
1011 };
1012
1013 static int cascade_filter_fn(struct stream_filter *filter,
1014                              const char *input, size_t *isize_p,
1015                              char *output, size_t *osize_p)
1016 {
1017         struct cascade_filter *cas = (struct cascade_filter *) filter;
1018         size_t filled = 0;
1019         size_t sz = *osize_p;
1020         size_t to_feed, remaining;
1021
1022         /*
1023          * input -- (one) --> buf -- (two) --> output
1024          */
1025         while (filled < sz) {
1026                 remaining = sz - filled;
1027
1028                 /* do we already have something to feed two with? */
1029                 if (cas->ptr < cas->end) {
1030                         to_feed = cas->end - cas->ptr;
1031                         if (stream_filter(cas->two,
1032                                           cas->buf + cas->ptr, &to_feed,
1033                                           output + filled, &remaining))
1034                                 return -1;
1035                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1036                         filled = sz - remaining;
1037                         continue;
1038                 }
1039
1040                 /* feed one from upstream and have it emit into our buffer */
1041                 to_feed = input ? *isize_p : 0;
1042                 if (input && !to_feed)
1043                         break;
1044                 remaining = sizeof(cas->buf);
1045                 if (stream_filter(cas->one,
1046                                   input, &to_feed,
1047                                   cas->buf, &remaining))
1048                         return -1;
1049                 cas->end = sizeof(cas->buf) - remaining;
1050                 cas->ptr = 0;
1051                 if (input) {
1052                         size_t fed = *isize_p - to_feed;
1053                         *isize_p -= fed;
1054                         input += fed;
1055                 }
1056
1057                 /* do we know that we drained one completely? */
1058                 if (input || cas->end)
1059                         continue;
1060
1061                 /* tell two to drain; we have nothing more to give it */
1062                 to_feed = 0;
1063                 remaining = sz - filled;
1064                 if (stream_filter(cas->two,
1065                                   NULL, &to_feed,
1066                                   output + filled, &remaining))
1067                         return -1;
1068                 if (remaining == (sz - filled))
1069                         break; /* completely drained two */
1070                 filled = sz - remaining;
1071         }
1072         *osize_p -= filled;
1073         return 0;
1074 }
1075
1076 static void cascade_free_fn(struct stream_filter *filter)
1077 {
1078         struct cascade_filter *cas = (struct cascade_filter *)filter;
1079         free_stream_filter(cas->one);
1080         free_stream_filter(cas->two);
1081         free(filter);
1082 }
1083
1084 static struct stream_filter_vtbl cascade_vtbl = {
1085         cascade_filter_fn,
1086         cascade_free_fn,
1087 };
1088
1089 static struct stream_filter *cascade_filter(struct stream_filter *one,
1090                                             struct stream_filter *two)
1091 {
1092         struct cascade_filter *cascade;
1093
1094         if (!one || is_null_stream_filter(one))
1095                 return two;
1096         if (!two || is_null_stream_filter(two))
1097                 return one;
1098
1099         cascade = xmalloc(sizeof(*cascade));
1100         cascade->one = one;
1101         cascade->two = two;
1102         cascade->end = cascade->ptr = 0;
1103         cascade->filter.vtbl = &cascade_vtbl;
1104         return (struct stream_filter *)cascade;
1105 }
1106
1107 /*
1108  * ident filter
1109  */
1110 #define IDENT_DRAINING (-1)
1111 #define IDENT_SKIPPING (-2)
1112 struct ident_filter {
1113         struct stream_filter filter;
1114         struct strbuf left;
1115         int state;
1116         char ident[45]; /* ": x40 $" */
1117 };
1118
1119 static int is_foreign_ident(const char *str)
1120 {
1121         int i;
1122
1123         if (!skip_prefix(str, "$Id: ", &str))
1124                 return 0;
1125         for (i = 0; str[i]; i++) {
1126                 if (isspace(str[i]) && str[i+1] != '$')
1127                         return 1;
1128         }
1129         return 0;
1130 }
1131
1132 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1133 {
1134         size_t to_drain = ident->left.len;
1135
1136         if (*osize_p < to_drain)
1137                 to_drain = *osize_p;
1138         if (to_drain) {
1139                 memcpy(*output_p, ident->left.buf, to_drain);
1140                 strbuf_remove(&ident->left, 0, to_drain);
1141                 *output_p += to_drain;
1142                 *osize_p -= to_drain;
1143         }
1144         if (!ident->left.len)
1145                 ident->state = 0;
1146 }
1147
1148 static int ident_filter_fn(struct stream_filter *filter,
1149                            const char *input, size_t *isize_p,
1150                            char *output, size_t *osize_p)
1151 {
1152         struct ident_filter *ident = (struct ident_filter *)filter;
1153         static const char head[] = "$Id";
1154
1155         if (!input) {
1156                 /* drain upon eof */
1157                 switch (ident->state) {
1158                 default:
1159                         strbuf_add(&ident->left, head, ident->state);
1160                 case IDENT_SKIPPING:
1161                         /* fallthru */
1162                 case IDENT_DRAINING:
1163                         ident_drain(ident, &output, osize_p);
1164                 }
1165                 return 0;
1166         }
1167
1168         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1169                 int ch;
1170
1171                 if (ident->state == IDENT_DRAINING) {
1172                         ident_drain(ident, &output, osize_p);
1173                         if (!*osize_p)
1174                                 break;
1175                         continue;
1176                 }
1177
1178                 ch = *(input++);
1179                 (*isize_p)--;
1180
1181                 if (ident->state == IDENT_SKIPPING) {
1182                         /*
1183                          * Skipping until '$' or LF, but keeping them
1184                          * in case it is a foreign ident.
1185                          */
1186                         strbuf_addch(&ident->left, ch);
1187                         if (ch != '\n' && ch != '$')
1188                                 continue;
1189                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1190                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1191                                 strbuf_addstr(&ident->left, ident->ident);
1192                         }
1193                         ident->state = IDENT_DRAINING;
1194                         continue;
1195                 }
1196
1197                 if (ident->state < sizeof(head) &&
1198                     head[ident->state] == ch) {
1199                         ident->state++;
1200                         continue;
1201                 }
1202
1203                 if (ident->state)
1204                         strbuf_add(&ident->left, head, ident->state);
1205                 if (ident->state == sizeof(head) - 1) {
1206                         if (ch != ':' && ch != '$') {
1207                                 strbuf_addch(&ident->left, ch);
1208                                 ident->state = 0;
1209                                 continue;
1210                         }
1211
1212                         if (ch == ':') {
1213                                 strbuf_addch(&ident->left, ch);
1214                                 ident->state = IDENT_SKIPPING;
1215                         } else {
1216                                 strbuf_addstr(&ident->left, ident->ident);
1217                                 ident->state = IDENT_DRAINING;
1218                         }
1219                         continue;
1220                 }
1221
1222                 strbuf_addch(&ident->left, ch);
1223                 ident->state = IDENT_DRAINING;
1224         }
1225         return 0;
1226 }
1227
1228 static void ident_free_fn(struct stream_filter *filter)
1229 {
1230         struct ident_filter *ident = (struct ident_filter *)filter;
1231         strbuf_release(&ident->left);
1232         free(filter);
1233 }
1234
1235 static struct stream_filter_vtbl ident_vtbl = {
1236         ident_filter_fn,
1237         ident_free_fn,
1238 };
1239
1240 static struct stream_filter *ident_filter(const unsigned char *sha1)
1241 {
1242         struct ident_filter *ident = xmalloc(sizeof(*ident));
1243
1244         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1245         strbuf_init(&ident->left, 0);
1246         ident->filter.vtbl = &ident_vtbl;
1247         ident->state = 0;
1248         return (struct stream_filter *)ident;
1249 }
1250
1251 /*
1252  * Return an appropriately constructed filter for the path, or NULL if
1253  * the contents cannot be filtered without reading the whole thing
1254  * in-core.
1255  *
1256  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1257  * large binary blob you would want us not to slurp into the memory!
1258  */
1259 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1260 {
1261         struct conv_attrs ca;
1262         enum crlf_action crlf_action;
1263         struct stream_filter *filter = NULL;
1264
1265         convert_attrs(&ca, path);
1266
1267         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1268                 return filter;
1269
1270         if (ca.ident)
1271                 filter = ident_filter(sha1);
1272
1273         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1274
1275         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1276             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1277                 filter = cascade_filter(filter, &null_filter_singleton);
1278
1279         else if (output_eol(crlf_action) == EOL_CRLF &&
1280                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1281                 filter = cascade_filter(filter, lf_to_crlf_filter());
1282
1283         return filter;
1284 }
1285
1286 void free_stream_filter(struct stream_filter *filter)
1287 {
1288         filter->vtbl->free(filter);
1289 }
1290
1291 int stream_filter(struct stream_filter *filter,
1292                   const char *input, size_t *isize_p,
1293                   char *output, size_t *osize_p)
1294 {
1295         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1296 }