Merge branch 'jl/submodule-report-new-path-once'
[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 *ep, *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 (prefixcmp(var, "filter.") || (ep = strrchr(var, '.')) == var + 6)
469                 return 0;
470         name = var + 7;
471         namelen = ep - name;
472         for (drv = user_convert; drv; drv = drv->next)
473                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
474                         break;
475         if (!drv) {
476                 drv = xcalloc(1, sizeof(struct convert_driver));
477                 drv->name = xmemdupz(name, namelen);
478                 *user_convert_tail = drv;
479                 user_convert_tail = &(drv->next);
480         }
481
482         ep++;
483
484         /*
485          * filter.<name>.smudge and filter.<name>.clean specifies
486          * the command line:
487          *
488          *      command-line
489          *
490          * The command-line will not be interpolated in any way.
491          */
492
493         if (!strcmp("smudge", ep))
494                 return git_config_string(&drv->smudge, var, value);
495
496         if (!strcmp("clean", ep))
497                 return git_config_string(&drv->clean, var, value);
498
499         if (!strcmp("required", ep)) {
500                 drv->required = git_config_bool(var, value);
501                 return 0;
502         }
503
504         return 0;
505 }
506
507 static int count_ident(const char *cp, unsigned long size)
508 {
509         /*
510          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
511          */
512         int cnt = 0;
513         char ch;
514
515         while (size) {
516                 ch = *cp++;
517                 size--;
518                 if (ch != '$')
519                         continue;
520                 if (size < 3)
521                         break;
522                 if (memcmp("Id", cp, 2))
523                         continue;
524                 ch = cp[2];
525                 cp += 3;
526                 size -= 3;
527                 if (ch == '$')
528                         cnt++; /* $Id$ */
529                 if (ch != ':')
530                         continue;
531
532                 /*
533                  * "$Id: ... "; scan up to the closing dollar sign and discard.
534                  */
535                 while (size) {
536                         ch = *cp++;
537                         size--;
538                         if (ch == '$') {
539                                 cnt++;
540                                 break;
541                         }
542                         if (ch == '\n')
543                                 break;
544                 }
545         }
546         return cnt;
547 }
548
549 static int ident_to_git(const char *path, const char *src, size_t len,
550                         struct strbuf *buf, int ident)
551 {
552         char *dst, *dollar;
553
554         if (!ident || (src && !count_ident(src, len)))
555                 return 0;
556
557         if (!buf)
558                 return 1;
559
560         /* only grow if not in place */
561         if (strbuf_avail(buf) + buf->len < len)
562                 strbuf_grow(buf, len - buf->len);
563         dst = buf->buf;
564         for (;;) {
565                 dollar = memchr(src, '$', len);
566                 if (!dollar)
567                         break;
568                 memmove(dst, src, dollar + 1 - src);
569                 dst += dollar + 1 - src;
570                 len -= dollar + 1 - src;
571                 src  = dollar + 1;
572
573                 if (len > 3 && !memcmp(src, "Id:", 3)) {
574                         dollar = memchr(src + 3, '$', len - 3);
575                         if (!dollar)
576                                 break;
577                         if (memchr(src + 3, '\n', dollar - src - 3)) {
578                                 /* Line break before the next dollar. */
579                                 continue;
580                         }
581
582                         memcpy(dst, "Id$", 3);
583                         dst += 3;
584                         len -= dollar + 1 - src;
585                         src  = dollar + 1;
586                 }
587         }
588         memmove(dst, src, len);
589         strbuf_setlen(buf, dst + len - buf->buf);
590         return 1;
591 }
592
593 static int ident_to_worktree(const char *path, const char *src, size_t len,
594                              struct strbuf *buf, int ident)
595 {
596         unsigned char sha1[20];
597         char *to_free = NULL, *dollar, *spc;
598         int cnt;
599
600         if (!ident)
601                 return 0;
602
603         cnt = count_ident(src, len);
604         if (!cnt)
605                 return 0;
606
607         /* are we "faking" in place editing ? */
608         if (src == buf->buf)
609                 to_free = strbuf_detach(buf, NULL);
610         hash_sha1_file(src, len, "blob", sha1);
611
612         strbuf_grow(buf, len + cnt * 43);
613         for (;;) {
614                 /* step 1: run to the next '$' */
615                 dollar = memchr(src, '$', len);
616                 if (!dollar)
617                         break;
618                 strbuf_add(buf, src, dollar + 1 - src);
619                 len -= dollar + 1 - src;
620                 src  = dollar + 1;
621
622                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
623                 if (len < 3 || memcmp("Id", src, 2))
624                         continue;
625
626                 /* step 3: skip over Id$ or Id:xxxxx$ */
627                 if (src[2] == '$') {
628                         src += 3;
629                         len -= 3;
630                 } else if (src[2] == ':') {
631                         /*
632                          * It's possible that an expanded Id has crept its way into the
633                          * repository, we cope with that by stripping the expansion out.
634                          * This is probably not a good idea, since it will cause changes
635                          * on checkout, which won't go away by stash, but let's keep it
636                          * for git-style ids.
637                          */
638                         dollar = memchr(src + 3, '$', len - 3);
639                         if (!dollar) {
640                                 /* incomplete keyword, no more '$', so just quit the loop */
641                                 break;
642                         }
643
644                         if (memchr(src + 3, '\n', dollar - src - 3)) {
645                                 /* Line break before the next dollar. */
646                                 continue;
647                         }
648
649                         spc = memchr(src + 4, ' ', dollar - src - 4);
650                         if (spc && spc < dollar-1) {
651                                 /* There are spaces in unexpected places.
652                                  * This is probably an id from some other
653                                  * versioning system. Keep it for now.
654                                  */
655                                 continue;
656                         }
657
658                         len -= dollar + 1 - src;
659                         src  = dollar + 1;
660                 } else {
661                         /* it wasn't a "Id$" or "Id:xxxx$" */
662                         continue;
663                 }
664
665                 /* step 4: substitute */
666                 strbuf_addstr(buf, "Id: ");
667                 strbuf_add(buf, sha1_to_hex(sha1), 40);
668                 strbuf_addstr(buf, " $");
669         }
670         strbuf_add(buf, src, len);
671
672         free(to_free);
673         return 1;
674 }
675
676 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
677 {
678         const char *value = check->value;
679
680         if (ATTR_TRUE(value))
681                 return CRLF_TEXT;
682         else if (ATTR_FALSE(value))
683                 return CRLF_BINARY;
684         else if (ATTR_UNSET(value))
685                 ;
686         else if (!strcmp(value, "input"))
687                 return CRLF_INPUT;
688         else if (!strcmp(value, "auto"))
689                 return CRLF_AUTO;
690         return CRLF_GUESS;
691 }
692
693 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
694 {
695         const char *value = check->value;
696
697         if (ATTR_UNSET(value))
698                 ;
699         else if (!strcmp(value, "lf"))
700                 return EOL_LF;
701         else if (!strcmp(value, "crlf"))
702                 return EOL_CRLF;
703         return EOL_UNSET;
704 }
705
706 static struct convert_driver *git_path_check_convert(const char *path,
707                                              struct git_attr_check *check)
708 {
709         const char *value = check->value;
710         struct convert_driver *drv;
711
712         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
713                 return NULL;
714         for (drv = user_convert; drv; drv = drv->next)
715                 if (!strcmp(value, drv->name))
716                         return drv;
717         return NULL;
718 }
719
720 static int git_path_check_ident(const char *path, struct git_attr_check *check)
721 {
722         const char *value = check->value;
723
724         return !!ATTR_TRUE(value);
725 }
726
727 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
728 {
729         if (text_attr == CRLF_BINARY)
730                 return CRLF_BINARY;
731         if (eol_attr == EOL_LF)
732                 return CRLF_INPUT;
733         if (eol_attr == EOL_CRLF)
734                 return CRLF_CRLF;
735         return text_attr;
736 }
737
738 struct conv_attrs {
739         struct convert_driver *drv;
740         enum crlf_action crlf_action;
741         enum eol eol_attr;
742         int ident;
743 };
744
745 static const char *conv_attr_name[] = {
746         "crlf", "ident", "filter", "eol", "text",
747 };
748 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
749
750 static void convert_attrs(struct conv_attrs *ca, const char *path)
751 {
752         int i;
753         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
754
755         if (!ccheck[0].attr) {
756                 for (i = 0; i < NUM_CONV_ATTRS; i++)
757                         ccheck[i].attr = git_attr(conv_attr_name[i]);
758                 user_convert_tail = &user_convert;
759                 git_config(read_convert_config, NULL);
760         }
761
762         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
763                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
764                 if (ca->crlf_action == CRLF_GUESS)
765                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
766                 ca->ident = git_path_check_ident(path, ccheck + 1);
767                 ca->drv = git_path_check_convert(path, ccheck + 2);
768                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
769         } else {
770                 ca->drv = NULL;
771                 ca->crlf_action = CRLF_GUESS;
772                 ca->eol_attr = EOL_UNSET;
773                 ca->ident = 0;
774         }
775 }
776
777 int convert_to_git(const char *path, const char *src, size_t len,
778                    struct strbuf *dst, enum safe_crlf checksafe)
779 {
780         int ret = 0;
781         const char *filter = NULL;
782         int required = 0;
783         struct conv_attrs ca;
784
785         convert_attrs(&ca, path);
786         if (ca.drv) {
787                 filter = ca.drv->clean;
788                 required = ca.drv->required;
789         }
790
791         ret |= apply_filter(path, src, len, dst, filter);
792         if (!ret && required)
793                 die("%s: clean filter '%s' failed", path, ca.drv->name);
794
795         if (ret && dst) {
796                 src = dst->buf;
797                 len = dst->len;
798         }
799         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
800         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
801         if (ret && dst) {
802                 src = dst->buf;
803                 len = dst->len;
804         }
805         return ret | ident_to_git(path, src, len, dst, ca.ident);
806 }
807
808 static int convert_to_working_tree_internal(const char *path, const char *src,
809                                             size_t len, struct strbuf *dst,
810                                             int normalizing)
811 {
812         int ret = 0, ret_filter = 0;
813         const char *filter = NULL;
814         int required = 0;
815         struct conv_attrs ca;
816
817         convert_attrs(&ca, path);
818         if (ca.drv) {
819                 filter = ca.drv->smudge;
820                 required = ca.drv->required;
821         }
822
823         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
824         if (ret) {
825                 src = dst->buf;
826                 len = dst->len;
827         }
828         /*
829          * CRLF conversion can be skipped if normalizing, unless there
830          * is a smudge filter.  The filter might expect CRLFs.
831          */
832         if (filter || !normalizing) {
833                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
834                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
835                 if (ret) {
836                         src = dst->buf;
837                         len = dst->len;
838                 }
839         }
840
841         ret_filter = apply_filter(path, src, len, dst, filter);
842         if (!ret_filter && required)
843                 die("%s: smudge filter %s failed", path, ca.drv->name);
844
845         return ret | ret_filter;
846 }
847
848 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
849 {
850         return convert_to_working_tree_internal(path, src, len, dst, 0);
851 }
852
853 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
854 {
855         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
856         if (ret) {
857                 src = dst->buf;
858                 len = dst->len;
859         }
860         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
861 }
862
863 /*****************************************************************
864  *
865  * Streaming converison support
866  *
867  *****************************************************************/
868
869 typedef int (*filter_fn)(struct stream_filter *,
870                          const char *input, size_t *isize_p,
871                          char *output, size_t *osize_p);
872 typedef void (*free_fn)(struct stream_filter *);
873
874 struct stream_filter_vtbl {
875         filter_fn filter;
876         free_fn free;
877 };
878
879 struct stream_filter {
880         struct stream_filter_vtbl *vtbl;
881 };
882
883 static int null_filter_fn(struct stream_filter *filter,
884                           const char *input, size_t *isize_p,
885                           char *output, size_t *osize_p)
886 {
887         size_t count;
888
889         if (!input)
890                 return 0; /* we do not keep any states */
891         count = *isize_p;
892         if (*osize_p < count)
893                 count = *osize_p;
894         if (count) {
895                 memmove(output, input, count);
896                 *isize_p -= count;
897                 *osize_p -= count;
898         }
899         return 0;
900 }
901
902 static void null_free_fn(struct stream_filter *filter)
903 {
904         ; /* nothing -- null instances are shared */
905 }
906
907 static struct stream_filter_vtbl null_vtbl = {
908         null_filter_fn,
909         null_free_fn,
910 };
911
912 static struct stream_filter null_filter_singleton = {
913         &null_vtbl,
914 };
915
916 int is_null_stream_filter(struct stream_filter *filter)
917 {
918         return filter == &null_filter_singleton;
919 }
920
921
922 /*
923  * LF-to-CRLF filter
924  */
925
926 struct lf_to_crlf_filter {
927         struct stream_filter filter;
928         unsigned has_held:1;
929         char held;
930 };
931
932 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
933                                 const char *input, size_t *isize_p,
934                                 char *output, size_t *osize_p)
935 {
936         size_t count, o = 0;
937         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
938
939         /*
940          * We may be holding onto the CR to see if it is followed by a
941          * LF, in which case we would need to go to the main loop.
942          * Otherwise, just emit it to the output stream.
943          */
944         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
945                 output[o++] = lf_to_crlf->held;
946                 lf_to_crlf->has_held = 0;
947         }
948
949         /* We are told to drain */
950         if (!input) {
951                 *osize_p -= o;
952                 return 0;
953         }
954
955         count = *isize_p;
956         if (count || lf_to_crlf->has_held) {
957                 size_t i;
958                 int was_cr = 0;
959
960                 if (lf_to_crlf->has_held) {
961                         was_cr = 1;
962                         lf_to_crlf->has_held = 0;
963                 }
964
965                 for (i = 0; o < *osize_p && i < count; i++) {
966                         char ch = input[i];
967
968                         if (ch == '\n') {
969                                 output[o++] = '\r';
970                         } else if (was_cr) {
971                                 /*
972                                  * Previous round saw CR and it is not followed
973                                  * by a LF; emit the CR before processing the
974                                  * current character.
975                                  */
976                                 output[o++] = '\r';
977                         }
978
979                         /*
980                          * We may have consumed the last output slot,
981                          * in which case we need to break out of this
982                          * loop; hold the current character before
983                          * returning.
984                          */
985                         if (*osize_p <= o) {
986                                 lf_to_crlf->has_held = 1;
987                                 lf_to_crlf->held = ch;
988                                 continue; /* break but increment i */
989                         }
990
991                         if (ch == '\r') {
992                                 was_cr = 1;
993                                 continue;
994                         }
995
996                         was_cr = 0;
997                         output[o++] = ch;
998                 }
999
1000                 *osize_p -= o;
1001                 *isize_p -= i;
1002
1003                 if (!lf_to_crlf->has_held && was_cr) {
1004                         lf_to_crlf->has_held = 1;
1005                         lf_to_crlf->held = '\r';
1006                 }
1007         }
1008         return 0;
1009 }
1010
1011 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1012 {
1013         free(filter);
1014 }
1015
1016 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1017         lf_to_crlf_filter_fn,
1018         lf_to_crlf_free_fn,
1019 };
1020
1021 static struct stream_filter *lf_to_crlf_filter(void)
1022 {
1023         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1024
1025         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1026         return (struct stream_filter *)lf_to_crlf;
1027 }
1028
1029 /*
1030  * Cascade filter
1031  */
1032 #define FILTER_BUFFER 1024
1033 struct cascade_filter {
1034         struct stream_filter filter;
1035         struct stream_filter *one;
1036         struct stream_filter *two;
1037         char buf[FILTER_BUFFER];
1038         int end, ptr;
1039 };
1040
1041 static int cascade_filter_fn(struct stream_filter *filter,
1042                              const char *input, size_t *isize_p,
1043                              char *output, size_t *osize_p)
1044 {
1045         struct cascade_filter *cas = (struct cascade_filter *) filter;
1046         size_t filled = 0;
1047         size_t sz = *osize_p;
1048         size_t to_feed, remaining;
1049
1050         /*
1051          * input -- (one) --> buf -- (two) --> output
1052          */
1053         while (filled < sz) {
1054                 remaining = sz - filled;
1055
1056                 /* do we already have something to feed two with? */
1057                 if (cas->ptr < cas->end) {
1058                         to_feed = cas->end - cas->ptr;
1059                         if (stream_filter(cas->two,
1060                                           cas->buf + cas->ptr, &to_feed,
1061                                           output + filled, &remaining))
1062                                 return -1;
1063                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1064                         filled = sz - remaining;
1065                         continue;
1066                 }
1067
1068                 /* feed one from upstream and have it emit into our buffer */
1069                 to_feed = input ? *isize_p : 0;
1070                 if (input && !to_feed)
1071                         break;
1072                 remaining = sizeof(cas->buf);
1073                 if (stream_filter(cas->one,
1074                                   input, &to_feed,
1075                                   cas->buf, &remaining))
1076                         return -1;
1077                 cas->end = sizeof(cas->buf) - remaining;
1078                 cas->ptr = 0;
1079                 if (input) {
1080                         size_t fed = *isize_p - to_feed;
1081                         *isize_p -= fed;
1082                         input += fed;
1083                 }
1084
1085                 /* do we know that we drained one completely? */
1086                 if (input || cas->end)
1087                         continue;
1088
1089                 /* tell two to drain; we have nothing more to give it */
1090                 to_feed = 0;
1091                 remaining = sz - filled;
1092                 if (stream_filter(cas->two,
1093                                   NULL, &to_feed,
1094                                   output + filled, &remaining))
1095                         return -1;
1096                 if (remaining == (sz - filled))
1097                         break; /* completely drained two */
1098                 filled = sz - remaining;
1099         }
1100         *osize_p -= filled;
1101         return 0;
1102 }
1103
1104 static void cascade_free_fn(struct stream_filter *filter)
1105 {
1106         struct cascade_filter *cas = (struct cascade_filter *)filter;
1107         free_stream_filter(cas->one);
1108         free_stream_filter(cas->two);
1109         free(filter);
1110 }
1111
1112 static struct stream_filter_vtbl cascade_vtbl = {
1113         cascade_filter_fn,
1114         cascade_free_fn,
1115 };
1116
1117 static struct stream_filter *cascade_filter(struct stream_filter *one,
1118                                             struct stream_filter *two)
1119 {
1120         struct cascade_filter *cascade;
1121
1122         if (!one || is_null_stream_filter(one))
1123                 return two;
1124         if (!two || is_null_stream_filter(two))
1125                 return one;
1126
1127         cascade = xmalloc(sizeof(*cascade));
1128         cascade->one = one;
1129         cascade->two = two;
1130         cascade->end = cascade->ptr = 0;
1131         cascade->filter.vtbl = &cascade_vtbl;
1132         return (struct stream_filter *)cascade;
1133 }
1134
1135 /*
1136  * ident filter
1137  */
1138 #define IDENT_DRAINING (-1)
1139 #define IDENT_SKIPPING (-2)
1140 struct ident_filter {
1141         struct stream_filter filter;
1142         struct strbuf left;
1143         int state;
1144         char ident[45]; /* ": x40 $" */
1145 };
1146
1147 static int is_foreign_ident(const char *str)
1148 {
1149         int i;
1150
1151         if (prefixcmp(str, "$Id: "))
1152                 return 0;
1153         for (i = 5; str[i]; i++) {
1154                 if (isspace(str[i]) && str[i+1] != '$')
1155                         return 1;
1156         }
1157         return 0;
1158 }
1159
1160 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1161 {
1162         size_t to_drain = ident->left.len;
1163
1164         if (*osize_p < to_drain)
1165                 to_drain = *osize_p;
1166         if (to_drain) {
1167                 memcpy(*output_p, ident->left.buf, to_drain);
1168                 strbuf_remove(&ident->left, 0, to_drain);
1169                 *output_p += to_drain;
1170                 *osize_p -= to_drain;
1171         }
1172         if (!ident->left.len)
1173                 ident->state = 0;
1174 }
1175
1176 static int ident_filter_fn(struct stream_filter *filter,
1177                            const char *input, size_t *isize_p,
1178                            char *output, size_t *osize_p)
1179 {
1180         struct ident_filter *ident = (struct ident_filter *)filter;
1181         static const char head[] = "$Id";
1182
1183         if (!input) {
1184                 /* drain upon eof */
1185                 switch (ident->state) {
1186                 default:
1187                         strbuf_add(&ident->left, head, ident->state);
1188                 case IDENT_SKIPPING:
1189                         /* fallthru */
1190                 case IDENT_DRAINING:
1191                         ident_drain(ident, &output, osize_p);
1192                 }
1193                 return 0;
1194         }
1195
1196         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1197                 int ch;
1198
1199                 if (ident->state == IDENT_DRAINING) {
1200                         ident_drain(ident, &output, osize_p);
1201                         if (!*osize_p)
1202                                 break;
1203                         continue;
1204                 }
1205
1206                 ch = *(input++);
1207                 (*isize_p)--;
1208
1209                 if (ident->state == IDENT_SKIPPING) {
1210                         /*
1211                          * Skipping until '$' or LF, but keeping them
1212                          * in case it is a foreign ident.
1213                          */
1214                         strbuf_addch(&ident->left, ch);
1215                         if (ch != '\n' && ch != '$')
1216                                 continue;
1217                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1218                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1219                                 strbuf_addstr(&ident->left, ident->ident);
1220                         }
1221                         ident->state = IDENT_DRAINING;
1222                         continue;
1223                 }
1224
1225                 if (ident->state < sizeof(head) &&
1226                     head[ident->state] == ch) {
1227                         ident->state++;
1228                         continue;
1229                 }
1230
1231                 if (ident->state)
1232                         strbuf_add(&ident->left, head, ident->state);
1233                 if (ident->state == sizeof(head) - 1) {
1234                         if (ch != ':' && ch != '$') {
1235                                 strbuf_addch(&ident->left, ch);
1236                                 ident->state = 0;
1237                                 continue;
1238                         }
1239
1240                         if (ch == ':') {
1241                                 strbuf_addch(&ident->left, ch);
1242                                 ident->state = IDENT_SKIPPING;
1243                         } else {
1244                                 strbuf_addstr(&ident->left, ident->ident);
1245                                 ident->state = IDENT_DRAINING;
1246                         }
1247                         continue;
1248                 }
1249
1250                 strbuf_addch(&ident->left, ch);
1251                 ident->state = IDENT_DRAINING;
1252         }
1253         return 0;
1254 }
1255
1256 static void ident_free_fn(struct stream_filter *filter)
1257 {
1258         struct ident_filter *ident = (struct ident_filter *)filter;
1259         strbuf_release(&ident->left);
1260         free(filter);
1261 }
1262
1263 static struct stream_filter_vtbl ident_vtbl = {
1264         ident_filter_fn,
1265         ident_free_fn,
1266 };
1267
1268 static struct stream_filter *ident_filter(const unsigned char *sha1)
1269 {
1270         struct ident_filter *ident = xmalloc(sizeof(*ident));
1271
1272         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1273         strbuf_init(&ident->left, 0);
1274         ident->filter.vtbl = &ident_vtbl;
1275         ident->state = 0;
1276         return (struct stream_filter *)ident;
1277 }
1278
1279 /*
1280  * Return an appropriately constructed filter for the path, or NULL if
1281  * the contents cannot be filtered without reading the whole thing
1282  * in-core.
1283  *
1284  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1285  * large binary blob you would want us not to slurp into the memory!
1286  */
1287 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1288 {
1289         struct conv_attrs ca;
1290         enum crlf_action crlf_action;
1291         struct stream_filter *filter = NULL;
1292
1293         convert_attrs(&ca, path);
1294
1295         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1296                 return filter;
1297
1298         if (ca.ident)
1299                 filter = ident_filter(sha1);
1300
1301         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1302
1303         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1304             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1305                 filter = cascade_filter(filter, &null_filter_singleton);
1306
1307         else if (output_eol(crlf_action) == EOL_CRLF &&
1308                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1309                 filter = cascade_filter(filter, lf_to_crlf_filter());
1310
1311         return filter;
1312 }
1313
1314 void free_stream_filter(struct stream_filter *filter)
1315 {
1316         filter->vtbl->free(filter);
1317 }
1318
1319 int stream_filter(struct stream_filter *filter,
1320                   const char *input, size_t *isize_p,
1321                   char *output, size_t *osize_p)
1322 {
1323         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1324 }