replace: add a --raw mode for --edit
[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;
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         memset(&child_process, 0, sizeof(child_process));
348         child_process.argv = argv;
349         child_process.use_shell = 1;
350         child_process.in = -1;
351         child_process.out = out;
352
353         if (start_command(&child_process))
354                 return error("cannot fork to run external filter %s", params->cmd);
355
356         sigchain_push(SIGPIPE, SIG_IGN);
357
358         write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
359         if (close(child_process.in))
360                 write_err = 1;
361         if (write_err)
362                 error("cannot feed the input to external filter %s", params->cmd);
363
364         sigchain_pop(SIGPIPE);
365
366         status = finish_command(&child_process);
367         if (status)
368                 error("external filter %s failed %d", params->cmd, status);
369
370         strbuf_release(&cmd);
371         return (write_err || status);
372 }
373
374 static int apply_filter(const char *path, const char *src, size_t len,
375                         struct strbuf *dst, const char *cmd)
376 {
377         /*
378          * Create a pipeline to have the command filter the buffer's
379          * contents.
380          *
381          * (child --> cmd) --> us
382          */
383         int ret = 1;
384         struct strbuf nbuf = STRBUF_INIT;
385         struct async async;
386         struct filter_params params;
387
388         if (!cmd)
389                 return 0;
390
391         if (!dst)
392                 return 1;
393
394         memset(&async, 0, sizeof(async));
395         async.proc = filter_buffer;
396         async.data = &params;
397         async.out = -1;
398         params.src = src;
399         params.size = len;
400         params.cmd = cmd;
401         params.path = path;
402
403         fflush(NULL);
404         if (start_async(&async))
405                 return 0;       /* error was already reported */
406
407         if (strbuf_read(&nbuf, async.out, len) < 0) {
408                 error("read from external filter %s failed", cmd);
409                 ret = 0;
410         }
411         if (close(async.out)) {
412                 error("read from external filter %s failed", cmd);
413                 ret = 0;
414         }
415         if (finish_async(&async)) {
416                 error("external filter %s failed", cmd);
417                 ret = 0;
418         }
419
420         if (ret) {
421                 strbuf_swap(dst, &nbuf);
422         }
423         strbuf_release(&nbuf);
424         return ret;
425 }
426
427 static struct convert_driver {
428         const char *name;
429         struct convert_driver *next;
430         const char *smudge;
431         const char *clean;
432         int required;
433 } *user_convert, **user_convert_tail;
434
435 static int read_convert_config(const char *var, const char *value, void *cb)
436 {
437         const char *key, *name;
438         int namelen;
439         struct convert_driver *drv;
440
441         /*
442          * External conversion drivers are configured using
443          * "filter.<name>.variable".
444          */
445         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
446                 return 0;
447         for (drv = user_convert; drv; drv = drv->next)
448                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
449                         break;
450         if (!drv) {
451                 drv = xcalloc(1, sizeof(struct convert_driver));
452                 drv->name = xmemdupz(name, namelen);
453                 *user_convert_tail = drv;
454                 user_convert_tail = &(drv->next);
455         }
456
457         /*
458          * filter.<name>.smudge and filter.<name>.clean specifies
459          * the command line:
460          *
461          *      command-line
462          *
463          * The command-line will not be interpolated in any way.
464          */
465
466         if (!strcmp("smudge", key))
467                 return git_config_string(&drv->smudge, var, value);
468
469         if (!strcmp("clean", key))
470                 return git_config_string(&drv->clean, var, value);
471
472         if (!strcmp("required", key)) {
473                 drv->required = git_config_bool(var, value);
474                 return 0;
475         }
476
477         return 0;
478 }
479
480 static int count_ident(const char *cp, unsigned long size)
481 {
482         /*
483          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
484          */
485         int cnt = 0;
486         char ch;
487
488         while (size) {
489                 ch = *cp++;
490                 size--;
491                 if (ch != '$')
492                         continue;
493                 if (size < 3)
494                         break;
495                 if (memcmp("Id", cp, 2))
496                         continue;
497                 ch = cp[2];
498                 cp += 3;
499                 size -= 3;
500                 if (ch == '$')
501                         cnt++; /* $Id$ */
502                 if (ch != ':')
503                         continue;
504
505                 /*
506                  * "$Id: ... "; scan up to the closing dollar sign and discard.
507                  */
508                 while (size) {
509                         ch = *cp++;
510                         size--;
511                         if (ch == '$') {
512                                 cnt++;
513                                 break;
514                         }
515                         if (ch == '\n')
516                                 break;
517                 }
518         }
519         return cnt;
520 }
521
522 static int ident_to_git(const char *path, const char *src, size_t len,
523                         struct strbuf *buf, int ident)
524 {
525         char *dst, *dollar;
526
527         if (!ident || (src && !count_ident(src, len)))
528                 return 0;
529
530         if (!buf)
531                 return 1;
532
533         /* only grow if not in place */
534         if (strbuf_avail(buf) + buf->len < len)
535                 strbuf_grow(buf, len - buf->len);
536         dst = buf->buf;
537         for (;;) {
538                 dollar = memchr(src, '$', len);
539                 if (!dollar)
540                         break;
541                 memmove(dst, src, dollar + 1 - src);
542                 dst += dollar + 1 - src;
543                 len -= dollar + 1 - src;
544                 src  = dollar + 1;
545
546                 if (len > 3 && !memcmp(src, "Id:", 3)) {
547                         dollar = memchr(src + 3, '$', len - 3);
548                         if (!dollar)
549                                 break;
550                         if (memchr(src + 3, '\n', dollar - src - 3)) {
551                                 /* Line break before the next dollar. */
552                                 continue;
553                         }
554
555                         memcpy(dst, "Id$", 3);
556                         dst += 3;
557                         len -= dollar + 1 - src;
558                         src  = dollar + 1;
559                 }
560         }
561         memmove(dst, src, len);
562         strbuf_setlen(buf, dst + len - buf->buf);
563         return 1;
564 }
565
566 static int ident_to_worktree(const char *path, const char *src, size_t len,
567                              struct strbuf *buf, int ident)
568 {
569         unsigned char sha1[20];
570         char *to_free = NULL, *dollar, *spc;
571         int cnt;
572
573         if (!ident)
574                 return 0;
575
576         cnt = count_ident(src, len);
577         if (!cnt)
578                 return 0;
579
580         /* are we "faking" in place editing ? */
581         if (src == buf->buf)
582                 to_free = strbuf_detach(buf, NULL);
583         hash_sha1_file(src, len, "blob", sha1);
584
585         strbuf_grow(buf, len + cnt * 43);
586         for (;;) {
587                 /* step 1: run to the next '$' */
588                 dollar = memchr(src, '$', len);
589                 if (!dollar)
590                         break;
591                 strbuf_add(buf, src, dollar + 1 - src);
592                 len -= dollar + 1 - src;
593                 src  = dollar + 1;
594
595                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
596                 if (len < 3 || memcmp("Id", src, 2))
597                         continue;
598
599                 /* step 3: skip over Id$ or Id:xxxxx$ */
600                 if (src[2] == '$') {
601                         src += 3;
602                         len -= 3;
603                 } else if (src[2] == ':') {
604                         /*
605                          * It's possible that an expanded Id has crept its way into the
606                          * repository, we cope with that by stripping the expansion out.
607                          * This is probably not a good idea, since it will cause changes
608                          * on checkout, which won't go away by stash, but let's keep it
609                          * for git-style ids.
610                          */
611                         dollar = memchr(src + 3, '$', len - 3);
612                         if (!dollar) {
613                                 /* incomplete keyword, no more '$', so just quit the loop */
614                                 break;
615                         }
616
617                         if (memchr(src + 3, '\n', dollar - src - 3)) {
618                                 /* Line break before the next dollar. */
619                                 continue;
620                         }
621
622                         spc = memchr(src + 4, ' ', dollar - src - 4);
623                         if (spc && spc < dollar-1) {
624                                 /* There are spaces in unexpected places.
625                                  * This is probably an id from some other
626                                  * versioning system. Keep it for now.
627                                  */
628                                 continue;
629                         }
630
631                         len -= dollar + 1 - src;
632                         src  = dollar + 1;
633                 } else {
634                         /* it wasn't a "Id$" or "Id:xxxx$" */
635                         continue;
636                 }
637
638                 /* step 4: substitute */
639                 strbuf_addstr(buf, "Id: ");
640                 strbuf_add(buf, sha1_to_hex(sha1), 40);
641                 strbuf_addstr(buf, " $");
642         }
643         strbuf_add(buf, src, len);
644
645         free(to_free);
646         return 1;
647 }
648
649 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
650 {
651         const char *value = check->value;
652
653         if (ATTR_TRUE(value))
654                 return CRLF_TEXT;
655         else if (ATTR_FALSE(value))
656                 return CRLF_BINARY;
657         else if (ATTR_UNSET(value))
658                 ;
659         else if (!strcmp(value, "input"))
660                 return CRLF_INPUT;
661         else if (!strcmp(value, "auto"))
662                 return CRLF_AUTO;
663         return CRLF_GUESS;
664 }
665
666 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
667 {
668         const char *value = check->value;
669
670         if (ATTR_UNSET(value))
671                 ;
672         else if (!strcmp(value, "lf"))
673                 return EOL_LF;
674         else if (!strcmp(value, "crlf"))
675                 return EOL_CRLF;
676         return EOL_UNSET;
677 }
678
679 static struct convert_driver *git_path_check_convert(const char *path,
680                                              struct git_attr_check *check)
681 {
682         const char *value = check->value;
683         struct convert_driver *drv;
684
685         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
686                 return NULL;
687         for (drv = user_convert; drv; drv = drv->next)
688                 if (!strcmp(value, drv->name))
689                         return drv;
690         return NULL;
691 }
692
693 static int git_path_check_ident(const char *path, struct git_attr_check *check)
694 {
695         const char *value = check->value;
696
697         return !!ATTR_TRUE(value);
698 }
699
700 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
701 {
702         if (text_attr == CRLF_BINARY)
703                 return CRLF_BINARY;
704         if (eol_attr == EOL_LF)
705                 return CRLF_INPUT;
706         if (eol_attr == EOL_CRLF)
707                 return CRLF_CRLF;
708         return text_attr;
709 }
710
711 struct conv_attrs {
712         struct convert_driver *drv;
713         enum crlf_action crlf_action;
714         enum eol eol_attr;
715         int ident;
716 };
717
718 static const char *conv_attr_name[] = {
719         "crlf", "ident", "filter", "eol", "text",
720 };
721 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
722
723 static void convert_attrs(struct conv_attrs *ca, const char *path)
724 {
725         int i;
726         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
727
728         if (!ccheck[0].attr) {
729                 for (i = 0; i < NUM_CONV_ATTRS; i++)
730                         ccheck[i].attr = git_attr(conv_attr_name[i]);
731                 user_convert_tail = &user_convert;
732                 git_config(read_convert_config, NULL);
733         }
734
735         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
736                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
737                 if (ca->crlf_action == CRLF_GUESS)
738                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
739                 ca->ident = git_path_check_ident(path, ccheck + 1);
740                 ca->drv = git_path_check_convert(path, ccheck + 2);
741                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
742         } else {
743                 ca->drv = NULL;
744                 ca->crlf_action = CRLF_GUESS;
745                 ca->eol_attr = EOL_UNSET;
746                 ca->ident = 0;
747         }
748 }
749
750 int convert_to_git(const char *path, const char *src, size_t len,
751                    struct strbuf *dst, enum safe_crlf checksafe)
752 {
753         int ret = 0;
754         const char *filter = NULL;
755         int required = 0;
756         struct conv_attrs ca;
757
758         convert_attrs(&ca, path);
759         if (ca.drv) {
760                 filter = ca.drv->clean;
761                 required = ca.drv->required;
762         }
763
764         ret |= apply_filter(path, src, len, dst, filter);
765         if (!ret && required)
766                 die("%s: clean filter '%s' failed", path, ca.drv->name);
767
768         if (ret && dst) {
769                 src = dst->buf;
770                 len = dst->len;
771         }
772         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
773         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
774         if (ret && dst) {
775                 src = dst->buf;
776                 len = dst->len;
777         }
778         return ret | ident_to_git(path, src, len, dst, ca.ident);
779 }
780
781 static int convert_to_working_tree_internal(const char *path, const char *src,
782                                             size_t len, struct strbuf *dst,
783                                             int normalizing)
784 {
785         int ret = 0, ret_filter = 0;
786         const char *filter = NULL;
787         int required = 0;
788         struct conv_attrs ca;
789
790         convert_attrs(&ca, path);
791         if (ca.drv) {
792                 filter = ca.drv->smudge;
793                 required = ca.drv->required;
794         }
795
796         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
797         if (ret) {
798                 src = dst->buf;
799                 len = dst->len;
800         }
801         /*
802          * CRLF conversion can be skipped if normalizing, unless there
803          * is a smudge filter.  The filter might expect CRLFs.
804          */
805         if (filter || !normalizing) {
806                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
807                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
808                 if (ret) {
809                         src = dst->buf;
810                         len = dst->len;
811                 }
812         }
813
814         ret_filter = apply_filter(path, src, len, dst, filter);
815         if (!ret_filter && required)
816                 die("%s: smudge filter %s failed", path, ca.drv->name);
817
818         return ret | ret_filter;
819 }
820
821 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
822 {
823         return convert_to_working_tree_internal(path, src, len, dst, 0);
824 }
825
826 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
827 {
828         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
829         if (ret) {
830                 src = dst->buf;
831                 len = dst->len;
832         }
833         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
834 }
835
836 /*****************************************************************
837  *
838  * Streaming conversion support
839  *
840  *****************************************************************/
841
842 typedef int (*filter_fn)(struct stream_filter *,
843                          const char *input, size_t *isize_p,
844                          char *output, size_t *osize_p);
845 typedef void (*free_fn)(struct stream_filter *);
846
847 struct stream_filter_vtbl {
848         filter_fn filter;
849         free_fn free;
850 };
851
852 struct stream_filter {
853         struct stream_filter_vtbl *vtbl;
854 };
855
856 static int null_filter_fn(struct stream_filter *filter,
857                           const char *input, size_t *isize_p,
858                           char *output, size_t *osize_p)
859 {
860         size_t count;
861
862         if (!input)
863                 return 0; /* we do not keep any states */
864         count = *isize_p;
865         if (*osize_p < count)
866                 count = *osize_p;
867         if (count) {
868                 memmove(output, input, count);
869                 *isize_p -= count;
870                 *osize_p -= count;
871         }
872         return 0;
873 }
874
875 static void null_free_fn(struct stream_filter *filter)
876 {
877         ; /* nothing -- null instances are shared */
878 }
879
880 static struct stream_filter_vtbl null_vtbl = {
881         null_filter_fn,
882         null_free_fn,
883 };
884
885 static struct stream_filter null_filter_singleton = {
886         &null_vtbl,
887 };
888
889 int is_null_stream_filter(struct stream_filter *filter)
890 {
891         return filter == &null_filter_singleton;
892 }
893
894
895 /*
896  * LF-to-CRLF filter
897  */
898
899 struct lf_to_crlf_filter {
900         struct stream_filter filter;
901         unsigned has_held:1;
902         char held;
903 };
904
905 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
906                                 const char *input, size_t *isize_p,
907                                 char *output, size_t *osize_p)
908 {
909         size_t count, o = 0;
910         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
911
912         /*
913          * We may be holding onto the CR to see if it is followed by a
914          * LF, in which case we would need to go to the main loop.
915          * Otherwise, just emit it to the output stream.
916          */
917         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
918                 output[o++] = lf_to_crlf->held;
919                 lf_to_crlf->has_held = 0;
920         }
921
922         /* We are told to drain */
923         if (!input) {
924                 *osize_p -= o;
925                 return 0;
926         }
927
928         count = *isize_p;
929         if (count || lf_to_crlf->has_held) {
930                 size_t i;
931                 int was_cr = 0;
932
933                 if (lf_to_crlf->has_held) {
934                         was_cr = 1;
935                         lf_to_crlf->has_held = 0;
936                 }
937
938                 for (i = 0; o < *osize_p && i < count; i++) {
939                         char ch = input[i];
940
941                         if (ch == '\n') {
942                                 output[o++] = '\r';
943                         } else if (was_cr) {
944                                 /*
945                                  * Previous round saw CR and it is not followed
946                                  * by a LF; emit the CR before processing the
947                                  * current character.
948                                  */
949                                 output[o++] = '\r';
950                         }
951
952                         /*
953                          * We may have consumed the last output slot,
954                          * in which case we need to break out of this
955                          * loop; hold the current character before
956                          * returning.
957                          */
958                         if (*osize_p <= o) {
959                                 lf_to_crlf->has_held = 1;
960                                 lf_to_crlf->held = ch;
961                                 continue; /* break but increment i */
962                         }
963
964                         if (ch == '\r') {
965                                 was_cr = 1;
966                                 continue;
967                         }
968
969                         was_cr = 0;
970                         output[o++] = ch;
971                 }
972
973                 *osize_p -= o;
974                 *isize_p -= i;
975
976                 if (!lf_to_crlf->has_held && was_cr) {
977                         lf_to_crlf->has_held = 1;
978                         lf_to_crlf->held = '\r';
979                 }
980         }
981         return 0;
982 }
983
984 static void lf_to_crlf_free_fn(struct stream_filter *filter)
985 {
986         free(filter);
987 }
988
989 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
990         lf_to_crlf_filter_fn,
991         lf_to_crlf_free_fn,
992 };
993
994 static struct stream_filter *lf_to_crlf_filter(void)
995 {
996         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
997
998         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
999         return (struct stream_filter *)lf_to_crlf;
1000 }
1001
1002 /*
1003  * Cascade filter
1004  */
1005 #define FILTER_BUFFER 1024
1006 struct cascade_filter {
1007         struct stream_filter filter;
1008         struct stream_filter *one;
1009         struct stream_filter *two;
1010         char buf[FILTER_BUFFER];
1011         int end, ptr;
1012 };
1013
1014 static int cascade_filter_fn(struct stream_filter *filter,
1015                              const char *input, size_t *isize_p,
1016                              char *output, size_t *osize_p)
1017 {
1018         struct cascade_filter *cas = (struct cascade_filter *) filter;
1019         size_t filled = 0;
1020         size_t sz = *osize_p;
1021         size_t to_feed, remaining;
1022
1023         /*
1024          * input -- (one) --> buf -- (two) --> output
1025          */
1026         while (filled < sz) {
1027                 remaining = sz - filled;
1028
1029                 /* do we already have something to feed two with? */
1030                 if (cas->ptr < cas->end) {
1031                         to_feed = cas->end - cas->ptr;
1032                         if (stream_filter(cas->two,
1033                                           cas->buf + cas->ptr, &to_feed,
1034                                           output + filled, &remaining))
1035                                 return -1;
1036                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1037                         filled = sz - remaining;
1038                         continue;
1039                 }
1040
1041                 /* feed one from upstream and have it emit into our buffer */
1042                 to_feed = input ? *isize_p : 0;
1043                 if (input && !to_feed)
1044                         break;
1045                 remaining = sizeof(cas->buf);
1046                 if (stream_filter(cas->one,
1047                                   input, &to_feed,
1048                                   cas->buf, &remaining))
1049                         return -1;
1050                 cas->end = sizeof(cas->buf) - remaining;
1051                 cas->ptr = 0;
1052                 if (input) {
1053                         size_t fed = *isize_p - to_feed;
1054                         *isize_p -= fed;
1055                         input += fed;
1056                 }
1057
1058                 /* do we know that we drained one completely? */
1059                 if (input || cas->end)
1060                         continue;
1061
1062                 /* tell two to drain; we have nothing more to give it */
1063                 to_feed = 0;
1064                 remaining = sz - filled;
1065                 if (stream_filter(cas->two,
1066                                   NULL, &to_feed,
1067                                   output + filled, &remaining))
1068                         return -1;
1069                 if (remaining == (sz - filled))
1070                         break; /* completely drained two */
1071                 filled = sz - remaining;
1072         }
1073         *osize_p -= filled;
1074         return 0;
1075 }
1076
1077 static void cascade_free_fn(struct stream_filter *filter)
1078 {
1079         struct cascade_filter *cas = (struct cascade_filter *)filter;
1080         free_stream_filter(cas->one);
1081         free_stream_filter(cas->two);
1082         free(filter);
1083 }
1084
1085 static struct stream_filter_vtbl cascade_vtbl = {
1086         cascade_filter_fn,
1087         cascade_free_fn,
1088 };
1089
1090 static struct stream_filter *cascade_filter(struct stream_filter *one,
1091                                             struct stream_filter *two)
1092 {
1093         struct cascade_filter *cascade;
1094
1095         if (!one || is_null_stream_filter(one))
1096                 return two;
1097         if (!two || is_null_stream_filter(two))
1098                 return one;
1099
1100         cascade = xmalloc(sizeof(*cascade));
1101         cascade->one = one;
1102         cascade->two = two;
1103         cascade->end = cascade->ptr = 0;
1104         cascade->filter.vtbl = &cascade_vtbl;
1105         return (struct stream_filter *)cascade;
1106 }
1107
1108 /*
1109  * ident filter
1110  */
1111 #define IDENT_DRAINING (-1)
1112 #define IDENT_SKIPPING (-2)
1113 struct ident_filter {
1114         struct stream_filter filter;
1115         struct strbuf left;
1116         int state;
1117         char ident[45]; /* ": x40 $" */
1118 };
1119
1120 static int is_foreign_ident(const char *str)
1121 {
1122         int i;
1123
1124         if (!starts_with(str, "$Id: "))
1125                 return 0;
1126         for (i = 5; str[i]; i++) {
1127                 if (isspace(str[i]) && str[i+1] != '$')
1128                         return 1;
1129         }
1130         return 0;
1131 }
1132
1133 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1134 {
1135         size_t to_drain = ident->left.len;
1136
1137         if (*osize_p < to_drain)
1138                 to_drain = *osize_p;
1139         if (to_drain) {
1140                 memcpy(*output_p, ident->left.buf, to_drain);
1141                 strbuf_remove(&ident->left, 0, to_drain);
1142                 *output_p += to_drain;
1143                 *osize_p -= to_drain;
1144         }
1145         if (!ident->left.len)
1146                 ident->state = 0;
1147 }
1148
1149 static int ident_filter_fn(struct stream_filter *filter,
1150                            const char *input, size_t *isize_p,
1151                            char *output, size_t *osize_p)
1152 {
1153         struct ident_filter *ident = (struct ident_filter *)filter;
1154         static const char head[] = "$Id";
1155
1156         if (!input) {
1157                 /* drain upon eof */
1158                 switch (ident->state) {
1159                 default:
1160                         strbuf_add(&ident->left, head, ident->state);
1161                 case IDENT_SKIPPING:
1162                         /* fallthru */
1163                 case IDENT_DRAINING:
1164                         ident_drain(ident, &output, osize_p);
1165                 }
1166                 return 0;
1167         }
1168
1169         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1170                 int ch;
1171
1172                 if (ident->state == IDENT_DRAINING) {
1173                         ident_drain(ident, &output, osize_p);
1174                         if (!*osize_p)
1175                                 break;
1176                         continue;
1177                 }
1178
1179                 ch = *(input++);
1180                 (*isize_p)--;
1181
1182                 if (ident->state == IDENT_SKIPPING) {
1183                         /*
1184                          * Skipping until '$' or LF, but keeping them
1185                          * in case it is a foreign ident.
1186                          */
1187                         strbuf_addch(&ident->left, ch);
1188                         if (ch != '\n' && ch != '$')
1189                                 continue;
1190                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1191                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1192                                 strbuf_addstr(&ident->left, ident->ident);
1193                         }
1194                         ident->state = IDENT_DRAINING;
1195                         continue;
1196                 }
1197
1198                 if (ident->state < sizeof(head) &&
1199                     head[ident->state] == ch) {
1200                         ident->state++;
1201                         continue;
1202                 }
1203
1204                 if (ident->state)
1205                         strbuf_add(&ident->left, head, ident->state);
1206                 if (ident->state == sizeof(head) - 1) {
1207                         if (ch != ':' && ch != '$') {
1208                                 strbuf_addch(&ident->left, ch);
1209                                 ident->state = 0;
1210                                 continue;
1211                         }
1212
1213                         if (ch == ':') {
1214                                 strbuf_addch(&ident->left, ch);
1215                                 ident->state = IDENT_SKIPPING;
1216                         } else {
1217                                 strbuf_addstr(&ident->left, ident->ident);
1218                                 ident->state = IDENT_DRAINING;
1219                         }
1220                         continue;
1221                 }
1222
1223                 strbuf_addch(&ident->left, ch);
1224                 ident->state = IDENT_DRAINING;
1225         }
1226         return 0;
1227 }
1228
1229 static void ident_free_fn(struct stream_filter *filter)
1230 {
1231         struct ident_filter *ident = (struct ident_filter *)filter;
1232         strbuf_release(&ident->left);
1233         free(filter);
1234 }
1235
1236 static struct stream_filter_vtbl ident_vtbl = {
1237         ident_filter_fn,
1238         ident_free_fn,
1239 };
1240
1241 static struct stream_filter *ident_filter(const unsigned char *sha1)
1242 {
1243         struct ident_filter *ident = xmalloc(sizeof(*ident));
1244
1245         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1246         strbuf_init(&ident->left, 0);
1247         ident->filter.vtbl = &ident_vtbl;
1248         ident->state = 0;
1249         return (struct stream_filter *)ident;
1250 }
1251
1252 /*
1253  * Return an appropriately constructed filter for the path, or NULL if
1254  * the contents cannot be filtered without reading the whole thing
1255  * in-core.
1256  *
1257  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1258  * large binary blob you would want us not to slurp into the memory!
1259  */
1260 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1261 {
1262         struct conv_attrs ca;
1263         enum crlf_action crlf_action;
1264         struct stream_filter *filter = NULL;
1265
1266         convert_attrs(&ca, path);
1267
1268         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1269                 return filter;
1270
1271         if (ca.ident)
1272                 filter = ident_filter(sha1);
1273
1274         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1275
1276         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1277             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1278                 filter = cascade_filter(filter, &null_filter_singleton);
1279
1280         else if (output_eol(crlf_action) == EOL_CRLF &&
1281                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1282                 filter = cascade_filter(filter, lf_to_crlf_filter());
1283
1284         return filter;
1285 }
1286
1287 void free_stream_filter(struct stream_filter *filter)
1288 {
1289         filter->vtbl->free(filter);
1290 }
1291
1292 int stream_filter(struct stream_filter *filter,
1293                   const char *input, size_t *isize_p,
1294                   char *output, size_t *osize_p)
1295 {
1296         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1297 }