Merge branch 'sp/stream-clean-filter'
[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         int fd;
316         const char *cmd;
317         const char *path;
318 };
319
320 static int filter_buffer_or_fd(int in, int out, void *data)
321 {
322         /*
323          * Spawn cmd and feed the buffer contents through its stdin.
324          */
325         struct child_process child_process = CHILD_PROCESS_INIT;
326         struct filter_params *params = (struct filter_params *)data;
327         int write_err, status;
328         const char *argv[] = { NULL, NULL };
329
330         /* apply % substitution to cmd */
331         struct strbuf cmd = STRBUF_INIT;
332         struct strbuf path = STRBUF_INIT;
333         struct strbuf_expand_dict_entry dict[] = {
334                 { "f", NULL, },
335                 { NULL, NULL, },
336         };
337
338         /* quote the path to preserve spaces, etc. */
339         sq_quote_buf(&path, params->path);
340         dict[0].value = path.buf;
341
342         /* expand all %f with the quoted path */
343         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
344         strbuf_release(&path);
345
346         argv[0] = cmd.buf;
347
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         if (params->src) {
359                 write_err = (write_in_full(child_process.in, params->src, params->size) < 0);
360         } else {
361                 write_err = copy_fd(params->fd, child_process.in);
362         }
363
364         if (close(child_process.in))
365                 write_err = 1;
366         if (write_err)
367                 error("cannot feed the input to external filter %s", params->cmd);
368
369         sigchain_pop(SIGPIPE);
370
371         status = finish_command(&child_process);
372         if (status)
373                 error("external filter %s failed %d", params->cmd, status);
374
375         strbuf_release(&cmd);
376         return (write_err || status);
377 }
378
379 static int apply_filter(const char *path, const char *src, size_t len, int fd,
380                         struct strbuf *dst, const char *cmd)
381 {
382         /*
383          * Create a pipeline to have the command filter the buffer's
384          * contents.
385          *
386          * (child --> cmd) --> us
387          */
388         int ret = 1;
389         struct strbuf nbuf = STRBUF_INIT;
390         struct async async;
391         struct filter_params params;
392
393         if (!cmd)
394                 return 0;
395
396         if (!dst)
397                 return 1;
398
399         memset(&async, 0, sizeof(async));
400         async.proc = filter_buffer_or_fd;
401         async.data = &params;
402         async.out = -1;
403         params.src = src;
404         params.size = len;
405         params.fd = fd;
406         params.cmd = cmd;
407         params.path = path;
408
409         fflush(NULL);
410         if (start_async(&async))
411                 return 0;       /* error was already reported */
412
413         if (strbuf_read(&nbuf, async.out, len) < 0) {
414                 error("read from external filter %s failed", cmd);
415                 ret = 0;
416         }
417         if (close(async.out)) {
418                 error("read from external filter %s failed", cmd);
419                 ret = 0;
420         }
421         if (finish_async(&async)) {
422                 error("external filter %s failed", cmd);
423                 ret = 0;
424         }
425
426         if (ret) {
427                 strbuf_swap(dst, &nbuf);
428         }
429         strbuf_release(&nbuf);
430         return ret;
431 }
432
433 static struct convert_driver {
434         const char *name;
435         struct convert_driver *next;
436         const char *smudge;
437         const char *clean;
438         int required;
439 } *user_convert, **user_convert_tail;
440
441 static int read_convert_config(const char *var, const char *value, void *cb)
442 {
443         const char *key, *name;
444         int namelen;
445         struct convert_driver *drv;
446
447         /*
448          * External conversion drivers are configured using
449          * "filter.<name>.variable".
450          */
451         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
452                 return 0;
453         for (drv = user_convert; drv; drv = drv->next)
454                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
455                         break;
456         if (!drv) {
457                 drv = xcalloc(1, sizeof(struct convert_driver));
458                 drv->name = xmemdupz(name, namelen);
459                 *user_convert_tail = drv;
460                 user_convert_tail = &(drv->next);
461         }
462
463         /*
464          * filter.<name>.smudge and filter.<name>.clean specifies
465          * the command line:
466          *
467          *      command-line
468          *
469          * The command-line will not be interpolated in any way.
470          */
471
472         if (!strcmp("smudge", key))
473                 return git_config_string(&drv->smudge, var, value);
474
475         if (!strcmp("clean", key))
476                 return git_config_string(&drv->clean, var, value);
477
478         if (!strcmp("required", key)) {
479                 drv->required = git_config_bool(var, value);
480                 return 0;
481         }
482
483         return 0;
484 }
485
486 static int count_ident(const char *cp, unsigned long size)
487 {
488         /*
489          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
490          */
491         int cnt = 0;
492         char ch;
493
494         while (size) {
495                 ch = *cp++;
496                 size--;
497                 if (ch != '$')
498                         continue;
499                 if (size < 3)
500                         break;
501                 if (memcmp("Id", cp, 2))
502                         continue;
503                 ch = cp[2];
504                 cp += 3;
505                 size -= 3;
506                 if (ch == '$')
507                         cnt++; /* $Id$ */
508                 if (ch != ':')
509                         continue;
510
511                 /*
512                  * "$Id: ... "; scan up to the closing dollar sign and discard.
513                  */
514                 while (size) {
515                         ch = *cp++;
516                         size--;
517                         if (ch == '$') {
518                                 cnt++;
519                                 break;
520                         }
521                         if (ch == '\n')
522                                 break;
523                 }
524         }
525         return cnt;
526 }
527
528 static int ident_to_git(const char *path, const char *src, size_t len,
529                         struct strbuf *buf, int ident)
530 {
531         char *dst, *dollar;
532
533         if (!ident || (src && !count_ident(src, len)))
534                 return 0;
535
536         if (!buf)
537                 return 1;
538
539         /* only grow if not in place */
540         if (strbuf_avail(buf) + buf->len < len)
541                 strbuf_grow(buf, len - buf->len);
542         dst = buf->buf;
543         for (;;) {
544                 dollar = memchr(src, '$', len);
545                 if (!dollar)
546                         break;
547                 memmove(dst, src, dollar + 1 - src);
548                 dst += dollar + 1 - src;
549                 len -= dollar + 1 - src;
550                 src  = dollar + 1;
551
552                 if (len > 3 && !memcmp(src, "Id:", 3)) {
553                         dollar = memchr(src + 3, '$', len - 3);
554                         if (!dollar)
555                                 break;
556                         if (memchr(src + 3, '\n', dollar - src - 3)) {
557                                 /* Line break before the next dollar. */
558                                 continue;
559                         }
560
561                         memcpy(dst, "Id$", 3);
562                         dst += 3;
563                         len -= dollar + 1 - src;
564                         src  = dollar + 1;
565                 }
566         }
567         memmove(dst, src, len);
568         strbuf_setlen(buf, dst + len - buf->buf);
569         return 1;
570 }
571
572 static int ident_to_worktree(const char *path, const char *src, size_t len,
573                              struct strbuf *buf, int ident)
574 {
575         unsigned char sha1[20];
576         char *to_free = NULL, *dollar, *spc;
577         int cnt;
578
579         if (!ident)
580                 return 0;
581
582         cnt = count_ident(src, len);
583         if (!cnt)
584                 return 0;
585
586         /* are we "faking" in place editing ? */
587         if (src == buf->buf)
588                 to_free = strbuf_detach(buf, NULL);
589         hash_sha1_file(src, len, "blob", sha1);
590
591         strbuf_grow(buf, len + cnt * 43);
592         for (;;) {
593                 /* step 1: run to the next '$' */
594                 dollar = memchr(src, '$', len);
595                 if (!dollar)
596                         break;
597                 strbuf_add(buf, src, dollar + 1 - src);
598                 len -= dollar + 1 - src;
599                 src  = dollar + 1;
600
601                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
602                 if (len < 3 || memcmp("Id", src, 2))
603                         continue;
604
605                 /* step 3: skip over Id$ or Id:xxxxx$ */
606                 if (src[2] == '$') {
607                         src += 3;
608                         len -= 3;
609                 } else if (src[2] == ':') {
610                         /*
611                          * It's possible that an expanded Id has crept its way into the
612                          * repository, we cope with that by stripping the expansion out.
613                          * This is probably not a good idea, since it will cause changes
614                          * on checkout, which won't go away by stash, but let's keep it
615                          * for git-style ids.
616                          */
617                         dollar = memchr(src + 3, '$', len - 3);
618                         if (!dollar) {
619                                 /* incomplete keyword, no more '$', so just quit the loop */
620                                 break;
621                         }
622
623                         if (memchr(src + 3, '\n', dollar - src - 3)) {
624                                 /* Line break before the next dollar. */
625                                 continue;
626                         }
627
628                         spc = memchr(src + 4, ' ', dollar - src - 4);
629                         if (spc && spc < dollar-1) {
630                                 /* There are spaces in unexpected places.
631                                  * This is probably an id from some other
632                                  * versioning system. Keep it for now.
633                                  */
634                                 continue;
635                         }
636
637                         len -= dollar + 1 - src;
638                         src  = dollar + 1;
639                 } else {
640                         /* it wasn't a "Id$" or "Id:xxxx$" */
641                         continue;
642                 }
643
644                 /* step 4: substitute */
645                 strbuf_addstr(buf, "Id: ");
646                 strbuf_add(buf, sha1_to_hex(sha1), 40);
647                 strbuf_addstr(buf, " $");
648         }
649         strbuf_add(buf, src, len);
650
651         free(to_free);
652         return 1;
653 }
654
655 static enum crlf_action git_path_check_crlf(const char *path, struct git_attr_check *check)
656 {
657         const char *value = check->value;
658
659         if (ATTR_TRUE(value))
660                 return CRLF_TEXT;
661         else if (ATTR_FALSE(value))
662                 return CRLF_BINARY;
663         else if (ATTR_UNSET(value))
664                 ;
665         else if (!strcmp(value, "input"))
666                 return CRLF_INPUT;
667         else if (!strcmp(value, "auto"))
668                 return CRLF_AUTO;
669         return CRLF_GUESS;
670 }
671
672 static enum eol git_path_check_eol(const char *path, struct git_attr_check *check)
673 {
674         const char *value = check->value;
675
676         if (ATTR_UNSET(value))
677                 ;
678         else if (!strcmp(value, "lf"))
679                 return EOL_LF;
680         else if (!strcmp(value, "crlf"))
681                 return EOL_CRLF;
682         return EOL_UNSET;
683 }
684
685 static struct convert_driver *git_path_check_convert(const char *path,
686                                              struct git_attr_check *check)
687 {
688         const char *value = check->value;
689         struct convert_driver *drv;
690
691         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
692                 return NULL;
693         for (drv = user_convert; drv; drv = drv->next)
694                 if (!strcmp(value, drv->name))
695                         return drv;
696         return NULL;
697 }
698
699 static int git_path_check_ident(const char *path, struct git_attr_check *check)
700 {
701         const char *value = check->value;
702
703         return !!ATTR_TRUE(value);
704 }
705
706 static enum crlf_action input_crlf_action(enum crlf_action text_attr, enum eol eol_attr)
707 {
708         if (text_attr == CRLF_BINARY)
709                 return CRLF_BINARY;
710         if (eol_attr == EOL_LF)
711                 return CRLF_INPUT;
712         if (eol_attr == EOL_CRLF)
713                 return CRLF_CRLF;
714         return text_attr;
715 }
716
717 struct conv_attrs {
718         struct convert_driver *drv;
719         enum crlf_action crlf_action;
720         enum eol eol_attr;
721         int ident;
722 };
723
724 static const char *conv_attr_name[] = {
725         "crlf", "ident", "filter", "eol", "text",
726 };
727 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
728
729 static void convert_attrs(struct conv_attrs *ca, const char *path)
730 {
731         int i;
732         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
733
734         if (!ccheck[0].attr) {
735                 for (i = 0; i < NUM_CONV_ATTRS; i++)
736                         ccheck[i].attr = git_attr(conv_attr_name[i]);
737                 user_convert_tail = &user_convert;
738                 git_config(read_convert_config, NULL);
739         }
740
741         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
742                 ca->crlf_action = git_path_check_crlf(path, ccheck + 4);
743                 if (ca->crlf_action == CRLF_GUESS)
744                         ca->crlf_action = git_path_check_crlf(path, ccheck + 0);
745                 ca->ident = git_path_check_ident(path, ccheck + 1);
746                 ca->drv = git_path_check_convert(path, ccheck + 2);
747                 ca->eol_attr = git_path_check_eol(path, ccheck + 3);
748         } else {
749                 ca->drv = NULL;
750                 ca->crlf_action = CRLF_GUESS;
751                 ca->eol_attr = EOL_UNSET;
752                 ca->ident = 0;
753         }
754 }
755
756 int would_convert_to_git_filter_fd(const char *path)
757 {
758         struct conv_attrs ca;
759
760         convert_attrs(&ca, path);
761         if (!ca.drv)
762                 return 0;
763
764         /*
765          * Apply a filter to an fd only if the filter is required to succeed.
766          * We must die if the filter fails, because the original data before
767          * filtering is not available.
768          */
769         if (!ca.drv->required)
770                 return 0;
771
772         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
773 }
774
775 int convert_to_git(const char *path, const char *src, size_t len,
776                    struct strbuf *dst, enum safe_crlf checksafe)
777 {
778         int ret = 0;
779         const char *filter = NULL;
780         int required = 0;
781         struct conv_attrs ca;
782
783         convert_attrs(&ca, path);
784         if (ca.drv) {
785                 filter = ca.drv->clean;
786                 required = ca.drv->required;
787         }
788
789         ret |= apply_filter(path, src, len, -1, dst, filter);
790         if (!ret && required)
791                 die("%s: clean filter '%s' failed", path, ca.drv->name);
792
793         if (ret && dst) {
794                 src = dst->buf;
795                 len = dst->len;
796         }
797         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
798         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
799         if (ret && dst) {
800                 src = dst->buf;
801                 len = dst->len;
802         }
803         return ret | ident_to_git(path, src, len, dst, ca.ident);
804 }
805
806 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
807                               enum safe_crlf checksafe)
808 {
809         struct conv_attrs ca;
810         convert_attrs(&ca, path);
811
812         assert(ca.drv);
813         assert(ca.drv->clean);
814
815         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
816                 die("%s: clean filter '%s' failed", path, ca.drv->name);
817
818         ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
819         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
820         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
821 }
822
823 static int convert_to_working_tree_internal(const char *path, const char *src,
824                                             size_t len, struct strbuf *dst,
825                                             int normalizing)
826 {
827         int ret = 0, ret_filter = 0;
828         const char *filter = NULL;
829         int required = 0;
830         struct conv_attrs ca;
831
832         convert_attrs(&ca, path);
833         if (ca.drv) {
834                 filter = ca.drv->smudge;
835                 required = ca.drv->required;
836         }
837
838         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
839         if (ret) {
840                 src = dst->buf;
841                 len = dst->len;
842         }
843         /*
844          * CRLF conversion can be skipped if normalizing, unless there
845          * is a smudge filter.  The filter might expect CRLFs.
846          */
847         if (filter || !normalizing) {
848                 ca.crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
849                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
850                 if (ret) {
851                         src = dst->buf;
852                         len = dst->len;
853                 }
854         }
855
856         ret_filter = apply_filter(path, src, len, -1, dst, filter);
857         if (!ret_filter && required)
858                 die("%s: smudge filter %s failed", path, ca.drv->name);
859
860         return ret | ret_filter;
861 }
862
863 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
864 {
865         return convert_to_working_tree_internal(path, src, len, dst, 0);
866 }
867
868 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
869 {
870         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
871         if (ret) {
872                 src = dst->buf;
873                 len = dst->len;
874         }
875         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_FALSE);
876 }
877
878 /*****************************************************************
879  *
880  * Streaming conversion support
881  *
882  *****************************************************************/
883
884 typedef int (*filter_fn)(struct stream_filter *,
885                          const char *input, size_t *isize_p,
886                          char *output, size_t *osize_p);
887 typedef void (*free_fn)(struct stream_filter *);
888
889 struct stream_filter_vtbl {
890         filter_fn filter;
891         free_fn free;
892 };
893
894 struct stream_filter {
895         struct stream_filter_vtbl *vtbl;
896 };
897
898 static int null_filter_fn(struct stream_filter *filter,
899                           const char *input, size_t *isize_p,
900                           char *output, size_t *osize_p)
901 {
902         size_t count;
903
904         if (!input)
905                 return 0; /* we do not keep any states */
906         count = *isize_p;
907         if (*osize_p < count)
908                 count = *osize_p;
909         if (count) {
910                 memmove(output, input, count);
911                 *isize_p -= count;
912                 *osize_p -= count;
913         }
914         return 0;
915 }
916
917 static void null_free_fn(struct stream_filter *filter)
918 {
919         ; /* nothing -- null instances are shared */
920 }
921
922 static struct stream_filter_vtbl null_vtbl = {
923         null_filter_fn,
924         null_free_fn,
925 };
926
927 static struct stream_filter null_filter_singleton = {
928         &null_vtbl,
929 };
930
931 int is_null_stream_filter(struct stream_filter *filter)
932 {
933         return filter == &null_filter_singleton;
934 }
935
936
937 /*
938  * LF-to-CRLF filter
939  */
940
941 struct lf_to_crlf_filter {
942         struct stream_filter filter;
943         unsigned has_held:1;
944         char held;
945 };
946
947 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
948                                 const char *input, size_t *isize_p,
949                                 char *output, size_t *osize_p)
950 {
951         size_t count, o = 0;
952         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
953
954         /*
955          * We may be holding onto the CR to see if it is followed by a
956          * LF, in which case we would need to go to the main loop.
957          * Otherwise, just emit it to the output stream.
958          */
959         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
960                 output[o++] = lf_to_crlf->held;
961                 lf_to_crlf->has_held = 0;
962         }
963
964         /* We are told to drain */
965         if (!input) {
966                 *osize_p -= o;
967                 return 0;
968         }
969
970         count = *isize_p;
971         if (count || lf_to_crlf->has_held) {
972                 size_t i;
973                 int was_cr = 0;
974
975                 if (lf_to_crlf->has_held) {
976                         was_cr = 1;
977                         lf_to_crlf->has_held = 0;
978                 }
979
980                 for (i = 0; o < *osize_p && i < count; i++) {
981                         char ch = input[i];
982
983                         if (ch == '\n') {
984                                 output[o++] = '\r';
985                         } else if (was_cr) {
986                                 /*
987                                  * Previous round saw CR and it is not followed
988                                  * by a LF; emit the CR before processing the
989                                  * current character.
990                                  */
991                                 output[o++] = '\r';
992                         }
993
994                         /*
995                          * We may have consumed the last output slot,
996                          * in which case we need to break out of this
997                          * loop; hold the current character before
998                          * returning.
999                          */
1000                         if (*osize_p <= o) {
1001                                 lf_to_crlf->has_held = 1;
1002                                 lf_to_crlf->held = ch;
1003                                 continue; /* break but increment i */
1004                         }
1005
1006                         if (ch == '\r') {
1007                                 was_cr = 1;
1008                                 continue;
1009                         }
1010
1011                         was_cr = 0;
1012                         output[o++] = ch;
1013                 }
1014
1015                 *osize_p -= o;
1016                 *isize_p -= i;
1017
1018                 if (!lf_to_crlf->has_held && was_cr) {
1019                         lf_to_crlf->has_held = 1;
1020                         lf_to_crlf->held = '\r';
1021                 }
1022         }
1023         return 0;
1024 }
1025
1026 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1027 {
1028         free(filter);
1029 }
1030
1031 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1032         lf_to_crlf_filter_fn,
1033         lf_to_crlf_free_fn,
1034 };
1035
1036 static struct stream_filter *lf_to_crlf_filter(void)
1037 {
1038         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1039
1040         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1041         return (struct stream_filter *)lf_to_crlf;
1042 }
1043
1044 /*
1045  * Cascade filter
1046  */
1047 #define FILTER_BUFFER 1024
1048 struct cascade_filter {
1049         struct stream_filter filter;
1050         struct stream_filter *one;
1051         struct stream_filter *two;
1052         char buf[FILTER_BUFFER];
1053         int end, ptr;
1054 };
1055
1056 static int cascade_filter_fn(struct stream_filter *filter,
1057                              const char *input, size_t *isize_p,
1058                              char *output, size_t *osize_p)
1059 {
1060         struct cascade_filter *cas = (struct cascade_filter *) filter;
1061         size_t filled = 0;
1062         size_t sz = *osize_p;
1063         size_t to_feed, remaining;
1064
1065         /*
1066          * input -- (one) --> buf -- (two) --> output
1067          */
1068         while (filled < sz) {
1069                 remaining = sz - filled;
1070
1071                 /* do we already have something to feed two with? */
1072                 if (cas->ptr < cas->end) {
1073                         to_feed = cas->end - cas->ptr;
1074                         if (stream_filter(cas->two,
1075                                           cas->buf + cas->ptr, &to_feed,
1076                                           output + filled, &remaining))
1077                                 return -1;
1078                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1079                         filled = sz - remaining;
1080                         continue;
1081                 }
1082
1083                 /* feed one from upstream and have it emit into our buffer */
1084                 to_feed = input ? *isize_p : 0;
1085                 if (input && !to_feed)
1086                         break;
1087                 remaining = sizeof(cas->buf);
1088                 if (stream_filter(cas->one,
1089                                   input, &to_feed,
1090                                   cas->buf, &remaining))
1091                         return -1;
1092                 cas->end = sizeof(cas->buf) - remaining;
1093                 cas->ptr = 0;
1094                 if (input) {
1095                         size_t fed = *isize_p - to_feed;
1096                         *isize_p -= fed;
1097                         input += fed;
1098                 }
1099
1100                 /* do we know that we drained one completely? */
1101                 if (input || cas->end)
1102                         continue;
1103
1104                 /* tell two to drain; we have nothing more to give it */
1105                 to_feed = 0;
1106                 remaining = sz - filled;
1107                 if (stream_filter(cas->two,
1108                                   NULL, &to_feed,
1109                                   output + filled, &remaining))
1110                         return -1;
1111                 if (remaining == (sz - filled))
1112                         break; /* completely drained two */
1113                 filled = sz - remaining;
1114         }
1115         *osize_p -= filled;
1116         return 0;
1117 }
1118
1119 static void cascade_free_fn(struct stream_filter *filter)
1120 {
1121         struct cascade_filter *cas = (struct cascade_filter *)filter;
1122         free_stream_filter(cas->one);
1123         free_stream_filter(cas->two);
1124         free(filter);
1125 }
1126
1127 static struct stream_filter_vtbl cascade_vtbl = {
1128         cascade_filter_fn,
1129         cascade_free_fn,
1130 };
1131
1132 static struct stream_filter *cascade_filter(struct stream_filter *one,
1133                                             struct stream_filter *two)
1134 {
1135         struct cascade_filter *cascade;
1136
1137         if (!one || is_null_stream_filter(one))
1138                 return two;
1139         if (!two || is_null_stream_filter(two))
1140                 return one;
1141
1142         cascade = xmalloc(sizeof(*cascade));
1143         cascade->one = one;
1144         cascade->two = two;
1145         cascade->end = cascade->ptr = 0;
1146         cascade->filter.vtbl = &cascade_vtbl;
1147         return (struct stream_filter *)cascade;
1148 }
1149
1150 /*
1151  * ident filter
1152  */
1153 #define IDENT_DRAINING (-1)
1154 #define IDENT_SKIPPING (-2)
1155 struct ident_filter {
1156         struct stream_filter filter;
1157         struct strbuf left;
1158         int state;
1159         char ident[45]; /* ": x40 $" */
1160 };
1161
1162 static int is_foreign_ident(const char *str)
1163 {
1164         int i;
1165
1166         if (!skip_prefix(str, "$Id: ", &str))
1167                 return 0;
1168         for (i = 0; str[i]; i++) {
1169                 if (isspace(str[i]) && str[i+1] != '$')
1170                         return 1;
1171         }
1172         return 0;
1173 }
1174
1175 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1176 {
1177         size_t to_drain = ident->left.len;
1178
1179         if (*osize_p < to_drain)
1180                 to_drain = *osize_p;
1181         if (to_drain) {
1182                 memcpy(*output_p, ident->left.buf, to_drain);
1183                 strbuf_remove(&ident->left, 0, to_drain);
1184                 *output_p += to_drain;
1185                 *osize_p -= to_drain;
1186         }
1187         if (!ident->left.len)
1188                 ident->state = 0;
1189 }
1190
1191 static int ident_filter_fn(struct stream_filter *filter,
1192                            const char *input, size_t *isize_p,
1193                            char *output, size_t *osize_p)
1194 {
1195         struct ident_filter *ident = (struct ident_filter *)filter;
1196         static const char head[] = "$Id";
1197
1198         if (!input) {
1199                 /* drain upon eof */
1200                 switch (ident->state) {
1201                 default:
1202                         strbuf_add(&ident->left, head, ident->state);
1203                 case IDENT_SKIPPING:
1204                         /* fallthru */
1205                 case IDENT_DRAINING:
1206                         ident_drain(ident, &output, osize_p);
1207                 }
1208                 return 0;
1209         }
1210
1211         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1212                 int ch;
1213
1214                 if (ident->state == IDENT_DRAINING) {
1215                         ident_drain(ident, &output, osize_p);
1216                         if (!*osize_p)
1217                                 break;
1218                         continue;
1219                 }
1220
1221                 ch = *(input++);
1222                 (*isize_p)--;
1223
1224                 if (ident->state == IDENT_SKIPPING) {
1225                         /*
1226                          * Skipping until '$' or LF, but keeping them
1227                          * in case it is a foreign ident.
1228                          */
1229                         strbuf_addch(&ident->left, ch);
1230                         if (ch != '\n' && ch != '$')
1231                                 continue;
1232                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1233                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1234                                 strbuf_addstr(&ident->left, ident->ident);
1235                         }
1236                         ident->state = IDENT_DRAINING;
1237                         continue;
1238                 }
1239
1240                 if (ident->state < sizeof(head) &&
1241                     head[ident->state] == ch) {
1242                         ident->state++;
1243                         continue;
1244                 }
1245
1246                 if (ident->state)
1247                         strbuf_add(&ident->left, head, ident->state);
1248                 if (ident->state == sizeof(head) - 1) {
1249                         if (ch != ':' && ch != '$') {
1250                                 strbuf_addch(&ident->left, ch);
1251                                 ident->state = 0;
1252                                 continue;
1253                         }
1254
1255                         if (ch == ':') {
1256                                 strbuf_addch(&ident->left, ch);
1257                                 ident->state = IDENT_SKIPPING;
1258                         } else {
1259                                 strbuf_addstr(&ident->left, ident->ident);
1260                                 ident->state = IDENT_DRAINING;
1261                         }
1262                         continue;
1263                 }
1264
1265                 strbuf_addch(&ident->left, ch);
1266                 ident->state = IDENT_DRAINING;
1267         }
1268         return 0;
1269 }
1270
1271 static void ident_free_fn(struct stream_filter *filter)
1272 {
1273         struct ident_filter *ident = (struct ident_filter *)filter;
1274         strbuf_release(&ident->left);
1275         free(filter);
1276 }
1277
1278 static struct stream_filter_vtbl ident_vtbl = {
1279         ident_filter_fn,
1280         ident_free_fn,
1281 };
1282
1283 static struct stream_filter *ident_filter(const unsigned char *sha1)
1284 {
1285         struct ident_filter *ident = xmalloc(sizeof(*ident));
1286
1287         sprintf(ident->ident, ": %s $", sha1_to_hex(sha1));
1288         strbuf_init(&ident->left, 0);
1289         ident->filter.vtbl = &ident_vtbl;
1290         ident->state = 0;
1291         return (struct stream_filter *)ident;
1292 }
1293
1294 /*
1295  * Return an appropriately constructed filter for the path, or NULL if
1296  * the contents cannot be filtered without reading the whole thing
1297  * in-core.
1298  *
1299  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1300  * large binary blob you would want us not to slurp into the memory!
1301  */
1302 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1303 {
1304         struct conv_attrs ca;
1305         enum crlf_action crlf_action;
1306         struct stream_filter *filter = NULL;
1307
1308         convert_attrs(&ca, path);
1309
1310         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1311                 return filter;
1312
1313         if (ca.ident)
1314                 filter = ident_filter(sha1);
1315
1316         crlf_action = input_crlf_action(ca.crlf_action, ca.eol_attr);
1317
1318         if ((crlf_action == CRLF_BINARY) || (crlf_action == CRLF_INPUT) ||
1319             (crlf_action == CRLF_GUESS && auto_crlf == AUTO_CRLF_FALSE))
1320                 filter = cascade_filter(filter, &null_filter_singleton);
1321
1322         else if (output_eol(crlf_action) == EOL_CRLF &&
1323                  !(crlf_action == CRLF_AUTO || crlf_action == CRLF_GUESS))
1324                 filter = cascade_filter(filter, lf_to_crlf_filter());
1325
1326         return filter;
1327 }
1328
1329 void free_stream_filter(struct stream_filter *filter)
1330 {
1331         filter->vtbl->free(filter);
1332 }
1333
1334 int stream_filter(struct stream_filter *filter,
1335                   const char *input, size_t *isize_p,
1336                   char *output, size_t *osize_p)
1337 {
1338         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1339 }