Merge branch 'js/prepare-sequencer'
[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 /* Stat bits: When BIN is set, the txt bits are unset */
17 #define CONVERT_STAT_BITS_TXT_LF    0x1
18 #define CONVERT_STAT_BITS_TXT_CRLF  0x2
19 #define CONVERT_STAT_BITS_BIN       0x4
20
21 enum crlf_action {
22         CRLF_UNDEFINED,
23         CRLF_BINARY,
24         CRLF_TEXT,
25         CRLF_TEXT_INPUT,
26         CRLF_TEXT_CRLF,
27         CRLF_AUTO,
28         CRLF_AUTO_INPUT,
29         CRLF_AUTO_CRLF
30 };
31
32 struct text_stat {
33         /* NUL, CR, LF and CRLF counts */
34         unsigned nul, lonecr, lonelf, crlf;
35
36         /* These are just approximations! */
37         unsigned printable, nonprintable;
38 };
39
40 static void gather_stats(const char *buf, unsigned long size, struct text_stat *stats)
41 {
42         unsigned long i;
43
44         memset(stats, 0, sizeof(*stats));
45
46         for (i = 0; i < size; i++) {
47                 unsigned char c = buf[i];
48                 if (c == '\r') {
49                         if (i+1 < size && buf[i+1] == '\n') {
50                                 stats->crlf++;
51                                 i++;
52                         } else
53                                 stats->lonecr++;
54                         continue;
55                 }
56                 if (c == '\n') {
57                         stats->lonelf++;
58                         continue;
59                 }
60                 if (c == 127)
61                         /* DEL */
62                         stats->nonprintable++;
63                 else if (c < 32) {
64                         switch (c) {
65                                 /* BS, HT, ESC and FF */
66                         case '\b': case '\t': case '\033': case '\014':
67                                 stats->printable++;
68                                 break;
69                         case 0:
70                                 stats->nul++;
71                                 /* fall through */
72                         default:
73                                 stats->nonprintable++;
74                         }
75                 }
76                 else
77                         stats->printable++;
78         }
79
80         /* If file ends with EOF then don't count this EOF as non-printable. */
81         if (size >= 1 && buf[size-1] == '\032')
82                 stats->nonprintable--;
83 }
84
85 /*
86  * The same heuristics as diff.c::mmfile_is_binary()
87  * We treat files with bare CR as binary
88  */
89 static int convert_is_binary(unsigned long size, const struct text_stat *stats)
90 {
91         if (stats->lonecr)
92                 return 1;
93         if (stats->nul)
94                 return 1;
95         if ((stats->printable >> 7) < stats->nonprintable)
96                 return 1;
97         return 0;
98 }
99
100 static unsigned int gather_convert_stats(const char *data, unsigned long size)
101 {
102         struct text_stat stats;
103         int ret = 0;
104         if (!data || !size)
105                 return 0;
106         gather_stats(data, size, &stats);
107         if (convert_is_binary(size, &stats))
108                 ret |= CONVERT_STAT_BITS_BIN;
109         if (stats.crlf)
110                 ret |= CONVERT_STAT_BITS_TXT_CRLF;
111         if (stats.lonelf)
112                 ret |=  CONVERT_STAT_BITS_TXT_LF;
113
114         return ret;
115 }
116
117 static const char *gather_convert_stats_ascii(const char *data, unsigned long size)
118 {
119         unsigned int convert_stats = gather_convert_stats(data, size);
120
121         if (convert_stats & CONVERT_STAT_BITS_BIN)
122                 return "-text";
123         switch (convert_stats) {
124         case CONVERT_STAT_BITS_TXT_LF:
125                 return "lf";
126         case CONVERT_STAT_BITS_TXT_CRLF:
127                 return "crlf";
128         case CONVERT_STAT_BITS_TXT_LF | CONVERT_STAT_BITS_TXT_CRLF:
129                 return "mixed";
130         default:
131                 return "none";
132         }
133 }
134
135 const char *get_cached_convert_stats_ascii(const char *path)
136 {
137         const char *ret;
138         unsigned long sz;
139         void *data = read_blob_data_from_cache(path, &sz);
140         ret = gather_convert_stats_ascii(data, sz);
141         free(data);
142         return ret;
143 }
144
145 const char *get_wt_convert_stats_ascii(const char *path)
146 {
147         const char *ret = "";
148         struct strbuf sb = STRBUF_INIT;
149         if (strbuf_read_file(&sb, path, 0) >= 0)
150                 ret = gather_convert_stats_ascii(sb.buf, sb.len);
151         strbuf_release(&sb);
152         return ret;
153 }
154
155 static int text_eol_is_crlf(void)
156 {
157         if (auto_crlf == AUTO_CRLF_TRUE)
158                 return 1;
159         else if (auto_crlf == AUTO_CRLF_INPUT)
160                 return 0;
161         if (core_eol == EOL_CRLF)
162                 return 1;
163         if (core_eol == EOL_UNSET && EOL_NATIVE == EOL_CRLF)
164                 return 1;
165         return 0;
166 }
167
168 static enum eol output_eol(enum crlf_action crlf_action)
169 {
170         switch (crlf_action) {
171         case CRLF_BINARY:
172                 return EOL_UNSET;
173         case CRLF_TEXT_CRLF:
174                 return EOL_CRLF;
175         case CRLF_TEXT_INPUT:
176                 return EOL_LF;
177         case CRLF_UNDEFINED:
178         case CRLF_AUTO_CRLF:
179                 return EOL_CRLF;
180         case CRLF_AUTO_INPUT:
181                 return EOL_LF;
182         case CRLF_TEXT:
183         case CRLF_AUTO:
184                 /* fall through */
185                 return text_eol_is_crlf() ? EOL_CRLF : EOL_LF;
186         }
187         warning("Illegal crlf_action %d\n", (int)crlf_action);
188         return core_eol;
189 }
190
191 static void check_safe_crlf(const char *path, enum crlf_action crlf_action,
192                             struct text_stat *old_stats, struct text_stat *new_stats,
193                             enum safe_crlf checksafe)
194 {
195         if (old_stats->crlf && !new_stats->crlf ) {
196                 /*
197                  * CRLFs would not be restored by checkout
198                  */
199                 if (checksafe == SAFE_CRLF_WARN)
200                         warning(_("CRLF will be replaced by LF in %s.\n"
201                                   "The file will have its original line"
202                                   " endings in your working directory."), path);
203                 else /* i.e. SAFE_CRLF_FAIL */
204                         die(_("CRLF would be replaced by LF in %s."), path);
205         } else if (old_stats->lonelf && !new_stats->lonelf ) {
206                 /*
207                  * CRLFs would be added by checkout
208                  */
209                 if (checksafe == SAFE_CRLF_WARN)
210                         warning(_("LF will be replaced by CRLF in %s.\n"
211                                   "The file will have its original line"
212                                   " endings in your working directory."), path);
213                 else /* i.e. SAFE_CRLF_FAIL */
214                         die(_("LF would be replaced by CRLF in %s"), path);
215         }
216 }
217
218 static int has_cr_in_index(const char *path)
219 {
220         unsigned long sz;
221         void *data;
222         int has_cr;
223
224         data = read_blob_data_from_cache(path, &sz);
225         if (!data)
226                 return 0;
227         has_cr = memchr(data, '\r', sz) != NULL;
228         free(data);
229         return has_cr;
230 }
231
232 static int will_convert_lf_to_crlf(size_t len, struct text_stat *stats,
233                                    enum crlf_action crlf_action)
234 {
235         if (output_eol(crlf_action) != EOL_CRLF)
236                 return 0;
237         /* No "naked" LF? Nothing to convert, regardless. */
238         if (!stats->lonelf)
239                 return 0;
240
241         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
242                 /* If we have any CR or CRLF line endings, we do not touch it */
243                 /* This is the new safer autocrlf-handling */
244                 if (stats->lonecr || stats->crlf)
245                         return 0;
246
247                 if (convert_is_binary(len, stats))
248                         return 0;
249         }
250         return 1;
251
252 }
253
254 static int crlf_to_git(const char *path, const char *src, size_t len,
255                        struct strbuf *buf,
256                        enum crlf_action crlf_action, enum safe_crlf checksafe)
257 {
258         struct text_stat stats;
259         char *dst;
260         int convert_crlf_into_lf;
261
262         if (crlf_action == CRLF_BINARY ||
263             (src && !len))
264                 return 0;
265
266         /*
267          * If we are doing a dry-run and have no source buffer, there is
268          * nothing to analyze; we must assume we would convert.
269          */
270         if (!buf && !src)
271                 return 1;
272
273         gather_stats(src, len, &stats);
274         /* Optimization: No CRLF? Nothing to convert, regardless. */
275         convert_crlf_into_lf = !!stats.crlf;
276
277         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
278                 if (convert_is_binary(len, &stats))
279                         return 0;
280                 /*
281                  * If the file in the index has any CR in it, do not convert.
282                  * This is the new safer autocrlf handling.
283                  */
284                 if (checksafe == SAFE_CRLF_RENORMALIZE)
285                         checksafe = SAFE_CRLF_FALSE;
286                 else if (has_cr_in_index(path))
287                         convert_crlf_into_lf = 0;
288         }
289         if (checksafe && len) {
290                 struct text_stat new_stats;
291                 memcpy(&new_stats, &stats, sizeof(new_stats));
292                 /* simulate "git add" */
293                 if (convert_crlf_into_lf) {
294                         new_stats.lonelf += new_stats.crlf;
295                         new_stats.crlf = 0;
296                 }
297                 /* simulate "git checkout" */
298                 if (will_convert_lf_to_crlf(len, &new_stats, crlf_action)) {
299                         new_stats.crlf += new_stats.lonelf;
300                         new_stats.lonelf = 0;
301                 }
302                 check_safe_crlf(path, crlf_action, &stats, &new_stats, checksafe);
303         }
304         if (!convert_crlf_into_lf)
305                 return 0;
306
307         /*
308          * At this point all of our source analysis is done, and we are sure we
309          * would convert. If we are in dry-run mode, we can give an answer.
310          */
311         if (!buf)
312                 return 1;
313
314         /* only grow if not in place */
315         if (strbuf_avail(buf) + buf->len < len)
316                 strbuf_grow(buf, len - buf->len);
317         dst = buf->buf;
318         if (crlf_action == CRLF_AUTO || crlf_action == CRLF_AUTO_INPUT || crlf_action == CRLF_AUTO_CRLF) {
319                 /*
320                  * If we guessed, we already know we rejected a file with
321                  * lone CR, and we can strip a CR without looking at what
322                  * follow it.
323                  */
324                 do {
325                         unsigned char c = *src++;
326                         if (c != '\r')
327                                 *dst++ = c;
328                 } while (--len);
329         } else {
330                 do {
331                         unsigned char c = *src++;
332                         if (! (c == '\r' && (1 < len && *src == '\n')))
333                                 *dst++ = c;
334                 } while (--len);
335         }
336         strbuf_setlen(buf, dst - buf->buf);
337         return 1;
338 }
339
340 static int crlf_to_worktree(const char *path, const char *src, size_t len,
341                             struct strbuf *buf, enum crlf_action crlf_action)
342 {
343         char *to_free = NULL;
344         struct text_stat stats;
345
346         if (!len || output_eol(crlf_action) != EOL_CRLF)
347                 return 0;
348
349         gather_stats(src, len, &stats);
350         if (!will_convert_lf_to_crlf(len, &stats, crlf_action))
351                 return 0;
352
353         /* are we "faking" in place editing ? */
354         if (src == buf->buf)
355                 to_free = strbuf_detach(buf, NULL);
356
357         strbuf_grow(buf, len + stats.lonelf);
358         for (;;) {
359                 const char *nl = memchr(src, '\n', len);
360                 if (!nl)
361                         break;
362                 if (nl > src && nl[-1] == '\r') {
363                         strbuf_add(buf, src, nl + 1 - src);
364                 } else {
365                         strbuf_add(buf, src, nl - src);
366                         strbuf_addstr(buf, "\r\n");
367                 }
368                 len -= nl + 1 - src;
369                 src  = nl + 1;
370         }
371         strbuf_add(buf, src, len);
372
373         free(to_free);
374         return 1;
375 }
376
377 struct filter_params {
378         const char *src;
379         unsigned long size;
380         int fd;
381         const char *cmd;
382         const char *path;
383 };
384
385 static int filter_buffer_or_fd(int in, int out, void *data)
386 {
387         /*
388          * Spawn cmd and feed the buffer contents through its stdin.
389          */
390         struct child_process child_process = CHILD_PROCESS_INIT;
391         struct filter_params *params = (struct filter_params *)data;
392         int write_err, status;
393         const char *argv[] = { NULL, NULL };
394
395         /* apply % substitution to cmd */
396         struct strbuf cmd = STRBUF_INIT;
397         struct strbuf path = STRBUF_INIT;
398         struct strbuf_expand_dict_entry dict[] = {
399                 { "f", NULL, },
400                 { NULL, NULL, },
401         };
402
403         /* quote the path to preserve spaces, etc. */
404         sq_quote_buf(&path, params->path);
405         dict[0].value = path.buf;
406
407         /* expand all %f with the quoted path */
408         strbuf_expand(&cmd, params->cmd, strbuf_expand_dict_cb, &dict);
409         strbuf_release(&path);
410
411         argv[0] = cmd.buf;
412
413         child_process.argv = argv;
414         child_process.use_shell = 1;
415         child_process.in = -1;
416         child_process.out = out;
417
418         if (start_command(&child_process))
419                 return error("cannot fork to run external filter %s", params->cmd);
420
421         sigchain_push(SIGPIPE, SIG_IGN);
422
423         if (params->src) {
424                 write_err = (write_in_full(child_process.in,
425                                            params->src, params->size) < 0);
426                 if (errno == EPIPE)
427                         write_err = 0;
428         } else {
429                 write_err = copy_fd(params->fd, child_process.in);
430                 if (write_err == COPY_WRITE_ERROR && errno == EPIPE)
431                         write_err = 0;
432         }
433
434         if (close(child_process.in))
435                 write_err = 1;
436         if (write_err)
437                 error("cannot feed the input to external filter %s", params->cmd);
438
439         sigchain_pop(SIGPIPE);
440
441         status = finish_command(&child_process);
442         if (status)
443                 error("external filter %s failed %d", params->cmd, status);
444
445         strbuf_release(&cmd);
446         return (write_err || status);
447 }
448
449 static int apply_filter(const char *path, const char *src, size_t len, int fd,
450                         struct strbuf *dst, const char *cmd)
451 {
452         /*
453          * Create a pipeline to have the command filter the buffer's
454          * contents.
455          *
456          * (child --> cmd) --> us
457          */
458         int ret = 1;
459         struct strbuf nbuf = STRBUF_INIT;
460         struct async async;
461         struct filter_params params;
462
463         if (!cmd || !*cmd)
464                 return 0;
465
466         if (!dst)
467                 return 1;
468
469         memset(&async, 0, sizeof(async));
470         async.proc = filter_buffer_or_fd;
471         async.data = &params;
472         async.out = -1;
473         params.src = src;
474         params.size = len;
475         params.fd = fd;
476         params.cmd = cmd;
477         params.path = path;
478
479         fflush(NULL);
480         if (start_async(&async))
481                 return 0;       /* error was already reported */
482
483         if (strbuf_read(&nbuf, async.out, len) < 0) {
484                 error("read from external filter %s failed", cmd);
485                 ret = 0;
486         }
487         if (close(async.out)) {
488                 error("read from external filter %s failed", cmd);
489                 ret = 0;
490         }
491         if (finish_async(&async)) {
492                 error("external filter %s failed", cmd);
493                 ret = 0;
494         }
495
496         if (ret) {
497                 strbuf_swap(dst, &nbuf);
498         }
499         strbuf_release(&nbuf);
500         return ret;
501 }
502
503 static struct convert_driver {
504         const char *name;
505         struct convert_driver *next;
506         const char *smudge;
507         const char *clean;
508         int required;
509 } *user_convert, **user_convert_tail;
510
511 static int read_convert_config(const char *var, const char *value, void *cb)
512 {
513         const char *key, *name;
514         int namelen;
515         struct convert_driver *drv;
516
517         /*
518          * External conversion drivers are configured using
519          * "filter.<name>.variable".
520          */
521         if (parse_config_key(var, "filter", &name, &namelen, &key) < 0 || !name)
522                 return 0;
523         for (drv = user_convert; drv; drv = drv->next)
524                 if (!strncmp(drv->name, name, namelen) && !drv->name[namelen])
525                         break;
526         if (!drv) {
527                 drv = xcalloc(1, sizeof(struct convert_driver));
528                 drv->name = xmemdupz(name, namelen);
529                 *user_convert_tail = drv;
530                 user_convert_tail = &(drv->next);
531         }
532
533         /*
534          * filter.<name>.smudge and filter.<name>.clean specifies
535          * the command line:
536          *
537          *      command-line
538          *
539          * The command-line will not be interpolated in any way.
540          */
541
542         if (!strcmp("smudge", key))
543                 return git_config_string(&drv->smudge, var, value);
544
545         if (!strcmp("clean", key))
546                 return git_config_string(&drv->clean, var, value);
547
548         if (!strcmp("required", key)) {
549                 drv->required = git_config_bool(var, value);
550                 return 0;
551         }
552
553         return 0;
554 }
555
556 static int count_ident(const char *cp, unsigned long size)
557 {
558         /*
559          * "$Id: 0000000000000000000000000000000000000000 $" <=> "$Id$"
560          */
561         int cnt = 0;
562         char ch;
563
564         while (size) {
565                 ch = *cp++;
566                 size--;
567                 if (ch != '$')
568                         continue;
569                 if (size < 3)
570                         break;
571                 if (memcmp("Id", cp, 2))
572                         continue;
573                 ch = cp[2];
574                 cp += 3;
575                 size -= 3;
576                 if (ch == '$')
577                         cnt++; /* $Id$ */
578                 if (ch != ':')
579                         continue;
580
581                 /*
582                  * "$Id: ... "; scan up to the closing dollar sign and discard.
583                  */
584                 while (size) {
585                         ch = *cp++;
586                         size--;
587                         if (ch == '$') {
588                                 cnt++;
589                                 break;
590                         }
591                         if (ch == '\n')
592                                 break;
593                 }
594         }
595         return cnt;
596 }
597
598 static int ident_to_git(const char *path, const char *src, size_t len,
599                         struct strbuf *buf, int ident)
600 {
601         char *dst, *dollar;
602
603         if (!ident || (src && !count_ident(src, len)))
604                 return 0;
605
606         if (!buf)
607                 return 1;
608
609         /* only grow if not in place */
610         if (strbuf_avail(buf) + buf->len < len)
611                 strbuf_grow(buf, len - buf->len);
612         dst = buf->buf;
613         for (;;) {
614                 dollar = memchr(src, '$', len);
615                 if (!dollar)
616                         break;
617                 memmove(dst, src, dollar + 1 - src);
618                 dst += dollar + 1 - src;
619                 len -= dollar + 1 - src;
620                 src  = dollar + 1;
621
622                 if (len > 3 && !memcmp(src, "Id:", 3)) {
623                         dollar = memchr(src + 3, '$', len - 3);
624                         if (!dollar)
625                                 break;
626                         if (memchr(src + 3, '\n', dollar - src - 3)) {
627                                 /* Line break before the next dollar. */
628                                 continue;
629                         }
630
631                         memcpy(dst, "Id$", 3);
632                         dst += 3;
633                         len -= dollar + 1 - src;
634                         src  = dollar + 1;
635                 }
636         }
637         memmove(dst, src, len);
638         strbuf_setlen(buf, dst + len - buf->buf);
639         return 1;
640 }
641
642 static int ident_to_worktree(const char *path, const char *src, size_t len,
643                              struct strbuf *buf, int ident)
644 {
645         unsigned char sha1[20];
646         char *to_free = NULL, *dollar, *spc;
647         int cnt;
648
649         if (!ident)
650                 return 0;
651
652         cnt = count_ident(src, len);
653         if (!cnt)
654                 return 0;
655
656         /* are we "faking" in place editing ? */
657         if (src == buf->buf)
658                 to_free = strbuf_detach(buf, NULL);
659         hash_sha1_file(src, len, "blob", sha1);
660
661         strbuf_grow(buf, len + cnt * 43);
662         for (;;) {
663                 /* step 1: run to the next '$' */
664                 dollar = memchr(src, '$', len);
665                 if (!dollar)
666                         break;
667                 strbuf_add(buf, src, dollar + 1 - src);
668                 len -= dollar + 1 - src;
669                 src  = dollar + 1;
670
671                 /* step 2: does it looks like a bit like Id:xxx$ or Id$ ? */
672                 if (len < 3 || memcmp("Id", src, 2))
673                         continue;
674
675                 /* step 3: skip over Id$ or Id:xxxxx$ */
676                 if (src[2] == '$') {
677                         src += 3;
678                         len -= 3;
679                 } else if (src[2] == ':') {
680                         /*
681                          * It's possible that an expanded Id has crept its way into the
682                          * repository, we cope with that by stripping the expansion out.
683                          * This is probably not a good idea, since it will cause changes
684                          * on checkout, which won't go away by stash, but let's keep it
685                          * for git-style ids.
686                          */
687                         dollar = memchr(src + 3, '$', len - 3);
688                         if (!dollar) {
689                                 /* incomplete keyword, no more '$', so just quit the loop */
690                                 break;
691                         }
692
693                         if (memchr(src + 3, '\n', dollar - src - 3)) {
694                                 /* Line break before the next dollar. */
695                                 continue;
696                         }
697
698                         spc = memchr(src + 4, ' ', dollar - src - 4);
699                         if (spc && spc < dollar-1) {
700                                 /* There are spaces in unexpected places.
701                                  * This is probably an id from some other
702                                  * versioning system. Keep it for now.
703                                  */
704                                 continue;
705                         }
706
707                         len -= dollar + 1 - src;
708                         src  = dollar + 1;
709                 } else {
710                         /* it wasn't a "Id$" or "Id:xxxx$" */
711                         continue;
712                 }
713
714                 /* step 4: substitute */
715                 strbuf_addstr(buf, "Id: ");
716                 strbuf_add(buf, sha1_to_hex(sha1), 40);
717                 strbuf_addstr(buf, " $");
718         }
719         strbuf_add(buf, src, len);
720
721         free(to_free);
722         return 1;
723 }
724
725 static enum crlf_action git_path_check_crlf(struct git_attr_check *check)
726 {
727         const char *value = check->value;
728
729         if (ATTR_TRUE(value))
730                 return CRLF_TEXT;
731         else if (ATTR_FALSE(value))
732                 return CRLF_BINARY;
733         else if (ATTR_UNSET(value))
734                 ;
735         else if (!strcmp(value, "input"))
736                 return CRLF_TEXT_INPUT;
737         else if (!strcmp(value, "auto"))
738                 return CRLF_AUTO;
739         return CRLF_UNDEFINED;
740 }
741
742 static enum eol git_path_check_eol(struct git_attr_check *check)
743 {
744         const char *value = check->value;
745
746         if (ATTR_UNSET(value))
747                 ;
748         else if (!strcmp(value, "lf"))
749                 return EOL_LF;
750         else if (!strcmp(value, "crlf"))
751                 return EOL_CRLF;
752         return EOL_UNSET;
753 }
754
755 static struct convert_driver *git_path_check_convert(struct git_attr_check *check)
756 {
757         const char *value = check->value;
758         struct convert_driver *drv;
759
760         if (ATTR_TRUE(value) || ATTR_FALSE(value) || ATTR_UNSET(value))
761                 return NULL;
762         for (drv = user_convert; drv; drv = drv->next)
763                 if (!strcmp(value, drv->name))
764                         return drv;
765         return NULL;
766 }
767
768 static int git_path_check_ident(struct git_attr_check *check)
769 {
770         const char *value = check->value;
771
772         return !!ATTR_TRUE(value);
773 }
774
775 struct conv_attrs {
776         struct convert_driver *drv;
777         enum crlf_action attr_action; /* What attr says */
778         enum crlf_action crlf_action; /* When no attr is set, use core.autocrlf */
779         int ident;
780 };
781
782 static const char *conv_attr_name[] = {
783         "crlf", "ident", "filter", "eol", "text",
784 };
785 #define NUM_CONV_ATTRS ARRAY_SIZE(conv_attr_name)
786
787 static void convert_attrs(struct conv_attrs *ca, const char *path)
788 {
789         int i;
790         static struct git_attr_check ccheck[NUM_CONV_ATTRS];
791
792         if (!ccheck[0].attr) {
793                 for (i = 0; i < NUM_CONV_ATTRS; i++)
794                         ccheck[i].attr = git_attr(conv_attr_name[i]);
795                 user_convert_tail = &user_convert;
796                 git_config(read_convert_config, NULL);
797         }
798
799         if (!git_check_attr(path, NUM_CONV_ATTRS, ccheck)) {
800                 ca->crlf_action = git_path_check_crlf(ccheck + 4);
801                 if (ca->crlf_action == CRLF_UNDEFINED)
802                         ca->crlf_action = git_path_check_crlf(ccheck + 0);
803                 ca->attr_action = ca->crlf_action;
804                 ca->ident = git_path_check_ident(ccheck + 1);
805                 ca->drv = git_path_check_convert(ccheck + 2);
806                 if (ca->crlf_action != CRLF_BINARY) {
807                         enum eol eol_attr = git_path_check_eol(ccheck + 3);
808                         if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_LF)
809                                 ca->crlf_action = CRLF_AUTO_INPUT;
810                         else if (ca->crlf_action == CRLF_AUTO && eol_attr == EOL_CRLF)
811                                 ca->crlf_action = CRLF_AUTO_CRLF;
812                         else if (eol_attr == EOL_LF)
813                                 ca->crlf_action = CRLF_TEXT_INPUT;
814                         else if (eol_attr == EOL_CRLF)
815                                 ca->crlf_action = CRLF_TEXT_CRLF;
816                 }
817                 ca->attr_action = ca->crlf_action;
818         } else {
819                 ca->drv = NULL;
820                 ca->crlf_action = CRLF_UNDEFINED;
821                 ca->ident = 0;
822         }
823         if (ca->crlf_action == CRLF_TEXT)
824                 ca->crlf_action = text_eol_is_crlf() ? CRLF_TEXT_CRLF : CRLF_TEXT_INPUT;
825         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_FALSE)
826                 ca->crlf_action = CRLF_BINARY;
827         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_TRUE)
828                 ca->crlf_action = CRLF_AUTO_CRLF;
829         if (ca->crlf_action == CRLF_UNDEFINED && auto_crlf == AUTO_CRLF_INPUT)
830                 ca->crlf_action = CRLF_AUTO_INPUT;
831 }
832
833 int would_convert_to_git_filter_fd(const char *path)
834 {
835         struct conv_attrs ca;
836
837         convert_attrs(&ca, path);
838         if (!ca.drv)
839                 return 0;
840
841         /*
842          * Apply a filter to an fd only if the filter is required to succeed.
843          * We must die if the filter fails, because the original data before
844          * filtering is not available.
845          */
846         if (!ca.drv->required)
847                 return 0;
848
849         return apply_filter(path, NULL, 0, -1, NULL, ca.drv->clean);
850 }
851
852 const char *get_convert_attr_ascii(const char *path)
853 {
854         struct conv_attrs ca;
855
856         convert_attrs(&ca, path);
857         switch (ca.attr_action) {
858         case CRLF_UNDEFINED:
859                 return "";
860         case CRLF_BINARY:
861                 return "-text";
862         case CRLF_TEXT:
863                 return "text";
864         case CRLF_TEXT_INPUT:
865                 return "text eol=lf";
866         case CRLF_TEXT_CRLF:
867                 return "text eol=crlf";
868         case CRLF_AUTO:
869                 return "text=auto";
870         case CRLF_AUTO_CRLF:
871                 return "text=auto eol=crlf";
872         case CRLF_AUTO_INPUT:
873                 return "text=auto eol=lf";
874         }
875         return "";
876 }
877
878 int convert_to_git(const char *path, const char *src, size_t len,
879                    struct strbuf *dst, enum safe_crlf checksafe)
880 {
881         int ret = 0;
882         const char *filter = NULL;
883         int required = 0;
884         struct conv_attrs ca;
885
886         convert_attrs(&ca, path);
887         if (ca.drv) {
888                 filter = ca.drv->clean;
889                 required = ca.drv->required;
890         }
891
892         ret |= apply_filter(path, src, len, -1, dst, filter);
893         if (!ret && required)
894                 die("%s: clean filter '%s' failed", path, ca.drv->name);
895
896         if (ret && dst) {
897                 src = dst->buf;
898                 len = dst->len;
899         }
900         ret |= crlf_to_git(path, src, len, dst, ca.crlf_action, checksafe);
901         if (ret && dst) {
902                 src = dst->buf;
903                 len = dst->len;
904         }
905         return ret | ident_to_git(path, src, len, dst, ca.ident);
906 }
907
908 void convert_to_git_filter_fd(const char *path, int fd, struct strbuf *dst,
909                               enum safe_crlf checksafe)
910 {
911         struct conv_attrs ca;
912         convert_attrs(&ca, path);
913
914         assert(ca.drv);
915         assert(ca.drv->clean);
916
917         if (!apply_filter(path, NULL, 0, fd, dst, ca.drv->clean))
918                 die("%s: clean filter '%s' failed", path, ca.drv->name);
919
920         crlf_to_git(path, dst->buf, dst->len, dst, ca.crlf_action, checksafe);
921         ident_to_git(path, dst->buf, dst->len, dst, ca.ident);
922 }
923
924 static int convert_to_working_tree_internal(const char *path, const char *src,
925                                             size_t len, struct strbuf *dst,
926                                             int normalizing)
927 {
928         int ret = 0, ret_filter = 0;
929         const char *filter = NULL;
930         int required = 0;
931         struct conv_attrs ca;
932
933         convert_attrs(&ca, path);
934         if (ca.drv) {
935                 filter = ca.drv->smudge;
936                 required = ca.drv->required;
937         }
938
939         ret |= ident_to_worktree(path, src, len, dst, ca.ident);
940         if (ret) {
941                 src = dst->buf;
942                 len = dst->len;
943         }
944         /*
945          * CRLF conversion can be skipped if normalizing, unless there
946          * is a smudge filter.  The filter might expect CRLFs.
947          */
948         if (filter || !normalizing) {
949                 ret |= crlf_to_worktree(path, src, len, dst, ca.crlf_action);
950                 if (ret) {
951                         src = dst->buf;
952                         len = dst->len;
953                 }
954         }
955
956         ret_filter = apply_filter(path, src, len, -1, dst, filter);
957         if (!ret_filter && required)
958                 die("%s: smudge filter %s failed", path, ca.drv->name);
959
960         return ret | ret_filter;
961 }
962
963 int convert_to_working_tree(const char *path, const char *src, size_t len, struct strbuf *dst)
964 {
965         return convert_to_working_tree_internal(path, src, len, dst, 0);
966 }
967
968 int renormalize_buffer(const char *path, const char *src, size_t len, struct strbuf *dst)
969 {
970         int ret = convert_to_working_tree_internal(path, src, len, dst, 1);
971         if (ret) {
972                 src = dst->buf;
973                 len = dst->len;
974         }
975         return ret | convert_to_git(path, src, len, dst, SAFE_CRLF_RENORMALIZE);
976 }
977
978 /*****************************************************************
979  *
980  * Streaming conversion support
981  *
982  *****************************************************************/
983
984 typedef int (*filter_fn)(struct stream_filter *,
985                          const char *input, size_t *isize_p,
986                          char *output, size_t *osize_p);
987 typedef void (*free_fn)(struct stream_filter *);
988
989 struct stream_filter_vtbl {
990         filter_fn filter;
991         free_fn free;
992 };
993
994 struct stream_filter {
995         struct stream_filter_vtbl *vtbl;
996 };
997
998 static int null_filter_fn(struct stream_filter *filter,
999                           const char *input, size_t *isize_p,
1000                           char *output, size_t *osize_p)
1001 {
1002         size_t count;
1003
1004         if (!input)
1005                 return 0; /* we do not keep any states */
1006         count = *isize_p;
1007         if (*osize_p < count)
1008                 count = *osize_p;
1009         if (count) {
1010                 memmove(output, input, count);
1011                 *isize_p -= count;
1012                 *osize_p -= count;
1013         }
1014         return 0;
1015 }
1016
1017 static void null_free_fn(struct stream_filter *filter)
1018 {
1019         ; /* nothing -- null instances are shared */
1020 }
1021
1022 static struct stream_filter_vtbl null_vtbl = {
1023         null_filter_fn,
1024         null_free_fn,
1025 };
1026
1027 static struct stream_filter null_filter_singleton = {
1028         &null_vtbl,
1029 };
1030
1031 int is_null_stream_filter(struct stream_filter *filter)
1032 {
1033         return filter == &null_filter_singleton;
1034 }
1035
1036
1037 /*
1038  * LF-to-CRLF filter
1039  */
1040
1041 struct lf_to_crlf_filter {
1042         struct stream_filter filter;
1043         unsigned has_held:1;
1044         char held;
1045 };
1046
1047 static int lf_to_crlf_filter_fn(struct stream_filter *filter,
1048                                 const char *input, size_t *isize_p,
1049                                 char *output, size_t *osize_p)
1050 {
1051         size_t count, o = 0;
1052         struct lf_to_crlf_filter *lf_to_crlf = (struct lf_to_crlf_filter *)filter;
1053
1054         /*
1055          * We may be holding onto the CR to see if it is followed by a
1056          * LF, in which case we would need to go to the main loop.
1057          * Otherwise, just emit it to the output stream.
1058          */
1059         if (lf_to_crlf->has_held && (lf_to_crlf->held != '\r' || !input)) {
1060                 output[o++] = lf_to_crlf->held;
1061                 lf_to_crlf->has_held = 0;
1062         }
1063
1064         /* We are told to drain */
1065         if (!input) {
1066                 *osize_p -= o;
1067                 return 0;
1068         }
1069
1070         count = *isize_p;
1071         if (count || lf_to_crlf->has_held) {
1072                 size_t i;
1073                 int was_cr = 0;
1074
1075                 if (lf_to_crlf->has_held) {
1076                         was_cr = 1;
1077                         lf_to_crlf->has_held = 0;
1078                 }
1079
1080                 for (i = 0; o < *osize_p && i < count; i++) {
1081                         char ch = input[i];
1082
1083                         if (ch == '\n') {
1084                                 output[o++] = '\r';
1085                         } else if (was_cr) {
1086                                 /*
1087                                  * Previous round saw CR and it is not followed
1088                                  * by a LF; emit the CR before processing the
1089                                  * current character.
1090                                  */
1091                                 output[o++] = '\r';
1092                         }
1093
1094                         /*
1095                          * We may have consumed the last output slot,
1096                          * in which case we need to break out of this
1097                          * loop; hold the current character before
1098                          * returning.
1099                          */
1100                         if (*osize_p <= o) {
1101                                 lf_to_crlf->has_held = 1;
1102                                 lf_to_crlf->held = ch;
1103                                 continue; /* break but increment i */
1104                         }
1105
1106                         if (ch == '\r') {
1107                                 was_cr = 1;
1108                                 continue;
1109                         }
1110
1111                         was_cr = 0;
1112                         output[o++] = ch;
1113                 }
1114
1115                 *osize_p -= o;
1116                 *isize_p -= i;
1117
1118                 if (!lf_to_crlf->has_held && was_cr) {
1119                         lf_to_crlf->has_held = 1;
1120                         lf_to_crlf->held = '\r';
1121                 }
1122         }
1123         return 0;
1124 }
1125
1126 static void lf_to_crlf_free_fn(struct stream_filter *filter)
1127 {
1128         free(filter);
1129 }
1130
1131 static struct stream_filter_vtbl lf_to_crlf_vtbl = {
1132         lf_to_crlf_filter_fn,
1133         lf_to_crlf_free_fn,
1134 };
1135
1136 static struct stream_filter *lf_to_crlf_filter(void)
1137 {
1138         struct lf_to_crlf_filter *lf_to_crlf = xcalloc(1, sizeof(*lf_to_crlf));
1139
1140         lf_to_crlf->filter.vtbl = &lf_to_crlf_vtbl;
1141         return (struct stream_filter *)lf_to_crlf;
1142 }
1143
1144 /*
1145  * Cascade filter
1146  */
1147 #define FILTER_BUFFER 1024
1148 struct cascade_filter {
1149         struct stream_filter filter;
1150         struct stream_filter *one;
1151         struct stream_filter *two;
1152         char buf[FILTER_BUFFER];
1153         int end, ptr;
1154 };
1155
1156 static int cascade_filter_fn(struct stream_filter *filter,
1157                              const char *input, size_t *isize_p,
1158                              char *output, size_t *osize_p)
1159 {
1160         struct cascade_filter *cas = (struct cascade_filter *) filter;
1161         size_t filled = 0;
1162         size_t sz = *osize_p;
1163         size_t to_feed, remaining;
1164
1165         /*
1166          * input -- (one) --> buf -- (two) --> output
1167          */
1168         while (filled < sz) {
1169                 remaining = sz - filled;
1170
1171                 /* do we already have something to feed two with? */
1172                 if (cas->ptr < cas->end) {
1173                         to_feed = cas->end - cas->ptr;
1174                         if (stream_filter(cas->two,
1175                                           cas->buf + cas->ptr, &to_feed,
1176                                           output + filled, &remaining))
1177                                 return -1;
1178                         cas->ptr += (cas->end - cas->ptr) - to_feed;
1179                         filled = sz - remaining;
1180                         continue;
1181                 }
1182
1183                 /* feed one from upstream and have it emit into our buffer */
1184                 to_feed = input ? *isize_p : 0;
1185                 if (input && !to_feed)
1186                         break;
1187                 remaining = sizeof(cas->buf);
1188                 if (stream_filter(cas->one,
1189                                   input, &to_feed,
1190                                   cas->buf, &remaining))
1191                         return -1;
1192                 cas->end = sizeof(cas->buf) - remaining;
1193                 cas->ptr = 0;
1194                 if (input) {
1195                         size_t fed = *isize_p - to_feed;
1196                         *isize_p -= fed;
1197                         input += fed;
1198                 }
1199
1200                 /* do we know that we drained one completely? */
1201                 if (input || cas->end)
1202                         continue;
1203
1204                 /* tell two to drain; we have nothing more to give it */
1205                 to_feed = 0;
1206                 remaining = sz - filled;
1207                 if (stream_filter(cas->two,
1208                                   NULL, &to_feed,
1209                                   output + filled, &remaining))
1210                         return -1;
1211                 if (remaining == (sz - filled))
1212                         break; /* completely drained two */
1213                 filled = sz - remaining;
1214         }
1215         *osize_p -= filled;
1216         return 0;
1217 }
1218
1219 static void cascade_free_fn(struct stream_filter *filter)
1220 {
1221         struct cascade_filter *cas = (struct cascade_filter *)filter;
1222         free_stream_filter(cas->one);
1223         free_stream_filter(cas->two);
1224         free(filter);
1225 }
1226
1227 static struct stream_filter_vtbl cascade_vtbl = {
1228         cascade_filter_fn,
1229         cascade_free_fn,
1230 };
1231
1232 static struct stream_filter *cascade_filter(struct stream_filter *one,
1233                                             struct stream_filter *two)
1234 {
1235         struct cascade_filter *cascade;
1236
1237         if (!one || is_null_stream_filter(one))
1238                 return two;
1239         if (!two || is_null_stream_filter(two))
1240                 return one;
1241
1242         cascade = xmalloc(sizeof(*cascade));
1243         cascade->one = one;
1244         cascade->two = two;
1245         cascade->end = cascade->ptr = 0;
1246         cascade->filter.vtbl = &cascade_vtbl;
1247         return (struct stream_filter *)cascade;
1248 }
1249
1250 /*
1251  * ident filter
1252  */
1253 #define IDENT_DRAINING (-1)
1254 #define IDENT_SKIPPING (-2)
1255 struct ident_filter {
1256         struct stream_filter filter;
1257         struct strbuf left;
1258         int state;
1259         char ident[45]; /* ": x40 $" */
1260 };
1261
1262 static int is_foreign_ident(const char *str)
1263 {
1264         int i;
1265
1266         if (!skip_prefix(str, "$Id: ", &str))
1267                 return 0;
1268         for (i = 0; str[i]; i++) {
1269                 if (isspace(str[i]) && str[i+1] != '$')
1270                         return 1;
1271         }
1272         return 0;
1273 }
1274
1275 static void ident_drain(struct ident_filter *ident, char **output_p, size_t *osize_p)
1276 {
1277         size_t to_drain = ident->left.len;
1278
1279         if (*osize_p < to_drain)
1280                 to_drain = *osize_p;
1281         if (to_drain) {
1282                 memcpy(*output_p, ident->left.buf, to_drain);
1283                 strbuf_remove(&ident->left, 0, to_drain);
1284                 *output_p += to_drain;
1285                 *osize_p -= to_drain;
1286         }
1287         if (!ident->left.len)
1288                 ident->state = 0;
1289 }
1290
1291 static int ident_filter_fn(struct stream_filter *filter,
1292                            const char *input, size_t *isize_p,
1293                            char *output, size_t *osize_p)
1294 {
1295         struct ident_filter *ident = (struct ident_filter *)filter;
1296         static const char head[] = "$Id";
1297
1298         if (!input) {
1299                 /* drain upon eof */
1300                 switch (ident->state) {
1301                 default:
1302                         strbuf_add(&ident->left, head, ident->state);
1303                 case IDENT_SKIPPING:
1304                         /* fallthru */
1305                 case IDENT_DRAINING:
1306                         ident_drain(ident, &output, osize_p);
1307                 }
1308                 return 0;
1309         }
1310
1311         while (*isize_p || (ident->state == IDENT_DRAINING)) {
1312                 int ch;
1313
1314                 if (ident->state == IDENT_DRAINING) {
1315                         ident_drain(ident, &output, osize_p);
1316                         if (!*osize_p)
1317                                 break;
1318                         continue;
1319                 }
1320
1321                 ch = *(input++);
1322                 (*isize_p)--;
1323
1324                 if (ident->state == IDENT_SKIPPING) {
1325                         /*
1326                          * Skipping until '$' or LF, but keeping them
1327                          * in case it is a foreign ident.
1328                          */
1329                         strbuf_addch(&ident->left, ch);
1330                         if (ch != '\n' && ch != '$')
1331                                 continue;
1332                         if (ch == '$' && !is_foreign_ident(ident->left.buf)) {
1333                                 strbuf_setlen(&ident->left, sizeof(head) - 1);
1334                                 strbuf_addstr(&ident->left, ident->ident);
1335                         }
1336                         ident->state = IDENT_DRAINING;
1337                         continue;
1338                 }
1339
1340                 if (ident->state < sizeof(head) &&
1341                     head[ident->state] == ch) {
1342                         ident->state++;
1343                         continue;
1344                 }
1345
1346                 if (ident->state)
1347                         strbuf_add(&ident->left, head, ident->state);
1348                 if (ident->state == sizeof(head) - 1) {
1349                         if (ch != ':' && ch != '$') {
1350                                 strbuf_addch(&ident->left, ch);
1351                                 ident->state = 0;
1352                                 continue;
1353                         }
1354
1355                         if (ch == ':') {
1356                                 strbuf_addch(&ident->left, ch);
1357                                 ident->state = IDENT_SKIPPING;
1358                         } else {
1359                                 strbuf_addstr(&ident->left, ident->ident);
1360                                 ident->state = IDENT_DRAINING;
1361                         }
1362                         continue;
1363                 }
1364
1365                 strbuf_addch(&ident->left, ch);
1366                 ident->state = IDENT_DRAINING;
1367         }
1368         return 0;
1369 }
1370
1371 static void ident_free_fn(struct stream_filter *filter)
1372 {
1373         struct ident_filter *ident = (struct ident_filter *)filter;
1374         strbuf_release(&ident->left);
1375         free(filter);
1376 }
1377
1378 static struct stream_filter_vtbl ident_vtbl = {
1379         ident_filter_fn,
1380         ident_free_fn,
1381 };
1382
1383 static struct stream_filter *ident_filter(const unsigned char *sha1)
1384 {
1385         struct ident_filter *ident = xmalloc(sizeof(*ident));
1386
1387         xsnprintf(ident->ident, sizeof(ident->ident),
1388                   ": %s $", sha1_to_hex(sha1));
1389         strbuf_init(&ident->left, 0);
1390         ident->filter.vtbl = &ident_vtbl;
1391         ident->state = 0;
1392         return (struct stream_filter *)ident;
1393 }
1394
1395 /*
1396  * Return an appropriately constructed filter for the path, or NULL if
1397  * the contents cannot be filtered without reading the whole thing
1398  * in-core.
1399  *
1400  * Note that you would be crazy to set CRLF, smuge/clean or ident to a
1401  * large binary blob you would want us not to slurp into the memory!
1402  */
1403 struct stream_filter *get_stream_filter(const char *path, const unsigned char *sha1)
1404 {
1405         struct conv_attrs ca;
1406         struct stream_filter *filter = NULL;
1407
1408         convert_attrs(&ca, path);
1409         if (ca.drv && (ca.drv->smudge || ca.drv->clean))
1410                 return NULL;
1411
1412         if (ca.crlf_action == CRLF_AUTO || ca.crlf_action == CRLF_AUTO_CRLF)
1413                 return NULL;
1414
1415         if (ca.ident)
1416                 filter = ident_filter(sha1);
1417
1418         if (output_eol(ca.crlf_action) == EOL_CRLF)
1419                 filter = cascade_filter(filter, lf_to_crlf_filter());
1420         else
1421                 filter = cascade_filter(filter, &null_filter_singleton);
1422
1423         return filter;
1424 }
1425
1426 void free_stream_filter(struct stream_filter *filter)
1427 {
1428         filter->vtbl->free(filter);
1429 }
1430
1431 int stream_filter(struct stream_filter *filter,
1432                   const char *input, size_t *isize_p,
1433                   char *output, size_t *osize_p)
1434 {
1435         return filter->vtbl->filter(filter, input, isize_p, output, osize_p);
1436 }