Implement a simple delta_base cache
[git] / diff-lib.c
1 /*
2  * Copyright (C) 2005 Junio C Hamano
3  */
4 #include "cache.h"
5 #include "quote.h"
6 #include "commit.h"
7 #include "diff.h"
8 #include "diffcore.h"
9 #include "revision.h"
10 #include "cache-tree.h"
11 #include "path-list.h"
12
13 /*
14  * diff-files
15  */
16
17 static int read_directory(const char *path, struct path_list *list)
18 {
19         DIR *dir;
20         struct dirent *e;
21
22         if (!(dir = opendir(path)))
23                 return error("Could not open directory %s", path);
24
25         while ((e = readdir(dir)))
26                 if (strcmp(".", e->d_name) && strcmp("..", e->d_name))
27                         path_list_insert(xstrdup(e->d_name), list);
28
29         closedir(dir);
30         return 0;
31 }
32
33 static int get_mode(const char *path, int *mode)
34 {
35         struct stat st;
36
37         if (!path || !strcmp(path, "/dev/null"))
38                 *mode = 0;
39         else if (!strcmp(path, "-"))
40                 *mode = ntohl(create_ce_mode(0666));
41         else if (stat(path, &st))
42                 return error("Could not access '%s'", path);
43         else
44                 *mode = st.st_mode;
45         return 0;
46 }
47
48 static int queue_diff(struct diff_options *o,
49                 const char *name1, const char *name2)
50 {
51         int mode1 = 0, mode2 = 0;
52
53         if (get_mode(name1, &mode1) || get_mode(name2, &mode2))
54                 return -1;
55
56         if (mode1 && mode2 && S_ISDIR(mode1) != S_ISDIR(mode2))
57                 return error("file/directory conflict: %s, %s", name1, name2);
58
59         if (S_ISDIR(mode1) || S_ISDIR(mode2)) {
60                 char buffer1[PATH_MAX], buffer2[PATH_MAX];
61                 struct path_list p1 = {NULL, 0, 0, 1}, p2 = {NULL, 0, 0, 1};
62                 int len1 = 0, len2 = 0, i1, i2, ret = 0;
63
64                 if (name1 && read_directory(name1, &p1))
65                         return -1;
66                 if (name2 && read_directory(name2, &p2)) {
67                         path_list_clear(&p1, 0);
68                         return -1;
69                 }
70
71                 if (name1) {
72                         len1 = strlen(name1);
73                         if (len1 > 0 && name1[len1 - 1] == '/')
74                                 len1--;
75                         memcpy(buffer1, name1, len1);
76                         buffer1[len1++] = '/';
77                 }
78
79                 if (name2) {
80                         len2 = strlen(name2);
81                         if (len2 > 0 && name2[len2 - 1] == '/')
82                                 len2--;
83                         memcpy(buffer2, name2, len2);
84                         buffer2[len2++] = '/';
85                 }
86
87                 for (i1 = i2 = 0; !ret && (i1 < p1.nr || i2 < p2.nr); ) {
88                         const char *n1, *n2;
89                         int comp;
90
91                         if (i1 == p1.nr)
92                                 comp = 1;
93                         else if (i2 == p2.nr)
94                                 comp = -1;
95                         else
96                                 comp = strcmp(p1.items[i1].path,
97                                         p2.items[i2].path);
98
99                         if (comp > 0)
100                                 n1 = NULL;
101                         else {
102                                 n1 = buffer1;
103                                 strncpy(buffer1 + len1, p1.items[i1++].path,
104                                                 PATH_MAX - len1);
105                         }
106
107                         if (comp < 0)
108                                 n2 = NULL;
109                         else {
110                                 n2 = buffer2;
111                                 strncpy(buffer2 + len2, p2.items[i2++].path,
112                                                 PATH_MAX - len2);
113                         }
114
115                         ret = queue_diff(o, n1, n2);
116                 }
117                 path_list_clear(&p1, 0);
118                 path_list_clear(&p2, 0);
119
120                 return ret;
121         } else {
122                 struct diff_filespec *d1, *d2;
123
124                 if (o->reverse_diff) {
125                         unsigned tmp;
126                         const char *tmp_c;
127                         tmp = mode1; mode1 = mode2; mode2 = tmp;
128                         tmp_c = name1; name1 = name2; name2 = tmp_c;
129                 }
130
131                 if (!name1)
132                         name1 = "/dev/null";
133                 if (!name2)
134                         name2 = "/dev/null";
135                 d1 = alloc_filespec(name1);
136                 d2 = alloc_filespec(name2);
137                 fill_filespec(d1, null_sha1, mode1);
138                 fill_filespec(d2, null_sha1, mode2);
139
140                 diff_queue(&diff_queued_diff, d1, d2);
141                 return 0;
142         }
143 }
144
145 static int is_in_index(const char *path)
146 {
147         int len = strlen(path);
148         int pos = cache_name_pos(path, len);
149         char c;
150
151         if (pos < 0)
152                 return 0;
153         if (strncmp(active_cache[pos]->name, path, len))
154                 return 0;
155         c = active_cache[pos]->name[len];
156         return c == '\0' || c == '/';
157 }
158
159 static int handle_diff_files_args(struct rev_info *revs,
160                 int argc, const char **argv, int *silent)
161 {
162         *silent = 0;
163
164         /* revs->max_count == -2 means --no-index */
165         while (1 < argc && argv[1][0] == '-') {
166                 if (!strcmp(argv[1], "--base"))
167                         revs->max_count = 1;
168                 else if (!strcmp(argv[1], "--ours"))
169                         revs->max_count = 2;
170                 else if (!strcmp(argv[1], "--theirs"))
171                         revs->max_count = 3;
172                 else if (!strcmp(argv[1], "-n") ||
173                                 !strcmp(argv[1], "--no-index"))
174                         revs->max_count = -2;
175                 else if (!strcmp(argv[1], "-q"))
176                         *silent = 1;
177                 else
178                         return error("invalid option: %s", argv[1]);
179                 argv++; argc--;
180         }
181
182         if (revs->max_count == -1 && revs->diffopt.nr_paths == 2) {
183                 /*
184                  * If two files are specified, and at least one is untracked,
185                  * default to no-index.
186                  */
187                 read_cache();
188                 if (!is_in_index(revs->diffopt.paths[0]) ||
189                                         !is_in_index(revs->diffopt.paths[1]))
190                         revs->max_count = -2;
191         }
192
193         /*
194          * Make sure there are NO revision (i.e. pending object) parameter,
195          * rev.max_count is reasonable (0 <= n <= 3),
196          * there is no other revision filtering parameters.
197          */
198         if (revs->pending.nr || revs->max_count > 3 ||
199             revs->min_age != -1 || revs->max_age != -1)
200                 return error("no revision allowed with diff-files");
201
202         if (revs->max_count == -1 &&
203             (revs->diffopt.output_format & DIFF_FORMAT_PATCH))
204                 revs->combine_merges = revs->dense_combined_merges = 1;
205
206         return 0;
207 }
208
209 static int is_outside_repo(const char *path, int nongit, const char *prefix)
210 {
211         int i;
212         if (nongit || !strcmp(path, "-") || path[0] == '/')
213                 return 1;
214         if (prefixcmp(path, "../"))
215                 return 0;
216         if (!prefix)
217                 return 1;
218         for (i = strlen(prefix); !prefixcmp(path, "../"); ) {
219                 while (i > 0 && prefix[i - 1] != '/')
220                         i--;
221                 if (--i < 0)
222                         return 1;
223                 path += 3;
224         }
225         return 0;
226 }
227
228 int setup_diff_no_index(struct rev_info *revs,
229                 int argc, const char ** argv, int nongit, const char *prefix)
230 {
231         int i;
232         for (i = 1; i < argc; i++)
233                 if (argv[i][0] != '-' || argv[i][1] == '\0')
234                         break;
235                 else if (!strcmp(argv[i], "--")) {
236                         i++;
237                         break;
238                 } else if (i < argc - 3 && !strcmp(argv[i], "--no-index")) {
239                         i = argc - 3;
240                         break;
241                 }
242         if (argc != i + 2 || (!is_outside_repo(argv[i + 1], nongit, prefix) &&
243                                 !is_outside_repo(argv[i], nongit, prefix)))
244                 return -1;
245
246         diff_setup(&revs->diffopt);
247         for (i = 1; i < argc - 2; )
248                 if (!strcmp(argv[i], "--no-index"))
249                         i++;
250                 else {
251                         int j = diff_opt_parse(&revs->diffopt,
252                                         argv + i, argc - i);
253                         if (!j)
254                                 die("invalid diff option/value: %s", argv[i]);
255                         i += j;
256                 }
257
258         if (prefix) {
259                 int len = strlen(prefix);
260
261                 revs->diffopt.paths = xcalloc(2, sizeof(char*));
262                 for (i = 0; i < 2; i++) {
263                         const char *p = argv[argc - 2 + i];
264                         /*
265                          * stdin should be spelled as '-'; if you have
266                          * path that is '-', spell it as ./-.
267                          */
268                         p = (strcmp(p, "-")
269                              ? xstrdup(prefix_filename(prefix, len, p))
270                              : p);
271                         revs->diffopt.paths[i] = p;
272                 }
273         }
274         else
275                 revs->diffopt.paths = argv + argc - 2;
276         revs->diffopt.nr_paths = 2;
277         revs->max_count = -2;
278         return 0;
279 }
280
281 int run_diff_files_cmd(struct rev_info *revs, int argc, const char **argv)
282 {
283         int silent_on_removed;
284
285         if (handle_diff_files_args(revs, argc, argv, &silent_on_removed))
286                 return -1;
287
288         if (revs->max_count == -2) {
289                 if (revs->diffopt.nr_paths != 2)
290                         return error("need two files/directories with --no-index");
291                 if (queue_diff(&revs->diffopt, revs->diffopt.paths[0],
292                                 revs->diffopt.paths[1]))
293                         return -1;
294                 diffcore_std(&revs->diffopt);
295                 diff_flush(&revs->diffopt);
296                 /*
297                  * The return code for --no-index imitates diff(1):
298                  * 0 = no changes, 1 = changes, else error
299                  */
300                 return revs->diffopt.found_changes;
301         }
302
303         if (read_cache() < 0) {
304                 perror("read_cache");
305                 return -1;
306         }
307         return run_diff_files(revs, silent_on_removed);
308 }
309
310 int run_diff_files(struct rev_info *revs, int silent_on_removed)
311 {
312         int entries, i;
313         int diff_unmerged_stage = revs->max_count;
314
315         if (diff_unmerged_stage < 0)
316                 diff_unmerged_stage = 2;
317         entries = active_nr;
318         for (i = 0; i < entries; i++) {
319                 struct stat st;
320                 unsigned int oldmode, newmode;
321                 struct cache_entry *ce = active_cache[i];
322                 int changed;
323
324                 if (!ce_path_match(ce, revs->prune_data))
325                         continue;
326
327                 if (ce_stage(ce)) {
328                         struct combine_diff_path *dpath;
329                         int num_compare_stages = 0;
330                         size_t path_len;
331
332                         path_len = ce_namelen(ce);
333
334                         dpath = xmalloc(combine_diff_path_size(5, path_len));
335                         dpath->path = (char *) &(dpath->parent[5]);
336
337                         dpath->next = NULL;
338                         dpath->len = path_len;
339                         memcpy(dpath->path, ce->name, path_len);
340                         dpath->path[path_len] = '\0';
341                         hashclr(dpath->sha1);
342                         memset(&(dpath->parent[0]), 0,
343                                sizeof(struct combine_diff_parent)*5);
344
345                         if (lstat(ce->name, &st) < 0) {
346                                 if (errno != ENOENT && errno != ENOTDIR) {
347                                         perror(ce->name);
348                                         continue;
349                                 }
350                                 if (silent_on_removed)
351                                         continue;
352                         }
353                         else
354                                 dpath->mode = canon_mode(st.st_mode);
355
356                         while (i < entries) {
357                                 struct cache_entry *nce = active_cache[i];
358                                 int stage;
359
360                                 if (strcmp(ce->name, nce->name))
361                                         break;
362
363                                 /* Stage #2 (ours) is the first parent,
364                                  * stage #3 (theirs) is the second.
365                                  */
366                                 stage = ce_stage(nce);
367                                 if (2 <= stage) {
368                                         int mode = ntohl(nce->ce_mode);
369                                         num_compare_stages++;
370                                         hashcpy(dpath->parent[stage-2].sha1, nce->sha1);
371                                         dpath->parent[stage-2].mode =
372                                                 canon_mode(mode);
373                                         dpath->parent[stage-2].status =
374                                                 DIFF_STATUS_MODIFIED;
375                                 }
376
377                                 /* diff against the proper unmerged stage */
378                                 if (stage == diff_unmerged_stage)
379                                         ce = nce;
380                                 i++;
381                         }
382                         /*
383                          * Compensate for loop update
384                          */
385                         i--;
386
387                         if (revs->combine_merges && num_compare_stages == 2) {
388                                 show_combined_diff(dpath, 2,
389                                                    revs->dense_combined_merges,
390                                                    revs);
391                                 free(dpath);
392                                 continue;
393                         }
394                         free(dpath);
395                         dpath = NULL;
396
397                         /*
398                          * Show the diff for the 'ce' if we found the one
399                          * from the desired stage.
400                          */
401                         diff_unmerge(&revs->diffopt, ce->name, 0, null_sha1);
402                         if (ce_stage(ce) != diff_unmerged_stage)
403                                 continue;
404                 }
405
406                 if (lstat(ce->name, &st) < 0) {
407                         if (errno != ENOENT && errno != ENOTDIR) {
408                                 perror(ce->name);
409                                 continue;
410                         }
411                         if (silent_on_removed)
412                                 continue;
413                         diff_addremove(&revs->diffopt, '-', ntohl(ce->ce_mode),
414                                        ce->sha1, ce->name, NULL);
415                         continue;
416                 }
417                 changed = ce_match_stat(ce, &st, 0);
418                 if (!changed && !revs->diffopt.find_copies_harder)
419                         continue;
420                 oldmode = ntohl(ce->ce_mode);
421
422                 newmode = canon_mode(st.st_mode);
423                 if (!trust_executable_bit &&
424                     S_ISREG(newmode) && S_ISREG(oldmode) &&
425                     ((newmode ^ oldmode) == 0111))
426                         newmode = oldmode;
427                 else if (!has_symlinks &&
428                     S_ISREG(newmode) && S_ISLNK(oldmode))
429                         newmode = oldmode;
430                 diff_change(&revs->diffopt, oldmode, newmode,
431                             ce->sha1, (changed ? null_sha1 : ce->sha1),
432                             ce->name, NULL);
433
434         }
435         diffcore_std(&revs->diffopt);
436         diff_flush(&revs->diffopt);
437         return 0;
438 }
439
440 /*
441  * diff-index
442  */
443
444 /* A file entry went away or appeared */
445 static void diff_index_show_file(struct rev_info *revs,
446                                  const char *prefix,
447                                  struct cache_entry *ce,
448                                  unsigned char *sha1, unsigned int mode)
449 {
450         diff_addremove(&revs->diffopt, prefix[0], ntohl(mode),
451                        sha1, ce->name, NULL);
452 }
453
454 static int get_stat_data(struct cache_entry *ce,
455                          unsigned char **sha1p,
456                          unsigned int *modep,
457                          int cached, int match_missing)
458 {
459         unsigned char *sha1 = ce->sha1;
460         unsigned int mode = ce->ce_mode;
461
462         if (!cached) {
463                 static unsigned char no_sha1[20];
464                 int changed;
465                 struct stat st;
466                 if (lstat(ce->name, &st) < 0) {
467                         if (errno == ENOENT && match_missing) {
468                                 *sha1p = sha1;
469                                 *modep = mode;
470                                 return 0;
471                         }
472                         return -1;
473                 }
474                 changed = ce_match_stat(ce, &st, 0);
475                 if (changed) {
476                         mode = ce_mode_from_stat(ce, st.st_mode);
477                         sha1 = no_sha1;
478                 }
479         }
480
481         *sha1p = sha1;
482         *modep = mode;
483         return 0;
484 }
485
486 static void show_new_file(struct rev_info *revs,
487                           struct cache_entry *new,
488                           int cached, int match_missing)
489 {
490         unsigned char *sha1;
491         unsigned int mode;
492
493         /* New file in the index: it might actually be different in
494          * the working copy.
495          */
496         if (get_stat_data(new, &sha1, &mode, cached, match_missing) < 0)
497                 return;
498
499         diff_index_show_file(revs, "+", new, sha1, mode);
500 }
501
502 static int show_modified(struct rev_info *revs,
503                          struct cache_entry *old,
504                          struct cache_entry *new,
505                          int report_missing,
506                          int cached, int match_missing)
507 {
508         unsigned int mode, oldmode;
509         unsigned char *sha1;
510
511         if (get_stat_data(new, &sha1, &mode, cached, match_missing) < 0) {
512                 if (report_missing)
513                         diff_index_show_file(revs, "-", old,
514                                              old->sha1, old->ce_mode);
515                 return -1;
516         }
517
518         if (revs->combine_merges && !cached &&
519             (hashcmp(sha1, old->sha1) || hashcmp(old->sha1, new->sha1))) {
520                 struct combine_diff_path *p;
521                 int pathlen = ce_namelen(new);
522
523                 p = xmalloc(combine_diff_path_size(2, pathlen));
524                 p->path = (char *) &p->parent[2];
525                 p->next = NULL;
526                 p->len = pathlen;
527                 memcpy(p->path, new->name, pathlen);
528                 p->path[pathlen] = 0;
529                 p->mode = ntohl(mode);
530                 hashclr(p->sha1);
531                 memset(p->parent, 0, 2 * sizeof(struct combine_diff_parent));
532                 p->parent[0].status = DIFF_STATUS_MODIFIED;
533                 p->parent[0].mode = ntohl(new->ce_mode);
534                 hashcpy(p->parent[0].sha1, new->sha1);
535                 p->parent[1].status = DIFF_STATUS_MODIFIED;
536                 p->parent[1].mode = ntohl(old->ce_mode);
537                 hashcpy(p->parent[1].sha1, old->sha1);
538                 show_combined_diff(p, 2, revs->dense_combined_merges, revs);
539                 free(p);
540                 return 0;
541         }
542
543         oldmode = old->ce_mode;
544         if (mode == oldmode && !hashcmp(sha1, old->sha1) &&
545             !revs->diffopt.find_copies_harder)
546                 return 0;
547
548         mode = ntohl(mode);
549         oldmode = ntohl(oldmode);
550
551         diff_change(&revs->diffopt, oldmode, mode,
552                     old->sha1, sha1, old->name, NULL);
553         return 0;
554 }
555
556 static int diff_cache(struct rev_info *revs,
557                       struct cache_entry **ac, int entries,
558                       const char **pathspec,
559                       int cached, int match_missing)
560 {
561         while (entries) {
562                 struct cache_entry *ce = *ac;
563                 int same = (entries > 1) && ce_same_name(ce, ac[1]);
564
565                 if (!ce_path_match(ce, pathspec))
566                         goto skip_entry;
567
568                 switch (ce_stage(ce)) {
569                 case 0:
570                         /* No stage 1 entry? That means it's a new file */
571                         if (!same) {
572                                 show_new_file(revs, ce, cached, match_missing);
573                                 break;
574                         }
575                         /* Show difference between old and new */
576                         show_modified(revs, ac[1], ce, 1,
577                                       cached, match_missing);
578                         break;
579                 case 1:
580                         /* No stage 3 (merge) entry?
581                          * That means it's been deleted.
582                          */
583                         if (!same) {
584                                 diff_index_show_file(revs, "-", ce,
585                                                      ce->sha1, ce->ce_mode);
586                                 break;
587                         }
588                         /* We come here with ce pointing at stage 1
589                          * (original tree) and ac[1] pointing at stage
590                          * 3 (unmerged).  show-modified with
591                          * report-missing set to false does not say the
592                          * file is deleted but reports true if work
593                          * tree does not have it, in which case we
594                          * fall through to report the unmerged state.
595                          * Otherwise, we show the differences between
596                          * the original tree and the work tree.
597                          */
598                         if (!cached &&
599                             !show_modified(revs, ce, ac[1], 0,
600                                            cached, match_missing))
601                                 break;
602                         diff_unmerge(&revs->diffopt, ce->name,
603                                      ntohl(ce->ce_mode), ce->sha1);
604                         break;
605                 case 3:
606                         diff_unmerge(&revs->diffopt, ce->name,
607                                      0, null_sha1);
608                         break;
609
610                 default:
611                         die("impossible cache entry stage");
612                 }
613
614 skip_entry:
615                 /*
616                  * Ignore all the different stages for this file,
617                  * we've handled the relevant cases now.
618                  */
619                 do {
620                         ac++;
621                         entries--;
622                 } while (entries && ce_same_name(ce, ac[0]));
623         }
624         return 0;
625 }
626
627 /*
628  * This turns all merge entries into "stage 3". That guarantees that
629  * when we read in the new tree (into "stage 1"), we won't lose sight
630  * of the fact that we had unmerged entries.
631  */
632 static void mark_merge_entries(void)
633 {
634         int i;
635         for (i = 0; i < active_nr; i++) {
636                 struct cache_entry *ce = active_cache[i];
637                 if (!ce_stage(ce))
638                         continue;
639                 ce->ce_flags |= htons(CE_STAGEMASK);
640         }
641 }
642
643 int run_diff_index(struct rev_info *revs, int cached)
644 {
645         int ret;
646         struct object *ent;
647         struct tree *tree;
648         const char *tree_name;
649         int match_missing = 0;
650
651         /* 
652          * Backward compatibility wart - "diff-index -m" does
653          * not mean "do not ignore merges", but totally different.
654          */
655         if (!revs->ignore_merges)
656                 match_missing = 1;
657
658         mark_merge_entries();
659
660         ent = revs->pending.objects[0].item;
661         tree_name = revs->pending.objects[0].name;
662         tree = parse_tree_indirect(ent->sha1);
663         if (!tree)
664                 return error("bad tree object %s", tree_name);
665         if (read_tree(tree, 1, revs->prune_data))
666                 return error("unable to read tree object %s", tree_name);
667         ret = diff_cache(revs, active_cache, active_nr, revs->prune_data,
668                          cached, match_missing);
669         diffcore_std(&revs->diffopt);
670         diff_flush(&revs->diffopt);
671         return ret;
672 }
673
674 int do_diff_cache(const unsigned char *tree_sha1, struct diff_options *opt)
675 {
676         struct tree *tree;
677         struct rev_info revs;
678         int i;
679         struct cache_entry **dst;
680         struct cache_entry *last = NULL;
681
682         /*
683          * This is used by git-blame to run diff-cache internally;
684          * it potentially needs to repeatedly run this, so we will
685          * start by removing the higher order entries the last round
686          * left behind.
687          */
688         dst = active_cache;
689         for (i = 0; i < active_nr; i++) {
690                 struct cache_entry *ce = active_cache[i];
691                 if (ce_stage(ce)) {
692                         if (last && !strcmp(ce->name, last->name))
693                                 continue;
694                         cache_tree_invalidate_path(active_cache_tree,
695                                                    ce->name);
696                         last = ce;
697                         ce->ce_mode = 0;
698                         ce->ce_flags &= ~htons(CE_STAGEMASK);
699                 }
700                 *dst++ = ce;
701         }
702         active_nr = dst - active_cache;
703
704         init_revisions(&revs, NULL);
705         revs.prune_data = opt->paths;
706         tree = parse_tree_indirect(tree_sha1);
707         if (!tree)
708                 die("bad tree object %s", sha1_to_hex(tree_sha1));
709         if (read_tree(tree, 1, opt->paths))
710                 return error("unable to read tree %s", sha1_to_hex(tree_sha1));
711         return diff_cache(&revs, active_cache, active_nr, revs.prune_data,
712                           1, 0);
713 }