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