xsize_t: avoid implementation defined behavior when len < 0
[git] / builtin / diff.c
1 /*
2  * Builtin "git diff"
3  *
4  * Copyright (c) 2006 Junio C Hamano
5  */
6 #define USE_THE_INDEX_COMPATIBILITY_MACROS
7 #include "cache.h"
8 #include "config.h"
9 #include "ewah/ewok.h"
10 #include "lockfile.h"
11 #include "color.h"
12 #include "commit.h"
13 #include "blob.h"
14 #include "tag.h"
15 #include "diff.h"
16 #include "diff-merges.h"
17 #include "diffcore.h"
18 #include "revision.h"
19 #include "log-tree.h"
20 #include "builtin.h"
21 #include "submodule.h"
22 #include "oid-array.h"
23
24 #define DIFF_NO_INDEX_EXPLICIT 1
25 #define DIFF_NO_INDEX_IMPLICIT 2
26
27 static const char builtin_diff_usage[] =
28 "git diff [<options>] [<commit>] [--] [<path>...]\n"
29 "   or: git diff [<options>] --cached [<commit>] [--] [<path>...]\n"
30 "   or: git diff [<options>] <commit> [--merge-base] [<commit>...] <commit> [--] [<path>...]\n"
31 "   or: git diff [<options>] <commit>...<commit>] [--] [<path>...]\n"
32 "   or: git diff [<options>] <blob> <blob>]\n"
33 "   or: git diff [<options>] --no-index [--] <path> <path>]\n"
34 COMMON_DIFF_OPTIONS_HELP;
35
36 static const char *blob_path(struct object_array_entry *entry)
37 {
38         return entry->path ? entry->path : entry->name;
39 }
40
41 static void stuff_change(struct diff_options *opt,
42                          unsigned old_mode, unsigned new_mode,
43                          const struct object_id *old_oid,
44                          const struct object_id *new_oid,
45                          int old_oid_valid,
46                          int new_oid_valid,
47                          const char *old_path,
48                          const char *new_path)
49 {
50         struct diff_filespec *one, *two;
51
52         if (!is_null_oid(old_oid) && !is_null_oid(new_oid) &&
53             oideq(old_oid, new_oid) && (old_mode == new_mode))
54                 return;
55
56         if (opt->flags.reverse_diff) {
57                 SWAP(old_mode, new_mode);
58                 SWAP(old_oid, new_oid);
59                 SWAP(old_path, new_path);
60         }
61
62         if (opt->prefix &&
63             (strncmp(old_path, opt->prefix, opt->prefix_length) ||
64              strncmp(new_path, opt->prefix, opt->prefix_length)))
65                 return;
66
67         one = alloc_filespec(old_path);
68         two = alloc_filespec(new_path);
69         fill_filespec(one, old_oid, old_oid_valid, old_mode);
70         fill_filespec(two, new_oid, new_oid_valid, new_mode);
71
72         diff_queue(&diff_queued_diff, one, two);
73 }
74
75 static int builtin_diff_b_f(struct rev_info *revs,
76                             int argc, const char **argv,
77                             struct object_array_entry **blob)
78 {
79         /* Blob vs file in the working tree*/
80         struct stat st;
81         const char *path;
82
83         if (argc > 1)
84                 usage(builtin_diff_usage);
85
86         GUARD_PATHSPEC(&revs->prune_data, PATHSPEC_FROMTOP | PATHSPEC_LITERAL);
87         path = revs->prune_data.items[0].match;
88
89         if (lstat(path, &st))
90                 die_errno(_("failed to stat '%s'"), path);
91         if (!(S_ISREG(st.st_mode) || S_ISLNK(st.st_mode)))
92                 die(_("'%s': not a regular file or symlink"), path);
93
94         diff_set_mnemonic_prefix(&revs->diffopt, "o/", "w/");
95
96         if (blob[0]->mode == S_IFINVALID)
97                 blob[0]->mode = canon_mode(st.st_mode);
98
99         stuff_change(&revs->diffopt,
100                      blob[0]->mode, canon_mode(st.st_mode),
101                      &blob[0]->item->oid, &null_oid,
102                      1, 0,
103                      blob[0]->path ? blob[0]->path : path,
104                      path);
105         diffcore_std(&revs->diffopt);
106         diff_flush(&revs->diffopt);
107         return 0;
108 }
109
110 static int builtin_diff_blobs(struct rev_info *revs,
111                               int argc, const char **argv,
112                               struct object_array_entry **blob)
113 {
114         const unsigned mode = canon_mode(S_IFREG | 0644);
115
116         if (argc > 1)
117                 usage(builtin_diff_usage);
118
119         if (blob[0]->mode == S_IFINVALID)
120                 blob[0]->mode = mode;
121
122         if (blob[1]->mode == S_IFINVALID)
123                 blob[1]->mode = mode;
124
125         stuff_change(&revs->diffopt,
126                      blob[0]->mode, blob[1]->mode,
127                      &blob[0]->item->oid, &blob[1]->item->oid,
128                      1, 1,
129                      blob_path(blob[0]), blob_path(blob[1]));
130         diffcore_std(&revs->diffopt);
131         diff_flush(&revs->diffopt);
132         return 0;
133 }
134
135 static int builtin_diff_index(struct rev_info *revs,
136                               int argc, const char **argv)
137 {
138         unsigned int option = 0;
139         while (1 < argc) {
140                 const char *arg = argv[1];
141                 if (!strcmp(arg, "--cached") || !strcmp(arg, "--staged"))
142                         option |= DIFF_INDEX_CACHED;
143                 else if (!strcmp(arg, "--merge-base"))
144                         option |= DIFF_INDEX_MERGE_BASE;
145                 else
146                         usage(builtin_diff_usage);
147                 argv++; argc--;
148         }
149         /*
150          * Make sure there is one revision (i.e. pending object),
151          * and there is no revision filtering parameters.
152          */
153         if (revs->pending.nr != 1 ||
154             revs->max_count != -1 || revs->min_age != -1 ||
155             revs->max_age != -1)
156                 usage(builtin_diff_usage);
157         if (!(option & DIFF_INDEX_CACHED)) {
158                 setup_work_tree();
159                 if (read_cache_preload(&revs->diffopt.pathspec) < 0) {
160                         perror("read_cache_preload");
161                         return -1;
162                 }
163         } else if (read_cache() < 0) {
164                 perror("read_cache");
165                 return -1;
166         }
167         return run_diff_index(revs, option);
168 }
169
170 static int builtin_diff_tree(struct rev_info *revs,
171                              int argc, const char **argv,
172                              struct object_array_entry *ent0,
173                              struct object_array_entry *ent1)
174 {
175         const struct object_id *(oid[2]);
176         struct object_id mb_oid;
177         int merge_base = 0;
178
179         while (1 < argc) {
180                 const char *arg = argv[1];
181                 if (!strcmp(arg, "--merge-base"))
182                         merge_base = 1;
183                 else
184                         usage(builtin_diff_usage);
185                 argv++; argc--;
186         }
187
188         if (merge_base) {
189                 diff_get_merge_base(revs, &mb_oid);
190                 oid[0] = &mb_oid;
191                 oid[1] = &revs->pending.objects[1].item->oid;
192         } else {
193                 int swap = 0;
194
195                 /*
196                  * We saw two trees, ent0 and ent1.  If ent1 is uninteresting,
197                  * swap them.
198                  */
199                 if (ent1->item->flags & UNINTERESTING)
200                         swap = 1;
201                 oid[swap] = &ent0->item->oid;
202                 oid[1 - swap] = &ent1->item->oid;
203         }
204         diff_tree_oid(oid[0], oid[1], "", &revs->diffopt);
205         log_tree_diff_flush(revs);
206         return 0;
207 }
208
209 static int builtin_diff_combined(struct rev_info *revs,
210                                  int argc, const char **argv,
211                                  struct object_array_entry *ent,
212                                  int ents)
213 {
214         struct oid_array parents = OID_ARRAY_INIT;
215         int i;
216
217         if (argc > 1)
218                 usage(builtin_diff_usage);
219
220         diff_merges_set_dense_combined_if_unset(revs);
221
222         for (i = 1; i < ents; i++)
223                 oid_array_append(&parents, &ent[i].item->oid);
224         diff_tree_combined(&ent[0].item->oid, &parents, revs);
225         oid_array_clear(&parents);
226         return 0;
227 }
228
229 static void refresh_index_quietly(void)
230 {
231         struct lock_file lock_file = LOCK_INIT;
232         int fd;
233
234         fd = hold_locked_index(&lock_file, 0);
235         if (fd < 0)
236                 return;
237         discard_cache();
238         read_cache();
239         refresh_cache(REFRESH_QUIET|REFRESH_UNMERGED);
240         repo_update_index_if_able(the_repository, &lock_file);
241 }
242
243 static int builtin_diff_files(struct rev_info *revs, int argc, const char **argv)
244 {
245         unsigned int options = 0;
246
247         while (1 < argc && argv[1][0] == '-') {
248                 if (!strcmp(argv[1], "--base"))
249                         revs->max_count = 1;
250                 else if (!strcmp(argv[1], "--ours"))
251                         revs->max_count = 2;
252                 else if (!strcmp(argv[1], "--theirs"))
253                         revs->max_count = 3;
254                 else if (!strcmp(argv[1], "-q"))
255                         options |= DIFF_SILENT_ON_REMOVED;
256                 else if (!strcmp(argv[1], "-h"))
257                         usage(builtin_diff_usage);
258                 else
259                         return error(_("invalid option: %s"), argv[1]);
260                 argv++; argc--;
261         }
262
263         /*
264          * "diff --base" should not combine merges because it was not
265          * asked to.  "diff -c" should not densify (if the user wants
266          * dense one, --cc can be explicitly asked for, or just rely
267          * on the default).
268          */
269         if (revs->max_count == -1 &&
270             (revs->diffopt.output_format & DIFF_FORMAT_PATCH))
271                 diff_merges_set_dense_combined_if_unset(revs);
272
273         setup_work_tree();
274         if (read_cache_preload(&revs->diffopt.pathspec) < 0) {
275                 perror("read_cache_preload");
276                 return -1;
277         }
278         return run_diff_files(revs, options);
279 }
280
281 struct symdiff {
282         struct bitmap *skip;
283         int warn;
284         const char *base, *left, *right;
285 };
286
287 /*
288  * Check for symmetric-difference arguments, and if present, arrange
289  * everything we need to know to handle them correctly.  As a bonus,
290  * weed out all bogus range-based revision specifications, e.g.,
291  * "git diff A..B C..D" or "git diff A..B C" get rejected.
292  *
293  * For an actual symmetric diff, *symdiff is set this way:
294  *
295  *  - its skip is non-NULL and marks *all* rev->pending.objects[i]
296  *    indices that the caller should ignore (extra merge bases, of
297  *    which there might be many, and A in A...B).  Note that the
298  *    chosen merge base and right side are NOT marked.
299  *  - warn is set if there are multiple merge bases.
300  *  - base, left, and right point to the names to use in a
301  *    warning about multiple merge bases.
302  *
303  * If there is no symmetric diff argument, sym->skip is NULL and
304  * sym->warn is cleared.  The remaining fields are not set.
305  */
306 static void symdiff_prepare(struct rev_info *rev, struct symdiff *sym)
307 {
308         int i, is_symdiff = 0, basecount = 0, othercount = 0;
309         int lpos = -1, rpos = -1, basepos = -1;
310         struct bitmap *map = NULL;
311
312         /*
313          * Use the whence fields to find merge bases and left and
314          * right parts of symmetric difference, so that we do not
315          * depend on the order that revisions are parsed.  If there
316          * are any revs that aren't from these sources, we have a
317          * "git diff C A...B" or "git diff A...B C" case.  Or we
318          * could even get "git diff A...B C...E", for instance.
319          *
320          * If we don't have just one merge base, we pick one
321          * at random.
322          *
323          * NB: REV_CMD_LEFT, REV_CMD_RIGHT are also used for A..B,
324          * so we must check for SYMMETRIC_LEFT too.  The two arrays
325          * rev->pending.objects and rev->cmdline.rev are parallel.
326          */
327         for (i = 0; i < rev->cmdline.nr; i++) {
328                 struct object *obj = rev->pending.objects[i].item;
329                 switch (rev->cmdline.rev[i].whence) {
330                 case REV_CMD_MERGE_BASE:
331                         if (basepos < 0)
332                                 basepos = i;
333                         basecount++;
334                         break;          /* do mark all bases */
335                 case REV_CMD_LEFT:
336                         if (lpos >= 0)
337                                 usage(builtin_diff_usage);
338                         lpos = i;
339                         if (obj->flags & SYMMETRIC_LEFT) {
340                                 is_symdiff = 1;
341                                 break;  /* do mark A */
342                         }
343                         continue;
344                 case REV_CMD_RIGHT:
345                         if (rpos >= 0)
346                                 usage(builtin_diff_usage);
347                         rpos = i;
348                         continue;       /* don't mark B */
349                 case REV_CMD_PARENTS_ONLY:
350                 case REV_CMD_REF:
351                 case REV_CMD_REV:
352                         othercount++;
353                         continue;
354                 }
355                 if (map == NULL)
356                         map = bitmap_new();
357                 bitmap_set(map, i);
358         }
359
360         /*
361          * Forbid any additional revs for both A...B and A..B.
362          */
363         if (lpos >= 0 && othercount > 0)
364                 usage(builtin_diff_usage);
365
366         if (!is_symdiff) {
367                 bitmap_free(map);
368                 sym->warn = 0;
369                 sym->skip = NULL;
370                 return;
371         }
372
373         sym->left = rev->pending.objects[lpos].name;
374         sym->right = rev->pending.objects[rpos].name;
375         if (basecount == 0)
376                 die(_("%s...%s: no merge base"), sym->left, sym->right);
377         sym->base = rev->pending.objects[basepos].name;
378         bitmap_unset(map, basepos);     /* unmark the base we want */
379         sym->warn = basecount > 1;
380         sym->skip = map;
381 }
382
383 int cmd_diff(int argc, const char **argv, const char *prefix)
384 {
385         int i;
386         struct rev_info rev;
387         struct object_array ent = OBJECT_ARRAY_INIT;
388         int blobs = 0, paths = 0;
389         struct object_array_entry *blob[2];
390         int nongit = 0, no_index = 0;
391         int result = 0;
392         struct symdiff sdiff;
393
394         /*
395          * We could get N tree-ish in the rev.pending_objects list.
396          * Also there could be M blobs there, and P pathspecs. --cached may
397          * also be present.
398          *
399          * N=0, M=0:
400          *      cache vs files (diff-files)
401          *
402          * N=0, M=0, --cached:
403          *      HEAD vs cache (diff-index --cached)
404          *
405          * N=0, M=2:
406          *      compare two random blobs.  P must be zero.
407          *
408          * N=0, M=1, P=1:
409          *      compare a blob with a working tree file.
410          *
411          * N=1, M=0:
412          *      tree vs files (diff-index)
413          *
414          * N=1, M=0, --cached:
415          *      tree vs cache (diff-index --cached)
416          *
417          * N=2, M=0:
418          *      tree vs tree (diff-tree)
419          *
420          * N=0, M=0, P=2:
421          *      compare two filesystem entities (aka --no-index).
422          *
423          * Other cases are errors.
424          */
425
426         /* Were we asked to do --no-index explicitly? */
427         for (i = 1; i < argc; i++) {
428                 if (!strcmp(argv[i], "--")) {
429                         i++;
430                         break;
431                 }
432                 if (!strcmp(argv[i], "--no-index"))
433                         no_index = DIFF_NO_INDEX_EXPLICIT;
434                 if (argv[i][0] != '-')
435                         break;
436         }
437
438         prefix = setup_git_directory_gently(&nongit);
439
440         if (!no_index) {
441                 /*
442                  * Treat git diff with at least one path outside of the
443                  * repo the same as if the command would have been executed
444                  * outside of a git repository.  In this case it behaves
445                  * the same way as "git diff --no-index <a> <b>", which acts
446                  * as a colourful "diff" replacement.
447                  */
448                 if (nongit || ((argc == i + 2) &&
449                                (!path_inside_repo(prefix, argv[i]) ||
450                                 !path_inside_repo(prefix, argv[i + 1]))))
451                         no_index = DIFF_NO_INDEX_IMPLICIT;
452         }
453
454         init_diff_ui_defaults();
455         git_config(git_diff_ui_config, NULL);
456         prefix = precompose_argv_prefix(argc, argv, prefix);
457
458         repo_init_revisions(the_repository, &rev, prefix);
459
460         /* Set up defaults that will apply to both no-index and regular diffs. */
461         rev.diffopt.stat_width = -1;
462         rev.diffopt.stat_graph_width = -1;
463         rev.diffopt.flags.allow_external = 1;
464         rev.diffopt.flags.allow_textconv = 1;
465
466         /* If this is a no-index diff, just run it and exit there. */
467         if (no_index)
468                 exit(diff_no_index(&rev, no_index == DIFF_NO_INDEX_IMPLICIT,
469                                    argc, argv));
470
471
472         /*
473          * Otherwise, we are doing the usual "git" diff; set up any
474          * further defaults that apply to regular diffs.
475          */
476         rev.diffopt.skip_stat_unmatch = !!diff_auto_refresh_index;
477
478         /*
479          * Default to intent-to-add entries invisible in the
480          * index. This makes them show up as new files in diff-files
481          * and not at all in diff-cached.
482          */
483         rev.diffopt.ita_invisible_in_index = 1;
484
485         if (nongit)
486                 die(_("Not a git repository"));
487         argc = setup_revisions(argc, argv, &rev, NULL);
488         if (!rev.diffopt.output_format) {
489                 rev.diffopt.output_format = DIFF_FORMAT_PATCH;
490                 diff_setup_done(&rev.diffopt);
491         }
492
493         rev.diffopt.flags.recursive = 1;
494         rev.diffopt.rotate_to_strict = 1;
495
496         setup_diff_pager(&rev.diffopt);
497
498         /*
499          * Do we have --cached and not have a pending object, then
500          * default to HEAD by hand.  Eek.
501          */
502         if (!rev.pending.nr) {
503                 int i;
504                 for (i = 1; i < argc; i++) {
505                         const char *arg = argv[i];
506                         if (!strcmp(arg, "--"))
507                                 break;
508                         else if (!strcmp(arg, "--cached") ||
509                                  !strcmp(arg, "--staged")) {
510                                 add_head_to_pending(&rev);
511                                 if (!rev.pending.nr) {
512                                         struct tree *tree;
513                                         tree = lookup_tree(the_repository,
514                                                            the_repository->hash_algo->empty_tree);
515                                         add_pending_object(&rev, &tree->object, "HEAD");
516                                 }
517                                 break;
518                         }
519                 }
520         }
521
522         symdiff_prepare(&rev, &sdiff);
523         for (i = 0; i < rev.pending.nr; i++) {
524                 struct object_array_entry *entry = &rev.pending.objects[i];
525                 struct object *obj = entry->item;
526                 const char *name = entry->name;
527                 int flags = (obj->flags & UNINTERESTING);
528                 if (!obj->parsed)
529                         obj = parse_object(the_repository, &obj->oid);
530                 obj = deref_tag(the_repository, obj, NULL, 0);
531                 if (!obj)
532                         die(_("invalid object '%s' given."), name);
533                 if (obj->type == OBJ_COMMIT)
534                         obj = &get_commit_tree(((struct commit *)obj))->object;
535
536                 if (obj->type == OBJ_TREE) {
537                         if (sdiff.skip && bitmap_get(sdiff.skip, i))
538                                 continue;
539                         obj->flags |= flags;
540                         add_object_array(obj, name, &ent);
541                 } else if (obj->type == OBJ_BLOB) {
542                         if (2 <= blobs)
543                                 die(_("more than two blobs given: '%s'"), name);
544                         blob[blobs] = entry;
545                         blobs++;
546
547                 } else {
548                         die(_("unhandled object '%s' given."), name);
549                 }
550         }
551         if (rev.prune_data.nr)
552                 paths += rev.prune_data.nr;
553
554         /*
555          * Now, do the arguments look reasonable?
556          */
557         if (!ent.nr) {
558                 switch (blobs) {
559                 case 0:
560                         result = builtin_diff_files(&rev, argc, argv);
561                         break;
562                 case 1:
563                         if (paths != 1)
564                                 usage(builtin_diff_usage);
565                         result = builtin_diff_b_f(&rev, argc, argv, blob);
566                         break;
567                 case 2:
568                         if (paths)
569                                 usage(builtin_diff_usage);
570                         result = builtin_diff_blobs(&rev, argc, argv, blob);
571                         break;
572                 default:
573                         usage(builtin_diff_usage);
574                 }
575         }
576         else if (blobs)
577                 usage(builtin_diff_usage);
578         else if (ent.nr == 1)
579                 result = builtin_diff_index(&rev, argc, argv);
580         else if (ent.nr == 2) {
581                 if (sdiff.warn)
582                         warning(_("%s...%s: multiple merge bases, using %s"),
583                                 sdiff.left, sdiff.right, sdiff.base);
584                 result = builtin_diff_tree(&rev, argc, argv,
585                                            &ent.objects[0], &ent.objects[1]);
586         } else
587                 result = builtin_diff_combined(&rev, argc, argv,
588                                                ent.objects, ent.nr);
589         result = diff_result_code(&rev.diffopt, result);
590         if (1 < rev.diffopt.skip_stat_unmatch)
591                 refresh_index_quietly();
592         UNLEAK(rev);
593         UNLEAK(ent);
594         UNLEAK(blob);
595         return result;
596 }