combine-diff: simplify intersect_paths() further
[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 "unpack-trees.h"
12 #include "refs.h"
13 #include "submodule.h"
14
15 /*
16  * diff-files
17  */
18
19 /*
20  * Has the work tree entity been removed?
21  *
22  * Return 1 if it was removed from the work tree, 0 if an entity to be
23  * compared with the cache entry ce still exists (the latter includes
24  * the case where a directory that is not a submodule repository
25  * exists for ce that is a submodule -- it is a submodule that is not
26  * checked out).  Return negative for an error.
27  */
28 static int check_removed(const struct cache_entry *ce, struct stat *st)
29 {
30         if (lstat(ce->name, st) < 0) {
31                 if (errno != ENOENT && errno != ENOTDIR)
32                         return -1;
33                 return 1;
34         }
35         if (has_symlink_leading_path(ce->name, ce_namelen(ce)))
36                 return 1;
37         if (S_ISDIR(st->st_mode)) {
38                 unsigned char sub[20];
39
40                 /*
41                  * If ce is already a gitlink, we can have a plain
42                  * directory (i.e. the submodule is not checked out),
43                  * or a checked out submodule.  Either case this is not
44                  * a case where something was removed from the work tree,
45                  * so we will return 0.
46                  *
47                  * Otherwise, if the directory is not a submodule
48                  * repository, that means ce which was a blob turned into
49                  * a directory --- the blob was removed!
50                  */
51                 if (!S_ISGITLINK(ce->ce_mode) &&
52                     resolve_gitlink_ref(ce->name, "HEAD", sub))
53                         return 1;
54         }
55         return 0;
56 }
57
58 /*
59  * Has a file changed or has a submodule new commits or a dirty work tree?
60  *
61  * Return 1 when changes are detected, 0 otherwise. If the DIRTY_SUBMODULES
62  * option is set, the caller does not only want to know if a submodule is
63  * modified at all but wants to know all the conditions that are met (new
64  * commits, untracked content and/or modified content).
65  */
66 static int match_stat_with_submodule(struct diff_options *diffopt,
67                                      const struct cache_entry *ce,
68                                      struct stat *st, unsigned ce_option,
69                                      unsigned *dirty_submodule)
70 {
71         int changed = ce_match_stat(ce, st, ce_option);
72         if (S_ISGITLINK(ce->ce_mode)) {
73                 unsigned orig_flags = diffopt->flags;
74                 if (!DIFF_OPT_TST(diffopt, OVERRIDE_SUBMODULE_CONFIG))
75                         set_diffopt_flags_from_submodule_config(diffopt, ce->name);
76                 if (DIFF_OPT_TST(diffopt, IGNORE_SUBMODULES))
77                         changed = 0;
78                 else if (!DIFF_OPT_TST(diffopt, IGNORE_DIRTY_SUBMODULES)
79                     && (!changed || DIFF_OPT_TST(diffopt, DIRTY_SUBMODULES)))
80                         *dirty_submodule = is_submodule_modified(ce->name, DIFF_OPT_TST(diffopt, IGNORE_UNTRACKED_IN_SUBMODULES));
81                 diffopt->flags = orig_flags;
82         }
83         return changed;
84 }
85
86 int run_diff_files(struct rev_info *revs, unsigned int option)
87 {
88         int entries, i;
89         int diff_unmerged_stage = revs->max_count;
90         unsigned ce_option = ((option & DIFF_RACY_IS_MODIFIED)
91                               ? CE_MATCH_RACY_IS_DIRTY : 0);
92
93         if (option & DIFF_SILENT_ON_REMOVED)
94                 handle_deprecated_show_diff_q(&revs->diffopt);
95
96         diff_set_mnemonic_prefix(&revs->diffopt, "i/", "w/");
97
98         if (diff_unmerged_stage < 0)
99                 diff_unmerged_stage = 2;
100         entries = active_nr;
101         for (i = 0; i < entries; i++) {
102                 struct stat st;
103                 unsigned int oldmode, newmode;
104                 struct cache_entry *ce = active_cache[i];
105                 int changed;
106                 unsigned dirty_submodule = 0;
107
108                 if (diff_can_quit_early(&revs->diffopt))
109                         break;
110
111                 if (!ce_path_match(ce, &revs->prune_data))
112                         continue;
113
114                 if (ce_stage(ce)) {
115                         struct combine_diff_path *dpath;
116                         struct diff_filepair *pair;
117                         unsigned int wt_mode = 0;
118                         int num_compare_stages = 0;
119                         size_t path_len;
120
121                         path_len = ce_namelen(ce);
122
123                         dpath = xmalloc(combine_diff_path_size(5, path_len));
124                         dpath->path = (char *) &(dpath->parent[5]);
125
126                         dpath->next = NULL;
127                         memcpy(dpath->path, ce->name, path_len);
128                         dpath->path[path_len] = '\0';
129                         hashclr(dpath->sha1);
130                         memset(&(dpath->parent[0]), 0,
131                                sizeof(struct combine_diff_parent)*5);
132
133                         changed = check_removed(ce, &st);
134                         if (!changed)
135                                 wt_mode = ce_mode_from_stat(ce, st.st_mode);
136                         else {
137                                 if (changed < 0) {
138                                         perror(ce->name);
139                                         continue;
140                                 }
141                                 wt_mode = 0;
142                         }
143                         dpath->mode = wt_mode;
144
145                         while (i < entries) {
146                                 struct cache_entry *nce = active_cache[i];
147                                 int stage;
148
149                                 if (strcmp(ce->name, nce->name))
150                                         break;
151
152                                 /* Stage #2 (ours) is the first parent,
153                                  * stage #3 (theirs) is the second.
154                                  */
155                                 stage = ce_stage(nce);
156                                 if (2 <= stage) {
157                                         int mode = nce->ce_mode;
158                                         num_compare_stages++;
159                                         hashcpy(dpath->parent[stage-2].sha1, nce->sha1);
160                                         dpath->parent[stage-2].mode = ce_mode_from_stat(nce, mode);
161                                         dpath->parent[stage-2].status =
162                                                 DIFF_STATUS_MODIFIED;
163                                 }
164
165                                 /* diff against the proper unmerged stage */
166                                 if (stage == diff_unmerged_stage)
167                                         ce = nce;
168                                 i++;
169                         }
170                         /*
171                          * Compensate for loop update
172                          */
173                         i--;
174
175                         if (revs->combine_merges && num_compare_stages == 2) {
176                                 show_combined_diff(dpath, 2,
177                                                    revs->dense_combined_merges,
178                                                    revs);
179                                 free(dpath);
180                                 continue;
181                         }
182                         free(dpath);
183                         dpath = NULL;
184
185                         /*
186                          * Show the diff for the 'ce' if we found the one
187                          * from the desired stage.
188                          */
189                         pair = diff_unmerge(&revs->diffopt, ce->name);
190                         if (wt_mode)
191                                 pair->two->mode = wt_mode;
192                         if (ce_stage(ce) != diff_unmerged_stage)
193                                 continue;
194                 }
195
196                 if (ce_uptodate(ce) || ce_skip_worktree(ce))
197                         continue;
198
199                 /* If CE_VALID is set, don't look at workdir for file removal */
200                 changed = (ce->ce_flags & CE_VALID) ? 0 : check_removed(ce, &st);
201                 if (changed) {
202                         if (changed < 0) {
203                                 perror(ce->name);
204                                 continue;
205                         }
206                         diff_addremove(&revs->diffopt, '-', ce->ce_mode,
207                                        ce->sha1, !is_null_sha1(ce->sha1),
208                                        ce->name, 0);
209                         continue;
210                 }
211                 changed = match_stat_with_submodule(&revs->diffopt, ce, &st,
212                                                     ce_option, &dirty_submodule);
213                 if (!changed && !dirty_submodule) {
214                         ce_mark_uptodate(ce);
215                         if (!DIFF_OPT_TST(&revs->diffopt, FIND_COPIES_HARDER))
216                                 continue;
217                 }
218                 oldmode = ce->ce_mode;
219                 newmode = ce_mode_from_stat(ce, st.st_mode);
220                 diff_change(&revs->diffopt, oldmode, newmode,
221                             ce->sha1, (changed ? null_sha1 : ce->sha1),
222                             !is_null_sha1(ce->sha1), (changed ? 0 : !is_null_sha1(ce->sha1)),
223                             ce->name, 0, dirty_submodule);
224
225         }
226         diffcore_std(&revs->diffopt);
227         diff_flush(&revs->diffopt);
228         return 0;
229 }
230
231 /*
232  * diff-index
233  */
234
235 /* A file entry went away or appeared */
236 static void diff_index_show_file(struct rev_info *revs,
237                                  const char *prefix,
238                                  const struct cache_entry *ce,
239                                  const unsigned char *sha1, int sha1_valid,
240                                  unsigned int mode,
241                                  unsigned dirty_submodule)
242 {
243         diff_addremove(&revs->diffopt, prefix[0], mode,
244                        sha1, sha1_valid, ce->name, dirty_submodule);
245 }
246
247 static int get_stat_data(const struct cache_entry *ce,
248                          const unsigned char **sha1p,
249                          unsigned int *modep,
250                          int cached, int match_missing,
251                          unsigned *dirty_submodule, struct diff_options *diffopt)
252 {
253         const unsigned char *sha1 = ce->sha1;
254         unsigned int mode = ce->ce_mode;
255
256         if (!cached && !ce_uptodate(ce)) {
257                 int changed;
258                 struct stat st;
259                 changed = check_removed(ce, &st);
260                 if (changed < 0)
261                         return -1;
262                 else if (changed) {
263                         if (match_missing) {
264                                 *sha1p = sha1;
265                                 *modep = mode;
266                                 return 0;
267                         }
268                         return -1;
269                 }
270                 changed = match_stat_with_submodule(diffopt, ce, &st,
271                                                     0, dirty_submodule);
272                 if (changed) {
273                         mode = ce_mode_from_stat(ce, st.st_mode);
274                         sha1 = null_sha1;
275                 }
276         }
277
278         *sha1p = sha1;
279         *modep = mode;
280         return 0;
281 }
282
283 static void show_new_file(struct rev_info *revs,
284                           const struct cache_entry *new,
285                           int cached, int match_missing)
286 {
287         const unsigned char *sha1;
288         unsigned int mode;
289         unsigned dirty_submodule = 0;
290
291         /*
292          * New file in the index: it might actually be different in
293          * the working tree.
294          */
295         if (get_stat_data(new, &sha1, &mode, cached, match_missing,
296             &dirty_submodule, &revs->diffopt) < 0)
297                 return;
298
299         diff_index_show_file(revs, "+", new, sha1, !is_null_sha1(sha1), mode, dirty_submodule);
300 }
301
302 static int show_modified(struct rev_info *revs,
303                          const struct cache_entry *old,
304                          const struct cache_entry *new,
305                          int report_missing,
306                          int cached, int match_missing)
307 {
308         unsigned int mode, oldmode;
309         const unsigned char *sha1;
310         unsigned dirty_submodule = 0;
311
312         if (get_stat_data(new, &sha1, &mode, cached, match_missing,
313                           &dirty_submodule, &revs->diffopt) < 0) {
314                 if (report_missing)
315                         diff_index_show_file(revs, "-", old,
316                                              old->sha1, 1, old->ce_mode, 0);
317                 return -1;
318         }
319
320         if (revs->combine_merges && !cached &&
321             (hashcmp(sha1, old->sha1) || hashcmp(old->sha1, new->sha1))) {
322                 struct combine_diff_path *p;
323                 int pathlen = ce_namelen(new);
324
325                 p = xmalloc(combine_diff_path_size(2, pathlen));
326                 p->path = (char *) &p->parent[2];
327                 p->next = NULL;
328                 memcpy(p->path, new->name, pathlen);
329                 p->path[pathlen] = 0;
330                 p->mode = mode;
331                 hashclr(p->sha1);
332                 memset(p->parent, 0, 2 * sizeof(struct combine_diff_parent));
333                 p->parent[0].status = DIFF_STATUS_MODIFIED;
334                 p->parent[0].mode = new->ce_mode;
335                 hashcpy(p->parent[0].sha1, new->sha1);
336                 p->parent[1].status = DIFF_STATUS_MODIFIED;
337                 p->parent[1].mode = old->ce_mode;
338                 hashcpy(p->parent[1].sha1, old->sha1);
339                 show_combined_diff(p, 2, revs->dense_combined_merges, revs);
340                 free(p);
341                 return 0;
342         }
343
344         oldmode = old->ce_mode;
345         if (mode == oldmode && !hashcmp(sha1, old->sha1) && !dirty_submodule &&
346             !DIFF_OPT_TST(&revs->diffopt, FIND_COPIES_HARDER))
347                 return 0;
348
349         diff_change(&revs->diffopt, oldmode, mode,
350                     old->sha1, sha1, 1, !is_null_sha1(sha1),
351                     old->name, 0, dirty_submodule);
352         return 0;
353 }
354
355 /*
356  * This gets a mix of an existing index and a tree, one pathname entry
357  * at a time. The index entry may be a single stage-0 one, but it could
358  * also be multiple unmerged entries (in which case idx_pos/idx_nr will
359  * give you the position and number of entries in the index).
360  */
361 static void do_oneway_diff(struct unpack_trees_options *o,
362                            const struct cache_entry *idx,
363                            const struct cache_entry *tree)
364 {
365         struct rev_info *revs = o->unpack_data;
366         int match_missing, cached;
367
368         /* if the entry is not checked out, don't examine work tree */
369         cached = o->index_only ||
370                 (idx && ((idx->ce_flags & CE_VALID) || ce_skip_worktree(idx)));
371         /*
372          * Backward compatibility wart - "diff-index -m" does
373          * not mean "do not ignore merges", but "match_missing".
374          *
375          * But with the revision flag parsing, that's found in
376          * "!revs->ignore_merges".
377          */
378         match_missing = !revs->ignore_merges;
379
380         if (cached && idx && ce_stage(idx)) {
381                 struct diff_filepair *pair;
382                 pair = diff_unmerge(&revs->diffopt, idx->name);
383                 if (tree)
384                         fill_filespec(pair->one, tree->sha1, 1, tree->ce_mode);
385                 return;
386         }
387
388         /*
389          * Something added to the tree?
390          */
391         if (!tree) {
392                 show_new_file(revs, idx, cached, match_missing);
393                 return;
394         }
395
396         /*
397          * Something removed from the tree?
398          */
399         if (!idx) {
400                 diff_index_show_file(revs, "-", tree, tree->sha1, 1, tree->ce_mode, 0);
401                 return;
402         }
403
404         /* Show difference between old and new */
405         show_modified(revs, tree, idx, 1, cached, match_missing);
406 }
407
408 /*
409  * The unpack_trees() interface is designed for merging, so
410  * the different source entries are designed primarily for
411  * the source trees, with the old index being really mainly
412  * used for being replaced by the result.
413  *
414  * For diffing, the index is more important, and we only have a
415  * single tree.
416  *
417  * We're supposed to advance o->pos to skip what we have already processed.
418  *
419  * This wrapper makes it all more readable, and takes care of all
420  * the fairly complex unpack_trees() semantic requirements, including
421  * the skipping, the path matching, the type conflict cases etc.
422  */
423 static int oneway_diff(const struct cache_entry * const *src,
424                        struct unpack_trees_options *o)
425 {
426         const struct cache_entry *idx = src[0];
427         const struct cache_entry *tree = src[1];
428         struct rev_info *revs = o->unpack_data;
429
430         /*
431          * Unpack-trees generates a DF/conflict entry if
432          * there was a directory in the index and a tree
433          * in the tree. From a diff standpoint, that's a
434          * delete of the tree and a create of the file.
435          */
436         if (tree == o->df_conflict_entry)
437                 tree = NULL;
438
439         if (ce_path_match(idx ? idx : tree, &revs->prune_data)) {
440                 do_oneway_diff(o, idx, tree);
441                 if (diff_can_quit_early(&revs->diffopt)) {
442                         o->exiting_early = 1;
443                         return -1;
444                 }
445         }
446
447         return 0;
448 }
449
450 static int diff_cache(struct rev_info *revs,
451                       const unsigned char *tree_sha1,
452                       const char *tree_name,
453                       int cached)
454 {
455         struct tree *tree;
456         struct tree_desc t;
457         struct unpack_trees_options opts;
458
459         tree = parse_tree_indirect(tree_sha1);
460         if (!tree)
461                 return error("bad tree object %s",
462                              tree_name ? tree_name : sha1_to_hex(tree_sha1));
463         memset(&opts, 0, sizeof(opts));
464         opts.head_idx = 1;
465         opts.index_only = cached;
466         opts.diff_index_cached = (cached &&
467                                   !DIFF_OPT_TST(&revs->diffopt, FIND_COPIES_HARDER));
468         opts.merge = 1;
469         opts.fn = oneway_diff;
470         opts.unpack_data = revs;
471         opts.src_index = &the_index;
472         opts.dst_index = NULL;
473         opts.pathspec = &revs->diffopt.pathspec;
474         opts.pathspec->recursive = 1;
475
476         init_tree_desc(&t, tree->buffer, tree->size);
477         return unpack_trees(1, &t, &opts);
478 }
479
480 int run_diff_index(struct rev_info *revs, int cached)
481 {
482         struct object_array_entry *ent;
483
484         ent = revs->pending.objects;
485         if (diff_cache(revs, ent->item->sha1, ent->name, cached))
486                 exit(128);
487
488         diff_set_mnemonic_prefix(&revs->diffopt, "c/", cached ? "i/" : "w/");
489         diffcore_fix_diff_index(&revs->diffopt);
490         diffcore_std(&revs->diffopt);
491         diff_flush(&revs->diffopt);
492         return 0;
493 }
494
495 int do_diff_cache(const unsigned char *tree_sha1, struct diff_options *opt)
496 {
497         struct rev_info revs;
498
499         init_revisions(&revs, NULL);
500         copy_pathspec(&revs.prune_data, &opt->pathspec);
501         revs.diffopt = *opt;
502
503         if (diff_cache(&revs, tree_sha1, NULL, 1))
504                 exit(128);
505         return 0;
506 }
507
508 int index_differs_from(const char *def, int diff_flags)
509 {
510         struct rev_info rev;
511         struct setup_revision_opt opt;
512
513         init_revisions(&rev, NULL);
514         memset(&opt, 0, sizeof(opt));
515         opt.def = def;
516         setup_revisions(0, NULL, &rev, &opt);
517         DIFF_OPT_SET(&rev.diffopt, QUICK);
518         DIFF_OPT_SET(&rev.diffopt, EXIT_WITH_STATUS);
519         rev.diffopt.flags |= diff_flags;
520         run_diff_index(&rev, 1);
521         if (rev.pending.alloc)
522                 free(rev.pending.objects);
523         return (DIFF_OPT_TST(&rev.diffopt, HAS_CHANGES) != 0);
524 }