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