revision traversal: --remove-empty fix.
[git] / revision.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "blob.h"
4 #include "tree.h"
5 #include "commit.h"
6 #include "diff.h"
7 #include "refs.h"
8 #include "revision.h"
9
10 static char *path_name(struct name_path *path, const char *name)
11 {
12         struct name_path *p;
13         char *n, *m;
14         int nlen = strlen(name);
15         int len = nlen + 1;
16
17         for (p = path; p; p = p->up) {
18                 if (p->elem_len)
19                         len += p->elem_len + 1;
20         }
21         n = xmalloc(len);
22         m = n + len - (nlen + 1);
23         strcpy(m, name);
24         for (p = path; p; p = p->up) {
25                 if (p->elem_len) {
26                         m -= p->elem_len + 1;
27                         memcpy(m, p->elem, p->elem_len);
28                         m[p->elem_len] = '/';
29                 }
30         }
31         return n;
32 }
33
34 struct object_list **add_object(struct object *obj,
35                                        struct object_list **p,
36                                        struct name_path *path,
37                                        const char *name)
38 {
39         struct object_list *entry = xmalloc(sizeof(*entry));
40         entry->item = obj;
41         entry->next = *p;
42         entry->name = path_name(path, name);
43         *p = entry;
44         return &entry->next;
45 }
46
47 static void mark_blob_uninteresting(struct blob *blob)
48 {
49         if (blob->object.flags & UNINTERESTING)
50                 return;
51         blob->object.flags |= UNINTERESTING;
52 }
53
54 void mark_tree_uninteresting(struct tree *tree)
55 {
56         struct object *obj = &tree->object;
57         struct tree_entry_list *entry;
58
59         if (obj->flags & UNINTERESTING)
60                 return;
61         obj->flags |= UNINTERESTING;
62         if (!has_sha1_file(obj->sha1))
63                 return;
64         if (parse_tree(tree) < 0)
65                 die("bad tree %s", sha1_to_hex(obj->sha1));
66         entry = tree->entries;
67         tree->entries = NULL;
68         while (entry) {
69                 struct tree_entry_list *next = entry->next;
70                 if (entry->directory)
71                         mark_tree_uninteresting(entry->item.tree);
72                 else
73                         mark_blob_uninteresting(entry->item.blob);
74                 free(entry);
75                 entry = next;
76         }
77 }
78
79 void mark_parents_uninteresting(struct commit *commit)
80 {
81         struct commit_list *parents = commit->parents;
82
83         while (parents) {
84                 struct commit *commit = parents->item;
85                 if (!(commit->object.flags & UNINTERESTING)) {
86                         commit->object.flags |= UNINTERESTING;
87
88                         /*
89                          * Normally we haven't parsed the parent
90                          * yet, so we won't have a parent of a parent
91                          * here. However, it may turn out that we've
92                          * reached this commit some other way (where it
93                          * wasn't uninteresting), in which case we need
94                          * to mark its parents recursively too..
95                          */
96                         if (commit->parents)
97                                 mark_parents_uninteresting(commit);
98                 }
99
100                 /*
101                  * A missing commit is ok iff its parent is marked
102                  * uninteresting.
103                  *
104                  * We just mark such a thing parsed, so that when
105                  * it is popped next time around, we won't be trying
106                  * to parse it and get an error.
107                  */
108                 if (!has_sha1_file(commit->object.sha1))
109                         commit->object.parsed = 1;
110                 parents = parents->next;
111         }
112 }
113
114 static void add_pending_object(struct rev_info *revs, struct object *obj, const char *name)
115 {
116         add_object(obj, &revs->pending_objects, NULL, name);
117 }
118
119 static struct commit *get_commit_reference(struct rev_info *revs, const char *name, const unsigned char *sha1, unsigned int flags)
120 {
121         struct object *object;
122
123         object = parse_object(sha1);
124         if (!object)
125                 die("bad object %s", name);
126
127         /*
128          * Tag object? Look what it points to..
129          */
130         while (object->type == tag_type) {
131                 struct tag *tag = (struct tag *) object;
132                 object->flags |= flags;
133                 if (revs->tag_objects && !(object->flags & UNINTERESTING))
134                         add_pending_object(revs, object, tag->tag);
135                 object = parse_object(tag->tagged->sha1);
136                 if (!object)
137                         die("bad object %s", sha1_to_hex(tag->tagged->sha1));
138         }
139
140         /*
141          * Commit object? Just return it, we'll do all the complex
142          * reachability crud.
143          */
144         if (object->type == commit_type) {
145                 struct commit *commit = (struct commit *)object;
146                 object->flags |= flags;
147                 if (parse_commit(commit) < 0)
148                         die("unable to parse commit %s", name);
149                 if (flags & UNINTERESTING) {
150                         mark_parents_uninteresting(commit);
151                         revs->limited = 1;
152                 }
153                 return commit;
154         }
155
156         /*
157          * Tree object? Either mark it uniniteresting, or add it
158          * to the list of objects to look at later..
159          */
160         if (object->type == tree_type) {
161                 struct tree *tree = (struct tree *)object;
162                 if (!revs->tree_objects)
163                         return NULL;
164                 if (flags & UNINTERESTING) {
165                         mark_tree_uninteresting(tree);
166                         return NULL;
167                 }
168                 add_pending_object(revs, object, "");
169                 return NULL;
170         }
171
172         /*
173          * Blob object? You know the drill by now..
174          */
175         if (object->type == blob_type) {
176                 struct blob *blob = (struct blob *)object;
177                 if (!revs->blob_objects)
178                         return NULL;
179                 if (flags & UNINTERESTING) {
180                         mark_blob_uninteresting(blob);
181                         return NULL;
182                 }
183                 add_pending_object(revs, object, "");
184                 return NULL;
185         }
186         die("%s is unknown object", name);
187 }
188
189 static int everybody_uninteresting(struct commit_list *orig)
190 {
191         struct commit_list *list = orig;
192         while (list) {
193                 struct commit *commit = list->item;
194                 list = list->next;
195                 if (commit->object.flags & UNINTERESTING)
196                         continue;
197                 return 0;
198         }
199         return 1;
200 }
201
202 #define TREE_SAME       0
203 #define TREE_NEW        1
204 #define TREE_DIFFERENT  2
205 static int tree_difference = TREE_SAME;
206
207 static void file_add_remove(struct diff_options *options,
208                     int addremove, unsigned mode,
209                     const unsigned char *sha1,
210                     const char *base, const char *path)
211 {
212         int diff = TREE_DIFFERENT;
213
214         /*
215          * Is it an add of a new file? It means that
216          * the old tree didn't have it at all, so we
217          * will turn "TREE_SAME" -> "TREE_NEW", but
218          * leave any "TREE_DIFFERENT" alone (and if
219          * it already was "TREE_NEW", we'll keep it
220          * "TREE_NEW" of course).
221          */
222         if (addremove == '+') {
223                 diff = tree_difference;
224                 if (diff != TREE_SAME)
225                         return;
226                 diff = TREE_NEW;
227         }
228         tree_difference = diff;
229 }
230
231 static void file_change(struct diff_options *options,
232                  unsigned old_mode, unsigned new_mode,
233                  const unsigned char *old_sha1,
234                  const unsigned char *new_sha1,
235                  const char *base, const char *path)
236 {
237         tree_difference = TREE_DIFFERENT;
238 }
239
240 static struct diff_options diff_opt = {
241         .recursive = 1,
242         .add_remove = file_add_remove,
243         .change = file_change,
244 };
245
246 static int compare_tree(struct tree *t1, struct tree *t2)
247 {
248         if (!t1)
249                 return TREE_NEW;
250         if (!t2)
251                 return TREE_DIFFERENT;
252         tree_difference = TREE_SAME;
253         if (diff_tree_sha1(t1->object.sha1, t2->object.sha1, "", &diff_opt) < 0)
254                 return TREE_DIFFERENT;
255         return tree_difference;
256 }
257
258 static int same_tree_as_empty(struct tree *t1)
259 {
260         int retval;
261         void *tree;
262         struct tree_desc empty, real;
263
264         if (!t1)
265                 return 0;
266
267         tree = read_object_with_reference(t1->object.sha1, "tree", &real.size, NULL);
268         if (!tree)
269                 return 0;
270         real.buf = tree;
271
272         empty.buf = "";
273         empty.size = 0;
274
275         tree_difference = 0;
276         retval = diff_tree(&empty, &real, "", &diff_opt);
277         free(tree);
278
279         return retval >= 0 && !tree_difference;
280 }
281
282 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
283 {
284         struct commit_list **pp, *parent;
285         int tree_changed = 0;
286
287         if (!commit->tree)
288                 return;
289
290         if (!commit->parents) {
291                 if (!same_tree_as_empty(commit->tree))
292                         commit->object.flags |= TREECHANGE;
293                 return;
294         }
295
296         pp = &commit->parents;
297         while ((parent = *pp) != NULL) {
298                 struct commit *p = parent->item;
299
300                 parse_commit(p);
301                 switch (compare_tree(p->tree, commit->tree)) {
302                 case TREE_SAME:
303                         if (p->object.flags & UNINTERESTING) {
304                                 /* Even if a merge with an uninteresting
305                                  * side branch brought the entire change
306                                  * we are interested in, we do not want
307                                  * to lose the other branches of this
308                                  * merge, so we just keep going.
309                                  */
310                                 pp = &parent->next;
311                                 continue;
312                         }
313                         parent->next = NULL;
314                         commit->parents = parent;
315                         return;
316
317                 case TREE_NEW:
318                         if (revs->remove_empty_trees &&
319                             same_tree_as_empty(p->tree)) {
320                                 /* We are adding all the specified paths from
321                                  * this parent, so the parents of it is
322                                  * not interesting, but the difference between
323                                  * this parent and us still is interesting.
324                                  */
325                                 p->object.flags |= UNINTERESTING;
326                         }
327                 /* fallthrough */
328                 case TREE_DIFFERENT:
329                         tree_changed = 1;
330                         pp = &parent->next;
331                         continue;
332                 }
333                 die("bad tree compare for commit %s", sha1_to_hex(commit->object.sha1));
334         }
335         if (tree_changed)
336                 commit->object.flags |= TREECHANGE;
337 }
338
339 static void add_parents_to_list(struct rev_info *revs, struct commit *commit, struct commit_list **list)
340 {
341         struct commit_list *parent = commit->parents;
342
343         /*
344          * If the commit is uninteresting, don't try to
345          * prune parents - we want the maximal uninteresting
346          * set.
347          *
348          * Normally we haven't parsed the parent
349          * yet, so we won't have a parent of a parent
350          * here. However, it may turn out that we've
351          * reached this commit some other way (where it
352          * wasn't uninteresting), in which case we need
353          * to mark its parents recursively too..
354          */
355         if (commit->object.flags & UNINTERESTING) {
356                 while (parent) {
357                         struct commit *p = parent->item;
358                         parent = parent->next;
359                         parse_commit(p);
360                         p->object.flags |= UNINTERESTING;
361                         if (p->parents)
362                                 mark_parents_uninteresting(p);
363                         if (p->object.flags & SEEN)
364                                 continue;
365                         p->object.flags |= SEEN;
366                         insert_by_date(p, list);
367                 }
368                 return;
369         }
370
371         /*
372          * Ok, the commit wasn't uninteresting. Try to
373          * simplify the commit history and find the parent
374          * that has no differences in the path set if one exists.
375          */
376         if (revs->paths)
377                 try_to_simplify_commit(revs, commit);
378
379         parent = commit->parents;
380         while (parent) {
381                 struct commit *p = parent->item;
382
383                 parent = parent->next;
384
385                 parse_commit(p);
386                 if (p->object.flags & SEEN)
387                         continue;
388                 p->object.flags |= SEEN;
389                 insert_by_date(p, list);
390         }
391 }
392
393 static void limit_list(struct rev_info *revs)
394 {
395         struct commit_list *list = revs->commits;
396         struct commit_list *newlist = NULL;
397         struct commit_list **p = &newlist;
398
399         if (revs->paths)
400                 diff_tree_setup_paths(revs->paths);
401
402         while (list) {
403                 struct commit_list *entry = list;
404                 struct commit *commit = list->item;
405                 struct object *obj = &commit->object;
406
407                 list = list->next;
408                 free(entry);
409
410                 if (revs->max_age != -1 && (commit->date < revs->max_age))
411                         obj->flags |= UNINTERESTING;
412                 if (revs->unpacked && has_sha1_pack(obj->sha1))
413                         obj->flags |= UNINTERESTING;
414                 add_parents_to_list(revs, commit, &list);
415                 if (obj->flags & UNINTERESTING) {
416                         mark_parents_uninteresting(commit);
417                         if (everybody_uninteresting(list))
418                                 break;
419                         continue;
420                 }
421                 if (revs->min_age != -1 && (commit->date > revs->min_age))
422                         continue;
423                 p = &commit_list_insert(commit, p)->next;
424         }
425         revs->commits = newlist;
426 }
427
428 static void add_one_commit(struct commit *commit, struct rev_info *revs)
429 {
430         if (!commit || (commit->object.flags & SEEN))
431                 return;
432         commit->object.flags |= SEEN;
433         commit_list_insert(commit, &revs->commits);
434 }
435
436 static int all_flags;
437 static struct rev_info *all_revs;
438
439 static int handle_one_ref(const char *path, const unsigned char *sha1)
440 {
441         struct commit *commit = get_commit_reference(all_revs, path, sha1, all_flags);
442         add_one_commit(commit, all_revs);
443         return 0;
444 }
445
446 static void handle_all(struct rev_info *revs, unsigned flags)
447 {
448         all_revs = revs;
449         all_flags = flags;
450         for_each_ref(handle_one_ref);
451 }
452
453 /*
454  * Parse revision information, filling in the "rev_info" structure,
455  * and removing the used arguments from the argument list.
456  *
457  * Returns the number of arguments left that weren't recognized
458  * (which are also moved to the head of the argument list)
459  */
460 int setup_revisions(int argc, const char **argv, struct rev_info *revs, const char *def)
461 {
462         int i, flags, seen_dashdash;
463         const char **unrecognized = argv + 1;
464         int left = 1;
465
466         memset(revs, 0, sizeof(*revs));
467         revs->lifo = 1;
468         revs->dense = 1;
469         revs->prefix = setup_git_directory();
470         revs->max_age = -1;
471         revs->min_age = -1;
472         revs->max_count = -1;
473
474         /* First, search for "--" */
475         seen_dashdash = 0;
476         for (i = 1; i < argc; i++) {
477                 const char *arg = argv[i];
478                 if (strcmp(arg, "--"))
479                         continue;
480                 argv[i] = NULL;
481                 argc = i;
482                 revs->paths = get_pathspec(revs->prefix, argv + i + 1);
483                 seen_dashdash = 1;
484                 break;
485         }
486
487         flags = 0;
488         for (i = 1; i < argc; i++) {
489                 struct commit *commit;
490                 const char *arg = argv[i];
491                 unsigned char sha1[20];
492                 char *dotdot;
493                 int local_flags;
494
495                 if (*arg == '-') {
496                         if (!strncmp(arg, "--max-count=", 12)) {
497                                 revs->max_count = atoi(arg + 12);
498                                 continue;
499                         }
500                         /* accept -<digit>, like traditilnal "head" */
501                         if ((*arg == '-') && isdigit(arg[1])) {
502                                 revs->max_count = atoi(arg + 1);
503                                 continue;
504                         }
505                         if (!strcmp(arg, "-n")) {
506                                 if (argc <= i + 1)
507                                         die("-n requires an argument");
508                                 revs->max_count = atoi(argv[++i]);
509                                 continue;
510                         }
511                         if (!strncmp(arg,"-n",2)) {
512                                 revs->max_count = atoi(arg + 2);
513                                 continue;
514                         }
515                         if (!strncmp(arg, "--max-age=", 10)) {
516                                 revs->max_age = atoi(arg + 10);
517                                 revs->limited = 1;
518                                 continue;
519                         }
520                         if (!strncmp(arg, "--min-age=", 10)) {
521                                 revs->min_age = atoi(arg + 10);
522                                 revs->limited = 1;
523                                 continue;
524                         }
525                         if (!strncmp(arg, "--since=", 8)) {
526                                 revs->max_age = approxidate(arg + 8);
527                                 revs->limited = 1;
528                                 continue;
529                         }
530                         if (!strncmp(arg, "--after=", 8)) {
531                                 revs->max_age = approxidate(arg + 8);
532                                 revs->limited = 1;
533                                 continue;
534                         }
535                         if (!strncmp(arg, "--before=", 9)) {
536                                 revs->min_age = approxidate(arg + 9);
537                                 revs->limited = 1;
538                                 continue;
539                         }
540                         if (!strncmp(arg, "--until=", 8)) {
541                                 revs->min_age = approxidate(arg + 8);
542                                 revs->limited = 1;
543                                 continue;
544                         }
545                         if (!strcmp(arg, "--all")) {
546                                 handle_all(revs, flags);
547                                 continue;
548                         }
549                         if (!strcmp(arg, "--not")) {
550                                 flags ^= UNINTERESTING;
551                                 continue;
552                         }
553                         if (!strcmp(arg, "--default")) {
554                                 if (++i >= argc)
555                                         die("bad --default argument");
556                                 def = argv[i];
557                                 continue;
558                         }
559                         if (!strcmp(arg, "--topo-order")) {
560                                 revs->topo_order = 1;
561                                 revs->limited = 1;
562                                 continue;
563                         }
564                         if (!strcmp(arg, "--date-order")) {
565                                 revs->lifo = 0;
566                                 revs->topo_order = 1;
567                                 revs->limited = 1;
568                                 continue;
569                         }
570                         if (!strcmp(arg, "--dense")) {
571                                 revs->dense = 1;
572                                 continue;
573                         }
574                         if (!strcmp(arg, "--sparse")) {
575                                 revs->dense = 0;
576                                 continue;
577                         }
578                         if (!strcmp(arg, "--remove-empty")) {
579                                 revs->remove_empty_trees = 1;
580                                 continue;
581                         }
582                         if (!strncmp(arg, "--no-merges", 11)) {
583                                 revs->no_merges = 1;
584                                 continue;
585                         }
586                         if (!strcmp(arg, "--objects")) {
587                                 revs->tag_objects = 1;
588                                 revs->tree_objects = 1;
589                                 revs->blob_objects = 1;
590                                 continue;
591                         }
592                         if (!strcmp(arg, "--objects-edge")) {
593                                 revs->tag_objects = 1;
594                                 revs->tree_objects = 1;
595                                 revs->blob_objects = 1;
596                                 revs->edge_hint = 1;
597                                 continue;
598                         }
599                         if (!strcmp(arg, "--unpacked")) {
600                                 revs->unpacked = 1;
601                                 revs->limited = 1;
602                                 continue;
603                         }
604                         *unrecognized++ = arg;
605                         left++;
606                         continue;
607                 }
608                 dotdot = strstr(arg, "..");
609                 if (dotdot) {
610                         unsigned char from_sha1[20];
611                         char *next = dotdot + 2;
612                         *dotdot = 0;
613                         if (!*next)
614                                 next = "HEAD";
615                         if (!get_sha1(arg, from_sha1) && !get_sha1(next, sha1)) {
616                                 struct commit *exclude;
617                                 struct commit *include;
618
619                                 exclude = get_commit_reference(revs, arg, from_sha1, flags ^ UNINTERESTING);
620                                 include = get_commit_reference(revs, next, sha1, flags);
621                                 if (!exclude || !include)
622                                         die("Invalid revision range %s..%s", arg, next);
623                                 add_one_commit(exclude, revs);
624                                 add_one_commit(include, revs);
625                                 continue;
626                         }
627                         *dotdot = '.';
628                 }
629                 local_flags = 0;
630                 if (*arg == '^') {
631                         local_flags = UNINTERESTING;
632                         arg++;
633                 }
634                 if (get_sha1(arg, sha1) < 0) {
635                         struct stat st;
636                         int j;
637
638                         if (seen_dashdash || local_flags)
639                                 die("bad revision '%s'", arg);
640
641                         /* If we didn't have a "--", all filenames must exist */
642                         for (j = i; j < argc; j++) {
643                                 if (lstat(argv[j], &st) < 0)
644                                         die("'%s': %s", arg, strerror(errno));
645                         }
646                         revs->paths = get_pathspec(revs->prefix, argv + i);
647                         break;
648                 }
649                 commit = get_commit_reference(revs, arg, sha1, flags ^ local_flags);
650                 add_one_commit(commit, revs);
651         }
652         if (def && !revs->commits) {
653                 unsigned char sha1[20];
654                 struct commit *commit;
655                 if (get_sha1(def, sha1) < 0)
656                         die("bad default revision '%s'", def);
657                 commit = get_commit_reference(revs, def, sha1, 0);
658                 add_one_commit(commit, revs);
659         }
660         if (revs->paths)
661                 revs->limited = 1;
662         return left;
663 }
664
665 void prepare_revision_walk(struct rev_info *revs)
666 {
667         sort_by_date(&revs->commits);
668         if (revs->limited)
669                 limit_list(revs);
670         if (revs->topo_order)
671                 sort_in_topological_order(&revs->commits, revs->lifo);
672 }
673
674 static int rewrite_one(struct commit **pp)
675 {
676         for (;;) {
677                 struct commit *p = *pp;
678                 if (p->object.flags & (TREECHANGE | UNINTERESTING))
679                         return 0;
680                 if (!p->parents)
681                         return -1;
682                 *pp = p->parents->item;
683         }
684 }
685
686 static void rewrite_parents(struct commit *commit)
687 {
688         struct commit_list **pp = &commit->parents;
689         while (*pp) {
690                 struct commit_list *parent = *pp;
691                 if (rewrite_one(&parent->item) < 0) {
692                         *pp = parent->next;
693                         continue;
694                 }
695                 pp = &parent->next;
696         }
697 }
698
699 struct commit *get_revision(struct rev_info *revs)
700 {
701         struct commit_list *list = revs->commits;
702
703         if (!list)
704                 return NULL;
705
706         /* Check the max_count ... */
707         switch (revs->max_count) {
708         case -1:
709                 break;
710         case 0:
711                 return NULL;
712         default:
713                 revs->max_count--;
714         }
715
716         do {
717                 struct commit *commit = revs->commits->item;
718
719                 if (commit->object.flags & (UNINTERESTING|SHOWN))
720                         goto next;
721                 if (revs->min_age != -1 && (commit->date > revs->min_age))
722                         goto next;
723                 if (revs->max_age != -1 && (commit->date < revs->max_age))
724                         return NULL;
725                 if (revs->no_merges && commit->parents && commit->parents->next)
726                         goto next;
727                 if (revs->paths && revs->dense) {
728                         if (!(commit->object.flags & TREECHANGE))
729                                 goto next;
730                         rewrite_parents(commit);
731                 }
732                 /* More to go? */
733                 if (revs->max_count)
734                         pop_most_recent_commit(&revs->commits, SEEN);
735                 commit->object.flags |= SHOWN;
736                 return commit;
737 next:
738                 pop_most_recent_commit(&revs->commits, SEEN);
739         } while (revs->commits);
740         return NULL;
741 }