revision traversal: --remove-empty fix (take #2).
[git/gitweb.git] / revision.c
blob73fba5d3daa0a3447827bf85af513c0ad033a27e
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"
10 static char *path_name(struct name_path *path, const char *name)
12 struct name_path *p;
13 char *n, *m;
14 int nlen = strlen(name);
15 int len = nlen + 1;
17 for (p = path; p; p = p->up) {
18 if (p->elem_len)
19 len += p->elem_len + 1;
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] = '/';
31 return n;
34 struct object_list **add_object(struct object *obj,
35 struct object_list **p,
36 struct name_path *path,
37 const char *name)
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;
47 static void mark_blob_uninteresting(struct blob *blob)
49 if (blob->object.flags & UNINTERESTING)
50 return;
51 blob->object.flags |= UNINTERESTING;
54 void mark_tree_uninteresting(struct tree *tree)
56 struct object *obj = &tree->object;
57 struct tree_entry_list *entry;
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;
79 void mark_parents_uninteresting(struct commit *commit)
81 struct commit_list *parents = commit->parents;
83 while (parents) {
84 struct commit *commit = parents->item;
85 if (!(commit->object.flags & UNINTERESTING)) {
86 commit->object.flags |= UNINTERESTING;
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..
96 if (commit->parents)
97 mark_parents_uninteresting(commit);
101 * A missing commit is ok iff its parent is marked
102 * uninteresting.
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.
108 if (!has_sha1_file(commit->object.sha1))
109 commit->object.parsed = 1;
110 parents = parents->next;
114 static void add_pending_object(struct rev_info *revs, struct object *obj, const char *name)
116 add_object(obj, &revs->pending_objects, NULL, name);
119 static struct commit *get_commit_reference(struct rev_info *revs, const char *name, const unsigned char *sha1, unsigned int flags)
121 struct object *object;
123 object = parse_object(sha1);
124 if (!object)
125 die("bad object %s", name);
128 * Tag object? Look what it points to..
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));
141 * Commit object? Just return it, we'll do all the complex
142 * reachability crud.
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;
153 return commit;
157 * Tree object? Either mark it uniniteresting, or add it
158 * to the list of objects to look at later..
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;
168 add_pending_object(revs, object, "");
169 return NULL;
173 * Blob object? You know the drill by now..
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;
183 add_pending_object(revs, object, "");
184 return NULL;
186 die("%s is unknown object", name);
189 static int everybody_uninteresting(struct commit_list *orig)
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;
199 return 1;
202 #define TREE_SAME 0
203 #define TREE_NEW 1
204 #define TREE_DIFFERENT 2
205 static int tree_difference = TREE_SAME;
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)
212 int diff = TREE_DIFFERENT;
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).
222 if (addremove == '+') {
223 diff = tree_difference;
224 if (diff != TREE_SAME)
225 return;
226 diff = TREE_NEW;
228 tree_difference = diff;
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)
237 tree_difference = TREE_DIFFERENT;
240 static struct diff_options diff_opt = {
241 .recursive = 1,
242 .add_remove = file_add_remove,
243 .change = file_change,
246 static int compare_tree(struct tree *t1, struct tree *t2)
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;
258 static int same_tree_as_empty(struct tree *t1)
260 int retval;
261 void *tree;
262 struct tree_desc empty, real;
264 if (!t1)
265 return 0;
267 tree = read_object_with_reference(t1->object.sha1, "tree", &real.size, NULL);
268 if (!tree)
269 return 0;
270 real.buf = tree;
272 empty.buf = "";
273 empty.size = 0;
275 tree_difference = 0;
276 retval = diff_tree(&empty, &real, "", &diff_opt);
277 free(tree);
279 return retval >= 0 && !tree_difference;
282 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
284 struct commit_list **pp, *parent;
285 int tree_changed = 0;
287 if (!commit->tree)
288 return;
290 if (!commit->parents) {
291 if (!same_tree_as_empty(commit->tree))
292 commit->object.flags |= TREECHANGE;
293 return;
296 pp = &commit->parents;
297 while ((parent = *pp) != NULL) {
298 struct commit *p = parent->item;
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.
310 pp = &parent->next;
311 continue;
313 parent->next = NULL;
314 commit->parents = parent;
315 return;
317 case TREE_NEW:
318 if (revs->remove_empty_trees &&
319 same_tree_as_empty(p->tree)) {
320 /* We are adding all the specified
321 * paths from this parent, so the
322 * history beyond this parent is not
323 * interesting. Remove its parents
324 * (they are grandparents for us).
325 * IOW, we pretend this parent is a
326 * "root" commit.
328 parse_commit(p);
329 p->parents = NULL;
331 /* fallthrough */
332 case TREE_DIFFERENT:
333 tree_changed = 1;
334 pp = &parent->next;
335 continue;
337 die("bad tree compare for commit %s", sha1_to_hex(commit->object.sha1));
339 if (tree_changed)
340 commit->object.flags |= TREECHANGE;
343 static void add_parents_to_list(struct rev_info *revs, struct commit *commit, struct commit_list **list)
345 struct commit_list *parent = commit->parents;
348 * If the commit is uninteresting, don't try to
349 * prune parents - we want the maximal uninteresting
350 * set.
352 * Normally we haven't parsed the parent
353 * yet, so we won't have a parent of a parent
354 * here. However, it may turn out that we've
355 * reached this commit some other way (where it
356 * wasn't uninteresting), in which case we need
357 * to mark its parents recursively too..
359 if (commit->object.flags & UNINTERESTING) {
360 while (parent) {
361 struct commit *p = parent->item;
362 parent = parent->next;
363 parse_commit(p);
364 p->object.flags |= UNINTERESTING;
365 if (p->parents)
366 mark_parents_uninteresting(p);
367 if (p->object.flags & SEEN)
368 continue;
369 p->object.flags |= SEEN;
370 insert_by_date(p, list);
372 return;
376 * Ok, the commit wasn't uninteresting. Try to
377 * simplify the commit history and find the parent
378 * that has no differences in the path set if one exists.
380 if (revs->paths)
381 try_to_simplify_commit(revs, commit);
383 parent = commit->parents;
384 while (parent) {
385 struct commit *p = parent->item;
387 parent = parent->next;
389 parse_commit(p);
390 if (p->object.flags & SEEN)
391 continue;
392 p->object.flags |= SEEN;
393 insert_by_date(p, list);
397 static void limit_list(struct rev_info *revs)
399 struct commit_list *list = revs->commits;
400 struct commit_list *newlist = NULL;
401 struct commit_list **p = &newlist;
403 if (revs->paths)
404 diff_tree_setup_paths(revs->paths);
406 while (list) {
407 struct commit_list *entry = list;
408 struct commit *commit = list->item;
409 struct object *obj = &commit->object;
411 list = list->next;
412 free(entry);
414 if (revs->max_age != -1 && (commit->date < revs->max_age))
415 obj->flags |= UNINTERESTING;
416 if (revs->unpacked && has_sha1_pack(obj->sha1))
417 obj->flags |= UNINTERESTING;
418 add_parents_to_list(revs, commit, &list);
419 if (obj->flags & UNINTERESTING) {
420 mark_parents_uninteresting(commit);
421 if (everybody_uninteresting(list))
422 break;
423 continue;
425 if (revs->min_age != -1 && (commit->date > revs->min_age))
426 continue;
427 p = &commit_list_insert(commit, p)->next;
429 revs->commits = newlist;
432 static void add_one_commit(struct commit *commit, struct rev_info *revs)
434 if (!commit || (commit->object.flags & SEEN))
435 return;
436 commit->object.flags |= SEEN;
437 commit_list_insert(commit, &revs->commits);
440 static int all_flags;
441 static struct rev_info *all_revs;
443 static int handle_one_ref(const char *path, const unsigned char *sha1)
445 struct commit *commit = get_commit_reference(all_revs, path, sha1, all_flags);
446 add_one_commit(commit, all_revs);
447 return 0;
450 static void handle_all(struct rev_info *revs, unsigned flags)
452 all_revs = revs;
453 all_flags = flags;
454 for_each_ref(handle_one_ref);
458 * Parse revision information, filling in the "rev_info" structure,
459 * and removing the used arguments from the argument list.
461 * Returns the number of arguments left that weren't recognized
462 * (which are also moved to the head of the argument list)
464 int setup_revisions(int argc, const char **argv, struct rev_info *revs, const char *def)
466 int i, flags, seen_dashdash;
467 const char **unrecognized = argv + 1;
468 int left = 1;
470 memset(revs, 0, sizeof(*revs));
471 revs->lifo = 1;
472 revs->dense = 1;
473 revs->prefix = setup_git_directory();
474 revs->max_age = -1;
475 revs->min_age = -1;
476 revs->max_count = -1;
478 /* First, search for "--" */
479 seen_dashdash = 0;
480 for (i = 1; i < argc; i++) {
481 const char *arg = argv[i];
482 if (strcmp(arg, "--"))
483 continue;
484 argv[i] = NULL;
485 argc = i;
486 revs->paths = get_pathspec(revs->prefix, argv + i + 1);
487 seen_dashdash = 1;
488 break;
491 flags = 0;
492 for (i = 1; i < argc; i++) {
493 struct commit *commit;
494 const char *arg = argv[i];
495 unsigned char sha1[20];
496 char *dotdot;
497 int local_flags;
499 if (*arg == '-') {
500 if (!strncmp(arg, "--max-count=", 12)) {
501 revs->max_count = atoi(arg + 12);
502 continue;
504 /* accept -<digit>, like traditilnal "head" */
505 if ((*arg == '-') && isdigit(arg[1])) {
506 revs->max_count = atoi(arg + 1);
507 continue;
509 if (!strcmp(arg, "-n")) {
510 if (argc <= i + 1)
511 die("-n requires an argument");
512 revs->max_count = atoi(argv[++i]);
513 continue;
515 if (!strncmp(arg,"-n",2)) {
516 revs->max_count = atoi(arg + 2);
517 continue;
519 if (!strncmp(arg, "--max-age=", 10)) {
520 revs->max_age = atoi(arg + 10);
521 revs->limited = 1;
522 continue;
524 if (!strncmp(arg, "--min-age=", 10)) {
525 revs->min_age = atoi(arg + 10);
526 revs->limited = 1;
527 continue;
529 if (!strncmp(arg, "--since=", 8)) {
530 revs->max_age = approxidate(arg + 8);
531 revs->limited = 1;
532 continue;
534 if (!strncmp(arg, "--after=", 8)) {
535 revs->max_age = approxidate(arg + 8);
536 revs->limited = 1;
537 continue;
539 if (!strncmp(arg, "--before=", 9)) {
540 revs->min_age = approxidate(arg + 9);
541 revs->limited = 1;
542 continue;
544 if (!strncmp(arg, "--until=", 8)) {
545 revs->min_age = approxidate(arg + 8);
546 revs->limited = 1;
547 continue;
549 if (!strcmp(arg, "--all")) {
550 handle_all(revs, flags);
551 continue;
553 if (!strcmp(arg, "--not")) {
554 flags ^= UNINTERESTING;
555 continue;
557 if (!strcmp(arg, "--default")) {
558 if (++i >= argc)
559 die("bad --default argument");
560 def = argv[i];
561 continue;
563 if (!strcmp(arg, "--topo-order")) {
564 revs->topo_order = 1;
565 revs->limited = 1;
566 continue;
568 if (!strcmp(arg, "--date-order")) {
569 revs->lifo = 0;
570 revs->topo_order = 1;
571 revs->limited = 1;
572 continue;
574 if (!strcmp(arg, "--dense")) {
575 revs->dense = 1;
576 continue;
578 if (!strcmp(arg, "--sparse")) {
579 revs->dense = 0;
580 continue;
582 if (!strcmp(arg, "--remove-empty")) {
583 revs->remove_empty_trees = 1;
584 continue;
586 if (!strncmp(arg, "--no-merges", 11)) {
587 revs->no_merges = 1;
588 continue;
590 if (!strcmp(arg, "--objects")) {
591 revs->tag_objects = 1;
592 revs->tree_objects = 1;
593 revs->blob_objects = 1;
594 continue;
596 if (!strcmp(arg, "--objects-edge")) {
597 revs->tag_objects = 1;
598 revs->tree_objects = 1;
599 revs->blob_objects = 1;
600 revs->edge_hint = 1;
601 continue;
603 if (!strcmp(arg, "--unpacked")) {
604 revs->unpacked = 1;
605 revs->limited = 1;
606 continue;
608 *unrecognized++ = arg;
609 left++;
610 continue;
612 dotdot = strstr(arg, "..");
613 if (dotdot) {
614 unsigned char from_sha1[20];
615 char *next = dotdot + 2;
616 *dotdot = 0;
617 if (!*next)
618 next = "HEAD";
619 if (!get_sha1(arg, from_sha1) && !get_sha1(next, sha1)) {
620 struct commit *exclude;
621 struct commit *include;
623 exclude = get_commit_reference(revs, arg, from_sha1, flags ^ UNINTERESTING);
624 include = get_commit_reference(revs, next, sha1, flags);
625 if (!exclude || !include)
626 die("Invalid revision range %s..%s", arg, next);
627 add_one_commit(exclude, revs);
628 add_one_commit(include, revs);
629 continue;
631 *dotdot = '.';
633 local_flags = 0;
634 if (*arg == '^') {
635 local_flags = UNINTERESTING;
636 arg++;
638 if (get_sha1(arg, sha1) < 0) {
639 struct stat st;
640 int j;
642 if (seen_dashdash || local_flags)
643 die("bad revision '%s'", arg);
645 /* If we didn't have a "--", all filenames must exist */
646 for (j = i; j < argc; j++) {
647 if (lstat(argv[j], &st) < 0)
648 die("'%s': %s", arg, strerror(errno));
650 revs->paths = get_pathspec(revs->prefix, argv + i);
651 break;
653 commit = get_commit_reference(revs, arg, sha1, flags ^ local_flags);
654 add_one_commit(commit, revs);
656 if (def && !revs->commits) {
657 unsigned char sha1[20];
658 struct commit *commit;
659 if (get_sha1(def, sha1) < 0)
660 die("bad default revision '%s'", def);
661 commit = get_commit_reference(revs, def, sha1, 0);
662 add_one_commit(commit, revs);
664 if (revs->paths)
665 revs->limited = 1;
666 return left;
669 void prepare_revision_walk(struct rev_info *revs)
671 sort_by_date(&revs->commits);
672 if (revs->limited)
673 limit_list(revs);
674 if (revs->topo_order)
675 sort_in_topological_order(&revs->commits, revs->lifo);
678 static int rewrite_one(struct commit **pp)
680 for (;;) {
681 struct commit *p = *pp;
682 if (p->object.flags & (TREECHANGE | UNINTERESTING))
683 return 0;
684 if (!p->parents)
685 return -1;
686 *pp = p->parents->item;
690 static void rewrite_parents(struct commit *commit)
692 struct commit_list **pp = &commit->parents;
693 while (*pp) {
694 struct commit_list *parent = *pp;
695 if (rewrite_one(&parent->item) < 0) {
696 *pp = parent->next;
697 continue;
699 pp = &parent->next;
703 struct commit *get_revision(struct rev_info *revs)
705 struct commit_list *list = revs->commits;
707 if (!list)
708 return NULL;
710 /* Check the max_count ... */
711 switch (revs->max_count) {
712 case -1:
713 break;
714 case 0:
715 return NULL;
716 default:
717 revs->max_count--;
720 do {
721 struct commit *commit = revs->commits->item;
723 if (commit->object.flags & (UNINTERESTING|SHOWN))
724 goto next;
725 if (revs->min_age != -1 && (commit->date > revs->min_age))
726 goto next;
727 if (revs->max_age != -1 && (commit->date < revs->max_age))
728 return NULL;
729 if (revs->no_merges && commit->parents && commit->parents->next)
730 goto next;
731 if (revs->paths && revs->dense) {
732 if (!(commit->object.flags & TREECHANGE))
733 goto next;
734 rewrite_parents(commit);
736 /* More to go? */
737 if (revs->max_count)
738 pop_most_recent_commit(&revs->commits, SEEN);
739 commit->object.flags |= SHOWN;
740 return commit;
741 next:
742 pop_most_recent_commit(&revs->commits, SEEN);
743 } while (revs->commits);
744 return NULL;