Merge branch 'js/skip-dashed-built-ins-from-config-mak'
[git] / revision.c
1 #include "cache.h"
2 #include "object-store.h"
3 #include "tag.h"
4 #include "blob.h"
5 #include "tree.h"
6 #include "commit.h"
7 #include "diff.h"
8 #include "refs.h"
9 #include "revision.h"
10 #include "repository.h"
11 #include "graph.h"
12 #include "grep.h"
13 #include "reflog-walk.h"
14 #include "patch-ids.h"
15 #include "decorate.h"
16 #include "log-tree.h"
17 #include "string-list.h"
18 #include "line-log.h"
19 #include "mailmap.h"
20 #include "commit-slab.h"
21 #include "dir.h"
22 #include "cache-tree.h"
23 #include "bisect.h"
24 #include "packfile.h"
25 #include "worktree.h"
26 #include "strvec.h"
27 #include "commit-reach.h"
28 #include "commit-graph.h"
29 #include "prio-queue.h"
30 #include "hashmap.h"
31 #include "utf8.h"
32 #include "bloom.h"
33 #include "json-writer.h"
34
35 volatile show_early_output_fn_t show_early_output;
36
37 static const char *term_bad;
38 static const char *term_good;
39
40 implement_shared_commit_slab(revision_sources, char *);
41
42 static inline int want_ancestry(const struct rev_info *revs);
43
44 void show_object_with_name(FILE *out, struct object *obj, const char *name)
45 {
46         const char *p;
47
48         fprintf(out, "%s ", oid_to_hex(&obj->oid));
49         for (p = name; *p && *p != '\n'; p++)
50                 fputc(*p, out);
51         fputc('\n', out);
52 }
53
54 static void mark_blob_uninteresting(struct blob *blob)
55 {
56         if (!blob)
57                 return;
58         if (blob->object.flags & UNINTERESTING)
59                 return;
60         blob->object.flags |= UNINTERESTING;
61 }
62
63 static void mark_tree_contents_uninteresting(struct repository *r,
64                                              struct tree *tree)
65 {
66         struct tree_desc desc;
67         struct name_entry entry;
68
69         if (parse_tree_gently(tree, 1) < 0)
70                 return;
71
72         init_tree_desc(&desc, tree->buffer, tree->size);
73         while (tree_entry(&desc, &entry)) {
74                 switch (object_type(entry.mode)) {
75                 case OBJ_TREE:
76                         mark_tree_uninteresting(r, lookup_tree(r, &entry.oid));
77                         break;
78                 case OBJ_BLOB:
79                         mark_blob_uninteresting(lookup_blob(r, &entry.oid));
80                         break;
81                 default:
82                         /* Subproject commit - not in this repository */
83                         break;
84                 }
85         }
86
87         /*
88          * We don't care about the tree any more
89          * after it has been marked uninteresting.
90          */
91         free_tree_buffer(tree);
92 }
93
94 void mark_tree_uninteresting(struct repository *r, struct tree *tree)
95 {
96         struct object *obj;
97
98         if (!tree)
99                 return;
100
101         obj = &tree->object;
102         if (obj->flags & UNINTERESTING)
103                 return;
104         obj->flags |= UNINTERESTING;
105         mark_tree_contents_uninteresting(r, tree);
106 }
107
108 struct path_and_oids_entry {
109         struct hashmap_entry ent;
110         char *path;
111         struct oidset trees;
112 };
113
114 static int path_and_oids_cmp(const void *hashmap_cmp_fn_data,
115                              const struct hashmap_entry *eptr,
116                              const struct hashmap_entry *entry_or_key,
117                              const void *keydata)
118 {
119         const struct path_and_oids_entry *e1, *e2;
120
121         e1 = container_of(eptr, const struct path_and_oids_entry, ent);
122         e2 = container_of(entry_or_key, const struct path_and_oids_entry, ent);
123
124         return strcmp(e1->path, e2->path);
125 }
126
127 static void paths_and_oids_clear(struct hashmap *map)
128 {
129         struct hashmap_iter iter;
130         struct path_and_oids_entry *entry;
131
132         hashmap_for_each_entry(map, &iter, entry, ent /* member name */) {
133                 oidset_clear(&entry->trees);
134                 free(entry->path);
135         }
136
137         hashmap_clear_and_free(map, struct path_and_oids_entry, ent);
138 }
139
140 static void paths_and_oids_insert(struct hashmap *map,
141                                   const char *path,
142                                   const struct object_id *oid)
143 {
144         int hash = strhash(path);
145         struct path_and_oids_entry key;
146         struct path_and_oids_entry *entry;
147
148         hashmap_entry_init(&key.ent, hash);
149
150         /* use a shallow copy for the lookup */
151         key.path = (char *)path;
152         oidset_init(&key.trees, 0);
153
154         entry = hashmap_get_entry(map, &key, ent, NULL);
155         if (!entry) {
156                 entry = xcalloc(1, sizeof(struct path_and_oids_entry));
157                 hashmap_entry_init(&entry->ent, hash);
158                 entry->path = xstrdup(key.path);
159                 oidset_init(&entry->trees, 16);
160                 hashmap_put(map, &entry->ent);
161         }
162
163         oidset_insert(&entry->trees, oid);
164 }
165
166 static void add_children_by_path(struct repository *r,
167                                  struct tree *tree,
168                                  struct hashmap *map)
169 {
170         struct tree_desc desc;
171         struct name_entry entry;
172
173         if (!tree)
174                 return;
175
176         if (parse_tree_gently(tree, 1) < 0)
177                 return;
178
179         init_tree_desc(&desc, tree->buffer, tree->size);
180         while (tree_entry(&desc, &entry)) {
181                 switch (object_type(entry.mode)) {
182                 case OBJ_TREE:
183                         paths_and_oids_insert(map, entry.path, &entry.oid);
184
185                         if (tree->object.flags & UNINTERESTING) {
186                                 struct tree *child = lookup_tree(r, &entry.oid);
187                                 if (child)
188                                         child->object.flags |= UNINTERESTING;
189                         }
190                         break;
191                 case OBJ_BLOB:
192                         if (tree->object.flags & UNINTERESTING) {
193                                 struct blob *child = lookup_blob(r, &entry.oid);
194                                 if (child)
195                                         child->object.flags |= UNINTERESTING;
196                         }
197                         break;
198                 default:
199                         /* Subproject commit - not in this repository */
200                         break;
201                 }
202         }
203
204         free_tree_buffer(tree);
205 }
206
207 void mark_trees_uninteresting_sparse(struct repository *r,
208                                      struct oidset *trees)
209 {
210         unsigned has_interesting = 0, has_uninteresting = 0;
211         struct hashmap map = HASHMAP_INIT(path_and_oids_cmp, NULL);
212         struct hashmap_iter map_iter;
213         struct path_and_oids_entry *entry;
214         struct object_id *oid;
215         struct oidset_iter iter;
216
217         oidset_iter_init(trees, &iter);
218         while ((!has_interesting || !has_uninteresting) &&
219                (oid = oidset_iter_next(&iter))) {
220                 struct tree *tree = lookup_tree(r, oid);
221
222                 if (!tree)
223                         continue;
224
225                 if (tree->object.flags & UNINTERESTING)
226                         has_uninteresting = 1;
227                 else
228                         has_interesting = 1;
229         }
230
231         /* Do not walk unless we have both types of trees. */
232         if (!has_uninteresting || !has_interesting)
233                 return;
234
235         oidset_iter_init(trees, &iter);
236         while ((oid = oidset_iter_next(&iter))) {
237                 struct tree *tree = lookup_tree(r, oid);
238                 add_children_by_path(r, tree, &map);
239         }
240
241         hashmap_for_each_entry(&map, &map_iter, entry, ent /* member name */)
242                 mark_trees_uninteresting_sparse(r, &entry->trees);
243
244         paths_and_oids_clear(&map);
245 }
246
247 struct commit_stack {
248         struct commit **items;
249         size_t nr, alloc;
250 };
251 #define COMMIT_STACK_INIT { NULL, 0, 0 }
252
253 static void commit_stack_push(struct commit_stack *stack, struct commit *commit)
254 {
255         ALLOC_GROW(stack->items, stack->nr + 1, stack->alloc);
256         stack->items[stack->nr++] = commit;
257 }
258
259 static struct commit *commit_stack_pop(struct commit_stack *stack)
260 {
261         return stack->nr ? stack->items[--stack->nr] : NULL;
262 }
263
264 static void commit_stack_clear(struct commit_stack *stack)
265 {
266         FREE_AND_NULL(stack->items);
267         stack->nr = stack->alloc = 0;
268 }
269
270 static void mark_one_parent_uninteresting(struct commit *commit,
271                                           struct commit_stack *pending)
272 {
273         struct commit_list *l;
274
275         if (commit->object.flags & UNINTERESTING)
276                 return;
277         commit->object.flags |= UNINTERESTING;
278
279         /*
280          * Normally we haven't parsed the parent
281          * yet, so we won't have a parent of a parent
282          * here. However, it may turn out that we've
283          * reached this commit some other way (where it
284          * wasn't uninteresting), in which case we need
285          * to mark its parents recursively too..
286          */
287         for (l = commit->parents; l; l = l->next)
288                 commit_stack_push(pending, l->item);
289 }
290
291 void mark_parents_uninteresting(struct commit *commit)
292 {
293         struct commit_stack pending = COMMIT_STACK_INIT;
294         struct commit_list *l;
295
296         for (l = commit->parents; l; l = l->next)
297                 mark_one_parent_uninteresting(l->item, &pending);
298
299         while (pending.nr > 0)
300                 mark_one_parent_uninteresting(commit_stack_pop(&pending),
301                                               &pending);
302
303         commit_stack_clear(&pending);
304 }
305
306 static void add_pending_object_with_path(struct rev_info *revs,
307                                          struct object *obj,
308                                          const char *name, unsigned mode,
309                                          const char *path)
310 {
311         struct interpret_branch_name_options options = { 0 };
312         if (!obj)
313                 return;
314         if (revs->no_walk && (obj->flags & UNINTERESTING))
315                 revs->no_walk = 0;
316         if (revs->reflog_info && obj->type == OBJ_COMMIT) {
317                 struct strbuf buf = STRBUF_INIT;
318                 int len = interpret_branch_name(name, 0, &buf, &options);
319
320                 if (0 < len && name[len] && buf.len)
321                         strbuf_addstr(&buf, name + len);
322                 add_reflog_for_walk(revs->reflog_info,
323                                     (struct commit *)obj,
324                                     buf.buf[0] ? buf.buf: name);
325                 strbuf_release(&buf);
326                 return; /* do not add the commit itself */
327         }
328         add_object_array_with_path(obj, name, &revs->pending, mode, path);
329 }
330
331 static void add_pending_object_with_mode(struct rev_info *revs,
332                                          struct object *obj,
333                                          const char *name, unsigned mode)
334 {
335         add_pending_object_with_path(revs, obj, name, mode, NULL);
336 }
337
338 void add_pending_object(struct rev_info *revs,
339                         struct object *obj, const char *name)
340 {
341         add_pending_object_with_mode(revs, obj, name, S_IFINVALID);
342 }
343
344 void add_head_to_pending(struct rev_info *revs)
345 {
346         struct object_id oid;
347         struct object *obj;
348         if (get_oid("HEAD", &oid))
349                 return;
350         obj = parse_object(revs->repo, &oid);
351         if (!obj)
352                 return;
353         add_pending_object(revs, obj, "HEAD");
354 }
355
356 static struct object *get_reference(struct rev_info *revs, const char *name,
357                                     const struct object_id *oid,
358                                     unsigned int flags)
359 {
360         struct object *object;
361
362         /*
363          * If the repository has commit graphs, repo_parse_commit() avoids
364          * reading the object buffer, so use it whenever possible.
365          */
366         if (oid_object_info(revs->repo, oid, NULL) == OBJ_COMMIT) {
367                 struct commit *c = lookup_commit(revs->repo, oid);
368                 if (!repo_parse_commit(revs->repo, c))
369                         object = (struct object *) c;
370                 else
371                         object = NULL;
372         } else {
373                 object = parse_object(revs->repo, oid);
374         }
375
376         if (!object) {
377                 if (revs->ignore_missing)
378                         return object;
379                 if (revs->exclude_promisor_objects && is_promisor_object(oid))
380                         return NULL;
381                 die("bad object %s", name);
382         }
383         object->flags |= flags;
384         return object;
385 }
386
387 void add_pending_oid(struct rev_info *revs, const char *name,
388                       const struct object_id *oid, unsigned int flags)
389 {
390         struct object *object = get_reference(revs, name, oid, flags);
391         add_pending_object(revs, object, name);
392 }
393
394 static struct commit *handle_commit(struct rev_info *revs,
395                                     struct object_array_entry *entry)
396 {
397         struct object *object = entry->item;
398         const char *name = entry->name;
399         const char *path = entry->path;
400         unsigned int mode = entry->mode;
401         unsigned long flags = object->flags;
402
403         /*
404          * Tag object? Look what it points to..
405          */
406         while (object->type == OBJ_TAG) {
407                 struct tag *tag = (struct tag *) object;
408                 if (revs->tag_objects && !(flags & UNINTERESTING))
409                         add_pending_object(revs, object, tag->tag);
410                 object = parse_object(revs->repo, get_tagged_oid(tag));
411                 if (!object) {
412                         if (revs->ignore_missing_links || (flags & UNINTERESTING))
413                                 return NULL;
414                         if (revs->exclude_promisor_objects &&
415                             is_promisor_object(&tag->tagged->oid))
416                                 return NULL;
417                         die("bad object %s", oid_to_hex(&tag->tagged->oid));
418                 }
419                 object->flags |= flags;
420                 /*
421                  * We'll handle the tagged object by looping or dropping
422                  * through to the non-tag handlers below. Do not
423                  * propagate path data from the tag's pending entry.
424                  */
425                 path = NULL;
426                 mode = 0;
427         }
428
429         /*
430          * Commit object? Just return it, we'll do all the complex
431          * reachability crud.
432          */
433         if (object->type == OBJ_COMMIT) {
434                 struct commit *commit = (struct commit *)object;
435
436                 if (repo_parse_commit(revs->repo, commit) < 0)
437                         die("unable to parse commit %s", name);
438                 if (flags & UNINTERESTING) {
439                         mark_parents_uninteresting(commit);
440
441                         if (!revs->topo_order || !generation_numbers_enabled(the_repository))
442                                 revs->limited = 1;
443                 }
444                 if (revs->sources) {
445                         char **slot = revision_sources_at(revs->sources, commit);
446
447                         if (!*slot)
448                                 *slot = xstrdup(name);
449                 }
450                 return commit;
451         }
452
453         /*
454          * Tree object? Either mark it uninteresting, or add it
455          * to the list of objects to look at later..
456          */
457         if (object->type == OBJ_TREE) {
458                 struct tree *tree = (struct tree *)object;
459                 if (!revs->tree_objects)
460                         return NULL;
461                 if (flags & UNINTERESTING) {
462                         mark_tree_contents_uninteresting(revs->repo, tree);
463                         return NULL;
464                 }
465                 add_pending_object_with_path(revs, object, name, mode, path);
466                 return NULL;
467         }
468
469         /*
470          * Blob object? You know the drill by now..
471          */
472         if (object->type == OBJ_BLOB) {
473                 if (!revs->blob_objects)
474                         return NULL;
475                 if (flags & UNINTERESTING)
476                         return NULL;
477                 add_pending_object_with_path(revs, object, name, mode, path);
478                 return NULL;
479         }
480         die("%s is unknown object", name);
481 }
482
483 static int everybody_uninteresting(struct commit_list *orig,
484                                    struct commit **interesting_cache)
485 {
486         struct commit_list *list = orig;
487
488         if (*interesting_cache) {
489                 struct commit *commit = *interesting_cache;
490                 if (!(commit->object.flags & UNINTERESTING))
491                         return 0;
492         }
493
494         while (list) {
495                 struct commit *commit = list->item;
496                 list = list->next;
497                 if (commit->object.flags & UNINTERESTING)
498                         continue;
499
500                 *interesting_cache = commit;
501                 return 0;
502         }
503         return 1;
504 }
505
506 /*
507  * A definition of "relevant" commit that we can use to simplify limited graphs
508  * by eliminating side branches.
509  *
510  * A "relevant" commit is one that is !UNINTERESTING (ie we are including it
511  * in our list), or that is a specified BOTTOM commit. Then after computing
512  * a limited list, during processing we can generally ignore boundary merges
513  * coming from outside the graph, (ie from irrelevant parents), and treat
514  * those merges as if they were single-parent. TREESAME is defined to consider
515  * only relevant parents, if any. If we are TREESAME to our on-graph parents,
516  * we don't care if we were !TREESAME to non-graph parents.
517  *
518  * Treating bottom commits as relevant ensures that a limited graph's
519  * connection to the actual bottom commit is not viewed as a side branch, but
520  * treated as part of the graph. For example:
521  *
522  *   ....Z...A---X---o---o---B
523  *        .     /
524  *         W---Y
525  *
526  * When computing "A..B", the A-X connection is at least as important as
527  * Y-X, despite A being flagged UNINTERESTING.
528  *
529  * And when computing --ancestry-path "A..B", the A-X connection is more
530  * important than Y-X, despite both A and Y being flagged UNINTERESTING.
531  */
532 static inline int relevant_commit(struct commit *commit)
533 {
534         return (commit->object.flags & (UNINTERESTING | BOTTOM)) != UNINTERESTING;
535 }
536
537 /*
538  * Return a single relevant commit from a parent list. If we are a TREESAME
539  * commit, and this selects one of our parents, then we can safely simplify to
540  * that parent.
541  */
542 static struct commit *one_relevant_parent(const struct rev_info *revs,
543                                           struct commit_list *orig)
544 {
545         struct commit_list *list = orig;
546         struct commit *relevant = NULL;
547
548         if (!orig)
549                 return NULL;
550
551         /*
552          * For 1-parent commits, or if first-parent-only, then return that
553          * first parent (even if not "relevant" by the above definition).
554          * TREESAME will have been set purely on that parent.
555          */
556         if (revs->first_parent_only || !orig->next)
557                 return orig->item;
558
559         /*
560          * For multi-parent commits, identify a sole relevant parent, if any.
561          * If we have only one relevant parent, then TREESAME will be set purely
562          * with regard to that parent, and we can simplify accordingly.
563          *
564          * If we have more than one relevant parent, or no relevant parents
565          * (and multiple irrelevant ones), then we can't select a parent here
566          * and return NULL.
567          */
568         while (list) {
569                 struct commit *commit = list->item;
570                 list = list->next;
571                 if (relevant_commit(commit)) {
572                         if (relevant)
573                                 return NULL;
574                         relevant = commit;
575                 }
576         }
577         return relevant;
578 }
579
580 /*
581  * The goal is to get REV_TREE_NEW as the result only if the
582  * diff consists of all '+' (and no other changes), REV_TREE_OLD
583  * if the whole diff is removal of old data, and otherwise
584  * REV_TREE_DIFFERENT (of course if the trees are the same we
585  * want REV_TREE_SAME).
586  *
587  * The only time we care about the distinction is when
588  * remove_empty_trees is in effect, in which case we care only about
589  * whether the whole change is REV_TREE_NEW, or if there's another type
590  * of change. Which means we can stop the diff early in either of these
591  * cases:
592  *
593  *   1. We're not using remove_empty_trees at all.
594  *
595  *   2. We saw anything except REV_TREE_NEW.
596  */
597 static int tree_difference = REV_TREE_SAME;
598
599 static void file_add_remove(struct diff_options *options,
600                     int addremove, unsigned mode,
601                     const struct object_id *oid,
602                     int oid_valid,
603                     const char *fullpath, unsigned dirty_submodule)
604 {
605         int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
606         struct rev_info *revs = options->change_fn_data;
607
608         tree_difference |= diff;
609         if (!revs->remove_empty_trees || tree_difference != REV_TREE_NEW)
610                 options->flags.has_changes = 1;
611 }
612
613 static void file_change(struct diff_options *options,
614                  unsigned old_mode, unsigned new_mode,
615                  const struct object_id *old_oid,
616                  const struct object_id *new_oid,
617                  int old_oid_valid, int new_oid_valid,
618                  const char *fullpath,
619                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
620 {
621         tree_difference = REV_TREE_DIFFERENT;
622         options->flags.has_changes = 1;
623 }
624
625 static int bloom_filter_atexit_registered;
626 static unsigned int count_bloom_filter_maybe;
627 static unsigned int count_bloom_filter_definitely_not;
628 static unsigned int count_bloom_filter_false_positive;
629 static unsigned int count_bloom_filter_not_present;
630
631 static void trace2_bloom_filter_statistics_atexit(void)
632 {
633         struct json_writer jw = JSON_WRITER_INIT;
634
635         jw_object_begin(&jw, 0);
636         jw_object_intmax(&jw, "filter_not_present", count_bloom_filter_not_present);
637         jw_object_intmax(&jw, "maybe", count_bloom_filter_maybe);
638         jw_object_intmax(&jw, "definitely_not", count_bloom_filter_definitely_not);
639         jw_object_intmax(&jw, "false_positive", count_bloom_filter_false_positive);
640         jw_end(&jw);
641
642         trace2_data_json("bloom", the_repository, "statistics", &jw);
643
644         jw_release(&jw);
645 }
646
647 static int forbid_bloom_filters(struct pathspec *spec)
648 {
649         if (spec->has_wildcard)
650                 return 1;
651         if (spec->nr > 1)
652                 return 1;
653         if (spec->magic & ~PATHSPEC_LITERAL)
654                 return 1;
655         if (spec->nr && (spec->items[0].magic & ~PATHSPEC_LITERAL))
656                 return 1;
657
658         return 0;
659 }
660
661 static void prepare_to_use_bloom_filter(struct rev_info *revs)
662 {
663         struct pathspec_item *pi;
664         char *path_alloc = NULL;
665         const char *path, *p;
666         size_t len;
667         int path_component_nr = 1;
668
669         if (!revs->commits)
670                 return;
671
672         if (forbid_bloom_filters(&revs->prune_data))
673                 return;
674
675         repo_parse_commit(revs->repo, revs->commits->item);
676
677         revs->bloom_filter_settings = get_bloom_filter_settings(revs->repo);
678         if (!revs->bloom_filter_settings)
679                 return;
680
681         if (!revs->pruning.pathspec.nr)
682                 return;
683
684         pi = &revs->pruning.pathspec.items[0];
685
686         /* remove single trailing slash from path, if needed */
687         if (pi->len > 0 && pi->match[pi->len - 1] == '/') {
688                 path_alloc = xmemdupz(pi->match, pi->len - 1);
689                 path = path_alloc;
690         } else
691                 path = pi->match;
692
693         len = strlen(path);
694         if (!len) {
695                 revs->bloom_filter_settings = NULL;
696                 free(path_alloc);
697                 return;
698         }
699
700         p = path;
701         while (*p) {
702                 /*
703                  * At this point, the path is normalized to use Unix-style
704                  * path separators. This is required due to how the
705                  * changed-path Bloom filters store the paths.
706                  */
707                 if (*p == '/')
708                         path_component_nr++;
709                 p++;
710         }
711
712         revs->bloom_keys_nr = path_component_nr;
713         ALLOC_ARRAY(revs->bloom_keys, revs->bloom_keys_nr);
714
715         fill_bloom_key(path, len, &revs->bloom_keys[0],
716                        revs->bloom_filter_settings);
717         path_component_nr = 1;
718
719         p = path + len - 1;
720         while (p > path) {
721                 if (*p == '/')
722                         fill_bloom_key(path, p - path,
723                                        &revs->bloom_keys[path_component_nr++],
724                                        revs->bloom_filter_settings);
725                 p--;
726         }
727
728         if (trace2_is_enabled() && !bloom_filter_atexit_registered) {
729                 atexit(trace2_bloom_filter_statistics_atexit);
730                 bloom_filter_atexit_registered = 1;
731         }
732
733         free(path_alloc);
734 }
735
736 static int check_maybe_different_in_bloom_filter(struct rev_info *revs,
737                                                  struct commit *commit)
738 {
739         struct bloom_filter *filter;
740         int result = 1, j;
741
742         if (!revs->repo->objects->commit_graph)
743                 return -1;
744
745         if (commit_graph_generation(commit) == GENERATION_NUMBER_INFINITY)
746                 return -1;
747
748         filter = get_bloom_filter(revs->repo, commit);
749
750         if (!filter) {
751                 count_bloom_filter_not_present++;
752                 return -1;
753         }
754
755         for (j = 0; result && j < revs->bloom_keys_nr; j++) {
756                 result = bloom_filter_contains(filter,
757                                                &revs->bloom_keys[j],
758                                                revs->bloom_filter_settings);
759         }
760
761         if (result)
762                 count_bloom_filter_maybe++;
763         else
764                 count_bloom_filter_definitely_not++;
765
766         return result;
767 }
768
769 static int rev_compare_tree(struct rev_info *revs,
770                             struct commit *parent, struct commit *commit, int nth_parent)
771 {
772         struct tree *t1 = get_commit_tree(parent);
773         struct tree *t2 = get_commit_tree(commit);
774         int bloom_ret = 1;
775
776         if (!t1)
777                 return REV_TREE_NEW;
778         if (!t2)
779                 return REV_TREE_OLD;
780
781         if (revs->simplify_by_decoration) {
782                 /*
783                  * If we are simplifying by decoration, then the commit
784                  * is worth showing if it has a tag pointing at it.
785                  */
786                 if (get_name_decoration(&commit->object))
787                         return REV_TREE_DIFFERENT;
788                 /*
789                  * A commit that is not pointed by a tag is uninteresting
790                  * if we are not limited by path.  This means that you will
791                  * see the usual "commits that touch the paths" plus any
792                  * tagged commit by specifying both --simplify-by-decoration
793                  * and pathspec.
794                  */
795                 if (!revs->prune_data.nr)
796                         return REV_TREE_SAME;
797         }
798
799         if (revs->bloom_keys_nr && !nth_parent) {
800                 bloom_ret = check_maybe_different_in_bloom_filter(revs, commit);
801
802                 if (bloom_ret == 0)
803                         return REV_TREE_SAME;
804         }
805
806         tree_difference = REV_TREE_SAME;
807         revs->pruning.flags.has_changes = 0;
808         diff_tree_oid(&t1->object.oid, &t2->object.oid, "", &revs->pruning);
809
810         if (!nth_parent)
811                 if (bloom_ret == 1 && tree_difference == REV_TREE_SAME)
812                         count_bloom_filter_false_positive++;
813
814         return tree_difference;
815 }
816
817 static int rev_same_tree_as_empty(struct rev_info *revs, struct commit *commit)
818 {
819         struct tree *t1 = get_commit_tree(commit);
820
821         if (!t1)
822                 return 0;
823
824         tree_difference = REV_TREE_SAME;
825         revs->pruning.flags.has_changes = 0;
826         diff_tree_oid(NULL, &t1->object.oid, "", &revs->pruning);
827
828         return tree_difference == REV_TREE_SAME;
829 }
830
831 struct treesame_state {
832         unsigned int nparents;
833         unsigned char treesame[FLEX_ARRAY];
834 };
835
836 static struct treesame_state *initialise_treesame(struct rev_info *revs, struct commit *commit)
837 {
838         unsigned n = commit_list_count(commit->parents);
839         struct treesame_state *st = xcalloc(1, st_add(sizeof(*st), n));
840         st->nparents = n;
841         add_decoration(&revs->treesame, &commit->object, st);
842         return st;
843 }
844
845 /*
846  * Must be called immediately after removing the nth_parent from a commit's
847  * parent list, if we are maintaining the per-parent treesame[] decoration.
848  * This does not recalculate the master TREESAME flag - update_treesame()
849  * should be called to update it after a sequence of treesame[] modifications
850  * that may have affected it.
851  */
852 static int compact_treesame(struct rev_info *revs, struct commit *commit, unsigned nth_parent)
853 {
854         struct treesame_state *st;
855         int old_same;
856
857         if (!commit->parents) {
858                 /*
859                  * Have just removed the only parent from a non-merge.
860                  * Different handling, as we lack decoration.
861                  */
862                 if (nth_parent != 0)
863                         die("compact_treesame %u", nth_parent);
864                 old_same = !!(commit->object.flags & TREESAME);
865                 if (rev_same_tree_as_empty(revs, commit))
866                         commit->object.flags |= TREESAME;
867                 else
868                         commit->object.flags &= ~TREESAME;
869                 return old_same;
870         }
871
872         st = lookup_decoration(&revs->treesame, &commit->object);
873         if (!st || nth_parent >= st->nparents)
874                 die("compact_treesame %u", nth_parent);
875
876         old_same = st->treesame[nth_parent];
877         memmove(st->treesame + nth_parent,
878                 st->treesame + nth_parent + 1,
879                 st->nparents - nth_parent - 1);
880
881         /*
882          * If we've just become a non-merge commit, update TREESAME
883          * immediately, and remove the no-longer-needed decoration.
884          * If still a merge, defer update until update_treesame().
885          */
886         if (--st->nparents == 1) {
887                 if (commit->parents->next)
888                         die("compact_treesame parents mismatch");
889                 if (st->treesame[0] && revs->dense)
890                         commit->object.flags |= TREESAME;
891                 else
892                         commit->object.flags &= ~TREESAME;
893                 free(add_decoration(&revs->treesame, &commit->object, NULL));
894         }
895
896         return old_same;
897 }
898
899 static unsigned update_treesame(struct rev_info *revs, struct commit *commit)
900 {
901         if (commit->parents && commit->parents->next) {
902                 unsigned n;
903                 struct treesame_state *st;
904                 struct commit_list *p;
905                 unsigned relevant_parents;
906                 unsigned relevant_change, irrelevant_change;
907
908                 st = lookup_decoration(&revs->treesame, &commit->object);
909                 if (!st)
910                         die("update_treesame %s", oid_to_hex(&commit->object.oid));
911                 relevant_parents = 0;
912                 relevant_change = irrelevant_change = 0;
913                 for (p = commit->parents, n = 0; p; n++, p = p->next) {
914                         if (relevant_commit(p->item)) {
915                                 relevant_change |= !st->treesame[n];
916                                 relevant_parents++;
917                         } else
918                                 irrelevant_change |= !st->treesame[n];
919                 }
920                 if (relevant_parents ? relevant_change : irrelevant_change)
921                         commit->object.flags &= ~TREESAME;
922                 else
923                         commit->object.flags |= TREESAME;
924         }
925
926         return commit->object.flags & TREESAME;
927 }
928
929 static inline int limiting_can_increase_treesame(const struct rev_info *revs)
930 {
931         /*
932          * TREESAME is irrelevant unless prune && dense;
933          * if simplify_history is set, we can't have a mixture of TREESAME and
934          *    !TREESAME INTERESTING parents (and we don't have treesame[]
935          *    decoration anyway);
936          * if first_parent_only is set, then the TREESAME flag is locked
937          *    against the first parent (and again we lack treesame[] decoration).
938          */
939         return revs->prune && revs->dense &&
940                !revs->simplify_history &&
941                !revs->first_parent_only;
942 }
943
944 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
945 {
946         struct commit_list **pp, *parent;
947         struct treesame_state *ts = NULL;
948         int relevant_change = 0, irrelevant_change = 0;
949         int relevant_parents, nth_parent;
950
951         /*
952          * If we don't do pruning, everything is interesting
953          */
954         if (!revs->prune)
955                 return;
956
957         if (!get_commit_tree(commit))
958                 return;
959
960         if (!commit->parents) {
961                 if (rev_same_tree_as_empty(revs, commit))
962                         commit->object.flags |= TREESAME;
963                 return;
964         }
965
966         /*
967          * Normal non-merge commit? If we don't want to make the
968          * history dense, we consider it always to be a change..
969          */
970         if (!revs->dense && !commit->parents->next)
971                 return;
972
973         for (pp = &commit->parents, nth_parent = 0, relevant_parents = 0;
974              (parent = *pp) != NULL;
975              pp = &parent->next, nth_parent++) {
976                 struct commit *p = parent->item;
977                 if (relevant_commit(p))
978                         relevant_parents++;
979
980                 if (nth_parent == 1) {
981                         /*
982                          * This our second loop iteration - so we now know
983                          * we're dealing with a merge.
984                          *
985                          * Do not compare with later parents when we care only about
986                          * the first parent chain, in order to avoid derailing the
987                          * traversal to follow a side branch that brought everything
988                          * in the path we are limited to by the pathspec.
989                          */
990                         if (revs->first_parent_only)
991                                 break;
992                         /*
993                          * If this will remain a potentially-simplifiable
994                          * merge, remember per-parent treesame if needed.
995                          * Initialise the array with the comparison from our
996                          * first iteration.
997                          */
998                         if (revs->treesame.name &&
999                             !revs->simplify_history &&
1000                             !(commit->object.flags & UNINTERESTING)) {
1001                                 ts = initialise_treesame(revs, commit);
1002                                 if (!(irrelevant_change || relevant_change))
1003                                         ts->treesame[0] = 1;
1004                         }
1005                 }
1006                 if (repo_parse_commit(revs->repo, p) < 0)
1007                         die("cannot simplify commit %s (because of %s)",
1008                             oid_to_hex(&commit->object.oid),
1009                             oid_to_hex(&p->object.oid));
1010                 switch (rev_compare_tree(revs, p, commit, nth_parent)) {
1011                 case REV_TREE_SAME:
1012                         if (!revs->simplify_history || !relevant_commit(p)) {
1013                                 /* Even if a merge with an uninteresting
1014                                  * side branch brought the entire change
1015                                  * we are interested in, we do not want
1016                                  * to lose the other branches of this
1017                                  * merge, so we just keep going.
1018                                  */
1019                                 if (ts)
1020                                         ts->treesame[nth_parent] = 1;
1021                                 continue;
1022                         }
1023                         parent->next = NULL;
1024                         commit->parents = parent;
1025
1026                         /*
1027                          * A merge commit is a "diversion" if it is not
1028                          * TREESAME to its first parent but is TREESAME
1029                          * to a later parent. In the simplified history,
1030                          * we "divert" the history walk to the later
1031                          * parent. These commits are shown when "show_pulls"
1032                          * is enabled, so do not mark the object as
1033                          * TREESAME here.
1034                          */
1035                         if (!revs->show_pulls || !nth_parent)
1036                                 commit->object.flags |= TREESAME;
1037
1038                         return;
1039
1040                 case REV_TREE_NEW:
1041                         if (revs->remove_empty_trees &&
1042                             rev_same_tree_as_empty(revs, p)) {
1043                                 /* We are adding all the specified
1044                                  * paths from this parent, so the
1045                                  * history beyond this parent is not
1046                                  * interesting.  Remove its parents
1047                                  * (they are grandparents for us).
1048                                  * IOW, we pretend this parent is a
1049                                  * "root" commit.
1050                                  */
1051                                 if (repo_parse_commit(revs->repo, p) < 0)
1052                                         die("cannot simplify commit %s (invalid %s)",
1053                                             oid_to_hex(&commit->object.oid),
1054                                             oid_to_hex(&p->object.oid));
1055                                 p->parents = NULL;
1056                         }
1057                 /* fallthrough */
1058                 case REV_TREE_OLD:
1059                 case REV_TREE_DIFFERENT:
1060                         if (relevant_commit(p))
1061                                 relevant_change = 1;
1062                         else
1063                                 irrelevant_change = 1;
1064
1065                         if (!nth_parent)
1066                                 commit->object.flags |= PULL_MERGE;
1067
1068                         continue;
1069                 }
1070                 die("bad tree compare for commit %s", oid_to_hex(&commit->object.oid));
1071         }
1072
1073         /*
1074          * TREESAME is straightforward for single-parent commits. For merge
1075          * commits, it is most useful to define it so that "irrelevant"
1076          * parents cannot make us !TREESAME - if we have any relevant
1077          * parents, then we only consider TREESAMEness with respect to them,
1078          * allowing irrelevant merges from uninteresting branches to be
1079          * simplified away. Only if we have only irrelevant parents do we
1080          * base TREESAME on them. Note that this logic is replicated in
1081          * update_treesame, which should be kept in sync.
1082          */
1083         if (relevant_parents ? !relevant_change : !irrelevant_change)
1084                 commit->object.flags |= TREESAME;
1085 }
1086
1087 static int process_parents(struct rev_info *revs, struct commit *commit,
1088                            struct commit_list **list, struct prio_queue *queue)
1089 {
1090         struct commit_list *parent = commit->parents;
1091         unsigned left_flag;
1092
1093         if (commit->object.flags & ADDED)
1094                 return 0;
1095         commit->object.flags |= ADDED;
1096
1097         if (revs->include_check &&
1098             !revs->include_check(commit, revs->include_check_data))
1099                 return 0;
1100
1101         /*
1102          * If the commit is uninteresting, don't try to
1103          * prune parents - we want the maximal uninteresting
1104          * set.
1105          *
1106          * Normally we haven't parsed the parent
1107          * yet, so we won't have a parent of a parent
1108          * here. However, it may turn out that we've
1109          * reached this commit some other way (where it
1110          * wasn't uninteresting), in which case we need
1111          * to mark its parents recursively too..
1112          */
1113         if (commit->object.flags & UNINTERESTING) {
1114                 while (parent) {
1115                         struct commit *p = parent->item;
1116                         parent = parent->next;
1117                         if (p)
1118                                 p->object.flags |= UNINTERESTING;
1119                         if (repo_parse_commit_gently(revs->repo, p, 1) < 0)
1120                                 continue;
1121                         if (p->parents)
1122                                 mark_parents_uninteresting(p);
1123                         if (p->object.flags & SEEN)
1124                                 continue;
1125                         p->object.flags |= SEEN;
1126                         if (list)
1127                                 commit_list_insert_by_date(p, list);
1128                         if (queue)
1129                                 prio_queue_put(queue, p);
1130                 }
1131                 return 0;
1132         }
1133
1134         /*
1135          * Ok, the commit wasn't uninteresting. Try to
1136          * simplify the commit history and find the parent
1137          * that has no differences in the path set if one exists.
1138          */
1139         try_to_simplify_commit(revs, commit);
1140
1141         if (revs->no_walk)
1142                 return 0;
1143
1144         left_flag = (commit->object.flags & SYMMETRIC_LEFT);
1145
1146         for (parent = commit->parents; parent; parent = parent->next) {
1147                 struct commit *p = parent->item;
1148                 int gently = revs->ignore_missing_links ||
1149                              revs->exclude_promisor_objects;
1150                 if (repo_parse_commit_gently(revs->repo, p, gently) < 0) {
1151                         if (revs->exclude_promisor_objects &&
1152                             is_promisor_object(&p->object.oid)) {
1153                                 if (revs->first_parent_only)
1154                                         break;
1155                                 continue;
1156                         }
1157                         return -1;
1158                 }
1159                 if (revs->sources) {
1160                         char **slot = revision_sources_at(revs->sources, p);
1161
1162                         if (!*slot)
1163                                 *slot = *revision_sources_at(revs->sources, commit);
1164                 }
1165                 p->object.flags |= left_flag;
1166                 if (!(p->object.flags & SEEN)) {
1167                         p->object.flags |= SEEN;
1168                         if (list)
1169                                 commit_list_insert_by_date(p, list);
1170                         if (queue)
1171                                 prio_queue_put(queue, p);
1172                 }
1173                 if (revs->first_parent_only)
1174                         break;
1175         }
1176         return 0;
1177 }
1178
1179 static void cherry_pick_list(struct commit_list *list, struct rev_info *revs)
1180 {
1181         struct commit_list *p;
1182         int left_count = 0, right_count = 0;
1183         int left_first;
1184         struct patch_ids ids;
1185         unsigned cherry_flag;
1186
1187         /* First count the commits on the left and on the right */
1188         for (p = list; p; p = p->next) {
1189                 struct commit *commit = p->item;
1190                 unsigned flags = commit->object.flags;
1191                 if (flags & BOUNDARY)
1192                         ;
1193                 else if (flags & SYMMETRIC_LEFT)
1194                         left_count++;
1195                 else
1196                         right_count++;
1197         }
1198
1199         if (!left_count || !right_count)
1200                 return;
1201
1202         left_first = left_count < right_count;
1203         init_patch_ids(revs->repo, &ids);
1204         ids.diffopts.pathspec = revs->diffopt.pathspec;
1205
1206         /* Compute patch-ids for one side */
1207         for (p = list; p; p = p->next) {
1208                 struct commit *commit = p->item;
1209                 unsigned flags = commit->object.flags;
1210
1211                 if (flags & BOUNDARY)
1212                         continue;
1213                 /*
1214                  * If we have fewer left, left_first is set and we omit
1215                  * commits on the right branch in this loop.  If we have
1216                  * fewer right, we skip the left ones.
1217                  */
1218                 if (left_first != !!(flags & SYMMETRIC_LEFT))
1219                         continue;
1220                 add_commit_patch_id(commit, &ids);
1221         }
1222
1223         /* either cherry_mark or cherry_pick are true */
1224         cherry_flag = revs->cherry_mark ? PATCHSAME : SHOWN;
1225
1226         /* Check the other side */
1227         for (p = list; p; p = p->next) {
1228                 struct commit *commit = p->item;
1229                 struct patch_id *id;
1230                 unsigned flags = commit->object.flags;
1231
1232                 if (flags & BOUNDARY)
1233                         continue;
1234                 /*
1235                  * If we have fewer left, left_first is set and we omit
1236                  * commits on the left branch in this loop.
1237                  */
1238                 if (left_first == !!(flags & SYMMETRIC_LEFT))
1239                         continue;
1240
1241                 /*
1242                  * Have we seen the same patch id?
1243                  */
1244                 id = patch_id_iter_first(commit, &ids);
1245                 if (!id)
1246                         continue;
1247
1248                 commit->object.flags |= cherry_flag;
1249                 do {
1250                         id->commit->object.flags |= cherry_flag;
1251                 } while ((id = patch_id_iter_next(id, &ids)));
1252         }
1253
1254         free_patch_ids(&ids);
1255 }
1256
1257 /* How many extra uninteresting commits we want to see.. */
1258 #define SLOP 5
1259
1260 static int still_interesting(struct commit_list *src, timestamp_t date, int slop,
1261                              struct commit **interesting_cache)
1262 {
1263         /*
1264          * No source list at all? We're definitely done..
1265          */
1266         if (!src)
1267                 return 0;
1268
1269         /*
1270          * Does the destination list contain entries with a date
1271          * before the source list? Definitely _not_ done.
1272          */
1273         if (date <= src->item->date)
1274                 return SLOP;
1275
1276         /*
1277          * Does the source list still have interesting commits in
1278          * it? Definitely not done..
1279          */
1280         if (!everybody_uninteresting(src, interesting_cache))
1281                 return SLOP;
1282
1283         /* Ok, we're closing in.. */
1284         return slop-1;
1285 }
1286
1287 /*
1288  * "rev-list --ancestry-path A..B" computes commits that are ancestors
1289  * of B but not ancestors of A but further limits the result to those
1290  * that are descendants of A.  This takes the list of bottom commits and
1291  * the result of "A..B" without --ancestry-path, and limits the latter
1292  * further to the ones that can reach one of the commits in "bottom".
1293  */
1294 static void limit_to_ancestry(struct commit_list *bottom, struct commit_list *list)
1295 {
1296         struct commit_list *p;
1297         struct commit_list *rlist = NULL;
1298         int made_progress;
1299
1300         /*
1301          * Reverse the list so that it will be likely that we would
1302          * process parents before children.
1303          */
1304         for (p = list; p; p = p->next)
1305                 commit_list_insert(p->item, &rlist);
1306
1307         for (p = bottom; p; p = p->next)
1308                 p->item->object.flags |= TMP_MARK;
1309
1310         /*
1311          * Mark the ones that can reach bottom commits in "list",
1312          * in a bottom-up fashion.
1313          */
1314         do {
1315                 made_progress = 0;
1316                 for (p = rlist; p; p = p->next) {
1317                         struct commit *c = p->item;
1318                         struct commit_list *parents;
1319                         if (c->object.flags & (TMP_MARK | UNINTERESTING))
1320                                 continue;
1321                         for (parents = c->parents;
1322                              parents;
1323                              parents = parents->next) {
1324                                 if (!(parents->item->object.flags & TMP_MARK))
1325                                         continue;
1326                                 c->object.flags |= TMP_MARK;
1327                                 made_progress = 1;
1328                                 break;
1329                         }
1330                 }
1331         } while (made_progress);
1332
1333         /*
1334          * NEEDSWORK: decide if we want to remove parents that are
1335          * not marked with TMP_MARK from commit->parents for commits
1336          * in the resulting list.  We may not want to do that, though.
1337          */
1338
1339         /*
1340          * The ones that are not marked with TMP_MARK are uninteresting
1341          */
1342         for (p = list; p; p = p->next) {
1343                 struct commit *c = p->item;
1344                 if (c->object.flags & TMP_MARK)
1345                         continue;
1346                 c->object.flags |= UNINTERESTING;
1347         }
1348
1349         /* We are done with the TMP_MARK */
1350         for (p = list; p; p = p->next)
1351                 p->item->object.flags &= ~TMP_MARK;
1352         for (p = bottom; p; p = p->next)
1353                 p->item->object.flags &= ~TMP_MARK;
1354         free_commit_list(rlist);
1355 }
1356
1357 /*
1358  * Before walking the history, keep the set of "negative" refs the
1359  * caller has asked to exclude.
1360  *
1361  * This is used to compute "rev-list --ancestry-path A..B", as we need
1362  * to filter the result of "A..B" further to the ones that can actually
1363  * reach A.
1364  */
1365 static struct commit_list *collect_bottom_commits(struct commit_list *list)
1366 {
1367         struct commit_list *elem, *bottom = NULL;
1368         for (elem = list; elem; elem = elem->next)
1369                 if (elem->item->object.flags & BOTTOM)
1370                         commit_list_insert(elem->item, &bottom);
1371         return bottom;
1372 }
1373
1374 /* Assumes either left_only or right_only is set */
1375 static void limit_left_right(struct commit_list *list, struct rev_info *revs)
1376 {
1377         struct commit_list *p;
1378
1379         for (p = list; p; p = p->next) {
1380                 struct commit *commit = p->item;
1381
1382                 if (revs->right_only) {
1383                         if (commit->object.flags & SYMMETRIC_LEFT)
1384                                 commit->object.flags |= SHOWN;
1385                 } else  /* revs->left_only is set */
1386                         if (!(commit->object.flags & SYMMETRIC_LEFT))
1387                                 commit->object.flags |= SHOWN;
1388         }
1389 }
1390
1391 static int limit_list(struct rev_info *revs)
1392 {
1393         int slop = SLOP;
1394         timestamp_t date = TIME_MAX;
1395         struct commit_list *list = revs->commits;
1396         struct commit_list *newlist = NULL;
1397         struct commit_list **p = &newlist;
1398         struct commit_list *bottom = NULL;
1399         struct commit *interesting_cache = NULL;
1400
1401         if (revs->ancestry_path) {
1402                 bottom = collect_bottom_commits(list);
1403                 if (!bottom)
1404                         die("--ancestry-path given but there are no bottom commits");
1405         }
1406
1407         while (list) {
1408                 struct commit *commit = pop_commit(&list);
1409                 struct object *obj = &commit->object;
1410                 show_early_output_fn_t show;
1411
1412                 if (commit == interesting_cache)
1413                         interesting_cache = NULL;
1414
1415                 if (revs->max_age != -1 && (commit->date < revs->max_age))
1416                         obj->flags |= UNINTERESTING;
1417                 if (process_parents(revs, commit, &list, NULL) < 0)
1418                         return -1;
1419                 if (obj->flags & UNINTERESTING) {
1420                         mark_parents_uninteresting(commit);
1421                         slop = still_interesting(list, date, slop, &interesting_cache);
1422                         if (slop)
1423                                 continue;
1424                         break;
1425                 }
1426                 if (revs->min_age != -1 && (commit->date > revs->min_age) &&
1427                     !revs->line_level_traverse)
1428                         continue;
1429                 date = commit->date;
1430                 p = &commit_list_insert(commit, p)->next;
1431
1432                 show = show_early_output;
1433                 if (!show)
1434                         continue;
1435
1436                 show(revs, newlist);
1437                 show_early_output = NULL;
1438         }
1439         if (revs->cherry_pick || revs->cherry_mark)
1440                 cherry_pick_list(newlist, revs);
1441
1442         if (revs->left_only || revs->right_only)
1443                 limit_left_right(newlist, revs);
1444
1445         if (bottom) {
1446                 limit_to_ancestry(bottom, newlist);
1447                 free_commit_list(bottom);
1448         }
1449
1450         /*
1451          * Check if any commits have become TREESAME by some of their parents
1452          * becoming UNINTERESTING.
1453          */
1454         if (limiting_can_increase_treesame(revs))
1455                 for (list = newlist; list; list = list->next) {
1456                         struct commit *c = list->item;
1457                         if (c->object.flags & (UNINTERESTING | TREESAME))
1458                                 continue;
1459                         update_treesame(revs, c);
1460                 }
1461
1462         revs->commits = newlist;
1463         return 0;
1464 }
1465
1466 /*
1467  * Add an entry to refs->cmdline with the specified information.
1468  * *name is copied.
1469  */
1470 static void add_rev_cmdline(struct rev_info *revs,
1471                             struct object *item,
1472                             const char *name,
1473                             int whence,
1474                             unsigned flags)
1475 {
1476         struct rev_cmdline_info *info = &revs->cmdline;
1477         unsigned int nr = info->nr;
1478
1479         ALLOC_GROW(info->rev, nr + 1, info->alloc);
1480         info->rev[nr].item = item;
1481         info->rev[nr].name = xstrdup(name);
1482         info->rev[nr].whence = whence;
1483         info->rev[nr].flags = flags;
1484         info->nr++;
1485 }
1486
1487 static void add_rev_cmdline_list(struct rev_info *revs,
1488                                  struct commit_list *commit_list,
1489                                  int whence,
1490                                  unsigned flags)
1491 {
1492         while (commit_list) {
1493                 struct object *object = &commit_list->item->object;
1494                 add_rev_cmdline(revs, object, oid_to_hex(&object->oid),
1495                                 whence, flags);
1496                 commit_list = commit_list->next;
1497         }
1498 }
1499
1500 struct all_refs_cb {
1501         int all_flags;
1502         int warned_bad_reflog;
1503         struct rev_info *all_revs;
1504         const char *name_for_errormsg;
1505         struct worktree *wt;
1506 };
1507
1508 int ref_excluded(struct string_list *ref_excludes, const char *path)
1509 {
1510         struct string_list_item *item;
1511
1512         if (!ref_excludes)
1513                 return 0;
1514         for_each_string_list_item(item, ref_excludes) {
1515                 if (!wildmatch(item->string, path, 0))
1516                         return 1;
1517         }
1518         return 0;
1519 }
1520
1521 static int handle_one_ref(const char *path, const struct object_id *oid,
1522                           int flag, void *cb_data)
1523 {
1524         struct all_refs_cb *cb = cb_data;
1525         struct object *object;
1526
1527         if (ref_excluded(cb->all_revs->ref_excludes, path))
1528             return 0;
1529
1530         object = get_reference(cb->all_revs, path, oid, cb->all_flags);
1531         add_rev_cmdline(cb->all_revs, object, path, REV_CMD_REF, cb->all_flags);
1532         add_pending_oid(cb->all_revs, path, oid, cb->all_flags);
1533         return 0;
1534 }
1535
1536 static void init_all_refs_cb(struct all_refs_cb *cb, struct rev_info *revs,
1537         unsigned flags)
1538 {
1539         cb->all_revs = revs;
1540         cb->all_flags = flags;
1541         revs->rev_input_given = 1;
1542         cb->wt = NULL;
1543 }
1544
1545 void clear_ref_exclusion(struct string_list **ref_excludes_p)
1546 {
1547         if (*ref_excludes_p) {
1548                 string_list_clear(*ref_excludes_p, 0);
1549                 free(*ref_excludes_p);
1550         }
1551         *ref_excludes_p = NULL;
1552 }
1553
1554 void add_ref_exclusion(struct string_list **ref_excludes_p, const char *exclude)
1555 {
1556         if (!*ref_excludes_p) {
1557                 *ref_excludes_p = xcalloc(1, sizeof(**ref_excludes_p));
1558                 (*ref_excludes_p)->strdup_strings = 1;
1559         }
1560         string_list_append(*ref_excludes_p, exclude);
1561 }
1562
1563 static void handle_refs(struct ref_store *refs,
1564                         struct rev_info *revs, unsigned flags,
1565                         int (*for_each)(struct ref_store *, each_ref_fn, void *))
1566 {
1567         struct all_refs_cb cb;
1568
1569         if (!refs) {
1570                 /* this could happen with uninitialized submodules */
1571                 return;
1572         }
1573
1574         init_all_refs_cb(&cb, revs, flags);
1575         for_each(refs, handle_one_ref, &cb);
1576 }
1577
1578 static void handle_one_reflog_commit(struct object_id *oid, void *cb_data)
1579 {
1580         struct all_refs_cb *cb = cb_data;
1581         if (!is_null_oid(oid)) {
1582                 struct object *o = parse_object(cb->all_revs->repo, oid);
1583                 if (o) {
1584                         o->flags |= cb->all_flags;
1585                         /* ??? CMDLINEFLAGS ??? */
1586                         add_pending_object(cb->all_revs, o, "");
1587                 }
1588                 else if (!cb->warned_bad_reflog) {
1589                         warning("reflog of '%s' references pruned commits",
1590                                 cb->name_for_errormsg);
1591                         cb->warned_bad_reflog = 1;
1592                 }
1593         }
1594 }
1595
1596 static int handle_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
1597                 const char *email, timestamp_t timestamp, int tz,
1598                 const char *message, void *cb_data)
1599 {
1600         handle_one_reflog_commit(ooid, cb_data);
1601         handle_one_reflog_commit(noid, cb_data);
1602         return 0;
1603 }
1604
1605 static int handle_one_reflog(const char *refname_in_wt,
1606                              const struct object_id *oid,
1607                              int flag, void *cb_data)
1608 {
1609         struct all_refs_cb *cb = cb_data;
1610         struct strbuf refname = STRBUF_INIT;
1611
1612         cb->warned_bad_reflog = 0;
1613         strbuf_worktree_ref(cb->wt, &refname, refname_in_wt);
1614         cb->name_for_errormsg = refname.buf;
1615         refs_for_each_reflog_ent(get_main_ref_store(the_repository),
1616                                  refname.buf,
1617                                  handle_one_reflog_ent, cb_data);
1618         strbuf_release(&refname);
1619         return 0;
1620 }
1621
1622 static void add_other_reflogs_to_pending(struct all_refs_cb *cb)
1623 {
1624         struct worktree **worktrees, **p;
1625
1626         worktrees = get_worktrees();
1627         for (p = worktrees; *p; p++) {
1628                 struct worktree *wt = *p;
1629
1630                 if (wt->is_current)
1631                         continue;
1632
1633                 cb->wt = wt;
1634                 refs_for_each_reflog(get_worktree_ref_store(wt),
1635                                      handle_one_reflog,
1636                                      cb);
1637         }
1638         free_worktrees(worktrees);
1639 }
1640
1641 void add_reflogs_to_pending(struct rev_info *revs, unsigned flags)
1642 {
1643         struct all_refs_cb cb;
1644
1645         cb.all_revs = revs;
1646         cb.all_flags = flags;
1647         cb.wt = NULL;
1648         for_each_reflog(handle_one_reflog, &cb);
1649
1650         if (!revs->single_worktree)
1651                 add_other_reflogs_to_pending(&cb);
1652 }
1653
1654 static void add_cache_tree(struct cache_tree *it, struct rev_info *revs,
1655                            struct strbuf *path, unsigned int flags)
1656 {
1657         size_t baselen = path->len;
1658         int i;
1659
1660         if (it->entry_count >= 0) {
1661                 struct tree *tree = lookup_tree(revs->repo, &it->oid);
1662                 tree->object.flags |= flags;
1663                 add_pending_object_with_path(revs, &tree->object, "",
1664                                              040000, path->buf);
1665         }
1666
1667         for (i = 0; i < it->subtree_nr; i++) {
1668                 struct cache_tree_sub *sub = it->down[i];
1669                 strbuf_addf(path, "%s%s", baselen ? "/" : "", sub->name);
1670                 add_cache_tree(sub->cache_tree, revs, path, flags);
1671                 strbuf_setlen(path, baselen);
1672         }
1673
1674 }
1675
1676 static void do_add_index_objects_to_pending(struct rev_info *revs,
1677                                             struct index_state *istate,
1678                                             unsigned int flags)
1679 {
1680         int i;
1681
1682         for (i = 0; i < istate->cache_nr; i++) {
1683                 struct cache_entry *ce = istate->cache[i];
1684                 struct blob *blob;
1685
1686                 if (S_ISGITLINK(ce->ce_mode))
1687                         continue;
1688
1689                 blob = lookup_blob(revs->repo, &ce->oid);
1690                 if (!blob)
1691                         die("unable to add index blob to traversal");
1692                 blob->object.flags |= flags;
1693                 add_pending_object_with_path(revs, &blob->object, "",
1694                                              ce->ce_mode, ce->name);
1695         }
1696
1697         if (istate->cache_tree) {
1698                 struct strbuf path = STRBUF_INIT;
1699                 add_cache_tree(istate->cache_tree, revs, &path, flags);
1700                 strbuf_release(&path);
1701         }
1702 }
1703
1704 void add_index_objects_to_pending(struct rev_info *revs, unsigned int flags)
1705 {
1706         struct worktree **worktrees, **p;
1707
1708         repo_read_index(revs->repo);
1709         do_add_index_objects_to_pending(revs, revs->repo->index, flags);
1710
1711         if (revs->single_worktree)
1712                 return;
1713
1714         worktrees = get_worktrees();
1715         for (p = worktrees; *p; p++) {
1716                 struct worktree *wt = *p;
1717                 struct index_state istate = { NULL };
1718
1719                 if (wt->is_current)
1720                         continue; /* current index already taken care of */
1721
1722                 if (read_index_from(&istate,
1723                                     worktree_git_path(wt, "index"),
1724                                     get_worktree_git_dir(wt)) > 0)
1725                         do_add_index_objects_to_pending(revs, &istate, flags);
1726                 discard_index(&istate);
1727         }
1728         free_worktrees(worktrees);
1729 }
1730
1731 struct add_alternate_refs_data {
1732         struct rev_info *revs;
1733         unsigned int flags;
1734 };
1735
1736 static void add_one_alternate_ref(const struct object_id *oid,
1737                                   void *vdata)
1738 {
1739         const char *name = ".alternate";
1740         struct add_alternate_refs_data *data = vdata;
1741         struct object *obj;
1742
1743         obj = get_reference(data->revs, name, oid, data->flags);
1744         add_rev_cmdline(data->revs, obj, name, REV_CMD_REV, data->flags);
1745         add_pending_object(data->revs, obj, name);
1746 }
1747
1748 static void add_alternate_refs_to_pending(struct rev_info *revs,
1749                                           unsigned int flags)
1750 {
1751         struct add_alternate_refs_data data;
1752         data.revs = revs;
1753         data.flags = flags;
1754         for_each_alternate_ref(add_one_alternate_ref, &data);
1755 }
1756
1757 static int add_parents_only(struct rev_info *revs, const char *arg_, int flags,
1758                             int exclude_parent)
1759 {
1760         struct object_id oid;
1761         struct object *it;
1762         struct commit *commit;
1763         struct commit_list *parents;
1764         int parent_number;
1765         const char *arg = arg_;
1766
1767         if (*arg == '^') {
1768                 flags ^= UNINTERESTING | BOTTOM;
1769                 arg++;
1770         }
1771         if (get_oid_committish(arg, &oid))
1772                 return 0;
1773         while (1) {
1774                 it = get_reference(revs, arg, &oid, 0);
1775                 if (!it && revs->ignore_missing)
1776                         return 0;
1777                 if (it->type != OBJ_TAG)
1778                         break;
1779                 if (!((struct tag*)it)->tagged)
1780                         return 0;
1781                 oidcpy(&oid, &((struct tag*)it)->tagged->oid);
1782         }
1783         if (it->type != OBJ_COMMIT)
1784                 return 0;
1785         commit = (struct commit *)it;
1786         if (exclude_parent &&
1787             exclude_parent > commit_list_count(commit->parents))
1788                 return 0;
1789         for (parents = commit->parents, parent_number = 1;
1790              parents;
1791              parents = parents->next, parent_number++) {
1792                 if (exclude_parent && parent_number != exclude_parent)
1793                         continue;
1794
1795                 it = &parents->item->object;
1796                 it->flags |= flags;
1797                 add_rev_cmdline(revs, it, arg_, REV_CMD_PARENTS_ONLY, flags);
1798                 add_pending_object(revs, it, arg);
1799         }
1800         return 1;
1801 }
1802
1803 void repo_init_revisions(struct repository *r,
1804                          struct rev_info *revs,
1805                          const char *prefix)
1806 {
1807         memset(revs, 0, sizeof(*revs));
1808
1809         revs->repo = r;
1810         revs->abbrev = DEFAULT_ABBREV;
1811         revs->ignore_merges = -1;
1812         revs->simplify_history = 1;
1813         revs->pruning.repo = r;
1814         revs->pruning.flags.recursive = 1;
1815         revs->pruning.flags.quick = 1;
1816         revs->pruning.add_remove = file_add_remove;
1817         revs->pruning.change = file_change;
1818         revs->pruning.change_fn_data = revs;
1819         revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1820         revs->dense = 1;
1821         revs->prefix = prefix;
1822         revs->max_age = -1;
1823         revs->min_age = -1;
1824         revs->skip_count = -1;
1825         revs->max_count = -1;
1826         revs->max_parents = -1;
1827         revs->expand_tabs_in_log = -1;
1828
1829         revs->commit_format = CMIT_FMT_DEFAULT;
1830         revs->expand_tabs_in_log_default = 8;
1831
1832         grep_init(&revs->grep_filter, revs->repo, prefix);
1833         revs->grep_filter.status_only = 1;
1834
1835         repo_diff_setup(revs->repo, &revs->diffopt);
1836         if (prefix && !revs->diffopt.prefix) {
1837                 revs->diffopt.prefix = prefix;
1838                 revs->diffopt.prefix_length = strlen(prefix);
1839         }
1840
1841         init_display_notes(&revs->notes_opt);
1842 }
1843
1844 static void add_pending_commit_list(struct rev_info *revs,
1845                                     struct commit_list *commit_list,
1846                                     unsigned int flags)
1847 {
1848         while (commit_list) {
1849                 struct object *object = &commit_list->item->object;
1850                 object->flags |= flags;
1851                 add_pending_object(revs, object, oid_to_hex(&object->oid));
1852                 commit_list = commit_list->next;
1853         }
1854 }
1855
1856 static void prepare_show_merge(struct rev_info *revs)
1857 {
1858         struct commit_list *bases;
1859         struct commit *head, *other;
1860         struct object_id oid;
1861         const char **prune = NULL;
1862         int i, prune_num = 1; /* counting terminating NULL */
1863         struct index_state *istate = revs->repo->index;
1864
1865         if (get_oid("HEAD", &oid))
1866                 die("--merge without HEAD?");
1867         head = lookup_commit_or_die(&oid, "HEAD");
1868         if (get_oid("MERGE_HEAD", &oid))
1869                 die("--merge without MERGE_HEAD?");
1870         other = lookup_commit_or_die(&oid, "MERGE_HEAD");
1871         add_pending_object(revs, &head->object, "HEAD");
1872         add_pending_object(revs, &other->object, "MERGE_HEAD");
1873         bases = get_merge_bases(head, other);
1874         add_rev_cmdline_list(revs, bases, REV_CMD_MERGE_BASE, UNINTERESTING | BOTTOM);
1875         add_pending_commit_list(revs, bases, UNINTERESTING | BOTTOM);
1876         free_commit_list(bases);
1877         head->object.flags |= SYMMETRIC_LEFT;
1878
1879         if (!istate->cache_nr)
1880                 repo_read_index(revs->repo);
1881         for (i = 0; i < istate->cache_nr; i++) {
1882                 const struct cache_entry *ce = istate->cache[i];
1883                 if (!ce_stage(ce))
1884                         continue;
1885                 if (ce_path_match(istate, ce, &revs->prune_data, NULL)) {
1886                         prune_num++;
1887                         REALLOC_ARRAY(prune, prune_num);
1888                         prune[prune_num-2] = ce->name;
1889                         prune[prune_num-1] = NULL;
1890                 }
1891                 while ((i+1 < istate->cache_nr) &&
1892                        ce_same_name(ce, istate->cache[i+1]))
1893                         i++;
1894         }
1895         clear_pathspec(&revs->prune_data);
1896         parse_pathspec(&revs->prune_data, PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
1897                        PATHSPEC_PREFER_FULL | PATHSPEC_LITERAL_PATH, "", prune);
1898         revs->limited = 1;
1899 }
1900
1901 static int dotdot_missing(const char *arg, char *dotdot,
1902                           struct rev_info *revs, int symmetric)
1903 {
1904         if (revs->ignore_missing)
1905                 return 0;
1906         /* de-munge so we report the full argument */
1907         *dotdot = '.';
1908         die(symmetric
1909             ? "Invalid symmetric difference expression %s"
1910             : "Invalid revision range %s", arg);
1911 }
1912
1913 static int handle_dotdot_1(const char *arg, char *dotdot,
1914                            struct rev_info *revs, int flags,
1915                            int cant_be_filename,
1916                            struct object_context *a_oc,
1917                            struct object_context *b_oc)
1918 {
1919         const char *a_name, *b_name;
1920         struct object_id a_oid, b_oid;
1921         struct object *a_obj, *b_obj;
1922         unsigned int a_flags, b_flags;
1923         int symmetric = 0;
1924         unsigned int flags_exclude = flags ^ (UNINTERESTING | BOTTOM);
1925         unsigned int oc_flags = GET_OID_COMMITTISH | GET_OID_RECORD_PATH;
1926
1927         a_name = arg;
1928         if (!*a_name)
1929                 a_name = "HEAD";
1930
1931         b_name = dotdot + 2;
1932         if (*b_name == '.') {
1933                 symmetric = 1;
1934                 b_name++;
1935         }
1936         if (!*b_name)
1937                 b_name = "HEAD";
1938
1939         if (get_oid_with_context(revs->repo, a_name, oc_flags, &a_oid, a_oc) ||
1940             get_oid_with_context(revs->repo, b_name, oc_flags, &b_oid, b_oc))
1941                 return -1;
1942
1943         if (!cant_be_filename) {
1944                 *dotdot = '.';
1945                 verify_non_filename(revs->prefix, arg);
1946                 *dotdot = '\0';
1947         }
1948
1949         a_obj = parse_object(revs->repo, &a_oid);
1950         b_obj = parse_object(revs->repo, &b_oid);
1951         if (!a_obj || !b_obj)
1952                 return dotdot_missing(arg, dotdot, revs, symmetric);
1953
1954         if (!symmetric) {
1955                 /* just A..B */
1956                 b_flags = flags;
1957                 a_flags = flags_exclude;
1958         } else {
1959                 /* A...B -- find merge bases between the two */
1960                 struct commit *a, *b;
1961                 struct commit_list *exclude;
1962
1963                 a = lookup_commit_reference(revs->repo, &a_obj->oid);
1964                 b = lookup_commit_reference(revs->repo, &b_obj->oid);
1965                 if (!a || !b)
1966                         return dotdot_missing(arg, dotdot, revs, symmetric);
1967
1968                 exclude = get_merge_bases(a, b);
1969                 add_rev_cmdline_list(revs, exclude, REV_CMD_MERGE_BASE,
1970                                      flags_exclude);
1971                 add_pending_commit_list(revs, exclude, flags_exclude);
1972                 free_commit_list(exclude);
1973
1974                 b_flags = flags;
1975                 a_flags = flags | SYMMETRIC_LEFT;
1976         }
1977
1978         a_obj->flags |= a_flags;
1979         b_obj->flags |= b_flags;
1980         add_rev_cmdline(revs, a_obj, a_name, REV_CMD_LEFT, a_flags);
1981         add_rev_cmdline(revs, b_obj, b_name, REV_CMD_RIGHT, b_flags);
1982         add_pending_object_with_path(revs, a_obj, a_name, a_oc->mode, a_oc->path);
1983         add_pending_object_with_path(revs, b_obj, b_name, b_oc->mode, b_oc->path);
1984         return 0;
1985 }
1986
1987 static int handle_dotdot(const char *arg,
1988                          struct rev_info *revs, int flags,
1989                          int cant_be_filename)
1990 {
1991         struct object_context a_oc, b_oc;
1992         char *dotdot = strstr(arg, "..");
1993         int ret;
1994
1995         if (!dotdot)
1996                 return -1;
1997
1998         memset(&a_oc, 0, sizeof(a_oc));
1999         memset(&b_oc, 0, sizeof(b_oc));
2000
2001         *dotdot = '\0';
2002         ret = handle_dotdot_1(arg, dotdot, revs, flags, cant_be_filename,
2003                               &a_oc, &b_oc);
2004         *dotdot = '.';
2005
2006         free(a_oc.path);
2007         free(b_oc.path);
2008
2009         return ret;
2010 }
2011
2012 static int handle_revision_arg_1(const char *arg_, struct rev_info *revs, int flags, unsigned revarg_opt)
2013 {
2014         struct object_context oc;
2015         char *mark;
2016         struct object *object;
2017         struct object_id oid;
2018         int local_flags;
2019         const char *arg = arg_;
2020         int cant_be_filename = revarg_opt & REVARG_CANNOT_BE_FILENAME;
2021         unsigned get_sha1_flags = GET_OID_RECORD_PATH;
2022
2023         flags = flags & UNINTERESTING ? flags | BOTTOM : flags & ~BOTTOM;
2024
2025         if (!cant_be_filename && !strcmp(arg, "..")) {
2026                 /*
2027                  * Just ".."?  That is not a range but the
2028                  * pathspec for the parent directory.
2029                  */
2030                 return -1;
2031         }
2032
2033         if (!handle_dotdot(arg, revs, flags, revarg_opt))
2034                 return 0;
2035
2036         mark = strstr(arg, "^@");
2037         if (mark && !mark[2]) {
2038                 *mark = 0;
2039                 if (add_parents_only(revs, arg, flags, 0))
2040                         return 0;
2041                 *mark = '^';
2042         }
2043         mark = strstr(arg, "^!");
2044         if (mark && !mark[2]) {
2045                 *mark = 0;
2046                 if (!add_parents_only(revs, arg, flags ^ (UNINTERESTING | BOTTOM), 0))
2047                         *mark = '^';
2048         }
2049         mark = strstr(arg, "^-");
2050         if (mark) {
2051                 int exclude_parent = 1;
2052
2053                 if (mark[2]) {
2054                         char *end;
2055                         exclude_parent = strtoul(mark + 2, &end, 10);
2056                         if (*end != '\0' || !exclude_parent)
2057                                 return -1;
2058                 }
2059
2060                 *mark = 0;
2061                 if (!add_parents_only(revs, arg, flags ^ (UNINTERESTING | BOTTOM), exclude_parent))
2062                         *mark = '^';
2063         }
2064
2065         local_flags = 0;
2066         if (*arg == '^') {
2067                 local_flags = UNINTERESTING | BOTTOM;
2068                 arg++;
2069         }
2070
2071         if (revarg_opt & REVARG_COMMITTISH)
2072                 get_sha1_flags |= GET_OID_COMMITTISH;
2073
2074         if (get_oid_with_context(revs->repo, arg, get_sha1_flags, &oid, &oc))
2075                 return revs->ignore_missing ? 0 : -1;
2076         if (!cant_be_filename)
2077                 verify_non_filename(revs->prefix, arg);
2078         object = get_reference(revs, arg, &oid, flags ^ local_flags);
2079         if (!object)
2080                 return revs->ignore_missing ? 0 : -1;
2081         add_rev_cmdline(revs, object, arg_, REV_CMD_REV, flags ^ local_flags);
2082         add_pending_object_with_path(revs, object, arg, oc.mode, oc.path);
2083         free(oc.path);
2084         return 0;
2085 }
2086
2087 int handle_revision_arg(const char *arg, struct rev_info *revs, int flags, unsigned revarg_opt)
2088 {
2089         int ret = handle_revision_arg_1(arg, revs, flags, revarg_opt);
2090         if (!ret)
2091                 revs->rev_input_given = 1;
2092         return ret;
2093 }
2094
2095 static void read_pathspec_from_stdin(struct strbuf *sb,
2096                                      struct strvec *prune)
2097 {
2098         while (strbuf_getline(sb, stdin) != EOF)
2099                 strvec_push(prune, sb->buf);
2100 }
2101
2102 static void read_revisions_from_stdin(struct rev_info *revs,
2103                                       struct strvec *prune)
2104 {
2105         struct strbuf sb;
2106         int seen_dashdash = 0;
2107         int save_warning;
2108
2109         save_warning = warn_on_object_refname_ambiguity;
2110         warn_on_object_refname_ambiguity = 0;
2111
2112         strbuf_init(&sb, 1000);
2113         while (strbuf_getline(&sb, stdin) != EOF) {
2114                 int len = sb.len;
2115                 if (!len)
2116                         break;
2117                 if (sb.buf[0] == '-') {
2118                         if (len == 2 && sb.buf[1] == '-') {
2119                                 seen_dashdash = 1;
2120                                 break;
2121                         }
2122                         die("options not supported in --stdin mode");
2123                 }
2124                 if (handle_revision_arg(sb.buf, revs, 0,
2125                                         REVARG_CANNOT_BE_FILENAME))
2126                         die("bad revision '%s'", sb.buf);
2127         }
2128         if (seen_dashdash)
2129                 read_pathspec_from_stdin(&sb, prune);
2130
2131         strbuf_release(&sb);
2132         warn_on_object_refname_ambiguity = save_warning;
2133 }
2134
2135 static void add_grep(struct rev_info *revs, const char *ptn, enum grep_pat_token what)
2136 {
2137         append_grep_pattern(&revs->grep_filter, ptn, "command line", 0, what);
2138 }
2139
2140 static void add_header_grep(struct rev_info *revs, enum grep_header_field field, const char *pattern)
2141 {
2142         append_header_grep_pattern(&revs->grep_filter, field, pattern);
2143 }
2144
2145 static void add_message_grep(struct rev_info *revs, const char *pattern)
2146 {
2147         add_grep(revs, pattern, GREP_PATTERN_BODY);
2148 }
2149
2150 static int handle_revision_opt(struct rev_info *revs, int argc, const char **argv,
2151                                int *unkc, const char **unkv,
2152                                const struct setup_revision_opt* opt)
2153 {
2154         const char *arg = argv[0];
2155         const char *optarg;
2156         int argcount;
2157         const unsigned hexsz = the_hash_algo->hexsz;
2158
2159         /* pseudo revision arguments */
2160         if (!strcmp(arg, "--all") || !strcmp(arg, "--branches") ||
2161             !strcmp(arg, "--tags") || !strcmp(arg, "--remotes") ||
2162             !strcmp(arg, "--reflog") || !strcmp(arg, "--not") ||
2163             !strcmp(arg, "--no-walk") || !strcmp(arg, "--do-walk") ||
2164             !strcmp(arg, "--bisect") || starts_with(arg, "--glob=") ||
2165             !strcmp(arg, "--indexed-objects") ||
2166             !strcmp(arg, "--alternate-refs") ||
2167             starts_with(arg, "--exclude=") ||
2168             starts_with(arg, "--branches=") || starts_with(arg, "--tags=") ||
2169             starts_with(arg, "--remotes=") || starts_with(arg, "--no-walk="))
2170         {
2171                 unkv[(*unkc)++] = arg;
2172                 return 1;
2173         }
2174
2175         if ((argcount = parse_long_opt("max-count", argv, &optarg))) {
2176                 revs->max_count = atoi(optarg);
2177                 revs->no_walk = 0;
2178                 return argcount;
2179         } else if ((argcount = parse_long_opt("skip", argv, &optarg))) {
2180                 revs->skip_count = atoi(optarg);
2181                 return argcount;
2182         } else if ((*arg == '-') && isdigit(arg[1])) {
2183                 /* accept -<digit>, like traditional "head" */
2184                 if (strtol_i(arg + 1, 10, &revs->max_count) < 0 ||
2185                     revs->max_count < 0)
2186                         die("'%s': not a non-negative integer", arg + 1);
2187                 revs->no_walk = 0;
2188         } else if (!strcmp(arg, "-n")) {
2189                 if (argc <= 1)
2190                         return error("-n requires an argument");
2191                 revs->max_count = atoi(argv[1]);
2192                 revs->no_walk = 0;
2193                 return 2;
2194         } else if (skip_prefix(arg, "-n", &optarg)) {
2195                 revs->max_count = atoi(optarg);
2196                 revs->no_walk = 0;
2197         } else if ((argcount = parse_long_opt("max-age", argv, &optarg))) {
2198                 revs->max_age = atoi(optarg);
2199                 return argcount;
2200         } else if ((argcount = parse_long_opt("since", argv, &optarg))) {
2201                 revs->max_age = approxidate(optarg);
2202                 return argcount;
2203         } else if ((argcount = parse_long_opt("after", argv, &optarg))) {
2204                 revs->max_age = approxidate(optarg);
2205                 return argcount;
2206         } else if ((argcount = parse_long_opt("min-age", argv, &optarg))) {
2207                 revs->min_age = atoi(optarg);
2208                 return argcount;
2209         } else if ((argcount = parse_long_opt("before", argv, &optarg))) {
2210                 revs->min_age = approxidate(optarg);
2211                 return argcount;
2212         } else if ((argcount = parse_long_opt("until", argv, &optarg))) {
2213                 revs->min_age = approxidate(optarg);
2214                 return argcount;
2215         } else if (!strcmp(arg, "--first-parent")) {
2216                 revs->first_parent_only = 1;
2217         } else if (!strcmp(arg, "--ancestry-path")) {
2218                 revs->ancestry_path = 1;
2219                 revs->simplify_history = 0;
2220                 revs->limited = 1;
2221         } else if (!strcmp(arg, "-g") || !strcmp(arg, "--walk-reflogs")) {
2222                 init_reflog_walk(&revs->reflog_info);
2223         } else if (!strcmp(arg, "--default")) {
2224                 if (argc <= 1)
2225                         return error("bad --default argument");
2226                 revs->def = argv[1];
2227                 return 2;
2228         } else if (!strcmp(arg, "--merge")) {
2229                 revs->show_merge = 1;
2230         } else if (!strcmp(arg, "--topo-order")) {
2231                 revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
2232                 revs->topo_order = 1;
2233         } else if (!strcmp(arg, "--simplify-merges")) {
2234                 revs->simplify_merges = 1;
2235                 revs->topo_order = 1;
2236                 revs->rewrite_parents = 1;
2237                 revs->simplify_history = 0;
2238                 revs->limited = 1;
2239         } else if (!strcmp(arg, "--simplify-by-decoration")) {
2240                 revs->simplify_merges = 1;
2241                 revs->topo_order = 1;
2242                 revs->rewrite_parents = 1;
2243                 revs->simplify_history = 0;
2244                 revs->simplify_by_decoration = 1;
2245                 revs->limited = 1;
2246                 revs->prune = 1;
2247         } else if (!strcmp(arg, "--date-order")) {
2248                 revs->sort_order = REV_SORT_BY_COMMIT_DATE;
2249                 revs->topo_order = 1;
2250         } else if (!strcmp(arg, "--author-date-order")) {
2251                 revs->sort_order = REV_SORT_BY_AUTHOR_DATE;
2252                 revs->topo_order = 1;
2253         } else if (!strcmp(arg, "--early-output")) {
2254                 revs->early_output = 100;
2255                 revs->topo_order = 1;
2256         } else if (skip_prefix(arg, "--early-output=", &optarg)) {
2257                 if (strtoul_ui(optarg, 10, &revs->early_output) < 0)
2258                         die("'%s': not a non-negative integer", optarg);
2259                 revs->topo_order = 1;
2260         } else if (!strcmp(arg, "--parents")) {
2261                 revs->rewrite_parents = 1;
2262                 revs->print_parents = 1;
2263         } else if (!strcmp(arg, "--dense")) {
2264                 revs->dense = 1;
2265         } else if (!strcmp(arg, "--sparse")) {
2266                 revs->dense = 0;
2267         } else if (!strcmp(arg, "--in-commit-order")) {
2268                 revs->tree_blobs_in_commit_order = 1;
2269         } else if (!strcmp(arg, "--remove-empty")) {
2270                 revs->remove_empty_trees = 1;
2271         } else if (!strcmp(arg, "--merges")) {
2272                 revs->min_parents = 2;
2273         } else if (!strcmp(arg, "--no-merges")) {
2274                 revs->max_parents = 1;
2275         } else if (skip_prefix(arg, "--min-parents=", &optarg)) {
2276                 revs->min_parents = atoi(optarg);
2277         } else if (!strcmp(arg, "--no-min-parents")) {
2278                 revs->min_parents = 0;
2279         } else if (skip_prefix(arg, "--max-parents=", &optarg)) {
2280                 revs->max_parents = atoi(optarg);
2281         } else if (!strcmp(arg, "--no-max-parents")) {
2282                 revs->max_parents = -1;
2283         } else if (!strcmp(arg, "--boundary")) {
2284                 revs->boundary = 1;
2285         } else if (!strcmp(arg, "--left-right")) {
2286                 revs->left_right = 1;
2287         } else if (!strcmp(arg, "--left-only")) {
2288                 if (revs->right_only)
2289                         die("--left-only is incompatible with --right-only"
2290                             " or --cherry");
2291                 revs->left_only = 1;
2292         } else if (!strcmp(arg, "--right-only")) {
2293                 if (revs->left_only)
2294                         die("--right-only is incompatible with --left-only");
2295                 revs->right_only = 1;
2296         } else if (!strcmp(arg, "--cherry")) {
2297                 if (revs->left_only)
2298                         die("--cherry is incompatible with --left-only");
2299                 revs->cherry_mark = 1;
2300                 revs->right_only = 1;
2301                 revs->max_parents = 1;
2302                 revs->limited = 1;
2303         } else if (!strcmp(arg, "--count")) {
2304                 revs->count = 1;
2305         } else if (!strcmp(arg, "--cherry-mark")) {
2306                 if (revs->cherry_pick)
2307                         die("--cherry-mark is incompatible with --cherry-pick");
2308                 revs->cherry_mark = 1;
2309                 revs->limited = 1; /* needs limit_list() */
2310         } else if (!strcmp(arg, "--cherry-pick")) {
2311                 if (revs->cherry_mark)
2312                         die("--cherry-pick is incompatible with --cherry-mark");
2313                 revs->cherry_pick = 1;
2314                 revs->limited = 1;
2315         } else if (!strcmp(arg, "--objects")) {
2316                 revs->tag_objects = 1;
2317                 revs->tree_objects = 1;
2318                 revs->blob_objects = 1;
2319         } else if (!strcmp(arg, "--objects-edge")) {
2320                 revs->tag_objects = 1;
2321                 revs->tree_objects = 1;
2322                 revs->blob_objects = 1;
2323                 revs->edge_hint = 1;
2324         } else if (!strcmp(arg, "--objects-edge-aggressive")) {
2325                 revs->tag_objects = 1;
2326                 revs->tree_objects = 1;
2327                 revs->blob_objects = 1;
2328                 revs->edge_hint = 1;
2329                 revs->edge_hint_aggressive = 1;
2330         } else if (!strcmp(arg, "--verify-objects")) {
2331                 revs->tag_objects = 1;
2332                 revs->tree_objects = 1;
2333                 revs->blob_objects = 1;
2334                 revs->verify_objects = 1;
2335         } else if (!strcmp(arg, "--unpacked")) {
2336                 revs->unpacked = 1;
2337         } else if (starts_with(arg, "--unpacked=")) {
2338                 die(_("--unpacked=<packfile> no longer supported"));
2339         } else if (!strcmp(arg, "-r")) {
2340                 revs->diff = 1;
2341                 revs->diffopt.flags.recursive = 1;
2342         } else if (!strcmp(arg, "-t")) {
2343                 revs->diff = 1;
2344                 revs->diffopt.flags.recursive = 1;
2345                 revs->diffopt.flags.tree_in_recursive = 1;
2346         } else if (!strcmp(arg, "-m")) {
2347                 /*
2348                  * To "diff-index", "-m" means "match missing", and to the "log"
2349                  * family of commands, it means "show full diff for merges". Set
2350                  * both fields appropriately.
2351                  */
2352                 revs->ignore_merges = 0;
2353                 revs->match_missing = 1;
2354         } else if ((argcount = parse_long_opt("diff-merges", argv, &optarg))) {
2355                 if (!strcmp(optarg, "off")) {
2356                         revs->ignore_merges = 1;
2357                 } else {
2358                         die(_("unknown value for --diff-merges: %s"), optarg);
2359                 }
2360                 return argcount;
2361         } else if (!strcmp(arg, "--no-diff-merges")) {
2362                 revs->ignore_merges = 1;
2363         } else if (!strcmp(arg, "-c")) {
2364                 revs->diff = 1;
2365                 revs->dense_combined_merges = 0;
2366                 revs->combine_merges = 1;
2367         } else if (!strcmp(arg, "--combined-all-paths")) {
2368                 revs->diff = 1;
2369                 revs->combined_all_paths = 1;
2370         } else if (!strcmp(arg, "--cc")) {
2371                 revs->diff = 1;
2372                 revs->dense_combined_merges = 1;
2373                 revs->combine_merges = 1;
2374         } else if (!strcmp(arg, "-v")) {
2375                 revs->verbose_header = 1;
2376         } else if (!strcmp(arg, "--pretty")) {
2377                 revs->verbose_header = 1;
2378                 revs->pretty_given = 1;
2379                 get_commit_format(NULL, revs);
2380         } else if (skip_prefix(arg, "--pretty=", &optarg) ||
2381                    skip_prefix(arg, "--format=", &optarg)) {
2382                 /*
2383                  * Detached form ("--pretty X" as opposed to "--pretty=X")
2384                  * not allowed, since the argument is optional.
2385                  */
2386                 revs->verbose_header = 1;
2387                 revs->pretty_given = 1;
2388                 get_commit_format(optarg, revs);
2389         } else if (!strcmp(arg, "--expand-tabs")) {
2390                 revs->expand_tabs_in_log = 8;
2391         } else if (!strcmp(arg, "--no-expand-tabs")) {
2392                 revs->expand_tabs_in_log = 0;
2393         } else if (skip_prefix(arg, "--expand-tabs=", &arg)) {
2394                 int val;
2395                 if (strtol_i(arg, 10, &val) < 0 || val < 0)
2396                         die("'%s': not a non-negative integer", arg);
2397                 revs->expand_tabs_in_log = val;
2398         } else if (!strcmp(arg, "--show-notes") || !strcmp(arg, "--notes")) {
2399                 enable_default_display_notes(&revs->notes_opt, &revs->show_notes);
2400                 revs->show_notes_given = 1;
2401         } else if (!strcmp(arg, "--show-signature")) {
2402                 revs->show_signature = 1;
2403         } else if (!strcmp(arg, "--no-show-signature")) {
2404                 revs->show_signature = 0;
2405         } else if (!strcmp(arg, "--show-linear-break")) {
2406                 revs->break_bar = "                    ..........";
2407                 revs->track_linear = 1;
2408                 revs->track_first_time = 1;
2409         } else if (skip_prefix(arg, "--show-linear-break=", &optarg)) {
2410                 revs->break_bar = xstrdup(optarg);
2411                 revs->track_linear = 1;
2412                 revs->track_first_time = 1;
2413         } else if (skip_prefix(arg, "--show-notes=", &optarg) ||
2414                    skip_prefix(arg, "--notes=", &optarg)) {
2415                 if (starts_with(arg, "--show-notes=") &&
2416                     revs->notes_opt.use_default_notes < 0)
2417                         revs->notes_opt.use_default_notes = 1;
2418                 enable_ref_display_notes(&revs->notes_opt, &revs->show_notes, optarg);
2419                 revs->show_notes_given = 1;
2420         } else if (!strcmp(arg, "--no-notes")) {
2421                 disable_display_notes(&revs->notes_opt, &revs->show_notes);
2422                 revs->show_notes_given = 1;
2423         } else if (!strcmp(arg, "--standard-notes")) {
2424                 revs->show_notes_given = 1;
2425                 revs->notes_opt.use_default_notes = 1;
2426         } else if (!strcmp(arg, "--no-standard-notes")) {
2427                 revs->notes_opt.use_default_notes = 0;
2428         } else if (!strcmp(arg, "--oneline")) {
2429                 revs->verbose_header = 1;
2430                 get_commit_format("oneline", revs);
2431                 revs->pretty_given = 1;
2432                 revs->abbrev_commit = 1;
2433         } else if (!strcmp(arg, "--graph")) {
2434                 revs->topo_order = 1;
2435                 revs->rewrite_parents = 1;
2436                 revs->graph = graph_init(revs);
2437         } else if (!strcmp(arg, "--encode-email-headers")) {
2438                 revs->encode_email_headers = 1;
2439         } else if (!strcmp(arg, "--no-encode-email-headers")) {
2440                 revs->encode_email_headers = 0;
2441         } else if (!strcmp(arg, "--root")) {
2442                 revs->show_root_diff = 1;
2443         } else if (!strcmp(arg, "--no-commit-id")) {
2444                 revs->no_commit_id = 1;
2445         } else if (!strcmp(arg, "--always")) {
2446                 revs->always_show_header = 1;
2447         } else if (!strcmp(arg, "--no-abbrev")) {
2448                 revs->abbrev = 0;
2449         } else if (!strcmp(arg, "--abbrev")) {
2450                 revs->abbrev = DEFAULT_ABBREV;
2451         } else if (skip_prefix(arg, "--abbrev=", &optarg)) {
2452                 revs->abbrev = strtoul(optarg, NULL, 10);
2453                 if (revs->abbrev < MINIMUM_ABBREV)
2454                         revs->abbrev = MINIMUM_ABBREV;
2455                 else if (revs->abbrev > hexsz)
2456                         revs->abbrev = hexsz;
2457         } else if (!strcmp(arg, "--abbrev-commit")) {
2458                 revs->abbrev_commit = 1;
2459                 revs->abbrev_commit_given = 1;
2460         } else if (!strcmp(arg, "--no-abbrev-commit")) {
2461                 revs->abbrev_commit = 0;
2462         } else if (!strcmp(arg, "--full-diff")) {
2463                 revs->diff = 1;
2464                 revs->full_diff = 1;
2465         } else if (!strcmp(arg, "--show-pulls")) {
2466                 revs->show_pulls = 1;
2467         } else if (!strcmp(arg, "--full-history")) {
2468                 revs->simplify_history = 0;
2469         } else if (!strcmp(arg, "--relative-date")) {
2470                 revs->date_mode.type = DATE_RELATIVE;
2471                 revs->date_mode_explicit = 1;
2472         } else if ((argcount = parse_long_opt("date", argv, &optarg))) {
2473                 parse_date_format(optarg, &revs->date_mode);
2474                 revs->date_mode_explicit = 1;
2475                 return argcount;
2476         } else if (!strcmp(arg, "--log-size")) {
2477                 revs->show_log_size = 1;
2478         }
2479         /*
2480          * Grepping the commit log
2481          */
2482         else if ((argcount = parse_long_opt("author", argv, &optarg))) {
2483                 add_header_grep(revs, GREP_HEADER_AUTHOR, optarg);
2484                 return argcount;
2485         } else if ((argcount = parse_long_opt("committer", argv, &optarg))) {
2486                 add_header_grep(revs, GREP_HEADER_COMMITTER, optarg);
2487                 return argcount;
2488         } else if ((argcount = parse_long_opt("grep-reflog", argv, &optarg))) {
2489                 add_header_grep(revs, GREP_HEADER_REFLOG, optarg);
2490                 return argcount;
2491         } else if ((argcount = parse_long_opt("grep", argv, &optarg))) {
2492                 add_message_grep(revs, optarg);
2493                 return argcount;
2494         } else if (!strcmp(arg, "--grep-debug")) {
2495                 revs->grep_filter.debug = 1;
2496         } else if (!strcmp(arg, "--basic-regexp")) {
2497                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_BRE;
2498         } else if (!strcmp(arg, "--extended-regexp") || !strcmp(arg, "-E")) {
2499                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_ERE;
2500         } else if (!strcmp(arg, "--regexp-ignore-case") || !strcmp(arg, "-i")) {
2501                 revs->grep_filter.ignore_case = 1;
2502                 revs->diffopt.pickaxe_opts |= DIFF_PICKAXE_IGNORE_CASE;
2503         } else if (!strcmp(arg, "--fixed-strings") || !strcmp(arg, "-F")) {
2504                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_FIXED;
2505         } else if (!strcmp(arg, "--perl-regexp") || !strcmp(arg, "-P")) {
2506                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_PCRE;
2507         } else if (!strcmp(arg, "--all-match")) {
2508                 revs->grep_filter.all_match = 1;
2509         } else if (!strcmp(arg, "--invert-grep")) {
2510                 revs->invert_grep = 1;
2511         } else if ((argcount = parse_long_opt("encoding", argv, &optarg))) {
2512                 if (strcmp(optarg, "none"))
2513                         git_log_output_encoding = xstrdup(optarg);
2514                 else
2515                         git_log_output_encoding = "";
2516                 return argcount;
2517         } else if (!strcmp(arg, "--reverse")) {
2518                 revs->reverse ^= 1;
2519         } else if (!strcmp(arg, "--children")) {
2520                 revs->children.name = "children";
2521                 revs->limited = 1;
2522         } else if (!strcmp(arg, "--ignore-missing")) {
2523                 revs->ignore_missing = 1;
2524         } else if (opt && opt->allow_exclude_promisor_objects &&
2525                    !strcmp(arg, "--exclude-promisor-objects")) {
2526                 if (fetch_if_missing)
2527                         BUG("exclude_promisor_objects can only be used when fetch_if_missing is 0");
2528                 revs->exclude_promisor_objects = 1;
2529         } else {
2530                 int opts = diff_opt_parse(&revs->diffopt, argv, argc, revs->prefix);
2531                 if (!opts)
2532                         unkv[(*unkc)++] = arg;
2533                 return opts;
2534         }
2535         if (revs->graph && revs->track_linear)
2536                 die("--show-linear-break and --graph are incompatible");
2537
2538         return 1;
2539 }
2540
2541 void parse_revision_opt(struct rev_info *revs, struct parse_opt_ctx_t *ctx,
2542                         const struct option *options,
2543                         const char * const usagestr[])
2544 {
2545         int n = handle_revision_opt(revs, ctx->argc, ctx->argv,
2546                                     &ctx->cpidx, ctx->out, NULL);
2547         if (n <= 0) {
2548                 error("unknown option `%s'", ctx->argv[0]);
2549                 usage_with_options(usagestr, options);
2550         }
2551         ctx->argv += n;
2552         ctx->argc -= n;
2553 }
2554
2555 static int for_each_bisect_ref(struct ref_store *refs, each_ref_fn fn,
2556                                void *cb_data, const char *term)
2557 {
2558         struct strbuf bisect_refs = STRBUF_INIT;
2559         int status;
2560         strbuf_addf(&bisect_refs, "refs/bisect/%s", term);
2561         status = refs_for_each_fullref_in(refs, bisect_refs.buf, fn, cb_data, 0);
2562         strbuf_release(&bisect_refs);
2563         return status;
2564 }
2565
2566 static int for_each_bad_bisect_ref(struct ref_store *refs, each_ref_fn fn, void *cb_data)
2567 {
2568         return for_each_bisect_ref(refs, fn, cb_data, term_bad);
2569 }
2570
2571 static int for_each_good_bisect_ref(struct ref_store *refs, each_ref_fn fn, void *cb_data)
2572 {
2573         return for_each_bisect_ref(refs, fn, cb_data, term_good);
2574 }
2575
2576 static int handle_revision_pseudo_opt(const char *submodule,
2577                                       struct rev_info *revs,
2578                                       const char **argv, int *flags)
2579 {
2580         const char *arg = argv[0];
2581         const char *optarg;
2582         struct ref_store *refs;
2583         int argcount;
2584
2585         if (submodule) {
2586                 /*
2587                  * We need some something like get_submodule_worktrees()
2588                  * before we can go through all worktrees of a submodule,
2589                  * .e.g with adding all HEADs from --all, which is not
2590                  * supported right now, so stick to single worktree.
2591                  */
2592                 if (!revs->single_worktree)
2593                         BUG("--single-worktree cannot be used together with submodule");
2594                 refs = get_submodule_ref_store(submodule);
2595         } else
2596                 refs = get_main_ref_store(revs->repo);
2597
2598         /*
2599          * NOTE!
2600          *
2601          * Commands like "git shortlog" will not accept the options below
2602          * unless parse_revision_opt queues them (as opposed to erroring
2603          * out).
2604          *
2605          * When implementing your new pseudo-option, remember to
2606          * register it in the list at the top of handle_revision_opt.
2607          */
2608         if (!strcmp(arg, "--all")) {
2609                 handle_refs(refs, revs, *flags, refs_for_each_ref);
2610                 handle_refs(refs, revs, *flags, refs_head_ref);
2611                 if (!revs->single_worktree) {
2612                         struct all_refs_cb cb;
2613
2614                         init_all_refs_cb(&cb, revs, *flags);
2615                         other_head_refs(handle_one_ref, &cb);
2616                 }
2617                 clear_ref_exclusion(&revs->ref_excludes);
2618         } else if (!strcmp(arg, "--branches")) {
2619                 handle_refs(refs, revs, *flags, refs_for_each_branch_ref);
2620                 clear_ref_exclusion(&revs->ref_excludes);
2621         } else if (!strcmp(arg, "--bisect")) {
2622                 read_bisect_terms(&term_bad, &term_good);
2623                 handle_refs(refs, revs, *flags, for_each_bad_bisect_ref);
2624                 handle_refs(refs, revs, *flags ^ (UNINTERESTING | BOTTOM),
2625                             for_each_good_bisect_ref);
2626                 revs->bisect = 1;
2627         } else if (!strcmp(arg, "--tags")) {
2628                 handle_refs(refs, revs, *flags, refs_for_each_tag_ref);
2629                 clear_ref_exclusion(&revs->ref_excludes);
2630         } else if (!strcmp(arg, "--remotes")) {
2631                 handle_refs(refs, revs, *flags, refs_for_each_remote_ref);
2632                 clear_ref_exclusion(&revs->ref_excludes);
2633         } else if ((argcount = parse_long_opt("glob", argv, &optarg))) {
2634                 struct all_refs_cb cb;
2635                 init_all_refs_cb(&cb, revs, *flags);
2636                 for_each_glob_ref(handle_one_ref, optarg, &cb);
2637                 clear_ref_exclusion(&revs->ref_excludes);
2638                 return argcount;
2639         } else if ((argcount = parse_long_opt("exclude", argv, &optarg))) {
2640                 add_ref_exclusion(&revs->ref_excludes, optarg);
2641                 return argcount;
2642         } else if (skip_prefix(arg, "--branches=", &optarg)) {
2643                 struct all_refs_cb cb;
2644                 init_all_refs_cb(&cb, revs, *flags);
2645                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/heads/", &cb);
2646                 clear_ref_exclusion(&revs->ref_excludes);
2647         } else if (skip_prefix(arg, "--tags=", &optarg)) {
2648                 struct all_refs_cb cb;
2649                 init_all_refs_cb(&cb, revs, *flags);
2650                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/tags/", &cb);
2651                 clear_ref_exclusion(&revs->ref_excludes);
2652         } else if (skip_prefix(arg, "--remotes=", &optarg)) {
2653                 struct all_refs_cb cb;
2654                 init_all_refs_cb(&cb, revs, *flags);
2655                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/remotes/", &cb);
2656                 clear_ref_exclusion(&revs->ref_excludes);
2657         } else if (!strcmp(arg, "--reflog")) {
2658                 add_reflogs_to_pending(revs, *flags);
2659         } else if (!strcmp(arg, "--indexed-objects")) {
2660                 add_index_objects_to_pending(revs, *flags);
2661         } else if (!strcmp(arg, "--alternate-refs")) {
2662                 add_alternate_refs_to_pending(revs, *flags);
2663         } else if (!strcmp(arg, "--not")) {
2664                 *flags ^= UNINTERESTING | BOTTOM;
2665         } else if (!strcmp(arg, "--no-walk")) {
2666                 revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2667         } else if (skip_prefix(arg, "--no-walk=", &optarg)) {
2668                 /*
2669                  * Detached form ("--no-walk X" as opposed to "--no-walk=X")
2670                  * not allowed, since the argument is optional.
2671                  */
2672                 if (!strcmp(optarg, "sorted"))
2673                         revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2674                 else if (!strcmp(optarg, "unsorted"))
2675                         revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
2676                 else
2677                         return error("invalid argument to --no-walk");
2678         } else if (!strcmp(arg, "--do-walk")) {
2679                 revs->no_walk = 0;
2680         } else if (!strcmp(arg, "--single-worktree")) {
2681                 revs->single_worktree = 1;
2682         } else {
2683                 return 0;
2684         }
2685
2686         return 1;
2687 }
2688
2689 static void NORETURN diagnose_missing_default(const char *def)
2690 {
2691         int flags;
2692         const char *refname;
2693
2694         refname = resolve_ref_unsafe(def, 0, NULL, &flags);
2695         if (!refname || !(flags & REF_ISSYMREF) || (flags & REF_ISBROKEN))
2696                 die(_("your current branch appears to be broken"));
2697
2698         skip_prefix(refname, "refs/heads/", &refname);
2699         die(_("your current branch '%s' does not have any commits yet"),
2700             refname);
2701 }
2702
2703 /*
2704  * Parse revision information, filling in the "rev_info" structure,
2705  * and removing the used arguments from the argument list.
2706  *
2707  * Returns the number of arguments left that weren't recognized
2708  * (which are also moved to the head of the argument list)
2709  */
2710 int setup_revisions(int argc, const char **argv, struct rev_info *revs, struct setup_revision_opt *opt)
2711 {
2712         int i, flags, left, seen_dashdash, revarg_opt;
2713         struct strvec prune_data = STRVEC_INIT;
2714         const char *submodule = NULL;
2715         int seen_end_of_options = 0;
2716
2717         if (opt)
2718                 submodule = opt->submodule;
2719
2720         /* First, search for "--" */
2721         if (opt && opt->assume_dashdash) {
2722                 seen_dashdash = 1;
2723         } else {
2724                 seen_dashdash = 0;
2725                 for (i = 1; i < argc; i++) {
2726                         const char *arg = argv[i];
2727                         if (strcmp(arg, "--"))
2728                                 continue;
2729                         argv[i] = NULL;
2730                         argc = i;
2731                         if (argv[i + 1])
2732                                 strvec_pushv(&prune_data, argv + i + 1);
2733                         seen_dashdash = 1;
2734                         break;
2735                 }
2736         }
2737
2738         /* Second, deal with arguments and options */
2739         flags = 0;
2740         revarg_opt = opt ? opt->revarg_opt : 0;
2741         if (seen_dashdash)
2742                 revarg_opt |= REVARG_CANNOT_BE_FILENAME;
2743         for (left = i = 1; i < argc; i++) {
2744                 const char *arg = argv[i];
2745                 if (!seen_end_of_options && *arg == '-') {
2746                         int opts;
2747
2748                         opts = handle_revision_pseudo_opt(submodule,
2749                                                 revs, argv + i,
2750                                                 &flags);
2751                         if (opts > 0) {
2752                                 i += opts - 1;
2753                                 continue;
2754                         }
2755
2756                         if (!strcmp(arg, "--stdin")) {
2757                                 if (revs->disable_stdin) {
2758                                         argv[left++] = arg;
2759                                         continue;
2760                                 }
2761                                 if (revs->read_from_stdin++)
2762                                         die("--stdin given twice?");
2763                                 read_revisions_from_stdin(revs, &prune_data);
2764                                 continue;
2765                         }
2766
2767                         if (!strcmp(arg, "--end-of-options")) {
2768                                 seen_end_of_options = 1;
2769                                 continue;
2770                         }
2771
2772                         opts = handle_revision_opt(revs, argc - i, argv + i,
2773                                                    &left, argv, opt);
2774                         if (opts > 0) {
2775                                 i += opts - 1;
2776                                 continue;
2777                         }
2778                         if (opts < 0)
2779                                 exit(128);
2780                         continue;
2781                 }
2782
2783
2784                 if (handle_revision_arg(arg, revs, flags, revarg_opt)) {
2785                         int j;
2786                         if (seen_dashdash || *arg == '^')
2787                                 die("bad revision '%s'", arg);
2788
2789                         /* If we didn't have a "--":
2790                          * (1) all filenames must exist;
2791                          * (2) all rev-args must not be interpretable
2792                          *     as a valid filename.
2793                          * but the latter we have checked in the main loop.
2794                          */
2795                         for (j = i; j < argc; j++)
2796                                 verify_filename(revs->prefix, argv[j], j == i);
2797
2798                         strvec_pushv(&prune_data, argv + i);
2799                         break;
2800                 }
2801         }
2802
2803         if (prune_data.nr) {
2804                 /*
2805                  * If we need to introduce the magic "a lone ':' means no
2806                  * pathspec whatsoever", here is the place to do so.
2807                  *
2808                  * if (prune_data.nr == 1 && !strcmp(prune_data[0], ":")) {
2809                  *      prune_data.nr = 0;
2810                  *      prune_data.alloc = 0;
2811                  *      free(prune_data.path);
2812                  *      prune_data.path = NULL;
2813                  * } else {
2814                  *      terminate prune_data.alloc with NULL and
2815                  *      call init_pathspec() to set revs->prune_data here.
2816                  * }
2817                  */
2818                 parse_pathspec(&revs->prune_data, 0, 0,
2819                                revs->prefix, prune_data.v);
2820         }
2821         strvec_clear(&prune_data);
2822
2823         if (revs->def == NULL)
2824                 revs->def = opt ? opt->def : NULL;
2825         if (opt && opt->tweak)
2826                 opt->tweak(revs, opt);
2827         if (revs->show_merge)
2828                 prepare_show_merge(revs);
2829         if (revs->def && !revs->pending.nr && !revs->rev_input_given) {
2830                 struct object_id oid;
2831                 struct object *object;
2832                 struct object_context oc;
2833                 if (get_oid_with_context(revs->repo, revs->def, 0, &oid, &oc))
2834                         diagnose_missing_default(revs->def);
2835                 object = get_reference(revs, revs->def, &oid, 0);
2836                 add_pending_object_with_mode(revs, object, revs->def, oc.mode);
2837         }
2838
2839         /* Did the user ask for any diff output? Run the diff! */
2840         if (revs->diffopt.output_format & ~DIFF_FORMAT_NO_OUTPUT)
2841                 revs->diff = 1;
2842
2843         /* Pickaxe, diff-filter and rename following need diffs */
2844         if ((revs->diffopt.pickaxe_opts & DIFF_PICKAXE_KINDS_MASK) ||
2845             revs->diffopt.filter ||
2846             revs->diffopt.flags.follow_renames)
2847                 revs->diff = 1;
2848
2849         if (revs->diffopt.objfind)
2850                 revs->simplify_history = 0;
2851
2852         if (revs->line_level_traverse) {
2853                 if (want_ancestry(revs))
2854                         revs->limited = 1;
2855                 revs->topo_order = 1;
2856         }
2857
2858         if (revs->topo_order && !generation_numbers_enabled(the_repository))
2859                 revs->limited = 1;
2860
2861         if (revs->prune_data.nr) {
2862                 copy_pathspec(&revs->pruning.pathspec, &revs->prune_data);
2863                 /* Can't prune commits with rename following: the paths change.. */
2864                 if (!revs->diffopt.flags.follow_renames)
2865                         revs->prune = 1;
2866                 if (!revs->full_diff)
2867                         copy_pathspec(&revs->diffopt.pathspec,
2868                                       &revs->prune_data);
2869         }
2870         if (revs->combine_merges && revs->ignore_merges < 0)
2871                 revs->ignore_merges = 0;
2872         if (revs->ignore_merges < 0)
2873                 revs->ignore_merges = 1;
2874         if (revs->combined_all_paths && !revs->combine_merges)
2875                 die("--combined-all-paths makes no sense without -c or --cc");
2876
2877         revs->diffopt.abbrev = revs->abbrev;
2878
2879         diff_setup_done(&revs->diffopt);
2880
2881         grep_commit_pattern_type(GREP_PATTERN_TYPE_UNSPECIFIED,
2882                                  &revs->grep_filter);
2883         if (!is_encoding_utf8(get_log_output_encoding()))
2884                 revs->grep_filter.ignore_locale = 1;
2885         compile_grep_patterns(&revs->grep_filter);
2886
2887         if (revs->reverse && revs->reflog_info)
2888                 die("cannot combine --reverse with --walk-reflogs");
2889         if (revs->reflog_info && revs->limited)
2890                 die("cannot combine --walk-reflogs with history-limiting options");
2891         if (revs->rewrite_parents && revs->children.name)
2892                 die("cannot combine --parents and --children");
2893
2894         /*
2895          * Limitations on the graph functionality
2896          */
2897         if (revs->reverse && revs->graph)
2898                 die("cannot combine --reverse with --graph");
2899
2900         if (revs->reflog_info && revs->graph)
2901                 die("cannot combine --walk-reflogs with --graph");
2902         if (revs->no_walk && revs->graph)
2903                 die("cannot combine --no-walk with --graph");
2904         if (!revs->reflog_info && revs->grep_filter.use_reflog_filter)
2905                 die("cannot use --grep-reflog without --walk-reflogs");
2906
2907         if (revs->line_level_traverse &&
2908             (revs->diffopt.output_format & ~(DIFF_FORMAT_PATCH | DIFF_FORMAT_NO_OUTPUT)))
2909                 die(_("-L does not yet support diff formats besides -p and -s"));
2910
2911         if (revs->expand_tabs_in_log < 0)
2912                 revs->expand_tabs_in_log = revs->expand_tabs_in_log_default;
2913
2914         return left;
2915 }
2916
2917 static void add_child(struct rev_info *revs, struct commit *parent, struct commit *child)
2918 {
2919         struct commit_list *l = xcalloc(1, sizeof(*l));
2920
2921         l->item = child;
2922         l->next = add_decoration(&revs->children, &parent->object, l);
2923 }
2924
2925 static int remove_duplicate_parents(struct rev_info *revs, struct commit *commit)
2926 {
2927         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2928         struct commit_list **pp, *p;
2929         int surviving_parents;
2930
2931         /* Examine existing parents while marking ones we have seen... */
2932         pp = &commit->parents;
2933         surviving_parents = 0;
2934         while ((p = *pp) != NULL) {
2935                 struct commit *parent = p->item;
2936                 if (parent->object.flags & TMP_MARK) {
2937                         *pp = p->next;
2938                         if (ts)
2939                                 compact_treesame(revs, commit, surviving_parents);
2940                         continue;
2941                 }
2942                 parent->object.flags |= TMP_MARK;
2943                 surviving_parents++;
2944                 pp = &p->next;
2945         }
2946         /* clear the temporary mark */
2947         for (p = commit->parents; p; p = p->next) {
2948                 p->item->object.flags &= ~TMP_MARK;
2949         }
2950         /* no update_treesame() - removing duplicates can't affect TREESAME */
2951         return surviving_parents;
2952 }
2953
2954 struct merge_simplify_state {
2955         struct commit *simplified;
2956 };
2957
2958 static struct merge_simplify_state *locate_simplify_state(struct rev_info *revs, struct commit *commit)
2959 {
2960         struct merge_simplify_state *st;
2961
2962         st = lookup_decoration(&revs->merge_simplification, &commit->object);
2963         if (!st) {
2964                 st = xcalloc(1, sizeof(*st));
2965                 add_decoration(&revs->merge_simplification, &commit->object, st);
2966         }
2967         return st;
2968 }
2969
2970 static int mark_redundant_parents(struct commit *commit)
2971 {
2972         struct commit_list *h = reduce_heads(commit->parents);
2973         int i = 0, marked = 0;
2974         struct commit_list *po, *pn;
2975
2976         /* Want these for sanity-checking only */
2977         int orig_cnt = commit_list_count(commit->parents);
2978         int cnt = commit_list_count(h);
2979
2980         /*
2981          * Not ready to remove items yet, just mark them for now, based
2982          * on the output of reduce_heads(). reduce_heads outputs the reduced
2983          * set in its original order, so this isn't too hard.
2984          */
2985         po = commit->parents;
2986         pn = h;
2987         while (po) {
2988                 if (pn && po->item == pn->item) {
2989                         pn = pn->next;
2990                         i++;
2991                 } else {
2992                         po->item->object.flags |= TMP_MARK;
2993                         marked++;
2994                 }
2995                 po=po->next;
2996         }
2997
2998         if (i != cnt || cnt+marked != orig_cnt)
2999                 die("mark_redundant_parents %d %d %d %d", orig_cnt, cnt, i, marked);
3000
3001         free_commit_list(h);
3002
3003         return marked;
3004 }
3005
3006 static int mark_treesame_root_parents(struct commit *commit)
3007 {
3008         struct commit_list *p;
3009         int marked = 0;
3010
3011         for (p = commit->parents; p; p = p->next) {
3012                 struct commit *parent = p->item;
3013                 if (!parent->parents && (parent->object.flags & TREESAME)) {
3014                         parent->object.flags |= TMP_MARK;
3015                         marked++;
3016                 }
3017         }
3018
3019         return marked;
3020 }
3021
3022 /*
3023  * Awkward naming - this means one parent we are TREESAME to.
3024  * cf mark_treesame_root_parents: root parents that are TREESAME (to an
3025  * empty tree). Better name suggestions?
3026  */
3027 static int leave_one_treesame_to_parent(struct rev_info *revs, struct commit *commit)
3028 {
3029         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
3030         struct commit *unmarked = NULL, *marked = NULL;
3031         struct commit_list *p;
3032         unsigned n;
3033
3034         for (p = commit->parents, n = 0; p; p = p->next, n++) {
3035                 if (ts->treesame[n]) {
3036                         if (p->item->object.flags & TMP_MARK) {
3037                                 if (!marked)
3038                                         marked = p->item;
3039                         } else {
3040                                 if (!unmarked) {
3041                                         unmarked = p->item;
3042                                         break;
3043                                 }
3044                         }
3045                 }
3046         }
3047
3048         /*
3049          * If we are TREESAME to a marked-for-deletion parent, but not to any
3050          * unmarked parents, unmark the first TREESAME parent. This is the
3051          * parent that the default simplify_history==1 scan would have followed,
3052          * and it doesn't make sense to omit that path when asking for a
3053          * simplified full history. Retaining it improves the chances of
3054          * understanding odd missed merges that took an old version of a file.
3055          *
3056          * Example:
3057          *
3058          *   I--------*X       A modified the file, but mainline merge X used
3059          *    \       /        "-s ours", so took the version from I. X is
3060          *     `-*A--'         TREESAME to I and !TREESAME to A.
3061          *
3062          * Default log from X would produce "I". Without this check,
3063          * --full-history --simplify-merges would produce "I-A-X", showing
3064          * the merge commit X and that it changed A, but not making clear that
3065          * it had just taken the I version. With this check, the topology above
3066          * is retained.
3067          *
3068          * Note that it is possible that the simplification chooses a different
3069          * TREESAME parent from the default, in which case this test doesn't
3070          * activate, and we _do_ drop the default parent. Example:
3071          *
3072          *   I------X         A modified the file, but it was reverted in B,
3073          *    \    /          meaning mainline merge X is TREESAME to both
3074          *    *A-*B           parents.
3075          *
3076          * Default log would produce "I" by following the first parent;
3077          * --full-history --simplify-merges will produce "I-A-B". But this is a
3078          * reasonable result - it presents a logical full history leading from
3079          * I to X, and X is not an important merge.
3080          */
3081         if (!unmarked && marked) {
3082                 marked->object.flags &= ~TMP_MARK;
3083                 return 1;
3084         }
3085
3086         return 0;
3087 }
3088
3089 static int remove_marked_parents(struct rev_info *revs, struct commit *commit)
3090 {
3091         struct commit_list **pp, *p;
3092         int nth_parent, removed = 0;
3093
3094         pp = &commit->parents;
3095         nth_parent = 0;
3096         while ((p = *pp) != NULL) {
3097                 struct commit *parent = p->item;
3098                 if (parent->object.flags & TMP_MARK) {
3099                         parent->object.flags &= ~TMP_MARK;
3100                         *pp = p->next;
3101                         free(p);
3102                         removed++;
3103                         compact_treesame(revs, commit, nth_parent);
3104                         continue;
3105                 }
3106                 pp = &p->next;
3107                 nth_parent++;
3108         }
3109
3110         /* Removing parents can only increase TREESAMEness */
3111         if (removed && !(commit->object.flags & TREESAME))
3112                 update_treesame(revs, commit);
3113
3114         return nth_parent;
3115 }
3116
3117 static struct commit_list **simplify_one(struct rev_info *revs, struct commit *commit, struct commit_list **tail)
3118 {
3119         struct commit_list *p;
3120         struct commit *parent;
3121         struct merge_simplify_state *st, *pst;
3122         int cnt;
3123
3124         st = locate_simplify_state(revs, commit);
3125
3126         /*
3127          * Have we handled this one?
3128          */
3129         if (st->simplified)
3130                 return tail;
3131
3132         /*
3133          * An UNINTERESTING commit simplifies to itself, so does a
3134          * root commit.  We do not rewrite parents of such commit
3135          * anyway.
3136          */
3137         if ((commit->object.flags & UNINTERESTING) || !commit->parents) {
3138                 st->simplified = commit;
3139                 return tail;
3140         }
3141
3142         /*
3143          * Do we know what commit all of our parents that matter
3144          * should be rewritten to?  Otherwise we are not ready to
3145          * rewrite this one yet.
3146          */
3147         for (cnt = 0, p = commit->parents; p; p = p->next) {
3148                 pst = locate_simplify_state(revs, p->item);
3149                 if (!pst->simplified) {
3150                         tail = &commit_list_insert(p->item, tail)->next;
3151                         cnt++;
3152                 }
3153                 if (revs->first_parent_only)
3154                         break;
3155         }
3156         if (cnt) {
3157                 tail = &commit_list_insert(commit, tail)->next;
3158                 return tail;
3159         }
3160
3161         /*
3162          * Rewrite our list of parents. Note that this cannot
3163          * affect our TREESAME flags in any way - a commit is
3164          * always TREESAME to its simplification.
3165          */
3166         for (p = commit->parents; p; p = p->next) {
3167                 pst = locate_simplify_state(revs, p->item);
3168                 p->item = pst->simplified;
3169                 if (revs->first_parent_only)
3170                         break;
3171         }
3172
3173         if (revs->first_parent_only)
3174                 cnt = 1;
3175         else
3176                 cnt = remove_duplicate_parents(revs, commit);
3177
3178         /*
3179          * It is possible that we are a merge and one side branch
3180          * does not have any commit that touches the given paths;
3181          * in such a case, the immediate parent from that branch
3182          * will be rewritten to be the merge base.
3183          *
3184          *      o----X          X: the commit we are looking at;
3185          *     /    /           o: a commit that touches the paths;
3186          * ---o----'
3187          *
3188          * Further, a merge of an independent branch that doesn't
3189          * touch the path will reduce to a treesame root parent:
3190          *
3191          *  ----o----X          X: the commit we are looking at;
3192          *          /           o: a commit that touches the paths;
3193          *         r            r: a root commit not touching the paths
3194          *
3195          * Detect and simplify both cases.
3196          */
3197         if (1 < cnt) {
3198                 int marked = mark_redundant_parents(commit);
3199                 marked += mark_treesame_root_parents(commit);
3200                 if (marked)
3201                         marked -= leave_one_treesame_to_parent(revs, commit);
3202                 if (marked)
3203                         cnt = remove_marked_parents(revs, commit);
3204         }
3205
3206         /*
3207          * A commit simplifies to itself if it is a root, if it is
3208          * UNINTERESTING, if it touches the given paths, or if it is a
3209          * merge and its parents don't simplify to one relevant commit
3210          * (the first two cases are already handled at the beginning of
3211          * this function).
3212          *
3213          * Otherwise, it simplifies to what its sole relevant parent
3214          * simplifies to.
3215          */
3216         if (!cnt ||
3217             (commit->object.flags & UNINTERESTING) ||
3218             !(commit->object.flags & TREESAME) ||
3219             (parent = one_relevant_parent(revs, commit->parents)) == NULL ||
3220             (revs->show_pulls && (commit->object.flags & PULL_MERGE)))
3221                 st->simplified = commit;
3222         else {
3223                 pst = locate_simplify_state(revs, parent);
3224                 st->simplified = pst->simplified;
3225         }
3226         return tail;
3227 }
3228
3229 static void simplify_merges(struct rev_info *revs)
3230 {
3231         struct commit_list *list, *next;
3232         struct commit_list *yet_to_do, **tail;
3233         struct commit *commit;
3234
3235         if (!revs->prune)
3236                 return;
3237
3238         /* feed the list reversed */
3239         yet_to_do = NULL;
3240         for (list = revs->commits; list; list = next) {
3241                 commit = list->item;
3242                 next = list->next;
3243                 /*
3244                  * Do not free(list) here yet; the original list
3245                  * is used later in this function.
3246                  */
3247                 commit_list_insert(commit, &yet_to_do);
3248         }
3249         while (yet_to_do) {
3250                 list = yet_to_do;
3251                 yet_to_do = NULL;
3252                 tail = &yet_to_do;
3253                 while (list) {
3254                         commit = pop_commit(&list);
3255                         tail = simplify_one(revs, commit, tail);
3256                 }
3257         }
3258
3259         /* clean up the result, removing the simplified ones */
3260         list = revs->commits;
3261         revs->commits = NULL;
3262         tail = &revs->commits;
3263         while (list) {
3264                 struct merge_simplify_state *st;
3265
3266                 commit = pop_commit(&list);
3267                 st = locate_simplify_state(revs, commit);
3268                 if (st->simplified == commit)
3269                         tail = &commit_list_insert(commit, tail)->next;
3270         }
3271 }
3272
3273 static void set_children(struct rev_info *revs)
3274 {
3275         struct commit_list *l;
3276         for (l = revs->commits; l; l = l->next) {
3277                 struct commit *commit = l->item;
3278                 struct commit_list *p;
3279
3280                 for (p = commit->parents; p; p = p->next)
3281                         add_child(revs, p->item, commit);
3282         }
3283 }
3284
3285 void reset_revision_walk(void)
3286 {
3287         clear_object_flags(SEEN | ADDED | SHOWN | TOPO_WALK_EXPLORED | TOPO_WALK_INDEGREE);
3288 }
3289
3290 static int mark_uninteresting(const struct object_id *oid,
3291                               struct packed_git *pack,
3292                               uint32_t pos,
3293                               void *cb)
3294 {
3295         struct rev_info *revs = cb;
3296         struct object *o = parse_object(revs->repo, oid);
3297         o->flags |= UNINTERESTING | SEEN;
3298         return 0;
3299 }
3300
3301 define_commit_slab(indegree_slab, int);
3302 define_commit_slab(author_date_slab, timestamp_t);
3303
3304 struct topo_walk_info {
3305         uint32_t min_generation;
3306         struct prio_queue explore_queue;
3307         struct prio_queue indegree_queue;
3308         struct prio_queue topo_queue;
3309         struct indegree_slab indegree;
3310         struct author_date_slab author_date;
3311 };
3312
3313 static int topo_walk_atexit_registered;
3314 static unsigned int count_explore_walked;
3315 static unsigned int count_indegree_walked;
3316 static unsigned int count_topo_walked;
3317
3318 static void trace2_topo_walk_statistics_atexit(void)
3319 {
3320         struct json_writer jw = JSON_WRITER_INIT;
3321
3322         jw_object_begin(&jw, 0);
3323         jw_object_intmax(&jw, "count_explore_walked", count_explore_walked);
3324         jw_object_intmax(&jw, "count_indegree_walked", count_indegree_walked);
3325         jw_object_intmax(&jw, "count_topo_walked", count_topo_walked);
3326         jw_end(&jw);
3327
3328         trace2_data_json("topo_walk", the_repository, "statistics", &jw);
3329
3330         jw_release(&jw);
3331 }
3332
3333 static inline void test_flag_and_insert(struct prio_queue *q, struct commit *c, int flag)
3334 {
3335         if (c->object.flags & flag)
3336                 return;
3337
3338         c->object.flags |= flag;
3339         prio_queue_put(q, c);
3340 }
3341
3342 static void explore_walk_step(struct rev_info *revs)
3343 {
3344         struct topo_walk_info *info = revs->topo_walk_info;
3345         struct commit_list *p;
3346         struct commit *c = prio_queue_get(&info->explore_queue);
3347
3348         if (!c)
3349                 return;
3350
3351         if (repo_parse_commit_gently(revs->repo, c, 1) < 0)
3352                 return;
3353
3354         count_explore_walked++;
3355
3356         if (revs->sort_order == REV_SORT_BY_AUTHOR_DATE)
3357                 record_author_date(&info->author_date, c);
3358
3359         if (revs->max_age != -1 && (c->date < revs->max_age))
3360                 c->object.flags |= UNINTERESTING;
3361
3362         if (process_parents(revs, c, NULL, NULL) < 0)
3363                 return;
3364
3365         if (c->object.flags & UNINTERESTING)
3366                 mark_parents_uninteresting(c);
3367
3368         for (p = c->parents; p; p = p->next)
3369                 test_flag_and_insert(&info->explore_queue, p->item, TOPO_WALK_EXPLORED);
3370 }
3371
3372 static void explore_to_depth(struct rev_info *revs,
3373                              uint32_t gen_cutoff)
3374 {
3375         struct topo_walk_info *info = revs->topo_walk_info;
3376         struct commit *c;
3377         while ((c = prio_queue_peek(&info->explore_queue)) &&
3378                commit_graph_generation(c) >= gen_cutoff)
3379                 explore_walk_step(revs);
3380 }
3381
3382 static void indegree_walk_step(struct rev_info *revs)
3383 {
3384         struct commit_list *p;
3385         struct topo_walk_info *info = revs->topo_walk_info;
3386         struct commit *c = prio_queue_get(&info->indegree_queue);
3387
3388         if (!c)
3389                 return;
3390
3391         if (repo_parse_commit_gently(revs->repo, c, 1) < 0)
3392                 return;
3393
3394         count_indegree_walked++;
3395
3396         explore_to_depth(revs, commit_graph_generation(c));
3397
3398         for (p = c->parents; p; p = p->next) {
3399                 struct commit *parent = p->item;
3400                 int *pi = indegree_slab_at(&info->indegree, parent);
3401
3402                 if (*pi)
3403                         (*pi)++;
3404                 else
3405                         *pi = 2;
3406
3407                 test_flag_and_insert(&info->indegree_queue, parent, TOPO_WALK_INDEGREE);
3408
3409                 if (revs->first_parent_only)
3410                         return;
3411         }
3412 }
3413
3414 static void compute_indegrees_to_depth(struct rev_info *revs,
3415                                        uint32_t gen_cutoff)
3416 {
3417         struct topo_walk_info *info = revs->topo_walk_info;
3418         struct commit *c;
3419         while ((c = prio_queue_peek(&info->indegree_queue)) &&
3420                commit_graph_generation(c) >= gen_cutoff)
3421                 indegree_walk_step(revs);
3422 }
3423
3424 static void reset_topo_walk(struct rev_info *revs)
3425 {
3426         struct topo_walk_info *info = revs->topo_walk_info;
3427
3428         clear_prio_queue(&info->explore_queue);
3429         clear_prio_queue(&info->indegree_queue);
3430         clear_prio_queue(&info->topo_queue);
3431         clear_indegree_slab(&info->indegree);
3432         clear_author_date_slab(&info->author_date);
3433
3434         FREE_AND_NULL(revs->topo_walk_info);
3435 }
3436
3437 static void init_topo_walk(struct rev_info *revs)
3438 {
3439         struct topo_walk_info *info;
3440         struct commit_list *list;
3441         if (revs->topo_walk_info)
3442                 reset_topo_walk(revs);
3443
3444         revs->topo_walk_info = xmalloc(sizeof(struct topo_walk_info));
3445         info = revs->topo_walk_info;
3446         memset(info, 0, sizeof(struct topo_walk_info));
3447
3448         init_indegree_slab(&info->indegree);
3449         memset(&info->explore_queue, 0, sizeof(info->explore_queue));
3450         memset(&info->indegree_queue, 0, sizeof(info->indegree_queue));
3451         memset(&info->topo_queue, 0, sizeof(info->topo_queue));
3452
3453         switch (revs->sort_order) {
3454         default: /* REV_SORT_IN_GRAPH_ORDER */
3455                 info->topo_queue.compare = NULL;
3456                 break;
3457         case REV_SORT_BY_COMMIT_DATE:
3458                 info->topo_queue.compare = compare_commits_by_commit_date;
3459                 break;
3460         case REV_SORT_BY_AUTHOR_DATE:
3461                 init_author_date_slab(&info->author_date);
3462                 info->topo_queue.compare = compare_commits_by_author_date;
3463                 info->topo_queue.cb_data = &info->author_date;
3464                 break;
3465         }
3466
3467         info->explore_queue.compare = compare_commits_by_gen_then_commit_date;
3468         info->indegree_queue.compare = compare_commits_by_gen_then_commit_date;
3469
3470         info->min_generation = GENERATION_NUMBER_INFINITY;
3471         for (list = revs->commits; list; list = list->next) {
3472                 struct commit *c = list->item;
3473                 uint32_t generation;
3474
3475                 if (repo_parse_commit_gently(revs->repo, c, 1))
3476                         continue;
3477
3478                 test_flag_and_insert(&info->explore_queue, c, TOPO_WALK_EXPLORED);
3479                 test_flag_and_insert(&info->indegree_queue, c, TOPO_WALK_INDEGREE);
3480
3481                 generation = commit_graph_generation(c);
3482                 if (generation < info->min_generation)
3483                         info->min_generation = generation;
3484
3485                 *(indegree_slab_at(&info->indegree, c)) = 1;
3486
3487                 if (revs->sort_order == REV_SORT_BY_AUTHOR_DATE)
3488                         record_author_date(&info->author_date, c);
3489         }
3490         compute_indegrees_to_depth(revs, info->min_generation);
3491
3492         for (list = revs->commits; list; list = list->next) {
3493                 struct commit *c = list->item;
3494
3495                 if (*(indegree_slab_at(&info->indegree, c)) == 1)
3496                         prio_queue_put(&info->topo_queue, c);
3497         }
3498
3499         /*
3500          * This is unfortunate; the initial tips need to be shown
3501          * in the order given from the revision traversal machinery.
3502          */
3503         if (revs->sort_order == REV_SORT_IN_GRAPH_ORDER)
3504                 prio_queue_reverse(&info->topo_queue);
3505
3506         if (trace2_is_enabled() && !topo_walk_atexit_registered) {
3507                 atexit(trace2_topo_walk_statistics_atexit);
3508                 topo_walk_atexit_registered = 1;
3509         }
3510 }
3511
3512 static struct commit *next_topo_commit(struct rev_info *revs)
3513 {
3514         struct commit *c;
3515         struct topo_walk_info *info = revs->topo_walk_info;
3516
3517         /* pop next off of topo_queue */
3518         c = prio_queue_get(&info->topo_queue);
3519
3520         if (c)
3521                 *(indegree_slab_at(&info->indegree, c)) = 0;
3522
3523         return c;
3524 }
3525
3526 static void expand_topo_walk(struct rev_info *revs, struct commit *commit)
3527 {
3528         struct commit_list *p;
3529         struct topo_walk_info *info = revs->topo_walk_info;
3530         if (process_parents(revs, commit, NULL, NULL) < 0) {
3531                 if (!revs->ignore_missing_links)
3532                         die("Failed to traverse parents of commit %s",
3533                             oid_to_hex(&commit->object.oid));
3534         }
3535
3536         count_topo_walked++;
3537
3538         for (p = commit->parents; p; p = p->next) {
3539                 struct commit *parent = p->item;
3540                 int *pi;
3541                 uint32_t generation;
3542
3543                 if (parent->object.flags & UNINTERESTING)
3544                         continue;
3545
3546                 if (repo_parse_commit_gently(revs->repo, parent, 1) < 0)
3547                         continue;
3548
3549                 generation = commit_graph_generation(parent);
3550                 if (generation < info->min_generation) {
3551                         info->min_generation = generation;
3552                         compute_indegrees_to_depth(revs, info->min_generation);
3553                 }
3554
3555                 pi = indegree_slab_at(&info->indegree, parent);
3556
3557                 (*pi)--;
3558                 if (*pi == 1)
3559                         prio_queue_put(&info->topo_queue, parent);
3560
3561                 if (revs->first_parent_only)
3562                         return;
3563         }
3564 }
3565
3566 int prepare_revision_walk(struct rev_info *revs)
3567 {
3568         int i;
3569         struct object_array old_pending;
3570         struct commit_list **next = &revs->commits;
3571
3572         memcpy(&old_pending, &revs->pending, sizeof(old_pending));
3573         revs->pending.nr = 0;
3574         revs->pending.alloc = 0;
3575         revs->pending.objects = NULL;
3576         for (i = 0; i < old_pending.nr; i++) {
3577                 struct object_array_entry *e = old_pending.objects + i;
3578                 struct commit *commit = handle_commit(revs, e);
3579                 if (commit) {
3580                         if (!(commit->object.flags & SEEN)) {
3581                                 commit->object.flags |= SEEN;
3582                                 next = commit_list_append(commit, next);
3583                         }
3584                 }
3585         }
3586         object_array_clear(&old_pending);
3587
3588         /* Signal whether we need per-parent treesame decoration */
3589         if (revs->simplify_merges ||
3590             (revs->limited && limiting_can_increase_treesame(revs)))
3591                 revs->treesame.name = "treesame";
3592
3593         if (revs->exclude_promisor_objects) {
3594                 for_each_packed_object(mark_uninteresting, revs,
3595                                        FOR_EACH_OBJECT_PROMISOR_ONLY);
3596         }
3597
3598         if (!revs->reflog_info)
3599                 prepare_to_use_bloom_filter(revs);
3600         if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
3601                 commit_list_sort_by_date(&revs->commits);
3602         if (revs->no_walk)
3603                 return 0;
3604         if (revs->limited) {
3605                 if (limit_list(revs) < 0)
3606                         return -1;
3607                 if (revs->topo_order)
3608                         sort_in_topological_order(&revs->commits, revs->sort_order);
3609         } else if (revs->topo_order)
3610                 init_topo_walk(revs);
3611         if (revs->line_level_traverse && want_ancestry(revs))
3612                 /*
3613                  * At the moment we can only do line-level log with parent
3614                  * rewriting by performing this expensive pre-filtering step.
3615                  * If parent rewriting is not requested, then we rather
3616                  * perform the line-level log filtering during the regular
3617                  * history traversal.
3618                  */
3619                 line_log_filter(revs);
3620         if (revs->simplify_merges)
3621                 simplify_merges(revs);
3622         if (revs->children.name)
3623                 set_children(revs);
3624
3625         return 0;
3626 }
3627
3628 static enum rewrite_result rewrite_one_1(struct rev_info *revs,
3629                                          struct commit **pp,
3630                                          struct prio_queue *queue)
3631 {
3632         for (;;) {
3633                 struct commit *p = *pp;
3634                 if (!revs->limited)
3635                         if (process_parents(revs, p, NULL, queue) < 0)
3636                                 return rewrite_one_error;
3637                 if (p->object.flags & UNINTERESTING)
3638                         return rewrite_one_ok;
3639                 if (!(p->object.flags & TREESAME))
3640                         return rewrite_one_ok;
3641                 if (!p->parents)
3642                         return rewrite_one_noparents;
3643                 if ((p = one_relevant_parent(revs, p->parents)) == NULL)
3644                         return rewrite_one_ok;
3645                 *pp = p;
3646         }
3647 }
3648
3649 static void merge_queue_into_list(struct prio_queue *q, struct commit_list **list)
3650 {
3651         while (q->nr) {
3652                 struct commit *item = prio_queue_peek(q);
3653                 struct commit_list *p = *list;
3654
3655                 if (p && p->item->date >= item->date)
3656                         list = &p->next;
3657                 else {
3658                         p = commit_list_insert(item, list);
3659                         list = &p->next; /* skip newly added item */
3660                         prio_queue_get(q); /* pop item */
3661                 }
3662         }
3663 }
3664
3665 static enum rewrite_result rewrite_one(struct rev_info *revs, struct commit **pp)
3666 {
3667         struct prio_queue queue = { compare_commits_by_commit_date };
3668         enum rewrite_result ret = rewrite_one_1(revs, pp, &queue);
3669         merge_queue_into_list(&queue, &revs->commits);
3670         clear_prio_queue(&queue);
3671         return ret;
3672 }
3673
3674 int rewrite_parents(struct rev_info *revs, struct commit *commit,
3675         rewrite_parent_fn_t rewrite_parent)
3676 {
3677         struct commit_list **pp = &commit->parents;
3678         while (*pp) {
3679                 struct commit_list *parent = *pp;
3680                 switch (rewrite_parent(revs, &parent->item)) {
3681                 case rewrite_one_ok:
3682                         break;
3683                 case rewrite_one_noparents:
3684                         *pp = parent->next;
3685                         continue;
3686                 case rewrite_one_error:
3687                         return -1;
3688                 }
3689                 pp = &parent->next;
3690         }
3691         remove_duplicate_parents(revs, commit);
3692         return 0;
3693 }
3694
3695 static int commit_rewrite_person(struct strbuf *buf, const char *what, struct string_list *mailmap)
3696 {
3697         char *person, *endp;
3698         size_t len, namelen, maillen;
3699         const char *name;
3700         const char *mail;
3701         struct ident_split ident;
3702
3703         person = strstr(buf->buf, what);
3704         if (!person)
3705                 return 0;
3706
3707         person += strlen(what);
3708         endp = strchr(person, '\n');
3709         if (!endp)
3710                 return 0;
3711
3712         len = endp - person;
3713
3714         if (split_ident_line(&ident, person, len))
3715                 return 0;
3716
3717         mail = ident.mail_begin;
3718         maillen = ident.mail_end - ident.mail_begin;
3719         name = ident.name_begin;
3720         namelen = ident.name_end - ident.name_begin;
3721
3722         if (map_user(mailmap, &mail, &maillen, &name, &namelen)) {
3723                 struct strbuf namemail = STRBUF_INIT;
3724
3725                 strbuf_addf(&namemail, "%.*s <%.*s>",
3726                             (int)namelen, name, (int)maillen, mail);
3727
3728                 strbuf_splice(buf, ident.name_begin - buf->buf,
3729                               ident.mail_end - ident.name_begin + 1,
3730                               namemail.buf, namemail.len);
3731
3732                 strbuf_release(&namemail);
3733
3734                 return 1;
3735         }
3736
3737         return 0;
3738 }
3739
3740 static int commit_match(struct commit *commit, struct rev_info *opt)
3741 {
3742         int retval;
3743         const char *encoding;
3744         const char *message;
3745         struct strbuf buf = STRBUF_INIT;
3746
3747         if (!opt->grep_filter.pattern_list && !opt->grep_filter.header_list)
3748                 return 1;
3749
3750         /* Prepend "fake" headers as needed */
3751         if (opt->grep_filter.use_reflog_filter) {
3752                 strbuf_addstr(&buf, "reflog ");
3753                 get_reflog_message(&buf, opt->reflog_info);
3754                 strbuf_addch(&buf, '\n');
3755         }
3756
3757         /*
3758          * We grep in the user's output encoding, under the assumption that it
3759          * is the encoding they are most likely to write their grep pattern
3760          * for. In addition, it means we will match the "notes" encoding below,
3761          * so we will not end up with a buffer that has two different encodings
3762          * in it.
3763          */
3764         encoding = get_log_output_encoding();
3765         message = logmsg_reencode(commit, NULL, encoding);
3766
3767         /* Copy the commit to temporary if we are using "fake" headers */
3768         if (buf.len)
3769                 strbuf_addstr(&buf, message);
3770
3771         if (opt->grep_filter.header_list && opt->mailmap) {
3772                 if (!buf.len)
3773                         strbuf_addstr(&buf, message);
3774
3775                 commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
3776                 commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
3777         }
3778
3779         /* Append "fake" message parts as needed */
3780         if (opt->show_notes) {
3781                 if (!buf.len)
3782                         strbuf_addstr(&buf, message);
3783                 format_display_notes(&commit->object.oid, &buf, encoding, 1);
3784         }
3785
3786         /*
3787          * Find either in the original commit message, or in the temporary.
3788          * Note that we cast away the constness of "message" here. It is
3789          * const because it may come from the cached commit buffer. That's OK,
3790          * because we know that it is modifiable heap memory, and that while
3791          * grep_buffer may modify it for speed, it will restore any
3792          * changes before returning.
3793          */
3794         if (buf.len)
3795                 retval = grep_buffer(&opt->grep_filter, buf.buf, buf.len);
3796         else
3797                 retval = grep_buffer(&opt->grep_filter,
3798                                      (char *)message, strlen(message));
3799         strbuf_release(&buf);
3800         unuse_commit_buffer(commit, message);
3801         return opt->invert_grep ? !retval : retval;
3802 }
3803
3804 static inline int want_ancestry(const struct rev_info *revs)
3805 {
3806         return (revs->rewrite_parents || revs->children.name);
3807 }
3808
3809 /*
3810  * Return a timestamp to be used for --since/--until comparisons for this
3811  * commit, based on the revision options.
3812  */
3813 static timestamp_t comparison_date(const struct rev_info *revs,
3814                                    struct commit *commit)
3815 {
3816         return revs->reflog_info ?
3817                 get_reflog_timestamp(revs->reflog_info) :
3818                 commit->date;
3819 }
3820
3821 enum commit_action get_commit_action(struct rev_info *revs, struct commit *commit)
3822 {
3823         if (commit->object.flags & SHOWN)
3824                 return commit_ignore;
3825         if (revs->unpacked && has_object_pack(&commit->object.oid))
3826                 return commit_ignore;
3827         if (commit->object.flags & UNINTERESTING)
3828                 return commit_ignore;
3829         if (revs->line_level_traverse && !want_ancestry(revs)) {
3830                 /*
3831                  * In case of line-level log with parent rewriting
3832                  * prepare_revision_walk() already took care of all line-level
3833                  * log filtering, and there is nothing left to do here.
3834                  *
3835                  * If parent rewriting was not requested, then this is the
3836                  * place to perform the line-level log filtering.  Notably,
3837                  * this check, though expensive, must come before the other,
3838                  * cheaper filtering conditions, because the tracked line
3839                  * ranges must be adjusted even when the commit will end up
3840                  * being ignored based on other conditions.
3841                  */
3842                 if (!line_log_process_ranges_arbitrary_commit(revs, commit))
3843                         return commit_ignore;
3844         }
3845         if (revs->min_age != -1 &&
3846             comparison_date(revs, commit) > revs->min_age)
3847                         return commit_ignore;
3848         if (revs->min_parents || (revs->max_parents >= 0)) {
3849                 int n = commit_list_count(commit->parents);
3850                 if ((n < revs->min_parents) ||
3851                     ((revs->max_parents >= 0) && (n > revs->max_parents)))
3852                         return commit_ignore;
3853         }
3854         if (!commit_match(commit, revs))
3855                 return commit_ignore;
3856         if (revs->prune && revs->dense) {
3857                 /* Commit without changes? */
3858                 if (commit->object.flags & TREESAME) {
3859                         int n;
3860                         struct commit_list *p;
3861                         /* drop merges unless we want parenthood */
3862                         if (!want_ancestry(revs))
3863                                 return commit_ignore;
3864
3865                         if (revs->show_pulls && (commit->object.flags & PULL_MERGE))
3866                                 return commit_show;
3867
3868                         /*
3869                          * If we want ancestry, then need to keep any merges
3870                          * between relevant commits to tie together topology.
3871                          * For consistency with TREESAME and simplification
3872                          * use "relevant" here rather than just INTERESTING,
3873                          * to treat bottom commit(s) as part of the topology.
3874                          */
3875                         for (n = 0, p = commit->parents; p; p = p->next)
3876                                 if (relevant_commit(p->item))
3877                                         if (++n >= 2)
3878                                                 return commit_show;
3879                         return commit_ignore;
3880                 }
3881         }
3882         return commit_show;
3883 }
3884
3885 define_commit_slab(saved_parents, struct commit_list *);
3886
3887 #define EMPTY_PARENT_LIST ((struct commit_list *)-1)
3888
3889 /*
3890  * You may only call save_parents() once per commit (this is checked
3891  * for non-root commits).
3892  */
3893 static void save_parents(struct rev_info *revs, struct commit *commit)
3894 {
3895         struct commit_list **pp;
3896
3897         if (!revs->saved_parents_slab) {
3898                 revs->saved_parents_slab = xmalloc(sizeof(struct saved_parents));
3899                 init_saved_parents(revs->saved_parents_slab);
3900         }
3901
3902         pp = saved_parents_at(revs->saved_parents_slab, commit);
3903
3904         /*
3905          * When walking with reflogs, we may visit the same commit
3906          * several times: once for each appearance in the reflog.
3907          *
3908          * In this case, save_parents() will be called multiple times.
3909          * We want to keep only the first set of parents.  We need to
3910          * store a sentinel value for an empty (i.e., NULL) parent
3911          * list to distinguish it from a not-yet-saved list, however.
3912          */
3913         if (*pp)
3914                 return;
3915         if (commit->parents)
3916                 *pp = copy_commit_list(commit->parents);
3917         else
3918                 *pp = EMPTY_PARENT_LIST;
3919 }
3920
3921 static void free_saved_parents(struct rev_info *revs)
3922 {
3923         if (revs->saved_parents_slab)
3924                 clear_saved_parents(revs->saved_parents_slab);
3925 }
3926
3927 struct commit_list *get_saved_parents(struct rev_info *revs, const struct commit *commit)
3928 {
3929         struct commit_list *parents;
3930
3931         if (!revs->saved_parents_slab)
3932                 return commit->parents;
3933
3934         parents = *saved_parents_at(revs->saved_parents_slab, commit);
3935         if (parents == EMPTY_PARENT_LIST)
3936                 return NULL;
3937         return parents;
3938 }
3939
3940 enum commit_action simplify_commit(struct rev_info *revs, struct commit *commit)
3941 {
3942         enum commit_action action = get_commit_action(revs, commit);
3943
3944         if (action == commit_show &&
3945             revs->prune && revs->dense && want_ancestry(revs)) {
3946                 /*
3947                  * --full-diff on simplified parents is no good: it
3948                  * will show spurious changes from the commits that
3949                  * were elided.  So we save the parents on the side
3950                  * when --full-diff is in effect.
3951                  */
3952                 if (revs->full_diff)
3953                         save_parents(revs, commit);
3954                 if (rewrite_parents(revs, commit, rewrite_one) < 0)
3955                         return commit_error;
3956         }
3957         return action;
3958 }
3959
3960 static void track_linear(struct rev_info *revs, struct commit *commit)
3961 {
3962         if (revs->track_first_time) {
3963                 revs->linear = 1;
3964                 revs->track_first_time = 0;
3965         } else {
3966                 struct commit_list *p;
3967                 for (p = revs->previous_parents; p; p = p->next)
3968                         if (p->item == NULL || /* first commit */
3969                             oideq(&p->item->object.oid, &commit->object.oid))
3970                                 break;
3971                 revs->linear = p != NULL;
3972         }
3973         if (revs->reverse) {
3974                 if (revs->linear)
3975                         commit->object.flags |= TRACK_LINEAR;
3976         }
3977         free_commit_list(revs->previous_parents);
3978         revs->previous_parents = copy_commit_list(commit->parents);
3979 }
3980
3981 static struct commit *get_revision_1(struct rev_info *revs)
3982 {
3983         while (1) {
3984                 struct commit *commit;
3985
3986                 if (revs->reflog_info)
3987                         commit = next_reflog_entry(revs->reflog_info);
3988                 else if (revs->topo_walk_info)
3989                         commit = next_topo_commit(revs);
3990                 else
3991                         commit = pop_commit(&revs->commits);
3992
3993                 if (!commit)
3994                         return NULL;
3995
3996                 if (revs->reflog_info)
3997                         commit->object.flags &= ~(ADDED | SEEN | SHOWN);
3998
3999                 /*
4000                  * If we haven't done the list limiting, we need to look at
4001                  * the parents here. We also need to do the date-based limiting
4002                  * that we'd otherwise have done in limit_list().
4003                  */
4004                 if (!revs->limited) {
4005                         if (revs->max_age != -1 &&
4006                             comparison_date(revs, commit) < revs->max_age)
4007                                 continue;
4008
4009                         if (revs->reflog_info)
4010                                 try_to_simplify_commit(revs, commit);
4011                         else if (revs->topo_walk_info)
4012                                 expand_topo_walk(revs, commit);
4013                         else if (process_parents(revs, commit, &revs->commits, NULL) < 0) {
4014                                 if (!revs->ignore_missing_links)
4015                                         die("Failed to traverse parents of commit %s",
4016                                                 oid_to_hex(&commit->object.oid));
4017                         }
4018                 }
4019
4020                 switch (simplify_commit(revs, commit)) {
4021                 case commit_ignore:
4022                         continue;
4023                 case commit_error:
4024                         die("Failed to simplify parents of commit %s",
4025                             oid_to_hex(&commit->object.oid));
4026                 default:
4027                         if (revs->track_linear)
4028                                 track_linear(revs, commit);
4029                         return commit;
4030                 }
4031         }
4032 }
4033
4034 /*
4035  * Return true for entries that have not yet been shown.  (This is an
4036  * object_array_each_func_t.)
4037  */
4038 static int entry_unshown(struct object_array_entry *entry, void *cb_data_unused)
4039 {
4040         return !(entry->item->flags & SHOWN);
4041 }
4042
4043 /*
4044  * If array is on the verge of a realloc, garbage-collect any entries
4045  * that have already been shown to try to free up some space.
4046  */
4047 static void gc_boundary(struct object_array *array)
4048 {
4049         if (array->nr == array->alloc)
4050                 object_array_filter(array, entry_unshown, NULL);
4051 }
4052
4053 static void create_boundary_commit_list(struct rev_info *revs)
4054 {
4055         unsigned i;
4056         struct commit *c;
4057         struct object_array *array = &revs->boundary_commits;
4058         struct object_array_entry *objects = array->objects;
4059
4060         /*
4061          * If revs->commits is non-NULL at this point, an error occurred in
4062          * get_revision_1().  Ignore the error and continue printing the
4063          * boundary commits anyway.  (This is what the code has always
4064          * done.)
4065          */
4066         if (revs->commits) {
4067                 free_commit_list(revs->commits);
4068                 revs->commits = NULL;
4069         }
4070
4071         /*
4072          * Put all of the actual boundary commits from revs->boundary_commits
4073          * into revs->commits
4074          */
4075         for (i = 0; i < array->nr; i++) {
4076                 c = (struct commit *)(objects[i].item);
4077                 if (!c)
4078                         continue;
4079                 if (!(c->object.flags & CHILD_SHOWN))
4080                         continue;
4081                 if (c->object.flags & (SHOWN | BOUNDARY))
4082                         continue;
4083                 c->object.flags |= BOUNDARY;
4084                 commit_list_insert(c, &revs->commits);
4085         }
4086
4087         /*
4088          * If revs->topo_order is set, sort the boundary commits
4089          * in topological order
4090          */
4091         sort_in_topological_order(&revs->commits, revs->sort_order);
4092 }
4093
4094 static struct commit *get_revision_internal(struct rev_info *revs)
4095 {
4096         struct commit *c = NULL;
4097         struct commit_list *l;
4098
4099         if (revs->boundary == 2) {
4100                 /*
4101                  * All of the normal commits have already been returned,
4102                  * and we are now returning boundary commits.
4103                  * create_boundary_commit_list() has populated
4104                  * revs->commits with the remaining commits to return.
4105                  */
4106                 c = pop_commit(&revs->commits);
4107                 if (c)
4108                         c->object.flags |= SHOWN;
4109                 return c;
4110         }
4111
4112         /*
4113          * If our max_count counter has reached zero, then we are done. We
4114          * don't simply return NULL because we still might need to show
4115          * boundary commits. But we want to avoid calling get_revision_1, which
4116          * might do a considerable amount of work finding the next commit only
4117          * for us to throw it away.
4118          *
4119          * If it is non-zero, then either we don't have a max_count at all
4120          * (-1), or it is still counting, in which case we decrement.
4121          */
4122         if (revs->max_count) {
4123                 c = get_revision_1(revs);
4124                 if (c) {
4125                         while (revs->skip_count > 0) {
4126                                 revs->skip_count--;
4127                                 c = get_revision_1(revs);
4128                                 if (!c)
4129                                         break;
4130                         }
4131                 }
4132
4133                 if (revs->max_count > 0)
4134                         revs->max_count--;
4135         }
4136
4137         if (c)
4138                 c->object.flags |= SHOWN;
4139
4140         if (!revs->boundary)
4141                 return c;
4142
4143         if (!c) {
4144                 /*
4145                  * get_revision_1() runs out the commits, and
4146                  * we are done computing the boundaries.
4147                  * switch to boundary commits output mode.
4148                  */
4149                 revs->boundary = 2;
4150
4151                 /*
4152                  * Update revs->commits to contain the list of
4153                  * boundary commits.
4154                  */
4155                 create_boundary_commit_list(revs);
4156
4157                 return get_revision_internal(revs);
4158         }
4159
4160         /*
4161          * boundary commits are the commits that are parents of the
4162          * ones we got from get_revision_1() but they themselves are
4163          * not returned from get_revision_1().  Before returning
4164          * 'c', we need to mark its parents that they could be boundaries.
4165          */
4166
4167         for (l = c->parents; l; l = l->next) {
4168                 struct object *p;
4169                 p = &(l->item->object);
4170                 if (p->flags & (CHILD_SHOWN | SHOWN))
4171                         continue;
4172                 p->flags |= CHILD_SHOWN;
4173                 gc_boundary(&revs->boundary_commits);
4174                 add_object_array(p, NULL, &revs->boundary_commits);
4175         }
4176
4177         return c;
4178 }
4179
4180 struct commit *get_revision(struct rev_info *revs)
4181 {
4182         struct commit *c;
4183         struct commit_list *reversed;
4184
4185         if (revs->reverse) {
4186                 reversed = NULL;
4187                 while ((c = get_revision_internal(revs)))
4188                         commit_list_insert(c, &reversed);
4189                 revs->commits = reversed;
4190                 revs->reverse = 0;
4191                 revs->reverse_output_stage = 1;
4192         }
4193
4194         if (revs->reverse_output_stage) {
4195                 c = pop_commit(&revs->commits);
4196                 if (revs->track_linear)
4197                         revs->linear = !!(c && c->object.flags & TRACK_LINEAR);
4198                 return c;
4199         }
4200
4201         c = get_revision_internal(revs);
4202         if (c && revs->graph)
4203                 graph_update(revs->graph, c);
4204         if (!c) {
4205                 free_saved_parents(revs);
4206                 if (revs->previous_parents) {
4207                         free_commit_list(revs->previous_parents);
4208                         revs->previous_parents = NULL;
4209                 }
4210         }
4211         return c;
4212 }
4213
4214 const char *get_revision_mark(const struct rev_info *revs, const struct commit *commit)
4215 {
4216         if (commit->object.flags & BOUNDARY)
4217                 return "-";
4218         else if (commit->object.flags & UNINTERESTING)
4219                 return "^";
4220         else if (commit->object.flags & PATCHSAME)
4221                 return "=";
4222         else if (!revs || revs->left_right) {
4223                 if (commit->object.flags & SYMMETRIC_LEFT)
4224                         return "<";
4225                 else
4226                         return ">";
4227         } else if (revs->graph)
4228                 return "*";
4229         else if (revs->cherry_mark)
4230                 return "+";
4231         return "";
4232 }
4233
4234 void put_revision_mark(const struct rev_info *revs, const struct commit *commit)
4235 {
4236         const char *mark = get_revision_mark(revs, commit);
4237         if (!strlen(mark))
4238                 return;
4239         fputs(mark, stdout);
4240         putchar(' ');
4241 }