Merge branch 'jx/t5411-unique-filenames'
[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 "diff-merges.h"
9 #include "refs.h"
10 #include "revision.h"
11 #include "repository.h"
12 #include "graph.h"
13 #include "grep.h"
14 #include "reflog-walk.h"
15 #include "patch-ids.h"
16 #include "decorate.h"
17 #include "log-tree.h"
18 #include "string-list.h"
19 #include "line-log.h"
20 #include "mailmap.h"
21 #include "commit-slab.h"
22 #include "dir.h"
23 #include "cache-tree.h"
24 #include "bisect.h"
25 #include "packfile.h"
26 #include "worktree.h"
27 #include "strvec.h"
28 #include "commit-reach.h"
29 #include "commit-graph.h"
30 #include "prio-queue.h"
31 #include "hashmap.h"
32 #include "utf8.h"
33 #include "bloom.h"
34 #include "json-writer.h"
35
36 volatile show_early_output_fn_t show_early_output;
37
38 static const char *term_bad;
39 static const char *term_good;
40
41 implement_shared_commit_slab(revision_sources, char *);
42
43 static inline int want_ancestry(const struct rev_info *revs);
44
45 void show_object_with_name(FILE *out, struct object *obj, const char *name)
46 {
47         const char *p;
48
49         fprintf(out, "%s ", oid_to_hex(&obj->oid));
50         for (p = name; *p && *p != '\n'; p++)
51                 fputc(*p, out);
52         fputc('\n', out);
53 }
54
55 static void mark_blob_uninteresting(struct blob *blob)
56 {
57         if (!blob)
58                 return;
59         if (blob->object.flags & UNINTERESTING)
60                 return;
61         blob->object.flags |= UNINTERESTING;
62 }
63
64 static void mark_tree_contents_uninteresting(struct repository *r,
65                                              struct tree *tree)
66 {
67         struct tree_desc desc;
68         struct name_entry entry;
69
70         if (parse_tree_gently(tree, 1) < 0)
71                 return;
72
73         init_tree_desc(&desc, tree->buffer, tree->size);
74         while (tree_entry(&desc, &entry)) {
75                 switch (object_type(entry.mode)) {
76                 case OBJ_TREE:
77                         mark_tree_uninteresting(r, lookup_tree(r, &entry.oid));
78                         break;
79                 case OBJ_BLOB:
80                         mark_blob_uninteresting(lookup_blob(r, &entry.oid));
81                         break;
82                 default:
83                         /* Subproject commit - not in this repository */
84                         break;
85                 }
86         }
87
88         /*
89          * We don't care about the tree any more
90          * after it has been marked uninteresting.
91          */
92         free_tree_buffer(tree);
93 }
94
95 void mark_tree_uninteresting(struct repository *r, struct tree *tree)
96 {
97         struct object *obj;
98
99         if (!tree)
100                 return;
101
102         obj = &tree->object;
103         if (obj->flags & UNINTERESTING)
104                 return;
105         obj->flags |= UNINTERESTING;
106         mark_tree_contents_uninteresting(r, tree);
107 }
108
109 struct path_and_oids_entry {
110         struct hashmap_entry ent;
111         char *path;
112         struct oidset trees;
113 };
114
115 static int path_and_oids_cmp(const void *hashmap_cmp_fn_data,
116                              const struct hashmap_entry *eptr,
117                              const struct hashmap_entry *entry_or_key,
118                              const void *keydata)
119 {
120         const struct path_and_oids_entry *e1, *e2;
121
122         e1 = container_of(eptr, const struct path_and_oids_entry, ent);
123         e2 = container_of(entry_or_key, const struct path_and_oids_entry, ent);
124
125         return strcmp(e1->path, e2->path);
126 }
127
128 static void paths_and_oids_clear(struct hashmap *map)
129 {
130         struct hashmap_iter iter;
131         struct path_and_oids_entry *entry;
132
133         hashmap_for_each_entry(map, &iter, entry, ent /* member name */) {
134                 oidset_clear(&entry->trees);
135                 free(entry->path);
136         }
137
138         hashmap_clear_and_free(map, struct path_and_oids_entry, ent);
139 }
140
141 static void paths_and_oids_insert(struct hashmap *map,
142                                   const char *path,
143                                   const struct object_id *oid)
144 {
145         int hash = strhash(path);
146         struct path_and_oids_entry key;
147         struct path_and_oids_entry *entry;
148
149         hashmap_entry_init(&key.ent, hash);
150
151         /* use a shallow copy for the lookup */
152         key.path = (char *)path;
153         oidset_init(&key.trees, 0);
154
155         entry = hashmap_get_entry(map, &key, ent, NULL);
156         if (!entry) {
157                 entry = xcalloc(1, sizeof(struct path_and_oids_entry));
158                 hashmap_entry_init(&entry->ent, hash);
159                 entry->path = xstrdup(key.path);
160                 oidset_init(&entry->trees, 16);
161                 hashmap_put(map, &entry->ent);
162         }
163
164         oidset_insert(&entry->trees, oid);
165 }
166
167 static void add_children_by_path(struct repository *r,
168                                  struct tree *tree,
169                                  struct hashmap *map)
170 {
171         struct tree_desc desc;
172         struct name_entry entry;
173
174         if (!tree)
175                 return;
176
177         if (parse_tree_gently(tree, 1) < 0)
178                 return;
179
180         init_tree_desc(&desc, tree->buffer, tree->size);
181         while (tree_entry(&desc, &entry)) {
182                 switch (object_type(entry.mode)) {
183                 case OBJ_TREE:
184                         paths_and_oids_insert(map, entry.path, &entry.oid);
185
186                         if (tree->object.flags & UNINTERESTING) {
187                                 struct tree *child = lookup_tree(r, &entry.oid);
188                                 if (child)
189                                         child->object.flags |= UNINTERESTING;
190                         }
191                         break;
192                 case OBJ_BLOB:
193                         if (tree->object.flags & UNINTERESTING) {
194                                 struct blob *child = lookup_blob(r, &entry.oid);
195                                 if (child)
196                                         child->object.flags |= UNINTERESTING;
197                         }
198                         break;
199                 default:
200                         /* Subproject commit - not in this repository */
201                         break;
202                 }
203         }
204
205         free_tree_buffer(tree);
206 }
207
208 void mark_trees_uninteresting_sparse(struct repository *r,
209                                      struct oidset *trees)
210 {
211         unsigned has_interesting = 0, has_uninteresting = 0;
212         struct hashmap map = HASHMAP_INIT(path_and_oids_cmp, NULL);
213         struct hashmap_iter map_iter;
214         struct path_and_oids_entry *entry;
215         struct object_id *oid;
216         struct oidset_iter iter;
217
218         oidset_iter_init(trees, &iter);
219         while ((!has_interesting || !has_uninteresting) &&
220                (oid = oidset_iter_next(&iter))) {
221                 struct tree *tree = lookup_tree(r, oid);
222
223                 if (!tree)
224                         continue;
225
226                 if (tree->object.flags & UNINTERESTING)
227                         has_uninteresting = 1;
228                 else
229                         has_interesting = 1;
230         }
231
232         /* Do not walk unless we have both types of trees. */
233         if (!has_uninteresting || !has_interesting)
234                 return;
235
236         oidset_iter_init(trees, &iter);
237         while ((oid = oidset_iter_next(&iter))) {
238                 struct tree *tree = lookup_tree(r, oid);
239                 add_children_by_path(r, tree, &map);
240         }
241
242         hashmap_for_each_entry(&map, &map_iter, entry, ent /* member name */)
243                 mark_trees_uninteresting_sparse(r, &entry->trees);
244
245         paths_and_oids_clear(&map);
246 }
247
248 struct commit_stack {
249         struct commit **items;
250         size_t nr, alloc;
251 };
252 #define COMMIT_STACK_INIT { NULL, 0, 0 }
253
254 static void commit_stack_push(struct commit_stack *stack, struct commit *commit)
255 {
256         ALLOC_GROW(stack->items, stack->nr + 1, stack->alloc);
257         stack->items[stack->nr++] = commit;
258 }
259
260 static struct commit *commit_stack_pop(struct commit_stack *stack)
261 {
262         return stack->nr ? stack->items[--stack->nr] : NULL;
263 }
264
265 static void commit_stack_clear(struct commit_stack *stack)
266 {
267         FREE_AND_NULL(stack->items);
268         stack->nr = stack->alloc = 0;
269 }
270
271 static void mark_one_parent_uninteresting(struct commit *commit,
272                                           struct commit_stack *pending)
273 {
274         struct commit_list *l;
275
276         if (commit->object.flags & UNINTERESTING)
277                 return;
278         commit->object.flags |= UNINTERESTING;
279
280         /*
281          * Normally we haven't parsed the parent
282          * yet, so we won't have a parent of a parent
283          * here. However, it may turn out that we've
284          * reached this commit some other way (where it
285          * wasn't uninteresting), in which case we need
286          * to mark its parents recursively too..
287          */
288         for (l = commit->parents; l; l = l->next)
289                 commit_stack_push(pending, l->item);
290 }
291
292 void mark_parents_uninteresting(struct commit *commit)
293 {
294         struct commit_stack pending = COMMIT_STACK_INIT;
295         struct commit_list *l;
296
297         for (l = commit->parents; l; l = l->next)
298                 mark_one_parent_uninteresting(l->item, &pending);
299
300         while (pending.nr > 0)
301                 mark_one_parent_uninteresting(commit_stack_pop(&pending),
302                                               &pending);
303
304         commit_stack_clear(&pending);
305 }
306
307 static void add_pending_object_with_path(struct rev_info *revs,
308                                          struct object *obj,
309                                          const char *name, unsigned mode,
310                                          const char *path)
311 {
312         struct interpret_branch_name_options options = { 0 };
313         if (!obj)
314                 return;
315         if (revs->no_walk && (obj->flags & UNINTERESTING))
316                 revs->no_walk = 0;
317         if (revs->reflog_info && obj->type == OBJ_COMMIT) {
318                 struct strbuf buf = STRBUF_INIT;
319                 int len = interpret_branch_name(name, 0, &buf, &options);
320
321                 if (0 < len && name[len] && buf.len)
322                         strbuf_addstr(&buf, name + len);
323                 add_reflog_for_walk(revs->reflog_info,
324                                     (struct commit *)obj,
325                                     buf.buf[0] ? buf.buf: name);
326                 strbuf_release(&buf);
327                 return; /* do not add the commit itself */
328         }
329         add_object_array_with_path(obj, name, &revs->pending, mode, path);
330 }
331
332 static void add_pending_object_with_mode(struct rev_info *revs,
333                                          struct object *obj,
334                                          const char *name, unsigned mode)
335 {
336         add_pending_object_with_path(revs, obj, name, mode, NULL);
337 }
338
339 void add_pending_object(struct rev_info *revs,
340                         struct object *obj, const char *name)
341 {
342         add_pending_object_with_mode(revs, obj, name, S_IFINVALID);
343 }
344
345 void add_head_to_pending(struct rev_info *revs)
346 {
347         struct object_id oid;
348         struct object *obj;
349         if (get_oid("HEAD", &oid))
350                 return;
351         obj = parse_object(revs->repo, &oid);
352         if (!obj)
353                 return;
354         add_pending_object(revs, obj, "HEAD");
355 }
356
357 static struct object *get_reference(struct rev_info *revs, const char *name,
358                                     const struct object_id *oid,
359                                     unsigned int flags)
360 {
361         struct object *object;
362
363         /*
364          * If the repository has commit graphs, repo_parse_commit() avoids
365          * reading the object buffer, so use it whenever possible.
366          */
367         if (oid_object_info(revs->repo, oid, NULL) == OBJ_COMMIT) {
368                 struct commit *c = lookup_commit(revs->repo, oid);
369                 if (!repo_parse_commit(revs->repo, c))
370                         object = (struct object *) c;
371                 else
372                         object = NULL;
373         } else {
374                 object = parse_object(revs->repo, oid);
375         }
376
377         if (!object) {
378                 if (revs->ignore_missing)
379                         return object;
380                 if (revs->exclude_promisor_objects && is_promisor_object(oid))
381                         return NULL;
382                 die("bad object %s", name);
383         }
384         object->flags |= flags;
385         return object;
386 }
387
388 void add_pending_oid(struct rev_info *revs, const char *name,
389                       const struct object_id *oid, unsigned int flags)
390 {
391         struct object *object = get_reference(revs, name, oid, flags);
392         add_pending_object(revs, object, name);
393 }
394
395 static struct commit *handle_commit(struct rev_info *revs,
396                                     struct object_array_entry *entry)
397 {
398         struct object *object = entry->item;
399         const char *name = entry->name;
400         const char *path = entry->path;
401         unsigned int mode = entry->mode;
402         unsigned long flags = object->flags;
403
404         /*
405          * Tag object? Look what it points to..
406          */
407         while (object->type == OBJ_TAG) {
408                 struct tag *tag = (struct tag *) object;
409                 if (revs->tag_objects && !(flags & UNINTERESTING))
410                         add_pending_object(revs, object, tag->tag);
411                 object = parse_object(revs->repo, get_tagged_oid(tag));
412                 if (!object) {
413                         if (revs->ignore_missing_links || (flags & UNINTERESTING))
414                                 return NULL;
415                         if (revs->exclude_promisor_objects &&
416                             is_promisor_object(&tag->tagged->oid))
417                                 return NULL;
418                         die("bad object %s", oid_to_hex(&tag->tagged->oid));
419                 }
420                 object->flags |= flags;
421                 /*
422                  * We'll handle the tagged object by looping or dropping
423                  * through to the non-tag handlers below. Do not
424                  * propagate path data from the tag's pending entry.
425                  */
426                 path = NULL;
427                 mode = 0;
428         }
429
430         /*
431          * Commit object? Just return it, we'll do all the complex
432          * reachability crud.
433          */
434         if (object->type == OBJ_COMMIT) {
435                 struct commit *commit = (struct commit *)object;
436
437                 if (repo_parse_commit(revs->repo, commit) < 0)
438                         die("unable to parse commit %s", name);
439                 if (flags & UNINTERESTING) {
440                         mark_parents_uninteresting(commit);
441
442                         if (!revs->topo_order || !generation_numbers_enabled(the_repository))
443                                 revs->limited = 1;
444                 }
445                 if (revs->sources) {
446                         char **slot = revision_sources_at(revs->sources, commit);
447
448                         if (!*slot)
449                                 *slot = xstrdup(name);
450                 }
451                 return commit;
452         }
453
454         /*
455          * Tree object? Either mark it uninteresting, or add it
456          * to the list of objects to look at later..
457          */
458         if (object->type == OBJ_TREE) {
459                 struct tree *tree = (struct tree *)object;
460                 if (!revs->tree_objects)
461                         return NULL;
462                 if (flags & UNINTERESTING) {
463                         mark_tree_contents_uninteresting(revs->repo, tree);
464                         return NULL;
465                 }
466                 add_pending_object_with_path(revs, object, name, mode, path);
467                 return NULL;
468         }
469
470         /*
471          * Blob object? You know the drill by now..
472          */
473         if (object->type == OBJ_BLOB) {
474                 if (!revs->blob_objects)
475                         return NULL;
476                 if (flags & UNINTERESTING)
477                         return NULL;
478                 add_pending_object_with_path(revs, object, name, mode, path);
479                 return NULL;
480         }
481         die("%s is unknown object", name);
482 }
483
484 static int everybody_uninteresting(struct commit_list *orig,
485                                    struct commit **interesting_cache)
486 {
487         struct commit_list *list = orig;
488
489         if (*interesting_cache) {
490                 struct commit *commit = *interesting_cache;
491                 if (!(commit->object.flags & UNINTERESTING))
492                         return 0;
493         }
494
495         while (list) {
496                 struct commit *commit = list->item;
497                 list = list->next;
498                 if (commit->object.flags & UNINTERESTING)
499                         continue;
500
501                 *interesting_cache = commit;
502                 return 0;
503         }
504         return 1;
505 }
506
507 /*
508  * A definition of "relevant" commit that we can use to simplify limited graphs
509  * by eliminating side branches.
510  *
511  * A "relevant" commit is one that is !UNINTERESTING (ie we are including it
512  * in our list), or that is a specified BOTTOM commit. Then after computing
513  * a limited list, during processing we can generally ignore boundary merges
514  * coming from outside the graph, (ie from irrelevant parents), and treat
515  * those merges as if they were single-parent. TREESAME is defined to consider
516  * only relevant parents, if any. If we are TREESAME to our on-graph parents,
517  * we don't care if we were !TREESAME to non-graph parents.
518  *
519  * Treating bottom commits as relevant ensures that a limited graph's
520  * connection to the actual bottom commit is not viewed as a side branch, but
521  * treated as part of the graph. For example:
522  *
523  *   ....Z...A---X---o---o---B
524  *        .     /
525  *         W---Y
526  *
527  * When computing "A..B", the A-X connection is at least as important as
528  * Y-X, despite A being flagged UNINTERESTING.
529  *
530  * And when computing --ancestry-path "A..B", the A-X connection is more
531  * important than Y-X, despite both A and Y being flagged UNINTERESTING.
532  */
533 static inline int relevant_commit(struct commit *commit)
534 {
535         return (commit->object.flags & (UNINTERESTING | BOTTOM)) != UNINTERESTING;
536 }
537
538 /*
539  * Return a single relevant commit from a parent list. If we are a TREESAME
540  * commit, and this selects one of our parents, then we can safely simplify to
541  * that parent.
542  */
543 static struct commit *one_relevant_parent(const struct rev_info *revs,
544                                           struct commit_list *orig)
545 {
546         struct commit_list *list = orig;
547         struct commit *relevant = NULL;
548
549         if (!orig)
550                 return NULL;
551
552         /*
553          * For 1-parent commits, or if first-parent-only, then return that
554          * first parent (even if not "relevant" by the above definition).
555          * TREESAME will have been set purely on that parent.
556          */
557         if (revs->first_parent_only || !orig->next)
558                 return orig->item;
559
560         /*
561          * For multi-parent commits, identify a sole relevant parent, if any.
562          * If we have only one relevant parent, then TREESAME will be set purely
563          * with regard to that parent, and we can simplify accordingly.
564          *
565          * If we have more than one relevant parent, or no relevant parents
566          * (and multiple irrelevant ones), then we can't select a parent here
567          * and return NULL.
568          */
569         while (list) {
570                 struct commit *commit = list->item;
571                 list = list->next;
572                 if (relevant_commit(commit)) {
573                         if (relevant)
574                                 return NULL;
575                         relevant = commit;
576                 }
577         }
578         return relevant;
579 }
580
581 /*
582  * The goal is to get REV_TREE_NEW as the result only if the
583  * diff consists of all '+' (and no other changes), REV_TREE_OLD
584  * if the whole diff is removal of old data, and otherwise
585  * REV_TREE_DIFFERENT (of course if the trees are the same we
586  * want REV_TREE_SAME).
587  *
588  * The only time we care about the distinction is when
589  * remove_empty_trees is in effect, in which case we care only about
590  * whether the whole change is REV_TREE_NEW, or if there's another type
591  * of change. Which means we can stop the diff early in either of these
592  * cases:
593  *
594  *   1. We're not using remove_empty_trees at all.
595  *
596  *   2. We saw anything except REV_TREE_NEW.
597  */
598 static int tree_difference = REV_TREE_SAME;
599
600 static void file_add_remove(struct diff_options *options,
601                     int addremove, unsigned mode,
602                     const struct object_id *oid,
603                     int oid_valid,
604                     const char *fullpath, unsigned dirty_submodule)
605 {
606         int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
607         struct rev_info *revs = options->change_fn_data;
608
609         tree_difference |= diff;
610         if (!revs->remove_empty_trees || tree_difference != REV_TREE_NEW)
611                 options->flags.has_changes = 1;
612 }
613
614 static void file_change(struct diff_options *options,
615                  unsigned old_mode, unsigned new_mode,
616                  const struct object_id *old_oid,
617                  const struct object_id *new_oid,
618                  int old_oid_valid, int new_oid_valid,
619                  const char *fullpath,
620                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
621 {
622         tree_difference = REV_TREE_DIFFERENT;
623         options->flags.has_changes = 1;
624 }
625
626 static int bloom_filter_atexit_registered;
627 static unsigned int count_bloom_filter_maybe;
628 static unsigned int count_bloom_filter_definitely_not;
629 static unsigned int count_bloom_filter_false_positive;
630 static unsigned int count_bloom_filter_not_present;
631
632 static void trace2_bloom_filter_statistics_atexit(void)
633 {
634         struct json_writer jw = JSON_WRITER_INIT;
635
636         jw_object_begin(&jw, 0);
637         jw_object_intmax(&jw, "filter_not_present", count_bloom_filter_not_present);
638         jw_object_intmax(&jw, "maybe", count_bloom_filter_maybe);
639         jw_object_intmax(&jw, "definitely_not", count_bloom_filter_definitely_not);
640         jw_object_intmax(&jw, "false_positive", count_bloom_filter_false_positive);
641         jw_end(&jw);
642
643         trace2_data_json("bloom", the_repository, "statistics", &jw);
644
645         jw_release(&jw);
646 }
647
648 static int forbid_bloom_filters(struct pathspec *spec)
649 {
650         if (spec->has_wildcard)
651                 return 1;
652         if (spec->nr > 1)
653                 return 1;
654         if (spec->magic & ~PATHSPEC_LITERAL)
655                 return 1;
656         if (spec->nr && (spec->items[0].magic & ~PATHSPEC_LITERAL))
657                 return 1;
658
659         return 0;
660 }
661
662 static void prepare_to_use_bloom_filter(struct rev_info *revs)
663 {
664         struct pathspec_item *pi;
665         char *path_alloc = NULL;
666         const char *path, *p;
667         size_t len;
668         int path_component_nr = 1;
669
670         if (!revs->commits)
671                 return;
672
673         if (forbid_bloom_filters(&revs->prune_data))
674                 return;
675
676         repo_parse_commit(revs->repo, revs->commits->item);
677
678         revs->bloom_filter_settings = get_bloom_filter_settings(revs->repo);
679         if (!revs->bloom_filter_settings)
680                 return;
681
682         if (!revs->pruning.pathspec.nr)
683                 return;
684
685         pi = &revs->pruning.pathspec.items[0];
686
687         /* remove single trailing slash from path, if needed */
688         if (pi->len > 0 && pi->match[pi->len - 1] == '/') {
689                 path_alloc = xmemdupz(pi->match, pi->len - 1);
690                 path = path_alloc;
691         } else
692                 path = pi->match;
693
694         len = strlen(path);
695         if (!len) {
696                 revs->bloom_filter_settings = NULL;
697                 free(path_alloc);
698                 return;
699         }
700
701         p = path;
702         while (*p) {
703                 /*
704                  * At this point, the path is normalized to use Unix-style
705                  * path separators. This is required due to how the
706                  * changed-path Bloom filters store the paths.
707                  */
708                 if (*p == '/')
709                         path_component_nr++;
710                 p++;
711         }
712
713         revs->bloom_keys_nr = path_component_nr;
714         ALLOC_ARRAY(revs->bloom_keys, revs->bloom_keys_nr);
715
716         fill_bloom_key(path, len, &revs->bloom_keys[0],
717                        revs->bloom_filter_settings);
718         path_component_nr = 1;
719
720         p = path + len - 1;
721         while (p > path) {
722                 if (*p == '/')
723                         fill_bloom_key(path, p - path,
724                                        &revs->bloom_keys[path_component_nr++],
725                                        revs->bloom_filter_settings);
726                 p--;
727         }
728
729         if (trace2_is_enabled() && !bloom_filter_atexit_registered) {
730                 atexit(trace2_bloom_filter_statistics_atexit);
731                 bloom_filter_atexit_registered = 1;
732         }
733
734         free(path_alloc);
735 }
736
737 static int check_maybe_different_in_bloom_filter(struct rev_info *revs,
738                                                  struct commit *commit)
739 {
740         struct bloom_filter *filter;
741         int result = 1, j;
742
743         if (!revs->repo->objects->commit_graph)
744                 return -1;
745
746         if (commit_graph_generation(commit) == GENERATION_NUMBER_INFINITY)
747                 return -1;
748
749         filter = get_bloom_filter(revs->repo, commit);
750
751         if (!filter) {
752                 count_bloom_filter_not_present++;
753                 return -1;
754         }
755
756         for (j = 0; result && j < revs->bloom_keys_nr; j++) {
757                 result = bloom_filter_contains(filter,
758                                                &revs->bloom_keys[j],
759                                                revs->bloom_filter_settings);
760         }
761
762         if (result)
763                 count_bloom_filter_maybe++;
764         else
765                 count_bloom_filter_definitely_not++;
766
767         return result;
768 }
769
770 static int rev_compare_tree(struct rev_info *revs,
771                             struct commit *parent, struct commit *commit, int nth_parent)
772 {
773         struct tree *t1 = get_commit_tree(parent);
774         struct tree *t2 = get_commit_tree(commit);
775         int bloom_ret = 1;
776
777         if (!t1)
778                 return REV_TREE_NEW;
779         if (!t2)
780                 return REV_TREE_OLD;
781
782         if (revs->simplify_by_decoration) {
783                 /*
784                  * If we are simplifying by decoration, then the commit
785                  * is worth showing if it has a tag pointing at it.
786                  */
787                 if (get_name_decoration(&commit->object))
788                         return REV_TREE_DIFFERENT;
789                 /*
790                  * A commit that is not pointed by a tag is uninteresting
791                  * if we are not limited by path.  This means that you will
792                  * see the usual "commits that touch the paths" plus any
793                  * tagged commit by specifying both --simplify-by-decoration
794                  * and pathspec.
795                  */
796                 if (!revs->prune_data.nr)
797                         return REV_TREE_SAME;
798         }
799
800         if (revs->bloom_keys_nr && !nth_parent) {
801                 bloom_ret = check_maybe_different_in_bloom_filter(revs, commit);
802
803                 if (bloom_ret == 0)
804                         return REV_TREE_SAME;
805         }
806
807         tree_difference = REV_TREE_SAME;
808         revs->pruning.flags.has_changes = 0;
809         diff_tree_oid(&t1->object.oid, &t2->object.oid, "", &revs->pruning);
810
811         if (!nth_parent)
812                 if (bloom_ret == 1 && tree_difference == REV_TREE_SAME)
813                         count_bloom_filter_false_positive++;
814
815         return tree_difference;
816 }
817
818 static int rev_same_tree_as_empty(struct rev_info *revs, struct commit *commit)
819 {
820         struct tree *t1 = get_commit_tree(commit);
821
822         if (!t1)
823                 return 0;
824
825         tree_difference = REV_TREE_SAME;
826         revs->pruning.flags.has_changes = 0;
827         diff_tree_oid(NULL, &t1->object.oid, "", &revs->pruning);
828
829         return tree_difference == REV_TREE_SAME;
830 }
831
832 struct treesame_state {
833         unsigned int nparents;
834         unsigned char treesame[FLEX_ARRAY];
835 };
836
837 static struct treesame_state *initialise_treesame(struct rev_info *revs, struct commit *commit)
838 {
839         unsigned n = commit_list_count(commit->parents);
840         struct treesame_state *st = xcalloc(1, st_add(sizeof(*st), n));
841         st->nparents = n;
842         add_decoration(&revs->treesame, &commit->object, st);
843         return st;
844 }
845
846 /*
847  * Must be called immediately after removing the nth_parent from a commit's
848  * parent list, if we are maintaining the per-parent treesame[] decoration.
849  * This does not recalculate the master TREESAME flag - update_treesame()
850  * should be called to update it after a sequence of treesame[] modifications
851  * that may have affected it.
852  */
853 static int compact_treesame(struct rev_info *revs, struct commit *commit, unsigned nth_parent)
854 {
855         struct treesame_state *st;
856         int old_same;
857
858         if (!commit->parents) {
859                 /*
860                  * Have just removed the only parent from a non-merge.
861                  * Different handling, as we lack decoration.
862                  */
863                 if (nth_parent != 0)
864                         die("compact_treesame %u", nth_parent);
865                 old_same = !!(commit->object.flags & TREESAME);
866                 if (rev_same_tree_as_empty(revs, commit))
867                         commit->object.flags |= TREESAME;
868                 else
869                         commit->object.flags &= ~TREESAME;
870                 return old_same;
871         }
872
873         st = lookup_decoration(&revs->treesame, &commit->object);
874         if (!st || nth_parent >= st->nparents)
875                 die("compact_treesame %u", nth_parent);
876
877         old_same = st->treesame[nth_parent];
878         memmove(st->treesame + nth_parent,
879                 st->treesame + nth_parent + 1,
880                 st->nparents - nth_parent - 1);
881
882         /*
883          * If we've just become a non-merge commit, update TREESAME
884          * immediately, and remove the no-longer-needed decoration.
885          * If still a merge, defer update until update_treesame().
886          */
887         if (--st->nparents == 1) {
888                 if (commit->parents->next)
889                         die("compact_treesame parents mismatch");
890                 if (st->treesame[0] && revs->dense)
891                         commit->object.flags |= TREESAME;
892                 else
893                         commit->object.flags &= ~TREESAME;
894                 free(add_decoration(&revs->treesame, &commit->object, NULL));
895         }
896
897         return old_same;
898 }
899
900 static unsigned update_treesame(struct rev_info *revs, struct commit *commit)
901 {
902         if (commit->parents && commit->parents->next) {
903                 unsigned n;
904                 struct treesame_state *st;
905                 struct commit_list *p;
906                 unsigned relevant_parents;
907                 unsigned relevant_change, irrelevant_change;
908
909                 st = lookup_decoration(&revs->treesame, &commit->object);
910                 if (!st)
911                         die("update_treesame %s", oid_to_hex(&commit->object.oid));
912                 relevant_parents = 0;
913                 relevant_change = irrelevant_change = 0;
914                 for (p = commit->parents, n = 0; p; n++, p = p->next) {
915                         if (relevant_commit(p->item)) {
916                                 relevant_change |= !st->treesame[n];
917                                 relevant_parents++;
918                         } else
919                                 irrelevant_change |= !st->treesame[n];
920                 }
921                 if (relevant_parents ? relevant_change : irrelevant_change)
922                         commit->object.flags &= ~TREESAME;
923                 else
924                         commit->object.flags |= TREESAME;
925         }
926
927         return commit->object.flags & TREESAME;
928 }
929
930 static inline int limiting_can_increase_treesame(const struct rev_info *revs)
931 {
932         /*
933          * TREESAME is irrelevant unless prune && dense;
934          * if simplify_history is set, we can't have a mixture of TREESAME and
935          *    !TREESAME INTERESTING parents (and we don't have treesame[]
936          *    decoration anyway);
937          * if first_parent_only is set, then the TREESAME flag is locked
938          *    against the first parent (and again we lack treesame[] decoration).
939          */
940         return revs->prune && revs->dense &&
941                !revs->simplify_history &&
942                !revs->first_parent_only;
943 }
944
945 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
946 {
947         struct commit_list **pp, *parent;
948         struct treesame_state *ts = NULL;
949         int relevant_change = 0, irrelevant_change = 0;
950         int relevant_parents, nth_parent;
951
952         /*
953          * If we don't do pruning, everything is interesting
954          */
955         if (!revs->prune)
956                 return;
957
958         if (!get_commit_tree(commit))
959                 return;
960
961         if (!commit->parents) {
962                 if (rev_same_tree_as_empty(revs, commit))
963                         commit->object.flags |= TREESAME;
964                 return;
965         }
966
967         /*
968          * Normal non-merge commit? If we don't want to make the
969          * history dense, we consider it always to be a change..
970          */
971         if (!revs->dense && !commit->parents->next)
972                 return;
973
974         for (pp = &commit->parents, nth_parent = 0, relevant_parents = 0;
975              (parent = *pp) != NULL;
976              pp = &parent->next, nth_parent++) {
977                 struct commit *p = parent->item;
978                 if (relevant_commit(p))
979                         relevant_parents++;
980
981                 if (nth_parent == 1) {
982                         /*
983                          * This our second loop iteration - so we now know
984                          * we're dealing with a merge.
985                          *
986                          * Do not compare with later parents when we care only about
987                          * the first parent chain, in order to avoid derailing the
988                          * traversal to follow a side branch that brought everything
989                          * in the path we are limited to by the pathspec.
990                          */
991                         if (revs->first_parent_only)
992                                 break;
993                         /*
994                          * If this will remain a potentially-simplifiable
995                          * merge, remember per-parent treesame if needed.
996                          * Initialise the array with the comparison from our
997                          * first iteration.
998                          */
999                         if (revs->treesame.name &&
1000                             !revs->simplify_history &&
1001                             !(commit->object.flags & UNINTERESTING)) {
1002                                 ts = initialise_treesame(revs, commit);
1003                                 if (!(irrelevant_change || relevant_change))
1004                                         ts->treesame[0] = 1;
1005                         }
1006                 }
1007                 if (repo_parse_commit(revs->repo, p) < 0)
1008                         die("cannot simplify commit %s (because of %s)",
1009                             oid_to_hex(&commit->object.oid),
1010                             oid_to_hex(&p->object.oid));
1011                 switch (rev_compare_tree(revs, p, commit, nth_parent)) {
1012                 case REV_TREE_SAME:
1013                         if (!revs->simplify_history || !relevant_commit(p)) {
1014                                 /* Even if a merge with an uninteresting
1015                                  * side branch brought the entire change
1016                                  * we are interested in, we do not want
1017                                  * to lose the other branches of this
1018                                  * merge, so we just keep going.
1019                                  */
1020                                 if (ts)
1021                                         ts->treesame[nth_parent] = 1;
1022                                 continue;
1023                         }
1024                         parent->next = NULL;
1025                         commit->parents = parent;
1026
1027                         /*
1028                          * A merge commit is a "diversion" if it is not
1029                          * TREESAME to its first parent but is TREESAME
1030                          * to a later parent. In the simplified history,
1031                          * we "divert" the history walk to the later
1032                          * parent. These commits are shown when "show_pulls"
1033                          * is enabled, so do not mark the object as
1034                          * TREESAME here.
1035                          */
1036                         if (!revs->show_pulls || !nth_parent)
1037                                 commit->object.flags |= TREESAME;
1038
1039                         return;
1040
1041                 case REV_TREE_NEW:
1042                         if (revs->remove_empty_trees &&
1043                             rev_same_tree_as_empty(revs, p)) {
1044                                 /* We are adding all the specified
1045                                  * paths from this parent, so the
1046                                  * history beyond this parent is not
1047                                  * interesting.  Remove its parents
1048                                  * (they are grandparents for us).
1049                                  * IOW, we pretend this parent is a
1050                                  * "root" commit.
1051                                  */
1052                                 if (repo_parse_commit(revs->repo, p) < 0)
1053                                         die("cannot simplify commit %s (invalid %s)",
1054                                             oid_to_hex(&commit->object.oid),
1055                                             oid_to_hex(&p->object.oid));
1056                                 p->parents = NULL;
1057                         }
1058                 /* fallthrough */
1059                 case REV_TREE_OLD:
1060                 case REV_TREE_DIFFERENT:
1061                         if (relevant_commit(p))
1062                                 relevant_change = 1;
1063                         else
1064                                 irrelevant_change = 1;
1065
1066                         if (!nth_parent)
1067                                 commit->object.flags |= PULL_MERGE;
1068
1069                         continue;
1070                 }
1071                 die("bad tree compare for commit %s", oid_to_hex(&commit->object.oid));
1072         }
1073
1074         /*
1075          * TREESAME is straightforward for single-parent commits. For merge
1076          * commits, it is most useful to define it so that "irrelevant"
1077          * parents cannot make us !TREESAME - if we have any relevant
1078          * parents, then we only consider TREESAMEness with respect to them,
1079          * allowing irrelevant merges from uninteresting branches to be
1080          * simplified away. Only if we have only irrelevant parents do we
1081          * base TREESAME on them. Note that this logic is replicated in
1082          * update_treesame, which should be kept in sync.
1083          */
1084         if (relevant_parents ? !relevant_change : !irrelevant_change)
1085                 commit->object.flags |= TREESAME;
1086 }
1087
1088 static int process_parents(struct rev_info *revs, struct commit *commit,
1089                            struct commit_list **list, struct prio_queue *queue)
1090 {
1091         struct commit_list *parent = commit->parents;
1092         unsigned left_flag;
1093
1094         if (commit->object.flags & ADDED)
1095                 return 0;
1096         commit->object.flags |= ADDED;
1097
1098         if (revs->include_check &&
1099             !revs->include_check(commit, revs->include_check_data))
1100                 return 0;
1101
1102         /*
1103          * If the commit is uninteresting, don't try to
1104          * prune parents - we want the maximal uninteresting
1105          * set.
1106          *
1107          * Normally we haven't parsed the parent
1108          * yet, so we won't have a parent of a parent
1109          * here. However, it may turn out that we've
1110          * reached this commit some other way (where it
1111          * wasn't uninteresting), in which case we need
1112          * to mark its parents recursively too..
1113          */
1114         if (commit->object.flags & UNINTERESTING) {
1115                 while (parent) {
1116                         struct commit *p = parent->item;
1117                         parent = parent->next;
1118                         if (p)
1119                                 p->object.flags |= UNINTERESTING;
1120                         if (repo_parse_commit_gently(revs->repo, p, 1) < 0)
1121                                 continue;
1122                         if (p->parents)
1123                                 mark_parents_uninteresting(p);
1124                         if (p->object.flags & SEEN)
1125                                 continue;
1126                         p->object.flags |= SEEN;
1127                         if (list)
1128                                 commit_list_insert_by_date(p, list);
1129                         if (queue)
1130                                 prio_queue_put(queue, p);
1131                 }
1132                 return 0;
1133         }
1134
1135         /*
1136          * Ok, the commit wasn't uninteresting. Try to
1137          * simplify the commit history and find the parent
1138          * that has no differences in the path set if one exists.
1139          */
1140         try_to_simplify_commit(revs, commit);
1141
1142         if (revs->no_walk)
1143                 return 0;
1144
1145         left_flag = (commit->object.flags & SYMMETRIC_LEFT);
1146
1147         for (parent = commit->parents; parent; parent = parent->next) {
1148                 struct commit *p = parent->item;
1149                 int gently = revs->ignore_missing_links ||
1150                              revs->exclude_promisor_objects;
1151                 if (repo_parse_commit_gently(revs->repo, p, gently) < 0) {
1152                         if (revs->exclude_promisor_objects &&
1153                             is_promisor_object(&p->object.oid)) {
1154                                 if (revs->first_parent_only)
1155                                         break;
1156                                 continue;
1157                         }
1158                         return -1;
1159                 }
1160                 if (revs->sources) {
1161                         char **slot = revision_sources_at(revs->sources, p);
1162
1163                         if (!*slot)
1164                                 *slot = *revision_sources_at(revs->sources, commit);
1165                 }
1166                 p->object.flags |= left_flag;
1167                 if (!(p->object.flags & SEEN)) {
1168                         p->object.flags |= SEEN;
1169                         if (list)
1170                                 commit_list_insert_by_date(p, list);
1171                         if (queue)
1172                                 prio_queue_put(queue, p);
1173                 }
1174                 if (revs->first_parent_only)
1175                         break;
1176         }
1177         return 0;
1178 }
1179
1180 static void cherry_pick_list(struct commit_list *list, struct rev_info *revs)
1181 {
1182         struct commit_list *p;
1183         int left_count = 0, right_count = 0;
1184         int left_first;
1185         struct patch_ids ids;
1186         unsigned cherry_flag;
1187
1188         /* First count the commits on the left and on the right */
1189         for (p = list; p; p = p->next) {
1190                 struct commit *commit = p->item;
1191                 unsigned flags = commit->object.flags;
1192                 if (flags & BOUNDARY)
1193                         ;
1194                 else if (flags & SYMMETRIC_LEFT)
1195                         left_count++;
1196                 else
1197                         right_count++;
1198         }
1199
1200         if (!left_count || !right_count)
1201                 return;
1202
1203         left_first = left_count < right_count;
1204         init_patch_ids(revs->repo, &ids);
1205         ids.diffopts.pathspec = revs->diffopt.pathspec;
1206
1207         /* Compute patch-ids for one side */
1208         for (p = list; p; p = p->next) {
1209                 struct commit *commit = p->item;
1210                 unsigned flags = commit->object.flags;
1211
1212                 if (flags & BOUNDARY)
1213                         continue;
1214                 /*
1215                  * If we have fewer left, left_first is set and we omit
1216                  * commits on the right branch in this loop.  If we have
1217                  * fewer right, we skip the left ones.
1218                  */
1219                 if (left_first != !!(flags & SYMMETRIC_LEFT))
1220                         continue;
1221                 add_commit_patch_id(commit, &ids);
1222         }
1223
1224         /* either cherry_mark or cherry_pick are true */
1225         cherry_flag = revs->cherry_mark ? PATCHSAME : SHOWN;
1226
1227         /* Check the other side */
1228         for (p = list; p; p = p->next) {
1229                 struct commit *commit = p->item;
1230                 struct patch_id *id;
1231                 unsigned flags = commit->object.flags;
1232
1233                 if (flags & BOUNDARY)
1234                         continue;
1235                 /*
1236                  * If we have fewer left, left_first is set and we omit
1237                  * commits on the left branch in this loop.
1238                  */
1239                 if (left_first == !!(flags & SYMMETRIC_LEFT))
1240                         continue;
1241
1242                 /*
1243                  * Have we seen the same patch id?
1244                  */
1245                 id = patch_id_iter_first(commit, &ids);
1246                 if (!id)
1247                         continue;
1248
1249                 commit->object.flags |= cherry_flag;
1250                 do {
1251                         id->commit->object.flags |= cherry_flag;
1252                 } while ((id = patch_id_iter_next(id, &ids)));
1253         }
1254
1255         free_patch_ids(&ids);
1256 }
1257
1258 /* How many extra uninteresting commits we want to see.. */
1259 #define SLOP 5
1260
1261 static int still_interesting(struct commit_list *src, timestamp_t date, int slop,
1262                              struct commit **interesting_cache)
1263 {
1264         /*
1265          * No source list at all? We're definitely done..
1266          */
1267         if (!src)
1268                 return 0;
1269
1270         /*
1271          * Does the destination list contain entries with a date
1272          * before the source list? Definitely _not_ done.
1273          */
1274         if (date <= src->item->date)
1275                 return SLOP;
1276
1277         /*
1278          * Does the source list still have interesting commits in
1279          * it? Definitely not done..
1280          */
1281         if (!everybody_uninteresting(src, interesting_cache))
1282                 return SLOP;
1283
1284         /* Ok, we're closing in.. */
1285         return slop-1;
1286 }
1287
1288 /*
1289  * "rev-list --ancestry-path A..B" computes commits that are ancestors
1290  * of B but not ancestors of A but further limits the result to those
1291  * that are descendants of A.  This takes the list of bottom commits and
1292  * the result of "A..B" without --ancestry-path, and limits the latter
1293  * further to the ones that can reach one of the commits in "bottom".
1294  */
1295 static void limit_to_ancestry(struct commit_list *bottom, struct commit_list *list)
1296 {
1297         struct commit_list *p;
1298         struct commit_list *rlist = NULL;
1299         int made_progress;
1300
1301         /*
1302          * Reverse the list so that it will be likely that we would
1303          * process parents before children.
1304          */
1305         for (p = list; p; p = p->next)
1306                 commit_list_insert(p->item, &rlist);
1307
1308         for (p = bottom; p; p = p->next)
1309                 p->item->object.flags |= TMP_MARK;
1310
1311         /*
1312          * Mark the ones that can reach bottom commits in "list",
1313          * in a bottom-up fashion.
1314          */
1315         do {
1316                 made_progress = 0;
1317                 for (p = rlist; p; p = p->next) {
1318                         struct commit *c = p->item;
1319                         struct commit_list *parents;
1320                         if (c->object.flags & (TMP_MARK | UNINTERESTING))
1321                                 continue;
1322                         for (parents = c->parents;
1323                              parents;
1324                              parents = parents->next) {
1325                                 if (!(parents->item->object.flags & TMP_MARK))
1326                                         continue;
1327                                 c->object.flags |= TMP_MARK;
1328                                 made_progress = 1;
1329                                 break;
1330                         }
1331                 }
1332         } while (made_progress);
1333
1334         /*
1335          * NEEDSWORK: decide if we want to remove parents that are
1336          * not marked with TMP_MARK from commit->parents for commits
1337          * in the resulting list.  We may not want to do that, though.
1338          */
1339
1340         /*
1341          * The ones that are not marked with TMP_MARK are uninteresting
1342          */
1343         for (p = list; p; p = p->next) {
1344                 struct commit *c = p->item;
1345                 if (c->object.flags & TMP_MARK)
1346                         continue;
1347                 c->object.flags |= UNINTERESTING;
1348         }
1349
1350         /* We are done with the TMP_MARK */
1351         for (p = list; p; p = p->next)
1352                 p->item->object.flags &= ~TMP_MARK;
1353         for (p = bottom; p; p = p->next)
1354                 p->item->object.flags &= ~TMP_MARK;
1355         free_commit_list(rlist);
1356 }
1357
1358 /*
1359  * Before walking the history, keep the set of "negative" refs the
1360  * caller has asked to exclude.
1361  *
1362  * This is used to compute "rev-list --ancestry-path A..B", as we need
1363  * to filter the result of "A..B" further to the ones that can actually
1364  * reach A.
1365  */
1366 static struct commit_list *collect_bottom_commits(struct commit_list *list)
1367 {
1368         struct commit_list *elem, *bottom = NULL;
1369         for (elem = list; elem; elem = elem->next)
1370                 if (elem->item->object.flags & BOTTOM)
1371                         commit_list_insert(elem->item, &bottom);
1372         return bottom;
1373 }
1374
1375 /* Assumes either left_only or right_only is set */
1376 static void limit_left_right(struct commit_list *list, struct rev_info *revs)
1377 {
1378         struct commit_list *p;
1379
1380         for (p = list; p; p = p->next) {
1381                 struct commit *commit = p->item;
1382
1383                 if (revs->right_only) {
1384                         if (commit->object.flags & SYMMETRIC_LEFT)
1385                                 commit->object.flags |= SHOWN;
1386                 } else  /* revs->left_only is set */
1387                         if (!(commit->object.flags & SYMMETRIC_LEFT))
1388                                 commit->object.flags |= SHOWN;
1389         }
1390 }
1391
1392 static int limit_list(struct rev_info *revs)
1393 {
1394         int slop = SLOP;
1395         timestamp_t date = TIME_MAX;
1396         struct commit_list *list = revs->commits;
1397         struct commit_list *newlist = NULL;
1398         struct commit_list **p = &newlist;
1399         struct commit_list *bottom = NULL;
1400         struct commit *interesting_cache = NULL;
1401
1402         if (revs->ancestry_path) {
1403                 bottom = collect_bottom_commits(list);
1404                 if (!bottom)
1405                         die("--ancestry-path given but there are no bottom commits");
1406         }
1407
1408         while (list) {
1409                 struct commit *commit = pop_commit(&list);
1410                 struct object *obj = &commit->object;
1411                 show_early_output_fn_t show;
1412
1413                 if (commit == interesting_cache)
1414                         interesting_cache = NULL;
1415
1416                 if (revs->max_age != -1 && (commit->date < revs->max_age))
1417                         obj->flags |= UNINTERESTING;
1418                 if (process_parents(revs, commit, &list, NULL) < 0)
1419                         return -1;
1420                 if (obj->flags & UNINTERESTING) {
1421                         mark_parents_uninteresting(commit);
1422                         slop = still_interesting(list, date, slop, &interesting_cache);
1423                         if (slop)
1424                                 continue;
1425                         break;
1426                 }
1427                 if (revs->min_age != -1 && (commit->date > revs->min_age) &&
1428                     !revs->line_level_traverse)
1429                         continue;
1430                 date = commit->date;
1431                 p = &commit_list_insert(commit, p)->next;
1432
1433                 show = show_early_output;
1434                 if (!show)
1435                         continue;
1436
1437                 show(revs, newlist);
1438                 show_early_output = NULL;
1439         }
1440         if (revs->cherry_pick || revs->cherry_mark)
1441                 cherry_pick_list(newlist, revs);
1442
1443         if (revs->left_only || revs->right_only)
1444                 limit_left_right(newlist, revs);
1445
1446         if (bottom) {
1447                 limit_to_ancestry(bottom, newlist);
1448                 free_commit_list(bottom);
1449         }
1450
1451         /*
1452          * Check if any commits have become TREESAME by some of their parents
1453          * becoming UNINTERESTING.
1454          */
1455         if (limiting_can_increase_treesame(revs))
1456                 for (list = newlist; list; list = list->next) {
1457                         struct commit *c = list->item;
1458                         if (c->object.flags & (UNINTERESTING | TREESAME))
1459                                 continue;
1460                         update_treesame(revs, c);
1461                 }
1462
1463         revs->commits = newlist;
1464         return 0;
1465 }
1466
1467 /*
1468  * Add an entry to refs->cmdline with the specified information.
1469  * *name is copied.
1470  */
1471 static void add_rev_cmdline(struct rev_info *revs,
1472                             struct object *item,
1473                             const char *name,
1474                             int whence,
1475                             unsigned flags)
1476 {
1477         struct rev_cmdline_info *info = &revs->cmdline;
1478         unsigned int nr = info->nr;
1479
1480         ALLOC_GROW(info->rev, nr + 1, info->alloc);
1481         info->rev[nr].item = item;
1482         info->rev[nr].name = xstrdup(name);
1483         info->rev[nr].whence = whence;
1484         info->rev[nr].flags = flags;
1485         info->nr++;
1486 }
1487
1488 static void add_rev_cmdline_list(struct rev_info *revs,
1489                                  struct commit_list *commit_list,
1490                                  int whence,
1491                                  unsigned flags)
1492 {
1493         while (commit_list) {
1494                 struct object *object = &commit_list->item->object;
1495                 add_rev_cmdline(revs, object, oid_to_hex(&object->oid),
1496                                 whence, flags);
1497                 commit_list = commit_list->next;
1498         }
1499 }
1500
1501 struct all_refs_cb {
1502         int all_flags;
1503         int warned_bad_reflog;
1504         struct rev_info *all_revs;
1505         const char *name_for_errormsg;
1506         struct worktree *wt;
1507 };
1508
1509 int ref_excluded(struct string_list *ref_excludes, const char *path)
1510 {
1511         struct string_list_item *item;
1512
1513         if (!ref_excludes)
1514                 return 0;
1515         for_each_string_list_item(item, ref_excludes) {
1516                 if (!wildmatch(item->string, path, 0))
1517                         return 1;
1518         }
1519         return 0;
1520 }
1521
1522 static int handle_one_ref(const char *path, const struct object_id *oid,
1523                           int flag, void *cb_data)
1524 {
1525         struct all_refs_cb *cb = cb_data;
1526         struct object *object;
1527
1528         if (ref_excluded(cb->all_revs->ref_excludes, path))
1529             return 0;
1530
1531         object = get_reference(cb->all_revs, path, oid, cb->all_flags);
1532         add_rev_cmdline(cb->all_revs, object, path, REV_CMD_REF, cb->all_flags);
1533         add_pending_oid(cb->all_revs, path, oid, cb->all_flags);
1534         return 0;
1535 }
1536
1537 static void init_all_refs_cb(struct all_refs_cb *cb, struct rev_info *revs,
1538         unsigned flags)
1539 {
1540         cb->all_revs = revs;
1541         cb->all_flags = flags;
1542         revs->rev_input_given = 1;
1543         cb->wt = NULL;
1544 }
1545
1546 void clear_ref_exclusion(struct string_list **ref_excludes_p)
1547 {
1548         if (*ref_excludes_p) {
1549                 string_list_clear(*ref_excludes_p, 0);
1550                 free(*ref_excludes_p);
1551         }
1552         *ref_excludes_p = NULL;
1553 }
1554
1555 void add_ref_exclusion(struct string_list **ref_excludes_p, const char *exclude)
1556 {
1557         if (!*ref_excludes_p) {
1558                 *ref_excludes_p = xcalloc(1, sizeof(**ref_excludes_p));
1559                 (*ref_excludes_p)->strdup_strings = 1;
1560         }
1561         string_list_append(*ref_excludes_p, exclude);
1562 }
1563
1564 static void handle_refs(struct ref_store *refs,
1565                         struct rev_info *revs, unsigned flags,
1566                         int (*for_each)(struct ref_store *, each_ref_fn, void *))
1567 {
1568         struct all_refs_cb cb;
1569
1570         if (!refs) {
1571                 /* this could happen with uninitialized submodules */
1572                 return;
1573         }
1574
1575         init_all_refs_cb(&cb, revs, flags);
1576         for_each(refs, handle_one_ref, &cb);
1577 }
1578
1579 static void handle_one_reflog_commit(struct object_id *oid, void *cb_data)
1580 {
1581         struct all_refs_cb *cb = cb_data;
1582         if (!is_null_oid(oid)) {
1583                 struct object *o = parse_object(cb->all_revs->repo, oid);
1584                 if (o) {
1585                         o->flags |= cb->all_flags;
1586                         /* ??? CMDLINEFLAGS ??? */
1587                         add_pending_object(cb->all_revs, o, "");
1588                 }
1589                 else if (!cb->warned_bad_reflog) {
1590                         warning("reflog of '%s' references pruned commits",
1591                                 cb->name_for_errormsg);
1592                         cb->warned_bad_reflog = 1;
1593                 }
1594         }
1595 }
1596
1597 static int handle_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
1598                 const char *email, timestamp_t timestamp, int tz,
1599                 const char *message, void *cb_data)
1600 {
1601         handle_one_reflog_commit(ooid, cb_data);
1602         handle_one_reflog_commit(noid, cb_data);
1603         return 0;
1604 }
1605
1606 static int handle_one_reflog(const char *refname_in_wt,
1607                              const struct object_id *oid,
1608                              int flag, void *cb_data)
1609 {
1610         struct all_refs_cb *cb = cb_data;
1611         struct strbuf refname = STRBUF_INIT;
1612
1613         cb->warned_bad_reflog = 0;
1614         strbuf_worktree_ref(cb->wt, &refname, refname_in_wt);
1615         cb->name_for_errormsg = refname.buf;
1616         refs_for_each_reflog_ent(get_main_ref_store(the_repository),
1617                                  refname.buf,
1618                                  handle_one_reflog_ent, cb_data);
1619         strbuf_release(&refname);
1620         return 0;
1621 }
1622
1623 static void add_other_reflogs_to_pending(struct all_refs_cb *cb)
1624 {
1625         struct worktree **worktrees, **p;
1626
1627         worktrees = get_worktrees();
1628         for (p = worktrees; *p; p++) {
1629                 struct worktree *wt = *p;
1630
1631                 if (wt->is_current)
1632                         continue;
1633
1634                 cb->wt = wt;
1635                 refs_for_each_reflog(get_worktree_ref_store(wt),
1636                                      handle_one_reflog,
1637                                      cb);
1638         }
1639         free_worktrees(worktrees);
1640 }
1641
1642 void add_reflogs_to_pending(struct rev_info *revs, unsigned flags)
1643 {
1644         struct all_refs_cb cb;
1645
1646         cb.all_revs = revs;
1647         cb.all_flags = flags;
1648         cb.wt = NULL;
1649         for_each_reflog(handle_one_reflog, &cb);
1650
1651         if (!revs->single_worktree)
1652                 add_other_reflogs_to_pending(&cb);
1653 }
1654
1655 static void add_cache_tree(struct cache_tree *it, struct rev_info *revs,
1656                            struct strbuf *path, unsigned int flags)
1657 {
1658         size_t baselen = path->len;
1659         int i;
1660
1661         if (it->entry_count >= 0) {
1662                 struct tree *tree = lookup_tree(revs->repo, &it->oid);
1663                 tree->object.flags |= flags;
1664                 add_pending_object_with_path(revs, &tree->object, "",
1665                                              040000, path->buf);
1666         }
1667
1668         for (i = 0; i < it->subtree_nr; i++) {
1669                 struct cache_tree_sub *sub = it->down[i];
1670                 strbuf_addf(path, "%s%s", baselen ? "/" : "", sub->name);
1671                 add_cache_tree(sub->cache_tree, revs, path, flags);
1672                 strbuf_setlen(path, baselen);
1673         }
1674
1675 }
1676
1677 static void do_add_index_objects_to_pending(struct rev_info *revs,
1678                                             struct index_state *istate,
1679                                             unsigned int flags)
1680 {
1681         int i;
1682
1683         for (i = 0; i < istate->cache_nr; i++) {
1684                 struct cache_entry *ce = istate->cache[i];
1685                 struct blob *blob;
1686
1687                 if (S_ISGITLINK(ce->ce_mode))
1688                         continue;
1689
1690                 blob = lookup_blob(revs->repo, &ce->oid);
1691                 if (!blob)
1692                         die("unable to add index blob to traversal");
1693                 blob->object.flags |= flags;
1694                 add_pending_object_with_path(revs, &blob->object, "",
1695                                              ce->ce_mode, ce->name);
1696         }
1697
1698         if (istate->cache_tree) {
1699                 struct strbuf path = STRBUF_INIT;
1700                 add_cache_tree(istate->cache_tree, revs, &path, flags);
1701                 strbuf_release(&path);
1702         }
1703 }
1704
1705 void add_index_objects_to_pending(struct rev_info *revs, unsigned int flags)
1706 {
1707         struct worktree **worktrees, **p;
1708
1709         repo_read_index(revs->repo);
1710         do_add_index_objects_to_pending(revs, revs->repo->index, flags);
1711
1712         if (revs->single_worktree)
1713                 return;
1714
1715         worktrees = get_worktrees();
1716         for (p = worktrees; *p; p++) {
1717                 struct worktree *wt = *p;
1718                 struct index_state istate = { NULL };
1719
1720                 if (wt->is_current)
1721                         continue; /* current index already taken care of */
1722
1723                 if (read_index_from(&istate,
1724                                     worktree_git_path(wt, "index"),
1725                                     get_worktree_git_dir(wt)) > 0)
1726                         do_add_index_objects_to_pending(revs, &istate, flags);
1727                 discard_index(&istate);
1728         }
1729         free_worktrees(worktrees);
1730 }
1731
1732 struct add_alternate_refs_data {
1733         struct rev_info *revs;
1734         unsigned int flags;
1735 };
1736
1737 static void add_one_alternate_ref(const struct object_id *oid,
1738                                   void *vdata)
1739 {
1740         const char *name = ".alternate";
1741         struct add_alternate_refs_data *data = vdata;
1742         struct object *obj;
1743
1744         obj = get_reference(data->revs, name, oid, data->flags);
1745         add_rev_cmdline(data->revs, obj, name, REV_CMD_REV, data->flags);
1746         add_pending_object(data->revs, obj, name);
1747 }
1748
1749 static void add_alternate_refs_to_pending(struct rev_info *revs,
1750                                           unsigned int flags)
1751 {
1752         struct add_alternate_refs_data data;
1753         data.revs = revs;
1754         data.flags = flags;
1755         for_each_alternate_ref(add_one_alternate_ref, &data);
1756 }
1757
1758 static int add_parents_only(struct rev_info *revs, const char *arg_, int flags,
1759                             int exclude_parent)
1760 {
1761         struct object_id oid;
1762         struct object *it;
1763         struct commit *commit;
1764         struct commit_list *parents;
1765         int parent_number;
1766         const char *arg = arg_;
1767
1768         if (*arg == '^') {
1769                 flags ^= UNINTERESTING | BOTTOM;
1770                 arg++;
1771         }
1772         if (get_oid_committish(arg, &oid))
1773                 return 0;
1774         while (1) {
1775                 it = get_reference(revs, arg, &oid, 0);
1776                 if (!it && revs->ignore_missing)
1777                         return 0;
1778                 if (it->type != OBJ_TAG)
1779                         break;
1780                 if (!((struct tag*)it)->tagged)
1781                         return 0;
1782                 oidcpy(&oid, &((struct tag*)it)->tagged->oid);
1783         }
1784         if (it->type != OBJ_COMMIT)
1785                 return 0;
1786         commit = (struct commit *)it;
1787         if (exclude_parent &&
1788             exclude_parent > commit_list_count(commit->parents))
1789                 return 0;
1790         for (parents = commit->parents, parent_number = 1;
1791              parents;
1792              parents = parents->next, parent_number++) {
1793                 if (exclude_parent && parent_number != exclude_parent)
1794                         continue;
1795
1796                 it = &parents->item->object;
1797                 it->flags |= flags;
1798                 add_rev_cmdline(revs, it, arg_, REV_CMD_PARENTS_ONLY, flags);
1799                 add_pending_object(revs, it, arg);
1800         }
1801         return 1;
1802 }
1803
1804 void repo_init_revisions(struct repository *r,
1805                          struct rev_info *revs,
1806                          const char *prefix)
1807 {
1808         memset(revs, 0, sizeof(*revs));
1809
1810         revs->repo = r;
1811         revs->abbrev = DEFAULT_ABBREV;
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 ((argcount = diff_merges_parse_opts(revs, argv))) {
2347                 return argcount;
2348         } else if (!strcmp(arg, "-v")) {
2349                 revs->verbose_header = 1;
2350         } else if (!strcmp(arg, "--pretty")) {
2351                 revs->verbose_header = 1;
2352                 revs->pretty_given = 1;
2353                 get_commit_format(NULL, revs);
2354         } else if (skip_prefix(arg, "--pretty=", &optarg) ||
2355                    skip_prefix(arg, "--format=", &optarg)) {
2356                 /*
2357                  * Detached form ("--pretty X" as opposed to "--pretty=X")
2358                  * not allowed, since the argument is optional.
2359                  */
2360                 revs->verbose_header = 1;
2361                 revs->pretty_given = 1;
2362                 get_commit_format(optarg, revs);
2363         } else if (!strcmp(arg, "--expand-tabs")) {
2364                 revs->expand_tabs_in_log = 8;
2365         } else if (!strcmp(arg, "--no-expand-tabs")) {
2366                 revs->expand_tabs_in_log = 0;
2367         } else if (skip_prefix(arg, "--expand-tabs=", &arg)) {
2368                 int val;
2369                 if (strtol_i(arg, 10, &val) < 0 || val < 0)
2370                         die("'%s': not a non-negative integer", arg);
2371                 revs->expand_tabs_in_log = val;
2372         } else if (!strcmp(arg, "--show-notes") || !strcmp(arg, "--notes")) {
2373                 enable_default_display_notes(&revs->notes_opt, &revs->show_notes);
2374                 revs->show_notes_given = 1;
2375         } else if (!strcmp(arg, "--show-signature")) {
2376                 revs->show_signature = 1;
2377         } else if (!strcmp(arg, "--no-show-signature")) {
2378                 revs->show_signature = 0;
2379         } else if (!strcmp(arg, "--show-linear-break")) {
2380                 revs->break_bar = "                    ..........";
2381                 revs->track_linear = 1;
2382                 revs->track_first_time = 1;
2383         } else if (skip_prefix(arg, "--show-linear-break=", &optarg)) {
2384                 revs->break_bar = xstrdup(optarg);
2385                 revs->track_linear = 1;
2386                 revs->track_first_time = 1;
2387         } else if (skip_prefix(arg, "--show-notes=", &optarg) ||
2388                    skip_prefix(arg, "--notes=", &optarg)) {
2389                 if (starts_with(arg, "--show-notes=") &&
2390                     revs->notes_opt.use_default_notes < 0)
2391                         revs->notes_opt.use_default_notes = 1;
2392                 enable_ref_display_notes(&revs->notes_opt, &revs->show_notes, optarg);
2393                 revs->show_notes_given = 1;
2394         } else if (!strcmp(arg, "--no-notes")) {
2395                 disable_display_notes(&revs->notes_opt, &revs->show_notes);
2396                 revs->show_notes_given = 1;
2397         } else if (!strcmp(arg, "--standard-notes")) {
2398                 revs->show_notes_given = 1;
2399                 revs->notes_opt.use_default_notes = 1;
2400         } else if (!strcmp(arg, "--no-standard-notes")) {
2401                 revs->notes_opt.use_default_notes = 0;
2402         } else if (!strcmp(arg, "--oneline")) {
2403                 revs->verbose_header = 1;
2404                 get_commit_format("oneline", revs);
2405                 revs->pretty_given = 1;
2406                 revs->abbrev_commit = 1;
2407         } else if (!strcmp(arg, "--graph")) {
2408                 revs->topo_order = 1;
2409                 revs->rewrite_parents = 1;
2410                 revs->graph = graph_init(revs);
2411         } else if (!strcmp(arg, "--encode-email-headers")) {
2412                 revs->encode_email_headers = 1;
2413         } else if (!strcmp(arg, "--no-encode-email-headers")) {
2414                 revs->encode_email_headers = 0;
2415         } else if (!strcmp(arg, "--root")) {
2416                 revs->show_root_diff = 1;
2417         } else if (!strcmp(arg, "--no-commit-id")) {
2418                 revs->no_commit_id = 1;
2419         } else if (!strcmp(arg, "--always")) {
2420                 revs->always_show_header = 1;
2421         } else if (!strcmp(arg, "--no-abbrev")) {
2422                 revs->abbrev = 0;
2423         } else if (!strcmp(arg, "--abbrev")) {
2424                 revs->abbrev = DEFAULT_ABBREV;
2425         } else if (skip_prefix(arg, "--abbrev=", &optarg)) {
2426                 revs->abbrev = strtoul(optarg, NULL, 10);
2427                 if (revs->abbrev < MINIMUM_ABBREV)
2428                         revs->abbrev = MINIMUM_ABBREV;
2429                 else if (revs->abbrev > hexsz)
2430                         revs->abbrev = hexsz;
2431         } else if (!strcmp(arg, "--abbrev-commit")) {
2432                 revs->abbrev_commit = 1;
2433                 revs->abbrev_commit_given = 1;
2434         } else if (!strcmp(arg, "--no-abbrev-commit")) {
2435                 revs->abbrev_commit = 0;
2436         } else if (!strcmp(arg, "--full-diff")) {
2437                 revs->diff = 1;
2438                 revs->full_diff = 1;
2439         } else if (!strcmp(arg, "--show-pulls")) {
2440                 revs->show_pulls = 1;
2441         } else if (!strcmp(arg, "--full-history")) {
2442                 revs->simplify_history = 0;
2443         } else if (!strcmp(arg, "--relative-date")) {
2444                 revs->date_mode.type = DATE_RELATIVE;
2445                 revs->date_mode_explicit = 1;
2446         } else if ((argcount = parse_long_opt("date", argv, &optarg))) {
2447                 parse_date_format(optarg, &revs->date_mode);
2448                 revs->date_mode_explicit = 1;
2449                 return argcount;
2450         } else if (!strcmp(arg, "--log-size")) {
2451                 revs->show_log_size = 1;
2452         }
2453         /*
2454          * Grepping the commit log
2455          */
2456         else if ((argcount = parse_long_opt("author", argv, &optarg))) {
2457                 add_header_grep(revs, GREP_HEADER_AUTHOR, optarg);
2458                 return argcount;
2459         } else if ((argcount = parse_long_opt("committer", argv, &optarg))) {
2460                 add_header_grep(revs, GREP_HEADER_COMMITTER, optarg);
2461                 return argcount;
2462         } else if ((argcount = parse_long_opt("grep-reflog", argv, &optarg))) {
2463                 add_header_grep(revs, GREP_HEADER_REFLOG, optarg);
2464                 return argcount;
2465         } else if ((argcount = parse_long_opt("grep", argv, &optarg))) {
2466                 add_message_grep(revs, optarg);
2467                 return argcount;
2468         } else if (!strcmp(arg, "--basic-regexp")) {
2469                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_BRE;
2470         } else if (!strcmp(arg, "--extended-regexp") || !strcmp(arg, "-E")) {
2471                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_ERE;
2472         } else if (!strcmp(arg, "--regexp-ignore-case") || !strcmp(arg, "-i")) {
2473                 revs->grep_filter.ignore_case = 1;
2474                 revs->diffopt.pickaxe_opts |= DIFF_PICKAXE_IGNORE_CASE;
2475         } else if (!strcmp(arg, "--fixed-strings") || !strcmp(arg, "-F")) {
2476                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_FIXED;
2477         } else if (!strcmp(arg, "--perl-regexp") || !strcmp(arg, "-P")) {
2478                 revs->grep_filter.pattern_type_option = GREP_PATTERN_TYPE_PCRE;
2479         } else if (!strcmp(arg, "--all-match")) {
2480                 revs->grep_filter.all_match = 1;
2481         } else if (!strcmp(arg, "--invert-grep")) {
2482                 revs->invert_grep = 1;
2483         } else if ((argcount = parse_long_opt("encoding", argv, &optarg))) {
2484                 if (strcmp(optarg, "none"))
2485                         git_log_output_encoding = xstrdup(optarg);
2486                 else
2487                         git_log_output_encoding = "";
2488                 return argcount;
2489         } else if (!strcmp(arg, "--reverse")) {
2490                 revs->reverse ^= 1;
2491         } else if (!strcmp(arg, "--children")) {
2492                 revs->children.name = "children";
2493                 revs->limited = 1;
2494         } else if (!strcmp(arg, "--ignore-missing")) {
2495                 revs->ignore_missing = 1;
2496         } else if (opt && opt->allow_exclude_promisor_objects &&
2497                    !strcmp(arg, "--exclude-promisor-objects")) {
2498                 if (fetch_if_missing)
2499                         BUG("exclude_promisor_objects can only be used when fetch_if_missing is 0");
2500                 revs->exclude_promisor_objects = 1;
2501         } else {
2502                 int opts = diff_opt_parse(&revs->diffopt, argv, argc, revs->prefix);
2503                 if (!opts)
2504                         unkv[(*unkc)++] = arg;
2505                 return opts;
2506         }
2507         if (revs->graph && revs->track_linear)
2508                 die("--show-linear-break and --graph are incompatible");
2509
2510         return 1;
2511 }
2512
2513 void parse_revision_opt(struct rev_info *revs, struct parse_opt_ctx_t *ctx,
2514                         const struct option *options,
2515                         const char * const usagestr[])
2516 {
2517         int n = handle_revision_opt(revs, ctx->argc, ctx->argv,
2518                                     &ctx->cpidx, ctx->out, NULL);
2519         if (n <= 0) {
2520                 error("unknown option `%s'", ctx->argv[0]);
2521                 usage_with_options(usagestr, options);
2522         }
2523         ctx->argv += n;
2524         ctx->argc -= n;
2525 }
2526
2527 static int for_each_bisect_ref(struct ref_store *refs, each_ref_fn fn,
2528                                void *cb_data, const char *term)
2529 {
2530         struct strbuf bisect_refs = STRBUF_INIT;
2531         int status;
2532         strbuf_addf(&bisect_refs, "refs/bisect/%s", term);
2533         status = refs_for_each_fullref_in(refs, bisect_refs.buf, fn, cb_data, 0);
2534         strbuf_release(&bisect_refs);
2535         return status;
2536 }
2537
2538 static int for_each_bad_bisect_ref(struct ref_store *refs, each_ref_fn fn, void *cb_data)
2539 {
2540         return for_each_bisect_ref(refs, fn, cb_data, term_bad);
2541 }
2542
2543 static int for_each_good_bisect_ref(struct ref_store *refs, each_ref_fn fn, void *cb_data)
2544 {
2545         return for_each_bisect_ref(refs, fn, cb_data, term_good);
2546 }
2547
2548 static int handle_revision_pseudo_opt(const char *submodule,
2549                                       struct rev_info *revs,
2550                                       const char **argv, int *flags)
2551 {
2552         const char *arg = argv[0];
2553         const char *optarg;
2554         struct ref_store *refs;
2555         int argcount;
2556
2557         if (submodule) {
2558                 /*
2559                  * We need some something like get_submodule_worktrees()
2560                  * before we can go through all worktrees of a submodule,
2561                  * .e.g with adding all HEADs from --all, which is not
2562                  * supported right now, so stick to single worktree.
2563                  */
2564                 if (!revs->single_worktree)
2565                         BUG("--single-worktree cannot be used together with submodule");
2566                 refs = get_submodule_ref_store(submodule);
2567         } else
2568                 refs = get_main_ref_store(revs->repo);
2569
2570         /*
2571          * NOTE!
2572          *
2573          * Commands like "git shortlog" will not accept the options below
2574          * unless parse_revision_opt queues them (as opposed to erroring
2575          * out).
2576          *
2577          * When implementing your new pseudo-option, remember to
2578          * register it in the list at the top of handle_revision_opt.
2579          */
2580         if (!strcmp(arg, "--all")) {
2581                 handle_refs(refs, revs, *flags, refs_for_each_ref);
2582                 handle_refs(refs, revs, *flags, refs_head_ref);
2583                 if (!revs->single_worktree) {
2584                         struct all_refs_cb cb;
2585
2586                         init_all_refs_cb(&cb, revs, *flags);
2587                         other_head_refs(handle_one_ref, &cb);
2588                 }
2589                 clear_ref_exclusion(&revs->ref_excludes);
2590         } else if (!strcmp(arg, "--branches")) {
2591                 handle_refs(refs, revs, *flags, refs_for_each_branch_ref);
2592                 clear_ref_exclusion(&revs->ref_excludes);
2593         } else if (!strcmp(arg, "--bisect")) {
2594                 read_bisect_terms(&term_bad, &term_good);
2595                 handle_refs(refs, revs, *flags, for_each_bad_bisect_ref);
2596                 handle_refs(refs, revs, *flags ^ (UNINTERESTING | BOTTOM),
2597                             for_each_good_bisect_ref);
2598                 revs->bisect = 1;
2599         } else if (!strcmp(arg, "--tags")) {
2600                 handle_refs(refs, revs, *flags, refs_for_each_tag_ref);
2601                 clear_ref_exclusion(&revs->ref_excludes);
2602         } else if (!strcmp(arg, "--remotes")) {
2603                 handle_refs(refs, revs, *flags, refs_for_each_remote_ref);
2604                 clear_ref_exclusion(&revs->ref_excludes);
2605         } else if ((argcount = parse_long_opt("glob", argv, &optarg))) {
2606                 struct all_refs_cb cb;
2607                 init_all_refs_cb(&cb, revs, *flags);
2608                 for_each_glob_ref(handle_one_ref, optarg, &cb);
2609                 clear_ref_exclusion(&revs->ref_excludes);
2610                 return argcount;
2611         } else if ((argcount = parse_long_opt("exclude", argv, &optarg))) {
2612                 add_ref_exclusion(&revs->ref_excludes, optarg);
2613                 return argcount;
2614         } else if (skip_prefix(arg, "--branches=", &optarg)) {
2615                 struct all_refs_cb cb;
2616                 init_all_refs_cb(&cb, revs, *flags);
2617                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/heads/", &cb);
2618                 clear_ref_exclusion(&revs->ref_excludes);
2619         } else if (skip_prefix(arg, "--tags=", &optarg)) {
2620                 struct all_refs_cb cb;
2621                 init_all_refs_cb(&cb, revs, *flags);
2622                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/tags/", &cb);
2623                 clear_ref_exclusion(&revs->ref_excludes);
2624         } else if (skip_prefix(arg, "--remotes=", &optarg)) {
2625                 struct all_refs_cb cb;
2626                 init_all_refs_cb(&cb, revs, *flags);
2627                 for_each_glob_ref_in(handle_one_ref, optarg, "refs/remotes/", &cb);
2628                 clear_ref_exclusion(&revs->ref_excludes);
2629         } else if (!strcmp(arg, "--reflog")) {
2630                 add_reflogs_to_pending(revs, *flags);
2631         } else if (!strcmp(arg, "--indexed-objects")) {
2632                 add_index_objects_to_pending(revs, *flags);
2633         } else if (!strcmp(arg, "--alternate-refs")) {
2634                 add_alternate_refs_to_pending(revs, *flags);
2635         } else if (!strcmp(arg, "--not")) {
2636                 *flags ^= UNINTERESTING | BOTTOM;
2637         } else if (!strcmp(arg, "--no-walk")) {
2638                 revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2639         } else if (skip_prefix(arg, "--no-walk=", &optarg)) {
2640                 /*
2641                  * Detached form ("--no-walk X" as opposed to "--no-walk=X")
2642                  * not allowed, since the argument is optional.
2643                  */
2644                 if (!strcmp(optarg, "sorted"))
2645                         revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2646                 else if (!strcmp(optarg, "unsorted"))
2647                         revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
2648                 else
2649                         return error("invalid argument to --no-walk");
2650         } else if (!strcmp(arg, "--do-walk")) {
2651                 revs->no_walk = 0;
2652         } else if (!strcmp(arg, "--single-worktree")) {
2653                 revs->single_worktree = 1;
2654         } else {
2655                 return 0;
2656         }
2657
2658         return 1;
2659 }
2660
2661 static void NORETURN diagnose_missing_default(const char *def)
2662 {
2663         int flags;
2664         const char *refname;
2665
2666         refname = resolve_ref_unsafe(def, 0, NULL, &flags);
2667         if (!refname || !(flags & REF_ISSYMREF) || (flags & REF_ISBROKEN))
2668                 die(_("your current branch appears to be broken"));
2669
2670         skip_prefix(refname, "refs/heads/", &refname);
2671         die(_("your current branch '%s' does not have any commits yet"),
2672             refname);
2673 }
2674
2675 /*
2676  * Parse revision information, filling in the "rev_info" structure,
2677  * and removing the used arguments from the argument list.
2678  *
2679  * Returns the number of arguments left that weren't recognized
2680  * (which are also moved to the head of the argument list)
2681  */
2682 int setup_revisions(int argc, const char **argv, struct rev_info *revs, struct setup_revision_opt *opt)
2683 {
2684         int i, flags, left, seen_dashdash, revarg_opt;
2685         struct strvec prune_data = STRVEC_INIT;
2686         const char *submodule = NULL;
2687         int seen_end_of_options = 0;
2688
2689         if (opt)
2690                 submodule = opt->submodule;
2691
2692         /* First, search for "--" */
2693         if (opt && opt->assume_dashdash) {
2694                 seen_dashdash = 1;
2695         } else {
2696                 seen_dashdash = 0;
2697                 for (i = 1; i < argc; i++) {
2698                         const char *arg = argv[i];
2699                         if (strcmp(arg, "--"))
2700                                 continue;
2701                         argv[i] = NULL;
2702                         argc = i;
2703                         if (argv[i + 1])
2704                                 strvec_pushv(&prune_data, argv + i + 1);
2705                         seen_dashdash = 1;
2706                         break;
2707                 }
2708         }
2709
2710         /* Second, deal with arguments and options */
2711         flags = 0;
2712         revarg_opt = opt ? opt->revarg_opt : 0;
2713         if (seen_dashdash)
2714                 revarg_opt |= REVARG_CANNOT_BE_FILENAME;
2715         for (left = i = 1; i < argc; i++) {
2716                 const char *arg = argv[i];
2717                 if (!seen_end_of_options && *arg == '-') {
2718                         int opts;
2719
2720                         opts = handle_revision_pseudo_opt(submodule,
2721                                                 revs, argv + i,
2722                                                 &flags);
2723                         if (opts > 0) {
2724                                 i += opts - 1;
2725                                 continue;
2726                         }
2727
2728                         if (!strcmp(arg, "--stdin")) {
2729                                 if (revs->disable_stdin) {
2730                                         argv[left++] = arg;
2731                                         continue;
2732                                 }
2733                                 if (revs->read_from_stdin++)
2734                                         die("--stdin given twice?");
2735                                 read_revisions_from_stdin(revs, &prune_data);
2736                                 continue;
2737                         }
2738
2739                         if (!strcmp(arg, "--end-of-options")) {
2740                                 seen_end_of_options = 1;
2741                                 continue;
2742                         }
2743
2744                         opts = handle_revision_opt(revs, argc - i, argv + i,
2745                                                    &left, argv, opt);
2746                         if (opts > 0) {
2747                                 i += opts - 1;
2748                                 continue;
2749                         }
2750                         if (opts < 0)
2751                                 exit(128);
2752                         continue;
2753                 }
2754
2755
2756                 if (handle_revision_arg(arg, revs, flags, revarg_opt)) {
2757                         int j;
2758                         if (seen_dashdash || *arg == '^')
2759                                 die("bad revision '%s'", arg);
2760
2761                         /* If we didn't have a "--":
2762                          * (1) all filenames must exist;
2763                          * (2) all rev-args must not be interpretable
2764                          *     as a valid filename.
2765                          * but the latter we have checked in the main loop.
2766                          */
2767                         for (j = i; j < argc; j++)
2768                                 verify_filename(revs->prefix, argv[j], j == i);
2769
2770                         strvec_pushv(&prune_data, argv + i);
2771                         break;
2772                 }
2773         }
2774
2775         if (prune_data.nr) {
2776                 /*
2777                  * If we need to introduce the magic "a lone ':' means no
2778                  * pathspec whatsoever", here is the place to do so.
2779                  *
2780                  * if (prune_data.nr == 1 && !strcmp(prune_data[0], ":")) {
2781                  *      prune_data.nr = 0;
2782                  *      prune_data.alloc = 0;
2783                  *      free(prune_data.path);
2784                  *      prune_data.path = NULL;
2785                  * } else {
2786                  *      terminate prune_data.alloc with NULL and
2787                  *      call init_pathspec() to set revs->prune_data here.
2788                  * }
2789                  */
2790                 parse_pathspec(&revs->prune_data, 0, 0,
2791                                revs->prefix, prune_data.v);
2792         }
2793         strvec_clear(&prune_data);
2794
2795         if (revs->def == NULL)
2796                 revs->def = opt ? opt->def : NULL;
2797         if (opt && opt->tweak)
2798                 opt->tweak(revs, opt);
2799         if (revs->show_merge)
2800                 prepare_show_merge(revs);
2801         if (revs->def && !revs->pending.nr && !revs->rev_input_given) {
2802                 struct object_id oid;
2803                 struct object *object;
2804                 struct object_context oc;
2805                 if (get_oid_with_context(revs->repo, revs->def, 0, &oid, &oc))
2806                         diagnose_missing_default(revs->def);
2807                 object = get_reference(revs, revs->def, &oid, 0);
2808                 add_pending_object_with_mode(revs, object, revs->def, oc.mode);
2809         }
2810
2811         /* Did the user ask for any diff output? Run the diff! */
2812         if (revs->diffopt.output_format & ~DIFF_FORMAT_NO_OUTPUT)
2813                 revs->diff = 1;
2814
2815         /* Pickaxe, diff-filter and rename following need diffs */
2816         if ((revs->diffopt.pickaxe_opts & DIFF_PICKAXE_KINDS_MASK) ||
2817             revs->diffopt.filter ||
2818             revs->diffopt.flags.follow_renames)
2819                 revs->diff = 1;
2820
2821         if (revs->diffopt.objfind)
2822                 revs->simplify_history = 0;
2823
2824         if (revs->line_level_traverse) {
2825                 if (want_ancestry(revs))
2826                         revs->limited = 1;
2827                 revs->topo_order = 1;
2828         }
2829
2830         if (revs->topo_order && !generation_numbers_enabled(the_repository))
2831                 revs->limited = 1;
2832
2833         if (revs->prune_data.nr) {
2834                 copy_pathspec(&revs->pruning.pathspec, &revs->prune_data);
2835                 /* Can't prune commits with rename following: the paths change.. */
2836                 if (!revs->diffopt.flags.follow_renames)
2837                         revs->prune = 1;
2838                 if (!revs->full_diff)
2839                         copy_pathspec(&revs->diffopt.pathspec,
2840                                       &revs->prune_data);
2841         }
2842
2843         diff_merges_setup_revs(revs);
2844
2845         revs->diffopt.abbrev = revs->abbrev;
2846
2847         diff_setup_done(&revs->diffopt);
2848
2849         grep_commit_pattern_type(GREP_PATTERN_TYPE_UNSPECIFIED,
2850                                  &revs->grep_filter);
2851         if (!is_encoding_utf8(get_log_output_encoding()))
2852                 revs->grep_filter.ignore_locale = 1;
2853         compile_grep_patterns(&revs->grep_filter);
2854
2855         if (revs->reverse && revs->reflog_info)
2856                 die("cannot combine --reverse with --walk-reflogs");
2857         if (revs->reflog_info && revs->limited)
2858                 die("cannot combine --walk-reflogs with history-limiting options");
2859         if (revs->rewrite_parents && revs->children.name)
2860                 die("cannot combine --parents and --children");
2861
2862         /*
2863          * Limitations on the graph functionality
2864          */
2865         if (revs->reverse && revs->graph)
2866                 die("cannot combine --reverse with --graph");
2867
2868         if (revs->reflog_info && revs->graph)
2869                 die("cannot combine --walk-reflogs with --graph");
2870         if (revs->no_walk && revs->graph)
2871                 die("cannot combine --no-walk with --graph");
2872         if (!revs->reflog_info && revs->grep_filter.use_reflog_filter)
2873                 die("cannot use --grep-reflog without --walk-reflogs");
2874
2875         if (revs->line_level_traverse &&
2876             (revs->diffopt.output_format & ~(DIFF_FORMAT_PATCH | DIFF_FORMAT_NO_OUTPUT)))
2877                 die(_("-L does not yet support diff formats besides -p and -s"));
2878
2879         if (revs->expand_tabs_in_log < 0)
2880                 revs->expand_tabs_in_log = revs->expand_tabs_in_log_default;
2881
2882         return left;
2883 }
2884
2885 static void add_child(struct rev_info *revs, struct commit *parent, struct commit *child)
2886 {
2887         struct commit_list *l = xcalloc(1, sizeof(*l));
2888
2889         l->item = child;
2890         l->next = add_decoration(&revs->children, &parent->object, l);
2891 }
2892
2893 static int remove_duplicate_parents(struct rev_info *revs, struct commit *commit)
2894 {
2895         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2896         struct commit_list **pp, *p;
2897         int surviving_parents;
2898
2899         /* Examine existing parents while marking ones we have seen... */
2900         pp = &commit->parents;
2901         surviving_parents = 0;
2902         while ((p = *pp) != NULL) {
2903                 struct commit *parent = p->item;
2904                 if (parent->object.flags & TMP_MARK) {
2905                         *pp = p->next;
2906                         if (ts)
2907                                 compact_treesame(revs, commit, surviving_parents);
2908                         continue;
2909                 }
2910                 parent->object.flags |= TMP_MARK;
2911                 surviving_parents++;
2912                 pp = &p->next;
2913         }
2914         /* clear the temporary mark */
2915         for (p = commit->parents; p; p = p->next) {
2916                 p->item->object.flags &= ~TMP_MARK;
2917         }
2918         /* no update_treesame() - removing duplicates can't affect TREESAME */
2919         return surviving_parents;
2920 }
2921
2922 struct merge_simplify_state {
2923         struct commit *simplified;
2924 };
2925
2926 static struct merge_simplify_state *locate_simplify_state(struct rev_info *revs, struct commit *commit)
2927 {
2928         struct merge_simplify_state *st;
2929
2930         st = lookup_decoration(&revs->merge_simplification, &commit->object);
2931         if (!st) {
2932                 st = xcalloc(1, sizeof(*st));
2933                 add_decoration(&revs->merge_simplification, &commit->object, st);
2934         }
2935         return st;
2936 }
2937
2938 static int mark_redundant_parents(struct commit *commit)
2939 {
2940         struct commit_list *h = reduce_heads(commit->parents);
2941         int i = 0, marked = 0;
2942         struct commit_list *po, *pn;
2943
2944         /* Want these for sanity-checking only */
2945         int orig_cnt = commit_list_count(commit->parents);
2946         int cnt = commit_list_count(h);
2947
2948         /*
2949          * Not ready to remove items yet, just mark them for now, based
2950          * on the output of reduce_heads(). reduce_heads outputs the reduced
2951          * set in its original order, so this isn't too hard.
2952          */
2953         po = commit->parents;
2954         pn = h;
2955         while (po) {
2956                 if (pn && po->item == pn->item) {
2957                         pn = pn->next;
2958                         i++;
2959                 } else {
2960                         po->item->object.flags |= TMP_MARK;
2961                         marked++;
2962                 }
2963                 po=po->next;
2964         }
2965
2966         if (i != cnt || cnt+marked != orig_cnt)
2967                 die("mark_redundant_parents %d %d %d %d", orig_cnt, cnt, i, marked);
2968
2969         free_commit_list(h);
2970
2971         return marked;
2972 }
2973
2974 static int mark_treesame_root_parents(struct commit *commit)
2975 {
2976         struct commit_list *p;
2977         int marked = 0;
2978
2979         for (p = commit->parents; p; p = p->next) {
2980                 struct commit *parent = p->item;
2981                 if (!parent->parents && (parent->object.flags & TREESAME)) {
2982                         parent->object.flags |= TMP_MARK;
2983                         marked++;
2984                 }
2985         }
2986
2987         return marked;
2988 }
2989
2990 /*
2991  * Awkward naming - this means one parent we are TREESAME to.
2992  * cf mark_treesame_root_parents: root parents that are TREESAME (to an
2993  * empty tree). Better name suggestions?
2994  */
2995 static int leave_one_treesame_to_parent(struct rev_info *revs, struct commit *commit)
2996 {
2997         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2998         struct commit *unmarked = NULL, *marked = NULL;
2999         struct commit_list *p;
3000         unsigned n;
3001
3002         for (p = commit->parents, n = 0; p; p = p->next, n++) {
3003                 if (ts->treesame[n]) {
3004                         if (p->item->object.flags & TMP_MARK) {
3005                                 if (!marked)
3006                                         marked = p->item;
3007                         } else {
3008                                 if (!unmarked) {
3009                                         unmarked = p->item;
3010                                         break;
3011                                 }
3012                         }
3013                 }
3014         }
3015
3016         /*
3017          * If we are TREESAME to a marked-for-deletion parent, but not to any
3018          * unmarked parents, unmark the first TREESAME parent. This is the
3019          * parent that the default simplify_history==1 scan would have followed,
3020          * and it doesn't make sense to omit that path when asking for a
3021          * simplified full history. Retaining it improves the chances of
3022          * understanding odd missed merges that took an old version of a file.
3023          *
3024          * Example:
3025          *
3026          *   I--------*X       A modified the file, but mainline merge X used
3027          *    \       /        "-s ours", so took the version from I. X is
3028          *     `-*A--'         TREESAME to I and !TREESAME to A.
3029          *
3030          * Default log from X would produce "I". Without this check,
3031          * --full-history --simplify-merges would produce "I-A-X", showing
3032          * the merge commit X and that it changed A, but not making clear that
3033          * it had just taken the I version. With this check, the topology above
3034          * is retained.
3035          *
3036          * Note that it is possible that the simplification chooses a different
3037          * TREESAME parent from the default, in which case this test doesn't
3038          * activate, and we _do_ drop the default parent. Example:
3039          *
3040          *   I------X         A modified the file, but it was reverted in B,
3041          *    \    /          meaning mainline merge X is TREESAME to both
3042          *    *A-*B           parents.
3043          *
3044          * Default log would produce "I" by following the first parent;
3045          * --full-history --simplify-merges will produce "I-A-B". But this is a
3046          * reasonable result - it presents a logical full history leading from
3047          * I to X, and X is not an important merge.
3048          */
3049         if (!unmarked && marked) {
3050                 marked->object.flags &= ~TMP_MARK;
3051                 return 1;
3052         }
3053
3054         return 0;
3055 }
3056
3057 static int remove_marked_parents(struct rev_info *revs, struct commit *commit)
3058 {
3059         struct commit_list **pp, *p;
3060         int nth_parent, removed = 0;
3061
3062         pp = &commit->parents;
3063         nth_parent = 0;
3064         while ((p = *pp) != NULL) {
3065                 struct commit *parent = p->item;
3066                 if (parent->object.flags & TMP_MARK) {
3067                         parent->object.flags &= ~TMP_MARK;
3068                         *pp = p->next;
3069                         free(p);
3070                         removed++;
3071                         compact_treesame(revs, commit, nth_parent);
3072                         continue;
3073                 }
3074                 pp = &p->next;
3075                 nth_parent++;
3076         }
3077
3078         /* Removing parents can only increase TREESAMEness */
3079         if (removed && !(commit->object.flags & TREESAME))
3080                 update_treesame(revs, commit);
3081
3082         return nth_parent;
3083 }
3084
3085 static struct commit_list **simplify_one(struct rev_info *revs, struct commit *commit, struct commit_list **tail)
3086 {
3087         struct commit_list *p;
3088         struct commit *parent;
3089         struct merge_simplify_state *st, *pst;
3090         int cnt;
3091
3092         st = locate_simplify_state(revs, commit);
3093
3094         /*
3095          * Have we handled this one?
3096          */
3097         if (st->simplified)
3098                 return tail;
3099
3100         /*
3101          * An UNINTERESTING commit simplifies to itself, so does a
3102          * root commit.  We do not rewrite parents of such commit
3103          * anyway.
3104          */
3105         if ((commit->object.flags & UNINTERESTING) || !commit->parents) {
3106                 st->simplified = commit;
3107                 return tail;
3108         }
3109
3110         /*
3111          * Do we know what commit all of our parents that matter
3112          * should be rewritten to?  Otherwise we are not ready to
3113          * rewrite this one yet.
3114          */
3115         for (cnt = 0, p = commit->parents; p; p = p->next) {
3116                 pst = locate_simplify_state(revs, p->item);
3117                 if (!pst->simplified) {
3118                         tail = &commit_list_insert(p->item, tail)->next;
3119                         cnt++;
3120                 }
3121                 if (revs->first_parent_only)
3122                         break;
3123         }
3124         if (cnt) {
3125                 tail = &commit_list_insert(commit, tail)->next;
3126                 return tail;
3127         }
3128
3129         /*
3130          * Rewrite our list of parents. Note that this cannot
3131          * affect our TREESAME flags in any way - a commit is
3132          * always TREESAME to its simplification.
3133          */
3134         for (p = commit->parents; p; p = p->next) {
3135                 pst = locate_simplify_state(revs, p->item);
3136                 p->item = pst->simplified;
3137                 if (revs->first_parent_only)
3138                         break;
3139         }
3140
3141         if (revs->first_parent_only)
3142                 cnt = 1;
3143         else
3144                 cnt = remove_duplicate_parents(revs, commit);
3145
3146         /*
3147          * It is possible that we are a merge and one side branch
3148          * does not have any commit that touches the given paths;
3149          * in such a case, the immediate parent from that branch
3150          * will be rewritten to be the merge base.
3151          *
3152          *      o----X          X: the commit we are looking at;
3153          *     /    /           o: a commit that touches the paths;
3154          * ---o----'
3155          *
3156          * Further, a merge of an independent branch that doesn't
3157          * touch the path will reduce to a treesame root parent:
3158          *
3159          *  ----o----X          X: the commit we are looking at;
3160          *          /           o: a commit that touches the paths;
3161          *         r            r: a root commit not touching the paths
3162          *
3163          * Detect and simplify both cases.
3164          */
3165         if (1 < cnt) {
3166                 int marked = mark_redundant_parents(commit);
3167                 marked += mark_treesame_root_parents(commit);
3168                 if (marked)
3169                         marked -= leave_one_treesame_to_parent(revs, commit);
3170                 if (marked)
3171                         cnt = remove_marked_parents(revs, commit);
3172         }
3173
3174         /*
3175          * A commit simplifies to itself if it is a root, if it is
3176          * UNINTERESTING, if it touches the given paths, or if it is a
3177          * merge and its parents don't simplify to one relevant commit
3178          * (the first two cases are already handled at the beginning of
3179          * this function).
3180          *
3181          * Otherwise, it simplifies to what its sole relevant parent
3182          * simplifies to.
3183          */
3184         if (!cnt ||
3185             (commit->object.flags & UNINTERESTING) ||
3186             !(commit->object.flags & TREESAME) ||
3187             (parent = one_relevant_parent(revs, commit->parents)) == NULL ||
3188             (revs->show_pulls && (commit->object.flags & PULL_MERGE)))
3189                 st->simplified = commit;
3190         else {
3191                 pst = locate_simplify_state(revs, parent);
3192                 st->simplified = pst->simplified;
3193         }
3194         return tail;
3195 }
3196
3197 static void simplify_merges(struct rev_info *revs)
3198 {
3199         struct commit_list *list, *next;
3200         struct commit_list *yet_to_do, **tail;
3201         struct commit *commit;
3202
3203         if (!revs->prune)
3204                 return;
3205
3206         /* feed the list reversed */
3207         yet_to_do = NULL;
3208         for (list = revs->commits; list; list = next) {
3209                 commit = list->item;
3210                 next = list->next;
3211                 /*
3212                  * Do not free(list) here yet; the original list
3213                  * is used later in this function.
3214                  */
3215                 commit_list_insert(commit, &yet_to_do);
3216         }
3217         while (yet_to_do) {
3218                 list = yet_to_do;
3219                 yet_to_do = NULL;
3220                 tail = &yet_to_do;
3221                 while (list) {
3222                         commit = pop_commit(&list);
3223                         tail = simplify_one(revs, commit, tail);
3224                 }
3225         }
3226
3227         /* clean up the result, removing the simplified ones */
3228         list = revs->commits;
3229         revs->commits = NULL;
3230         tail = &revs->commits;
3231         while (list) {
3232                 struct merge_simplify_state *st;
3233
3234                 commit = pop_commit(&list);
3235                 st = locate_simplify_state(revs, commit);
3236                 if (st->simplified == commit)
3237                         tail = &commit_list_insert(commit, tail)->next;
3238         }
3239 }
3240
3241 static void set_children(struct rev_info *revs)
3242 {
3243         struct commit_list *l;
3244         for (l = revs->commits; l; l = l->next) {
3245                 struct commit *commit = l->item;
3246                 struct commit_list *p;
3247
3248                 for (p = commit->parents; p; p = p->next)
3249                         add_child(revs, p->item, commit);
3250         }
3251 }
3252
3253 void reset_revision_walk(void)
3254 {
3255         clear_object_flags(SEEN | ADDED | SHOWN | TOPO_WALK_EXPLORED | TOPO_WALK_INDEGREE);
3256 }
3257
3258 static int mark_uninteresting(const struct object_id *oid,
3259                               struct packed_git *pack,
3260                               uint32_t pos,
3261                               void *cb)
3262 {
3263         struct rev_info *revs = cb;
3264         struct object *o = parse_object(revs->repo, oid);
3265         o->flags |= UNINTERESTING | SEEN;
3266         return 0;
3267 }
3268
3269 define_commit_slab(indegree_slab, int);
3270 define_commit_slab(author_date_slab, timestamp_t);
3271
3272 struct topo_walk_info {
3273         timestamp_t min_generation;
3274         struct prio_queue explore_queue;
3275         struct prio_queue indegree_queue;
3276         struct prio_queue topo_queue;
3277         struct indegree_slab indegree;
3278         struct author_date_slab author_date;
3279 };
3280
3281 static int topo_walk_atexit_registered;
3282 static unsigned int count_explore_walked;
3283 static unsigned int count_indegree_walked;
3284 static unsigned int count_topo_walked;
3285
3286 static void trace2_topo_walk_statistics_atexit(void)
3287 {
3288         struct json_writer jw = JSON_WRITER_INIT;
3289
3290         jw_object_begin(&jw, 0);
3291         jw_object_intmax(&jw, "count_explore_walked", count_explore_walked);
3292         jw_object_intmax(&jw, "count_indegree_walked", count_indegree_walked);
3293         jw_object_intmax(&jw, "count_topo_walked", count_topo_walked);
3294         jw_end(&jw);
3295
3296         trace2_data_json("topo_walk", the_repository, "statistics", &jw);
3297
3298         jw_release(&jw);
3299 }
3300
3301 static inline void test_flag_and_insert(struct prio_queue *q, struct commit *c, int flag)
3302 {
3303         if (c->object.flags & flag)
3304                 return;
3305
3306         c->object.flags |= flag;
3307         prio_queue_put(q, c);
3308 }
3309
3310 static void explore_walk_step(struct rev_info *revs)
3311 {
3312         struct topo_walk_info *info = revs->topo_walk_info;
3313         struct commit_list *p;
3314         struct commit *c = prio_queue_get(&info->explore_queue);
3315
3316         if (!c)
3317                 return;
3318
3319         if (repo_parse_commit_gently(revs->repo, c, 1) < 0)
3320                 return;
3321
3322         count_explore_walked++;
3323
3324         if (revs->sort_order == REV_SORT_BY_AUTHOR_DATE)
3325                 record_author_date(&info->author_date, c);
3326
3327         if (revs->max_age != -1 && (c->date < revs->max_age))
3328                 c->object.flags |= UNINTERESTING;
3329
3330         if (process_parents(revs, c, NULL, NULL) < 0)
3331                 return;
3332
3333         if (c->object.flags & UNINTERESTING)
3334                 mark_parents_uninteresting(c);
3335
3336         for (p = c->parents; p; p = p->next)
3337                 test_flag_and_insert(&info->explore_queue, p->item, TOPO_WALK_EXPLORED);
3338 }
3339
3340 static void explore_to_depth(struct rev_info *revs,
3341                              timestamp_t gen_cutoff)
3342 {
3343         struct topo_walk_info *info = revs->topo_walk_info;
3344         struct commit *c;
3345         while ((c = prio_queue_peek(&info->explore_queue)) &&
3346                commit_graph_generation(c) >= gen_cutoff)
3347                 explore_walk_step(revs);
3348 }
3349
3350 static void indegree_walk_step(struct rev_info *revs)
3351 {
3352         struct commit_list *p;
3353         struct topo_walk_info *info = revs->topo_walk_info;
3354         struct commit *c = prio_queue_get(&info->indegree_queue);
3355
3356         if (!c)
3357                 return;
3358
3359         if (repo_parse_commit_gently(revs->repo, c, 1) < 0)
3360                 return;
3361
3362         count_indegree_walked++;
3363
3364         explore_to_depth(revs, commit_graph_generation(c));
3365
3366         for (p = c->parents; p; p = p->next) {
3367                 struct commit *parent = p->item;
3368                 int *pi = indegree_slab_at(&info->indegree, parent);
3369
3370                 if (repo_parse_commit_gently(revs->repo, parent, 1) < 0)
3371                         return;
3372
3373                 if (*pi)
3374                         (*pi)++;
3375                 else
3376                         *pi = 2;
3377
3378                 test_flag_and_insert(&info->indegree_queue, parent, TOPO_WALK_INDEGREE);
3379
3380                 if (revs->first_parent_only)
3381                         return;
3382         }
3383 }
3384
3385 static void compute_indegrees_to_depth(struct rev_info *revs,
3386                                        timestamp_t gen_cutoff)
3387 {
3388         struct topo_walk_info *info = revs->topo_walk_info;
3389         struct commit *c;
3390         while ((c = prio_queue_peek(&info->indegree_queue)) &&
3391                commit_graph_generation(c) >= gen_cutoff)
3392                 indegree_walk_step(revs);
3393 }
3394
3395 static void reset_topo_walk(struct rev_info *revs)
3396 {
3397         struct topo_walk_info *info = revs->topo_walk_info;
3398
3399         clear_prio_queue(&info->explore_queue);
3400         clear_prio_queue(&info->indegree_queue);
3401         clear_prio_queue(&info->topo_queue);
3402         clear_indegree_slab(&info->indegree);
3403         clear_author_date_slab(&info->author_date);
3404
3405         FREE_AND_NULL(revs->topo_walk_info);
3406 }
3407
3408 static void init_topo_walk(struct rev_info *revs)
3409 {
3410         struct topo_walk_info *info;
3411         struct commit_list *list;
3412         if (revs->topo_walk_info)
3413                 reset_topo_walk(revs);
3414
3415         revs->topo_walk_info = xmalloc(sizeof(struct topo_walk_info));
3416         info = revs->topo_walk_info;
3417         memset(info, 0, sizeof(struct topo_walk_info));
3418
3419         init_indegree_slab(&info->indegree);
3420         memset(&info->explore_queue, 0, sizeof(info->explore_queue));
3421         memset(&info->indegree_queue, 0, sizeof(info->indegree_queue));
3422         memset(&info->topo_queue, 0, sizeof(info->topo_queue));
3423
3424         switch (revs->sort_order) {
3425         default: /* REV_SORT_IN_GRAPH_ORDER */
3426                 info->topo_queue.compare = NULL;
3427                 break;
3428         case REV_SORT_BY_COMMIT_DATE:
3429                 info->topo_queue.compare = compare_commits_by_commit_date;
3430                 break;
3431         case REV_SORT_BY_AUTHOR_DATE:
3432                 init_author_date_slab(&info->author_date);
3433                 info->topo_queue.compare = compare_commits_by_author_date;
3434                 info->topo_queue.cb_data = &info->author_date;
3435                 break;
3436         }
3437
3438         info->explore_queue.compare = compare_commits_by_gen_then_commit_date;
3439         info->indegree_queue.compare = compare_commits_by_gen_then_commit_date;
3440
3441         info->min_generation = GENERATION_NUMBER_INFINITY;
3442         for (list = revs->commits; list; list = list->next) {
3443                 struct commit *c = list->item;
3444                 timestamp_t generation;
3445
3446                 if (repo_parse_commit_gently(revs->repo, c, 1))
3447                         continue;
3448
3449                 test_flag_and_insert(&info->explore_queue, c, TOPO_WALK_EXPLORED);
3450                 test_flag_and_insert(&info->indegree_queue, c, TOPO_WALK_INDEGREE);
3451
3452                 generation = commit_graph_generation(c);
3453                 if (generation < info->min_generation)
3454                         info->min_generation = generation;
3455
3456                 *(indegree_slab_at(&info->indegree, c)) = 1;
3457
3458                 if (revs->sort_order == REV_SORT_BY_AUTHOR_DATE)
3459                         record_author_date(&info->author_date, c);
3460         }
3461         compute_indegrees_to_depth(revs, info->min_generation);
3462
3463         for (list = revs->commits; list; list = list->next) {
3464                 struct commit *c = list->item;
3465
3466                 if (*(indegree_slab_at(&info->indegree, c)) == 1)
3467                         prio_queue_put(&info->topo_queue, c);
3468         }
3469
3470         /*
3471          * This is unfortunate; the initial tips need to be shown
3472          * in the order given from the revision traversal machinery.
3473          */
3474         if (revs->sort_order == REV_SORT_IN_GRAPH_ORDER)
3475                 prio_queue_reverse(&info->topo_queue);
3476
3477         if (trace2_is_enabled() && !topo_walk_atexit_registered) {
3478                 atexit(trace2_topo_walk_statistics_atexit);
3479                 topo_walk_atexit_registered = 1;
3480         }
3481 }
3482
3483 static struct commit *next_topo_commit(struct rev_info *revs)
3484 {
3485         struct commit *c;
3486         struct topo_walk_info *info = revs->topo_walk_info;
3487
3488         /* pop next off of topo_queue */
3489         c = prio_queue_get(&info->topo_queue);
3490
3491         if (c)
3492                 *(indegree_slab_at(&info->indegree, c)) = 0;
3493
3494         return c;
3495 }
3496
3497 static void expand_topo_walk(struct rev_info *revs, struct commit *commit)
3498 {
3499         struct commit_list *p;
3500         struct topo_walk_info *info = revs->topo_walk_info;
3501         if (process_parents(revs, commit, NULL, NULL) < 0) {
3502                 if (!revs->ignore_missing_links)
3503                         die("Failed to traverse parents of commit %s",
3504                             oid_to_hex(&commit->object.oid));
3505         }
3506
3507         count_topo_walked++;
3508
3509         for (p = commit->parents; p; p = p->next) {
3510                 struct commit *parent = p->item;
3511                 int *pi;
3512                 timestamp_t generation;
3513
3514                 if (parent->object.flags & UNINTERESTING)
3515                         continue;
3516
3517                 if (repo_parse_commit_gently(revs->repo, parent, 1) < 0)
3518                         continue;
3519
3520                 generation = commit_graph_generation(parent);
3521                 if (generation < info->min_generation) {
3522                         info->min_generation = generation;
3523                         compute_indegrees_to_depth(revs, info->min_generation);
3524                 }
3525
3526                 pi = indegree_slab_at(&info->indegree, parent);
3527
3528                 (*pi)--;
3529                 if (*pi == 1)
3530                         prio_queue_put(&info->topo_queue, parent);
3531
3532                 if (revs->first_parent_only)
3533                         return;
3534         }
3535 }
3536
3537 int prepare_revision_walk(struct rev_info *revs)
3538 {
3539         int i;
3540         struct object_array old_pending;
3541         struct commit_list **next = &revs->commits;
3542
3543         memcpy(&old_pending, &revs->pending, sizeof(old_pending));
3544         revs->pending.nr = 0;
3545         revs->pending.alloc = 0;
3546         revs->pending.objects = NULL;
3547         for (i = 0; i < old_pending.nr; i++) {
3548                 struct object_array_entry *e = old_pending.objects + i;
3549                 struct commit *commit = handle_commit(revs, e);
3550                 if (commit) {
3551                         if (!(commit->object.flags & SEEN)) {
3552                                 commit->object.flags |= SEEN;
3553                                 next = commit_list_append(commit, next);
3554                         }
3555                 }
3556         }
3557         object_array_clear(&old_pending);
3558
3559         /* Signal whether we need per-parent treesame decoration */
3560         if (revs->simplify_merges ||
3561             (revs->limited && limiting_can_increase_treesame(revs)))
3562                 revs->treesame.name = "treesame";
3563
3564         if (revs->exclude_promisor_objects) {
3565                 for_each_packed_object(mark_uninteresting, revs,
3566                                        FOR_EACH_OBJECT_PROMISOR_ONLY);
3567         }
3568
3569         if (!revs->reflog_info)
3570                 prepare_to_use_bloom_filter(revs);
3571         if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
3572                 commit_list_sort_by_date(&revs->commits);
3573         if (revs->no_walk)
3574                 return 0;
3575         if (revs->limited) {
3576                 if (limit_list(revs) < 0)
3577                         return -1;
3578                 if (revs->topo_order)
3579                         sort_in_topological_order(&revs->commits, revs->sort_order);
3580         } else if (revs->topo_order)
3581                 init_topo_walk(revs);
3582         if (revs->line_level_traverse && want_ancestry(revs))
3583                 /*
3584                  * At the moment we can only do line-level log with parent
3585                  * rewriting by performing this expensive pre-filtering step.
3586                  * If parent rewriting is not requested, then we rather
3587                  * perform the line-level log filtering during the regular
3588                  * history traversal.
3589                  */
3590                 line_log_filter(revs);
3591         if (revs->simplify_merges)
3592                 simplify_merges(revs);
3593         if (revs->children.name)
3594                 set_children(revs);
3595
3596         return 0;
3597 }
3598
3599 static enum rewrite_result rewrite_one_1(struct rev_info *revs,
3600                                          struct commit **pp,
3601                                          struct prio_queue *queue)
3602 {
3603         for (;;) {
3604                 struct commit *p = *pp;
3605                 if (!revs->limited)
3606                         if (process_parents(revs, p, NULL, queue) < 0)
3607                                 return rewrite_one_error;
3608                 if (p->object.flags & UNINTERESTING)
3609                         return rewrite_one_ok;
3610                 if (!(p->object.flags & TREESAME))
3611                         return rewrite_one_ok;
3612                 if (!p->parents)
3613                         return rewrite_one_noparents;
3614                 if ((p = one_relevant_parent(revs, p->parents)) == NULL)
3615                         return rewrite_one_ok;
3616                 *pp = p;
3617         }
3618 }
3619
3620 static void merge_queue_into_list(struct prio_queue *q, struct commit_list **list)
3621 {
3622         while (q->nr) {
3623                 struct commit *item = prio_queue_peek(q);
3624                 struct commit_list *p = *list;
3625
3626                 if (p && p->item->date >= item->date)
3627                         list = &p->next;
3628                 else {
3629                         p = commit_list_insert(item, list);
3630                         list = &p->next; /* skip newly added item */
3631                         prio_queue_get(q); /* pop item */
3632                 }
3633         }
3634 }
3635
3636 static enum rewrite_result rewrite_one(struct rev_info *revs, struct commit **pp)
3637 {
3638         struct prio_queue queue = { compare_commits_by_commit_date };
3639         enum rewrite_result ret = rewrite_one_1(revs, pp, &queue);
3640         merge_queue_into_list(&queue, &revs->commits);
3641         clear_prio_queue(&queue);
3642         return ret;
3643 }
3644
3645 int rewrite_parents(struct rev_info *revs, struct commit *commit,
3646         rewrite_parent_fn_t rewrite_parent)
3647 {
3648         struct commit_list **pp = &commit->parents;
3649         while (*pp) {
3650                 struct commit_list *parent = *pp;
3651                 switch (rewrite_parent(revs, &parent->item)) {
3652                 case rewrite_one_ok:
3653                         break;
3654                 case rewrite_one_noparents:
3655                         *pp = parent->next;
3656                         continue;
3657                 case rewrite_one_error:
3658                         return -1;
3659                 }
3660                 pp = &parent->next;
3661         }
3662         remove_duplicate_parents(revs, commit);
3663         return 0;
3664 }
3665
3666 static int commit_rewrite_person(struct strbuf *buf, const char *what, struct string_list *mailmap)
3667 {
3668         char *person, *endp;
3669         size_t len, namelen, maillen;
3670         const char *name;
3671         const char *mail;
3672         struct ident_split ident;
3673
3674         person = strstr(buf->buf, what);
3675         if (!person)
3676                 return 0;
3677
3678         person += strlen(what);
3679         endp = strchr(person, '\n');
3680         if (!endp)
3681                 return 0;
3682
3683         len = endp - person;
3684
3685         if (split_ident_line(&ident, person, len))
3686                 return 0;
3687
3688         mail = ident.mail_begin;
3689         maillen = ident.mail_end - ident.mail_begin;
3690         name = ident.name_begin;
3691         namelen = ident.name_end - ident.name_begin;
3692
3693         if (map_user(mailmap, &mail, &maillen, &name, &namelen)) {
3694                 struct strbuf namemail = STRBUF_INIT;
3695
3696                 strbuf_addf(&namemail, "%.*s <%.*s>",
3697                             (int)namelen, name, (int)maillen, mail);
3698
3699                 strbuf_splice(buf, ident.name_begin - buf->buf,
3700                               ident.mail_end - ident.name_begin + 1,
3701                               namemail.buf, namemail.len);
3702
3703                 strbuf_release(&namemail);
3704
3705                 return 1;
3706         }
3707
3708         return 0;
3709 }
3710
3711 static int commit_match(struct commit *commit, struct rev_info *opt)
3712 {
3713         int retval;
3714         const char *encoding;
3715         const char *message;
3716         struct strbuf buf = STRBUF_INIT;
3717
3718         if (!opt->grep_filter.pattern_list && !opt->grep_filter.header_list)
3719                 return 1;
3720
3721         /* Prepend "fake" headers as needed */
3722         if (opt->grep_filter.use_reflog_filter) {
3723                 strbuf_addstr(&buf, "reflog ");
3724                 get_reflog_message(&buf, opt->reflog_info);
3725                 strbuf_addch(&buf, '\n');
3726         }
3727
3728         /*
3729          * We grep in the user's output encoding, under the assumption that it
3730          * is the encoding they are most likely to write their grep pattern
3731          * for. In addition, it means we will match the "notes" encoding below,
3732          * so we will not end up with a buffer that has two different encodings
3733          * in it.
3734          */
3735         encoding = get_log_output_encoding();
3736         message = logmsg_reencode(commit, NULL, encoding);
3737
3738         /* Copy the commit to temporary if we are using "fake" headers */
3739         if (buf.len)
3740                 strbuf_addstr(&buf, message);
3741
3742         if (opt->grep_filter.header_list && opt->mailmap) {
3743                 if (!buf.len)
3744                         strbuf_addstr(&buf, message);
3745
3746                 commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
3747                 commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
3748         }
3749
3750         /* Append "fake" message parts as needed */
3751         if (opt->show_notes) {
3752                 if (!buf.len)
3753                         strbuf_addstr(&buf, message);
3754                 format_display_notes(&commit->object.oid, &buf, encoding, 1);
3755         }
3756
3757         /*
3758          * Find either in the original commit message, or in the temporary.
3759          * Note that we cast away the constness of "message" here. It is
3760          * const because it may come from the cached commit buffer. That's OK,
3761          * because we know that it is modifiable heap memory, and that while
3762          * grep_buffer may modify it for speed, it will restore any
3763          * changes before returning.
3764          */
3765         if (buf.len)
3766                 retval = grep_buffer(&opt->grep_filter, buf.buf, buf.len);
3767         else
3768                 retval = grep_buffer(&opt->grep_filter,
3769                                      (char *)message, strlen(message));
3770         strbuf_release(&buf);
3771         unuse_commit_buffer(commit, message);
3772         return opt->invert_grep ? !retval : retval;
3773 }
3774
3775 static inline int want_ancestry(const struct rev_info *revs)
3776 {
3777         return (revs->rewrite_parents || revs->children.name);
3778 }
3779
3780 /*
3781  * Return a timestamp to be used for --since/--until comparisons for this
3782  * commit, based on the revision options.
3783  */
3784 static timestamp_t comparison_date(const struct rev_info *revs,
3785                                    struct commit *commit)
3786 {
3787         return revs->reflog_info ?
3788                 get_reflog_timestamp(revs->reflog_info) :
3789                 commit->date;
3790 }
3791
3792 enum commit_action get_commit_action(struct rev_info *revs, struct commit *commit)
3793 {
3794         if (commit->object.flags & SHOWN)
3795                 return commit_ignore;
3796         if (revs->unpacked && has_object_pack(&commit->object.oid))
3797                 return commit_ignore;
3798         if (commit->object.flags & UNINTERESTING)
3799                 return commit_ignore;
3800         if (revs->line_level_traverse && !want_ancestry(revs)) {
3801                 /*
3802                  * In case of line-level log with parent rewriting
3803                  * prepare_revision_walk() already took care of all line-level
3804                  * log filtering, and there is nothing left to do here.
3805                  *
3806                  * If parent rewriting was not requested, then this is the
3807                  * place to perform the line-level log filtering.  Notably,
3808                  * this check, though expensive, must come before the other,
3809                  * cheaper filtering conditions, because the tracked line
3810                  * ranges must be adjusted even when the commit will end up
3811                  * being ignored based on other conditions.
3812                  */
3813                 if (!line_log_process_ranges_arbitrary_commit(revs, commit))
3814                         return commit_ignore;
3815         }
3816         if (revs->min_age != -1 &&
3817             comparison_date(revs, commit) > revs->min_age)
3818                         return commit_ignore;
3819         if (revs->min_parents || (revs->max_parents >= 0)) {
3820                 int n = commit_list_count(commit->parents);
3821                 if ((n < revs->min_parents) ||
3822                     ((revs->max_parents >= 0) && (n > revs->max_parents)))
3823                         return commit_ignore;
3824         }
3825         if (!commit_match(commit, revs))
3826                 return commit_ignore;
3827         if (revs->prune && revs->dense) {
3828                 /* Commit without changes? */
3829                 if (commit->object.flags & TREESAME) {
3830                         int n;
3831                         struct commit_list *p;
3832                         /* drop merges unless we want parenthood */
3833                         if (!want_ancestry(revs))
3834                                 return commit_ignore;
3835
3836                         if (revs->show_pulls && (commit->object.flags & PULL_MERGE))
3837                                 return commit_show;
3838
3839                         /*
3840                          * If we want ancestry, then need to keep any merges
3841                          * between relevant commits to tie together topology.
3842                          * For consistency with TREESAME and simplification
3843                          * use "relevant" here rather than just INTERESTING,
3844                          * to treat bottom commit(s) as part of the topology.
3845                          */
3846                         for (n = 0, p = commit->parents; p; p = p->next)
3847                                 if (relevant_commit(p->item))
3848                                         if (++n >= 2)
3849                                                 return commit_show;
3850                         return commit_ignore;
3851                 }
3852         }
3853         return commit_show;
3854 }
3855
3856 define_commit_slab(saved_parents, struct commit_list *);
3857
3858 #define EMPTY_PARENT_LIST ((struct commit_list *)-1)
3859
3860 /*
3861  * You may only call save_parents() once per commit (this is checked
3862  * for non-root commits).
3863  */
3864 static void save_parents(struct rev_info *revs, struct commit *commit)
3865 {
3866         struct commit_list **pp;
3867
3868         if (!revs->saved_parents_slab) {
3869                 revs->saved_parents_slab = xmalloc(sizeof(struct saved_parents));
3870                 init_saved_parents(revs->saved_parents_slab);
3871         }
3872
3873         pp = saved_parents_at(revs->saved_parents_slab, commit);
3874
3875         /*
3876          * When walking with reflogs, we may visit the same commit
3877          * several times: once for each appearance in the reflog.
3878          *
3879          * In this case, save_parents() will be called multiple times.
3880          * We want to keep only the first set of parents.  We need to
3881          * store a sentinel value for an empty (i.e., NULL) parent
3882          * list to distinguish it from a not-yet-saved list, however.
3883          */
3884         if (*pp)
3885                 return;
3886         if (commit->parents)
3887                 *pp = copy_commit_list(commit->parents);
3888         else
3889                 *pp = EMPTY_PARENT_LIST;
3890 }
3891
3892 static void free_saved_parents(struct rev_info *revs)
3893 {
3894         if (revs->saved_parents_slab)
3895                 clear_saved_parents(revs->saved_parents_slab);
3896 }
3897
3898 struct commit_list *get_saved_parents(struct rev_info *revs, const struct commit *commit)
3899 {
3900         struct commit_list *parents;
3901
3902         if (!revs->saved_parents_slab)
3903                 return commit->parents;
3904
3905         parents = *saved_parents_at(revs->saved_parents_slab, commit);
3906         if (parents == EMPTY_PARENT_LIST)
3907                 return NULL;
3908         return parents;
3909 }
3910
3911 enum commit_action simplify_commit(struct rev_info *revs, struct commit *commit)
3912 {
3913         enum commit_action action = get_commit_action(revs, commit);
3914
3915         if (action == commit_show &&
3916             revs->prune && revs->dense && want_ancestry(revs)) {
3917                 /*
3918                  * --full-diff on simplified parents is no good: it
3919                  * will show spurious changes from the commits that
3920                  * were elided.  So we save the parents on the side
3921                  * when --full-diff is in effect.
3922                  */
3923                 if (revs->full_diff)
3924                         save_parents(revs, commit);
3925                 if (rewrite_parents(revs, commit, rewrite_one) < 0)
3926                         return commit_error;
3927         }
3928         return action;
3929 }
3930
3931 static void track_linear(struct rev_info *revs, struct commit *commit)
3932 {
3933         if (revs->track_first_time) {
3934                 revs->linear = 1;
3935                 revs->track_first_time = 0;
3936         } else {
3937                 struct commit_list *p;
3938                 for (p = revs->previous_parents; p; p = p->next)
3939                         if (p->item == NULL || /* first commit */
3940                             oideq(&p->item->object.oid, &commit->object.oid))
3941                                 break;
3942                 revs->linear = p != NULL;
3943         }
3944         if (revs->reverse) {
3945                 if (revs->linear)
3946                         commit->object.flags |= TRACK_LINEAR;
3947         }
3948         free_commit_list(revs->previous_parents);
3949         revs->previous_parents = copy_commit_list(commit->parents);
3950 }
3951
3952 static struct commit *get_revision_1(struct rev_info *revs)
3953 {
3954         while (1) {
3955                 struct commit *commit;
3956
3957                 if (revs->reflog_info)
3958                         commit = next_reflog_entry(revs->reflog_info);
3959                 else if (revs->topo_walk_info)
3960                         commit = next_topo_commit(revs);
3961                 else
3962                         commit = pop_commit(&revs->commits);
3963
3964                 if (!commit)
3965                         return NULL;
3966
3967                 if (revs->reflog_info)
3968                         commit->object.flags &= ~(ADDED | SEEN | SHOWN);
3969
3970                 /*
3971                  * If we haven't done the list limiting, we need to look at
3972                  * the parents here. We also need to do the date-based limiting
3973                  * that we'd otherwise have done in limit_list().
3974                  */
3975                 if (!revs->limited) {
3976                         if (revs->max_age != -1 &&
3977                             comparison_date(revs, commit) < revs->max_age)
3978                                 continue;
3979
3980                         if (revs->reflog_info)
3981                                 try_to_simplify_commit(revs, commit);
3982                         else if (revs->topo_walk_info)
3983                                 expand_topo_walk(revs, commit);
3984                         else if (process_parents(revs, commit, &revs->commits, NULL) < 0) {
3985                                 if (!revs->ignore_missing_links)
3986                                         die("Failed to traverse parents of commit %s",
3987                                                 oid_to_hex(&commit->object.oid));
3988                         }
3989                 }
3990
3991                 switch (simplify_commit(revs, commit)) {
3992                 case commit_ignore:
3993                         continue;
3994                 case commit_error:
3995                         die("Failed to simplify parents of commit %s",
3996                             oid_to_hex(&commit->object.oid));
3997                 default:
3998                         if (revs->track_linear)
3999                                 track_linear(revs, commit);
4000                         return commit;
4001                 }
4002         }
4003 }
4004
4005 /*
4006  * Return true for entries that have not yet been shown.  (This is an
4007  * object_array_each_func_t.)
4008  */
4009 static int entry_unshown(struct object_array_entry *entry, void *cb_data_unused)
4010 {
4011         return !(entry->item->flags & SHOWN);
4012 }
4013
4014 /*
4015  * If array is on the verge of a realloc, garbage-collect any entries
4016  * that have already been shown to try to free up some space.
4017  */
4018 static void gc_boundary(struct object_array *array)
4019 {
4020         if (array->nr == array->alloc)
4021                 object_array_filter(array, entry_unshown, NULL);
4022 }
4023
4024 static void create_boundary_commit_list(struct rev_info *revs)
4025 {
4026         unsigned i;
4027         struct commit *c;
4028         struct object_array *array = &revs->boundary_commits;
4029         struct object_array_entry *objects = array->objects;
4030
4031         /*
4032          * If revs->commits is non-NULL at this point, an error occurred in
4033          * get_revision_1().  Ignore the error and continue printing the
4034          * boundary commits anyway.  (This is what the code has always
4035          * done.)
4036          */
4037         if (revs->commits) {
4038                 free_commit_list(revs->commits);
4039                 revs->commits = NULL;
4040         }
4041
4042         /*
4043          * Put all of the actual boundary commits from revs->boundary_commits
4044          * into revs->commits
4045          */
4046         for (i = 0; i < array->nr; i++) {
4047                 c = (struct commit *)(objects[i].item);
4048                 if (!c)
4049                         continue;
4050                 if (!(c->object.flags & CHILD_SHOWN))
4051                         continue;
4052                 if (c->object.flags & (SHOWN | BOUNDARY))
4053                         continue;
4054                 c->object.flags |= BOUNDARY;
4055                 commit_list_insert(c, &revs->commits);
4056         }
4057
4058         /*
4059          * If revs->topo_order is set, sort the boundary commits
4060          * in topological order
4061          */
4062         sort_in_topological_order(&revs->commits, revs->sort_order);
4063 }
4064
4065 static struct commit *get_revision_internal(struct rev_info *revs)
4066 {
4067         struct commit *c = NULL;
4068         struct commit_list *l;
4069
4070         if (revs->boundary == 2) {
4071                 /*
4072                  * All of the normal commits have already been returned,
4073                  * and we are now returning boundary commits.
4074                  * create_boundary_commit_list() has populated
4075                  * revs->commits with the remaining commits to return.
4076                  */
4077                 c = pop_commit(&revs->commits);
4078                 if (c)
4079                         c->object.flags |= SHOWN;
4080                 return c;
4081         }
4082
4083         /*
4084          * If our max_count counter has reached zero, then we are done. We
4085          * don't simply return NULL because we still might need to show
4086          * boundary commits. But we want to avoid calling get_revision_1, which
4087          * might do a considerable amount of work finding the next commit only
4088          * for us to throw it away.
4089          *
4090          * If it is non-zero, then either we don't have a max_count at all
4091          * (-1), or it is still counting, in which case we decrement.
4092          */
4093         if (revs->max_count) {
4094                 c = get_revision_1(revs);
4095                 if (c) {
4096                         while (revs->skip_count > 0) {
4097                                 revs->skip_count--;
4098                                 c = get_revision_1(revs);
4099                                 if (!c)
4100                                         break;
4101                         }
4102                 }
4103
4104                 if (revs->max_count > 0)
4105                         revs->max_count--;
4106         }
4107
4108         if (c)
4109                 c->object.flags |= SHOWN;
4110
4111         if (!revs->boundary)
4112                 return c;
4113
4114         if (!c) {
4115                 /*
4116                  * get_revision_1() runs out the commits, and
4117                  * we are done computing the boundaries.
4118                  * switch to boundary commits output mode.
4119                  */
4120                 revs->boundary = 2;
4121
4122                 /*
4123                  * Update revs->commits to contain the list of
4124                  * boundary commits.
4125                  */
4126                 create_boundary_commit_list(revs);
4127
4128                 return get_revision_internal(revs);
4129         }
4130
4131         /*
4132          * boundary commits are the commits that are parents of the
4133          * ones we got from get_revision_1() but they themselves are
4134          * not returned from get_revision_1().  Before returning
4135          * 'c', we need to mark its parents that they could be boundaries.
4136          */
4137
4138         for (l = c->parents; l; l = l->next) {
4139                 struct object *p;
4140                 p = &(l->item->object);
4141                 if (p->flags & (CHILD_SHOWN | SHOWN))
4142                         continue;
4143                 p->flags |= CHILD_SHOWN;
4144                 gc_boundary(&revs->boundary_commits);
4145                 add_object_array(p, NULL, &revs->boundary_commits);
4146         }
4147
4148         return c;
4149 }
4150
4151 struct commit *get_revision(struct rev_info *revs)
4152 {
4153         struct commit *c;
4154         struct commit_list *reversed;
4155
4156         if (revs->reverse) {
4157                 reversed = NULL;
4158                 while ((c = get_revision_internal(revs)))
4159                         commit_list_insert(c, &reversed);
4160                 revs->commits = reversed;
4161                 revs->reverse = 0;
4162                 revs->reverse_output_stage = 1;
4163         }
4164
4165         if (revs->reverse_output_stage) {
4166                 c = pop_commit(&revs->commits);
4167                 if (revs->track_linear)
4168                         revs->linear = !!(c && c->object.flags & TRACK_LINEAR);
4169                 return c;
4170         }
4171
4172         c = get_revision_internal(revs);
4173         if (c && revs->graph)
4174                 graph_update(revs->graph, c);
4175         if (!c) {
4176                 free_saved_parents(revs);
4177                 if (revs->previous_parents) {
4178                         free_commit_list(revs->previous_parents);
4179                         revs->previous_parents = NULL;
4180                 }
4181         }
4182         return c;
4183 }
4184
4185 const char *get_revision_mark(const struct rev_info *revs, const struct commit *commit)
4186 {
4187         if (commit->object.flags & BOUNDARY)
4188                 return "-";
4189         else if (commit->object.flags & UNINTERESTING)
4190                 return "^";
4191         else if (commit->object.flags & PATCHSAME)
4192                 return "=";
4193         else if (!revs || revs->left_right) {
4194                 if (commit->object.flags & SYMMETRIC_LEFT)
4195                         return "<";
4196                 else
4197                         return ">";
4198         } else if (revs->graph)
4199                 return "*";
4200         else if (revs->cherry_mark)
4201                 return "+";
4202         return "";
4203 }
4204
4205 void put_revision_mark(const struct rev_info *revs, const struct commit *commit)
4206 {
4207         const char *mark = get_revision_mark(revs, commit);
4208         if (!strlen(mark))
4209                 return;
4210         fputs(mark, stdout);
4211         putchar(' ');
4212 }