strbuf: avoid calling strbuf_grow() twice in strbuf_addbuf()
[git] / revision.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "blob.h"
4 #include "tree.h"
5 #include "commit.h"
6 #include "diff.h"
7 #include "refs.h"
8 #include "revision.h"
9 #include "graph.h"
10 #include "grep.h"
11 #include "reflog-walk.h"
12 #include "patch-ids.h"
13 #include "decorate.h"
14 #include "log-tree.h"
15 #include "string-list.h"
16 #include "line-log.h"
17 #include "mailmap.h"
18 #include "commit-slab.h"
19 #include "dir.h"
20 #include "cache-tree.h"
21 #include "bisect.h"
22
23 volatile show_early_output_fn_t show_early_output;
24
25 static const char *term_bad;
26 static const char *term_good;
27
28 void show_object_with_name(FILE *out, struct object *obj, const char *name)
29 {
30         const char *p;
31
32         fprintf(out, "%s ", sha1_to_hex(obj->sha1));
33         for (p = name; *p && *p != '\n'; p++)
34                 fputc(*p, out);
35         fputc('\n', out);
36 }
37
38 static void mark_blob_uninteresting(struct blob *blob)
39 {
40         if (!blob)
41                 return;
42         if (blob->object.flags & UNINTERESTING)
43                 return;
44         blob->object.flags |= UNINTERESTING;
45 }
46
47 static void mark_tree_contents_uninteresting(struct tree *tree)
48 {
49         struct tree_desc desc;
50         struct name_entry entry;
51         struct object *obj = &tree->object;
52
53         if (!has_sha1_file(obj->sha1))
54                 return;
55         if (parse_tree(tree) < 0)
56                 die("bad tree %s", sha1_to_hex(obj->sha1));
57
58         init_tree_desc(&desc, tree->buffer, tree->size);
59         while (tree_entry(&desc, &entry)) {
60                 switch (object_type(entry.mode)) {
61                 case OBJ_TREE:
62                         mark_tree_uninteresting(lookup_tree(entry.sha1));
63                         break;
64                 case OBJ_BLOB:
65                         mark_blob_uninteresting(lookup_blob(entry.sha1));
66                         break;
67                 default:
68                         /* Subproject commit - not in this repository */
69                         break;
70                 }
71         }
72
73         /*
74          * We don't care about the tree any more
75          * after it has been marked uninteresting.
76          */
77         free_tree_buffer(tree);
78 }
79
80 void mark_tree_uninteresting(struct tree *tree)
81 {
82         struct object *obj;
83
84         if (!tree)
85                 return;
86
87         obj = &tree->object;
88         if (obj->flags & UNINTERESTING)
89                 return;
90         obj->flags |= UNINTERESTING;
91         mark_tree_contents_uninteresting(tree);
92 }
93
94 void mark_parents_uninteresting(struct commit *commit)
95 {
96         struct commit_list *parents = NULL, *l;
97
98         for (l = commit->parents; l; l = l->next)
99                 commit_list_insert(l->item, &parents);
100
101         while (parents) {
102                 struct commit *commit = pop_commit(&parents);
103
104                 while (commit) {
105                         /*
106                          * A missing commit is ok iff its parent is marked
107                          * uninteresting.
108                          *
109                          * We just mark such a thing parsed, so that when
110                          * it is popped next time around, we won't be trying
111                          * to parse it and get an error.
112                          */
113                         if (!has_sha1_file(commit->object.sha1))
114                                 commit->object.parsed = 1;
115
116                         if (commit->object.flags & UNINTERESTING)
117                                 break;
118
119                         commit->object.flags |= UNINTERESTING;
120
121                         /*
122                          * Normally we haven't parsed the parent
123                          * yet, so we won't have a parent of a parent
124                          * here. However, it may turn out that we've
125                          * reached this commit some other way (where it
126                          * wasn't uninteresting), in which case we need
127                          * to mark its parents recursively too..
128                          */
129                         if (!commit->parents)
130                                 break;
131
132                         for (l = commit->parents->next; l; l = l->next)
133                                 commit_list_insert(l->item, &parents);
134                         commit = commit->parents->item;
135                 }
136         }
137 }
138
139 static void add_pending_object_with_path(struct rev_info *revs,
140                                          struct object *obj,
141                                          const char *name, unsigned mode,
142                                          const char *path)
143 {
144         if (!obj)
145                 return;
146         if (revs->no_walk && (obj->flags & UNINTERESTING))
147                 revs->no_walk = 0;
148         if (revs->reflog_info && obj->type == OBJ_COMMIT) {
149                 struct strbuf buf = STRBUF_INIT;
150                 int len = interpret_branch_name(name, 0, &buf);
151                 int st;
152
153                 if (0 < len && name[len] && buf.len)
154                         strbuf_addstr(&buf, name + len);
155                 st = add_reflog_for_walk(revs->reflog_info,
156                                          (struct commit *)obj,
157                                          buf.buf[0] ? buf.buf: name);
158                 strbuf_release(&buf);
159                 if (st)
160                         return;
161         }
162         add_object_array_with_path(obj, name, &revs->pending, mode, path);
163 }
164
165 static void add_pending_object_with_mode(struct rev_info *revs,
166                                          struct object *obj,
167                                          const char *name, unsigned mode)
168 {
169         add_pending_object_with_path(revs, obj, name, mode, NULL);
170 }
171
172 void add_pending_object(struct rev_info *revs,
173                         struct object *obj, const char *name)
174 {
175         add_pending_object_with_mode(revs, obj, name, S_IFINVALID);
176 }
177
178 void add_head_to_pending(struct rev_info *revs)
179 {
180         unsigned char sha1[20];
181         struct object *obj;
182         if (get_sha1("HEAD", sha1))
183                 return;
184         obj = parse_object(sha1);
185         if (!obj)
186                 return;
187         add_pending_object(revs, obj, "HEAD");
188 }
189
190 static struct object *get_reference(struct rev_info *revs, const char *name,
191                                     const unsigned char *sha1,
192                                     unsigned int flags)
193 {
194         struct object *object;
195
196         object = parse_object(sha1);
197         if (!object) {
198                 if (revs->ignore_missing)
199                         return object;
200                 die("bad object %s", name);
201         }
202         object->flags |= flags;
203         return object;
204 }
205
206 void add_pending_sha1(struct rev_info *revs, const char *name,
207                       const unsigned char *sha1, unsigned int flags)
208 {
209         struct object *object = get_reference(revs, name, sha1, flags);
210         add_pending_object(revs, object, name);
211 }
212
213 static struct commit *handle_commit(struct rev_info *revs,
214                                     struct object_array_entry *entry)
215 {
216         struct object *object = entry->item;
217         const char *name = entry->name;
218         const char *path = entry->path;
219         unsigned int mode = entry->mode;
220         unsigned long flags = object->flags;
221
222         /*
223          * Tag object? Look what it points to..
224          */
225         while (object->type == OBJ_TAG) {
226                 struct tag *tag = (struct tag *) object;
227                 if (revs->tag_objects && !(flags & UNINTERESTING))
228                         add_pending_object(revs, object, tag->tag);
229                 if (!tag->tagged)
230                         die("bad tag");
231                 object = parse_object(tag->tagged->sha1);
232                 if (!object) {
233                         if (flags & UNINTERESTING)
234                                 return NULL;
235                         die("bad object %s", sha1_to_hex(tag->tagged->sha1));
236                 }
237                 object->flags |= flags;
238                 /*
239                  * We'll handle the tagged object by looping or dropping
240                  * through to the non-tag handlers below. Do not
241                  * propagate path data from the tag's pending entry.
242                  */
243                 path = NULL;
244                 mode = 0;
245         }
246
247         /*
248          * Commit object? Just return it, we'll do all the complex
249          * reachability crud.
250          */
251         if (object->type == OBJ_COMMIT) {
252                 struct commit *commit = (struct commit *)object;
253                 if (parse_commit(commit) < 0)
254                         die("unable to parse commit %s", name);
255                 if (flags & UNINTERESTING) {
256                         mark_parents_uninteresting(commit);
257                         revs->limited = 1;
258                 }
259                 if (revs->show_source && !commit->util)
260                         commit->util = xstrdup(name);
261                 return commit;
262         }
263
264         /*
265          * Tree object? Either mark it uninteresting, or add it
266          * to the list of objects to look at later..
267          */
268         if (object->type == OBJ_TREE) {
269                 struct tree *tree = (struct tree *)object;
270                 if (!revs->tree_objects)
271                         return NULL;
272                 if (flags & UNINTERESTING) {
273                         mark_tree_contents_uninteresting(tree);
274                         return NULL;
275                 }
276                 add_pending_object_with_path(revs, object, name, mode, path);
277                 return NULL;
278         }
279
280         /*
281          * Blob object? You know the drill by now..
282          */
283         if (object->type == OBJ_BLOB) {
284                 if (!revs->blob_objects)
285                         return NULL;
286                 if (flags & UNINTERESTING)
287                         return NULL;
288                 add_pending_object_with_path(revs, object, name, mode, path);
289                 return NULL;
290         }
291         die("%s is unknown object", name);
292 }
293
294 static int everybody_uninteresting(struct commit_list *orig,
295                                    struct commit **interesting_cache)
296 {
297         struct commit_list *list = orig;
298
299         if (*interesting_cache) {
300                 struct commit *commit = *interesting_cache;
301                 if (!(commit->object.flags & UNINTERESTING))
302                         return 0;
303         }
304
305         while (list) {
306                 struct commit *commit = list->item;
307                 list = list->next;
308                 if (commit->object.flags & UNINTERESTING)
309                         continue;
310
311                 *interesting_cache = commit;
312                 return 0;
313         }
314         return 1;
315 }
316
317 /*
318  * A definition of "relevant" commit that we can use to simplify limited graphs
319  * by eliminating side branches.
320  *
321  * A "relevant" commit is one that is !UNINTERESTING (ie we are including it
322  * in our list), or that is a specified BOTTOM commit. Then after computing
323  * a limited list, during processing we can generally ignore boundary merges
324  * coming from outside the graph, (ie from irrelevant parents), and treat
325  * those merges as if they were single-parent. TREESAME is defined to consider
326  * only relevant parents, if any. If we are TREESAME to our on-graph parents,
327  * we don't care if we were !TREESAME to non-graph parents.
328  *
329  * Treating bottom commits as relevant ensures that a limited graph's
330  * connection to the actual bottom commit is not viewed as a side branch, but
331  * treated as part of the graph. For example:
332  *
333  *   ....Z...A---X---o---o---B
334  *        .     /
335  *         W---Y
336  *
337  * When computing "A..B", the A-X connection is at least as important as
338  * Y-X, despite A being flagged UNINTERESTING.
339  *
340  * And when computing --ancestry-path "A..B", the A-X connection is more
341  * important than Y-X, despite both A and Y being flagged UNINTERESTING.
342  */
343 static inline int relevant_commit(struct commit *commit)
344 {
345         return (commit->object.flags & (UNINTERESTING | BOTTOM)) != UNINTERESTING;
346 }
347
348 /*
349  * Return a single relevant commit from a parent list. If we are a TREESAME
350  * commit, and this selects one of our parents, then we can safely simplify to
351  * that parent.
352  */
353 static struct commit *one_relevant_parent(const struct rev_info *revs,
354                                           struct commit_list *orig)
355 {
356         struct commit_list *list = orig;
357         struct commit *relevant = NULL;
358
359         if (!orig)
360                 return NULL;
361
362         /*
363          * For 1-parent commits, or if first-parent-only, then return that
364          * first parent (even if not "relevant" by the above definition).
365          * TREESAME will have been set purely on that parent.
366          */
367         if (revs->first_parent_only || !orig->next)
368                 return orig->item;
369
370         /*
371          * For multi-parent commits, identify a sole relevant parent, if any.
372          * If we have only one relevant parent, then TREESAME will be set purely
373          * with regard to that parent, and we can simplify accordingly.
374          *
375          * If we have more than one relevant parent, or no relevant parents
376          * (and multiple irrelevant ones), then we can't select a parent here
377          * and return NULL.
378          */
379         while (list) {
380                 struct commit *commit = list->item;
381                 list = list->next;
382                 if (relevant_commit(commit)) {
383                         if (relevant)
384                                 return NULL;
385                         relevant = commit;
386                 }
387         }
388         return relevant;
389 }
390
391 /*
392  * The goal is to get REV_TREE_NEW as the result only if the
393  * diff consists of all '+' (and no other changes), REV_TREE_OLD
394  * if the whole diff is removal of old data, and otherwise
395  * REV_TREE_DIFFERENT (of course if the trees are the same we
396  * want REV_TREE_SAME).
397  * That means that once we get to REV_TREE_DIFFERENT, we do not
398  * have to look any further.
399  */
400 static int tree_difference = REV_TREE_SAME;
401
402 static void file_add_remove(struct diff_options *options,
403                     int addremove, unsigned mode,
404                     const unsigned char *sha1,
405                     int sha1_valid,
406                     const char *fullpath, unsigned dirty_submodule)
407 {
408         int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
409
410         tree_difference |= diff;
411         if (tree_difference == REV_TREE_DIFFERENT)
412                 DIFF_OPT_SET(options, HAS_CHANGES);
413 }
414
415 static void file_change(struct diff_options *options,
416                  unsigned old_mode, unsigned new_mode,
417                  const unsigned char *old_sha1,
418                  const unsigned char *new_sha1,
419                  int old_sha1_valid, int new_sha1_valid,
420                  const char *fullpath,
421                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
422 {
423         tree_difference = REV_TREE_DIFFERENT;
424         DIFF_OPT_SET(options, HAS_CHANGES);
425 }
426
427 static int rev_compare_tree(struct rev_info *revs,
428                             struct commit *parent, struct commit *commit)
429 {
430         struct tree *t1 = parent->tree;
431         struct tree *t2 = commit->tree;
432
433         if (!t1)
434                 return REV_TREE_NEW;
435         if (!t2)
436                 return REV_TREE_OLD;
437
438         if (revs->simplify_by_decoration) {
439                 /*
440                  * If we are simplifying by decoration, then the commit
441                  * is worth showing if it has a tag pointing at it.
442                  */
443                 if (get_name_decoration(&commit->object))
444                         return REV_TREE_DIFFERENT;
445                 /*
446                  * A commit that is not pointed by a tag is uninteresting
447                  * if we are not limited by path.  This means that you will
448                  * see the usual "commits that touch the paths" plus any
449                  * tagged commit by specifying both --simplify-by-decoration
450                  * and pathspec.
451                  */
452                 if (!revs->prune_data.nr)
453                         return REV_TREE_SAME;
454         }
455
456         tree_difference = REV_TREE_SAME;
457         DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
458         if (diff_tree_sha1(t1->object.sha1, t2->object.sha1, "",
459                            &revs->pruning) < 0)
460                 return REV_TREE_DIFFERENT;
461         return tree_difference;
462 }
463
464 static int rev_same_tree_as_empty(struct rev_info *revs, struct commit *commit)
465 {
466         int retval;
467         struct tree *t1 = commit->tree;
468
469         if (!t1)
470                 return 0;
471
472         tree_difference = REV_TREE_SAME;
473         DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
474         retval = diff_tree_sha1(NULL, t1->object.sha1, "", &revs->pruning);
475
476         return retval >= 0 && (tree_difference == REV_TREE_SAME);
477 }
478
479 struct treesame_state {
480         unsigned int nparents;
481         unsigned char treesame[FLEX_ARRAY];
482 };
483
484 static struct treesame_state *initialise_treesame(struct rev_info *revs, struct commit *commit)
485 {
486         unsigned n = commit_list_count(commit->parents);
487         struct treesame_state *st = xcalloc(1, sizeof(*st) + n);
488         st->nparents = n;
489         add_decoration(&revs->treesame, &commit->object, st);
490         return st;
491 }
492
493 /*
494  * Must be called immediately after removing the nth_parent from a commit's
495  * parent list, if we are maintaining the per-parent treesame[] decoration.
496  * This does not recalculate the master TREESAME flag - update_treesame()
497  * should be called to update it after a sequence of treesame[] modifications
498  * that may have affected it.
499  */
500 static int compact_treesame(struct rev_info *revs, struct commit *commit, unsigned nth_parent)
501 {
502         struct treesame_state *st;
503         int old_same;
504
505         if (!commit->parents) {
506                 /*
507                  * Have just removed the only parent from a non-merge.
508                  * Different handling, as we lack decoration.
509                  */
510                 if (nth_parent != 0)
511                         die("compact_treesame %u", nth_parent);
512                 old_same = !!(commit->object.flags & TREESAME);
513                 if (rev_same_tree_as_empty(revs, commit))
514                         commit->object.flags |= TREESAME;
515                 else
516                         commit->object.flags &= ~TREESAME;
517                 return old_same;
518         }
519
520         st = lookup_decoration(&revs->treesame, &commit->object);
521         if (!st || nth_parent >= st->nparents)
522                 die("compact_treesame %u", nth_parent);
523
524         old_same = st->treesame[nth_parent];
525         memmove(st->treesame + nth_parent,
526                 st->treesame + nth_parent + 1,
527                 st->nparents - nth_parent - 1);
528
529         /*
530          * If we've just become a non-merge commit, update TREESAME
531          * immediately, and remove the no-longer-needed decoration.
532          * If still a merge, defer update until update_treesame().
533          */
534         if (--st->nparents == 1) {
535                 if (commit->parents->next)
536                         die("compact_treesame parents mismatch");
537                 if (st->treesame[0] && revs->dense)
538                         commit->object.flags |= TREESAME;
539                 else
540                         commit->object.flags &= ~TREESAME;
541                 free(add_decoration(&revs->treesame, &commit->object, NULL));
542         }
543
544         return old_same;
545 }
546
547 static unsigned update_treesame(struct rev_info *revs, struct commit *commit)
548 {
549         if (commit->parents && commit->parents->next) {
550                 unsigned n;
551                 struct treesame_state *st;
552                 struct commit_list *p;
553                 unsigned relevant_parents;
554                 unsigned relevant_change, irrelevant_change;
555
556                 st = lookup_decoration(&revs->treesame, &commit->object);
557                 if (!st)
558                         die("update_treesame %s", sha1_to_hex(commit->object.sha1));
559                 relevant_parents = 0;
560                 relevant_change = irrelevant_change = 0;
561                 for (p = commit->parents, n = 0; p; n++, p = p->next) {
562                         if (relevant_commit(p->item)) {
563                                 relevant_change |= !st->treesame[n];
564                                 relevant_parents++;
565                         } else
566                                 irrelevant_change |= !st->treesame[n];
567                 }
568                 if (relevant_parents ? relevant_change : irrelevant_change)
569                         commit->object.flags &= ~TREESAME;
570                 else
571                         commit->object.flags |= TREESAME;
572         }
573
574         return commit->object.flags & TREESAME;
575 }
576
577 static inline int limiting_can_increase_treesame(const struct rev_info *revs)
578 {
579         /*
580          * TREESAME is irrelevant unless prune && dense;
581          * if simplify_history is set, we can't have a mixture of TREESAME and
582          *    !TREESAME INTERESTING parents (and we don't have treesame[]
583          *    decoration anyway);
584          * if first_parent_only is set, then the TREESAME flag is locked
585          *    against the first parent (and again we lack treesame[] decoration).
586          */
587         return revs->prune && revs->dense &&
588                !revs->simplify_history &&
589                !revs->first_parent_only;
590 }
591
592 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
593 {
594         struct commit_list **pp, *parent;
595         struct treesame_state *ts = NULL;
596         int relevant_change = 0, irrelevant_change = 0;
597         int relevant_parents, nth_parent;
598
599         /*
600          * If we don't do pruning, everything is interesting
601          */
602         if (!revs->prune)
603                 return;
604
605         if (!commit->tree)
606                 return;
607
608         if (!commit->parents) {
609                 if (rev_same_tree_as_empty(revs, commit))
610                         commit->object.flags |= TREESAME;
611                 return;
612         }
613
614         /*
615          * Normal non-merge commit? If we don't want to make the
616          * history dense, we consider it always to be a change..
617          */
618         if (!revs->dense && !commit->parents->next)
619                 return;
620
621         for (pp = &commit->parents, nth_parent = 0, relevant_parents = 0;
622              (parent = *pp) != NULL;
623              pp = &parent->next, nth_parent++) {
624                 struct commit *p = parent->item;
625                 if (relevant_commit(p))
626                         relevant_parents++;
627
628                 if (nth_parent == 1) {
629                         /*
630                          * This our second loop iteration - so we now know
631                          * we're dealing with a merge.
632                          *
633                          * Do not compare with later parents when we care only about
634                          * the first parent chain, in order to avoid derailing the
635                          * traversal to follow a side branch that brought everything
636                          * in the path we are limited to by the pathspec.
637                          */
638                         if (revs->first_parent_only)
639                                 break;
640                         /*
641                          * If this will remain a potentially-simplifiable
642                          * merge, remember per-parent treesame if needed.
643                          * Initialise the array with the comparison from our
644                          * first iteration.
645                          */
646                         if (revs->treesame.name &&
647                             !revs->simplify_history &&
648                             !(commit->object.flags & UNINTERESTING)) {
649                                 ts = initialise_treesame(revs, commit);
650                                 if (!(irrelevant_change || relevant_change))
651                                         ts->treesame[0] = 1;
652                         }
653                 }
654                 if (parse_commit(p) < 0)
655                         die("cannot simplify commit %s (because of %s)",
656                             sha1_to_hex(commit->object.sha1),
657                             sha1_to_hex(p->object.sha1));
658                 switch (rev_compare_tree(revs, p, commit)) {
659                 case REV_TREE_SAME:
660                         if (!revs->simplify_history || !relevant_commit(p)) {
661                                 /* Even if a merge with an uninteresting
662                                  * side branch brought the entire change
663                                  * we are interested in, we do not want
664                                  * to lose the other branches of this
665                                  * merge, so we just keep going.
666                                  */
667                                 if (ts)
668                                         ts->treesame[nth_parent] = 1;
669                                 continue;
670                         }
671                         parent->next = NULL;
672                         commit->parents = parent;
673                         commit->object.flags |= TREESAME;
674                         return;
675
676                 case REV_TREE_NEW:
677                         if (revs->remove_empty_trees &&
678                             rev_same_tree_as_empty(revs, p)) {
679                                 /* We are adding all the specified
680                                  * paths from this parent, so the
681                                  * history beyond this parent is not
682                                  * interesting.  Remove its parents
683                                  * (they are grandparents for us).
684                                  * IOW, we pretend this parent is a
685                                  * "root" commit.
686                                  */
687                                 if (parse_commit(p) < 0)
688                                         die("cannot simplify commit %s (invalid %s)",
689                                             sha1_to_hex(commit->object.sha1),
690                                             sha1_to_hex(p->object.sha1));
691                                 p->parents = NULL;
692                         }
693                 /* fallthrough */
694                 case REV_TREE_OLD:
695                 case REV_TREE_DIFFERENT:
696                         if (relevant_commit(p))
697                                 relevant_change = 1;
698                         else
699                                 irrelevant_change = 1;
700                         continue;
701                 }
702                 die("bad tree compare for commit %s", sha1_to_hex(commit->object.sha1));
703         }
704
705         /*
706          * TREESAME is straightforward for single-parent commits. For merge
707          * commits, it is most useful to define it so that "irrelevant"
708          * parents cannot make us !TREESAME - if we have any relevant
709          * parents, then we only consider TREESAMEness with respect to them,
710          * allowing irrelevant merges from uninteresting branches to be
711          * simplified away. Only if we have only irrelevant parents do we
712          * base TREESAME on them. Note that this logic is replicated in
713          * update_treesame, which should be kept in sync.
714          */
715         if (relevant_parents ? !relevant_change : !irrelevant_change)
716                 commit->object.flags |= TREESAME;
717 }
718
719 static void commit_list_insert_by_date_cached(struct commit *p, struct commit_list **head,
720                     struct commit_list *cached_base, struct commit_list **cache)
721 {
722         struct commit_list *new_entry;
723
724         if (cached_base && p->date < cached_base->item->date)
725                 new_entry = commit_list_insert_by_date(p, &cached_base->next);
726         else
727                 new_entry = commit_list_insert_by_date(p, head);
728
729         if (cache && (!*cache || p->date < (*cache)->item->date))
730                 *cache = new_entry;
731 }
732
733 static int add_parents_to_list(struct rev_info *revs, struct commit *commit,
734                     struct commit_list **list, struct commit_list **cache_ptr)
735 {
736         struct commit_list *parent = commit->parents;
737         unsigned left_flag;
738         struct commit_list *cached_base = cache_ptr ? *cache_ptr : NULL;
739
740         if (commit->object.flags & ADDED)
741                 return 0;
742         commit->object.flags |= ADDED;
743
744         if (revs->include_check &&
745             !revs->include_check(commit, revs->include_check_data))
746                 return 0;
747
748         /*
749          * If the commit is uninteresting, don't try to
750          * prune parents - we want the maximal uninteresting
751          * set.
752          *
753          * Normally we haven't parsed the parent
754          * yet, so we won't have a parent of a parent
755          * here. However, it may turn out that we've
756          * reached this commit some other way (where it
757          * wasn't uninteresting), in which case we need
758          * to mark its parents recursively too..
759          */
760         if (commit->object.flags & UNINTERESTING) {
761                 while (parent) {
762                         struct commit *p = parent->item;
763                         parent = parent->next;
764                         if (p)
765                                 p->object.flags |= UNINTERESTING;
766                         if (parse_commit_gently(p, 1) < 0)
767                                 continue;
768                         if (p->parents)
769                                 mark_parents_uninteresting(p);
770                         if (p->object.flags & SEEN)
771                                 continue;
772                         p->object.flags |= SEEN;
773                         commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
774                 }
775                 return 0;
776         }
777
778         /*
779          * Ok, the commit wasn't uninteresting. Try to
780          * simplify the commit history and find the parent
781          * that has no differences in the path set if one exists.
782          */
783         try_to_simplify_commit(revs, commit);
784
785         if (revs->no_walk)
786                 return 0;
787
788         left_flag = (commit->object.flags & SYMMETRIC_LEFT);
789
790         for (parent = commit->parents; parent; parent = parent->next) {
791                 struct commit *p = parent->item;
792
793                 if (parse_commit_gently(p, revs->ignore_missing_links) < 0)
794                         return -1;
795                 if (revs->show_source && !p->util)
796                         p->util = commit->util;
797                 p->object.flags |= left_flag;
798                 if (!(p->object.flags & SEEN)) {
799                         p->object.flags |= SEEN;
800                         commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
801                 }
802                 if (revs->first_parent_only)
803                         break;
804         }
805         return 0;
806 }
807
808 static void cherry_pick_list(struct commit_list *list, struct rev_info *revs)
809 {
810         struct commit_list *p;
811         int left_count = 0, right_count = 0;
812         int left_first;
813         struct patch_ids ids;
814         unsigned cherry_flag;
815
816         /* First count the commits on the left and on the right */
817         for (p = list; p; p = p->next) {
818                 struct commit *commit = p->item;
819                 unsigned flags = commit->object.flags;
820                 if (flags & BOUNDARY)
821                         ;
822                 else if (flags & SYMMETRIC_LEFT)
823                         left_count++;
824                 else
825                         right_count++;
826         }
827
828         if (!left_count || !right_count)
829                 return;
830
831         left_first = left_count < right_count;
832         init_patch_ids(&ids);
833         ids.diffopts.pathspec = revs->diffopt.pathspec;
834
835         /* Compute patch-ids for one side */
836         for (p = list; p; p = p->next) {
837                 struct commit *commit = p->item;
838                 unsigned flags = commit->object.flags;
839
840                 if (flags & BOUNDARY)
841                         continue;
842                 /*
843                  * If we have fewer left, left_first is set and we omit
844                  * commits on the right branch in this loop.  If we have
845                  * fewer right, we skip the left ones.
846                  */
847                 if (left_first != !!(flags & SYMMETRIC_LEFT))
848                         continue;
849                 commit->util = add_commit_patch_id(commit, &ids);
850         }
851
852         /* either cherry_mark or cherry_pick are true */
853         cherry_flag = revs->cherry_mark ? PATCHSAME : SHOWN;
854
855         /* Check the other side */
856         for (p = list; p; p = p->next) {
857                 struct commit *commit = p->item;
858                 struct patch_id *id;
859                 unsigned flags = commit->object.flags;
860
861                 if (flags & BOUNDARY)
862                         continue;
863                 /*
864                  * If we have fewer left, left_first is set and we omit
865                  * commits on the left branch in this loop.
866                  */
867                 if (left_first == !!(flags & SYMMETRIC_LEFT))
868                         continue;
869
870                 /*
871                  * Have we seen the same patch id?
872                  */
873                 id = has_commit_patch_id(commit, &ids);
874                 if (!id)
875                         continue;
876                 id->seen = 1;
877                 commit->object.flags |= cherry_flag;
878         }
879
880         /* Now check the original side for seen ones */
881         for (p = list; p; p = p->next) {
882                 struct commit *commit = p->item;
883                 struct patch_id *ent;
884
885                 ent = commit->util;
886                 if (!ent)
887                         continue;
888                 if (ent->seen)
889                         commit->object.flags |= cherry_flag;
890                 commit->util = NULL;
891         }
892
893         free_patch_ids(&ids);
894 }
895
896 /* How many extra uninteresting commits we want to see.. */
897 #define SLOP 5
898
899 static int still_interesting(struct commit_list *src, unsigned long date, int slop,
900                              struct commit **interesting_cache)
901 {
902         /*
903          * No source list at all? We're definitely done..
904          */
905         if (!src)
906                 return 0;
907
908         /*
909          * Does the destination list contain entries with a date
910          * before the source list? Definitely _not_ done.
911          */
912         if (date <= src->item->date)
913                 return SLOP;
914
915         /*
916          * Does the source list still have interesting commits in
917          * it? Definitely not done..
918          */
919         if (!everybody_uninteresting(src, interesting_cache))
920                 return SLOP;
921
922         /* Ok, we're closing in.. */
923         return slop-1;
924 }
925
926 /*
927  * "rev-list --ancestry-path A..B" computes commits that are ancestors
928  * of B but not ancestors of A but further limits the result to those
929  * that are descendants of A.  This takes the list of bottom commits and
930  * the result of "A..B" without --ancestry-path, and limits the latter
931  * further to the ones that can reach one of the commits in "bottom".
932  */
933 static void limit_to_ancestry(struct commit_list *bottom, struct commit_list *list)
934 {
935         struct commit_list *p;
936         struct commit_list *rlist = NULL;
937         int made_progress;
938
939         /*
940          * Reverse the list so that it will be likely that we would
941          * process parents before children.
942          */
943         for (p = list; p; p = p->next)
944                 commit_list_insert(p->item, &rlist);
945
946         for (p = bottom; p; p = p->next)
947                 p->item->object.flags |= TMP_MARK;
948
949         /*
950          * Mark the ones that can reach bottom commits in "list",
951          * in a bottom-up fashion.
952          */
953         do {
954                 made_progress = 0;
955                 for (p = rlist; p; p = p->next) {
956                         struct commit *c = p->item;
957                         struct commit_list *parents;
958                         if (c->object.flags & (TMP_MARK | UNINTERESTING))
959                                 continue;
960                         for (parents = c->parents;
961                              parents;
962                              parents = parents->next) {
963                                 if (!(parents->item->object.flags & TMP_MARK))
964                                         continue;
965                                 c->object.flags |= TMP_MARK;
966                                 made_progress = 1;
967                                 break;
968                         }
969                 }
970         } while (made_progress);
971
972         /*
973          * NEEDSWORK: decide if we want to remove parents that are
974          * not marked with TMP_MARK from commit->parents for commits
975          * in the resulting list.  We may not want to do that, though.
976          */
977
978         /*
979          * The ones that are not marked with TMP_MARK are uninteresting
980          */
981         for (p = list; p; p = p->next) {
982                 struct commit *c = p->item;
983                 if (c->object.flags & TMP_MARK)
984                         continue;
985                 c->object.flags |= UNINTERESTING;
986         }
987
988         /* We are done with the TMP_MARK */
989         for (p = list; p; p = p->next)
990                 p->item->object.flags &= ~TMP_MARK;
991         for (p = bottom; p; p = p->next)
992                 p->item->object.flags &= ~TMP_MARK;
993         free_commit_list(rlist);
994 }
995
996 /*
997  * Before walking the history, keep the set of "negative" refs the
998  * caller has asked to exclude.
999  *
1000  * This is used to compute "rev-list --ancestry-path A..B", as we need
1001  * to filter the result of "A..B" further to the ones that can actually
1002  * reach A.
1003  */
1004 static struct commit_list *collect_bottom_commits(struct commit_list *list)
1005 {
1006         struct commit_list *elem, *bottom = NULL;
1007         for (elem = list; elem; elem = elem->next)
1008                 if (elem->item->object.flags & BOTTOM)
1009                         commit_list_insert(elem->item, &bottom);
1010         return bottom;
1011 }
1012
1013 /* Assumes either left_only or right_only is set */
1014 static void limit_left_right(struct commit_list *list, struct rev_info *revs)
1015 {
1016         struct commit_list *p;
1017
1018         for (p = list; p; p = p->next) {
1019                 struct commit *commit = p->item;
1020
1021                 if (revs->right_only) {
1022                         if (commit->object.flags & SYMMETRIC_LEFT)
1023                                 commit->object.flags |= SHOWN;
1024                 } else  /* revs->left_only is set */
1025                         if (!(commit->object.flags & SYMMETRIC_LEFT))
1026                                 commit->object.flags |= SHOWN;
1027         }
1028 }
1029
1030 static int limit_list(struct rev_info *revs)
1031 {
1032         int slop = SLOP;
1033         unsigned long date = ~0ul;
1034         struct commit_list *list = revs->commits;
1035         struct commit_list *newlist = NULL;
1036         struct commit_list **p = &newlist;
1037         struct commit_list *bottom = NULL;
1038         struct commit *interesting_cache = NULL;
1039
1040         if (revs->ancestry_path) {
1041                 bottom = collect_bottom_commits(list);
1042                 if (!bottom)
1043                         die("--ancestry-path given but there are no bottom commits");
1044         }
1045
1046         while (list) {
1047                 struct commit *commit = pop_commit(&list);
1048                 struct object *obj = &commit->object;
1049                 show_early_output_fn_t show;
1050
1051                 if (commit == interesting_cache)
1052                         interesting_cache = NULL;
1053
1054                 if (revs->max_age != -1 && (commit->date < revs->max_age))
1055                         obj->flags |= UNINTERESTING;
1056                 if (add_parents_to_list(revs, commit, &list, NULL) < 0)
1057                         return -1;
1058                 if (obj->flags & UNINTERESTING) {
1059                         mark_parents_uninteresting(commit);
1060                         if (revs->show_all)
1061                                 p = &commit_list_insert(commit, p)->next;
1062                         slop = still_interesting(list, date, slop, &interesting_cache);
1063                         if (slop)
1064                                 continue;
1065                         /* If showing all, add the whole pending list to the end */
1066                         if (revs->show_all)
1067                                 *p = list;
1068                         break;
1069                 }
1070                 if (revs->min_age != -1 && (commit->date > revs->min_age))
1071                         continue;
1072                 date = commit->date;
1073                 p = &commit_list_insert(commit, p)->next;
1074
1075                 show = show_early_output;
1076                 if (!show)
1077                         continue;
1078
1079                 show(revs, newlist);
1080                 show_early_output = NULL;
1081         }
1082         if (revs->cherry_pick || revs->cherry_mark)
1083                 cherry_pick_list(newlist, revs);
1084
1085         if (revs->left_only || revs->right_only)
1086                 limit_left_right(newlist, revs);
1087
1088         if (bottom) {
1089                 limit_to_ancestry(bottom, newlist);
1090                 free_commit_list(bottom);
1091         }
1092
1093         /*
1094          * Check if any commits have become TREESAME by some of their parents
1095          * becoming UNINTERESTING.
1096          */
1097         if (limiting_can_increase_treesame(revs))
1098                 for (list = newlist; list; list = list->next) {
1099                         struct commit *c = list->item;
1100                         if (c->object.flags & (UNINTERESTING | TREESAME))
1101                                 continue;
1102                         update_treesame(revs, c);
1103                 }
1104
1105         revs->commits = newlist;
1106         return 0;
1107 }
1108
1109 /*
1110  * Add an entry to refs->cmdline with the specified information.
1111  * *name is copied.
1112  */
1113 static void add_rev_cmdline(struct rev_info *revs,
1114                             struct object *item,
1115                             const char *name,
1116                             int whence,
1117                             unsigned flags)
1118 {
1119         struct rev_cmdline_info *info = &revs->cmdline;
1120         int nr = info->nr;
1121
1122         ALLOC_GROW(info->rev, nr + 1, info->alloc);
1123         info->rev[nr].item = item;
1124         info->rev[nr].name = xstrdup(name);
1125         info->rev[nr].whence = whence;
1126         info->rev[nr].flags = flags;
1127         info->nr++;
1128 }
1129
1130 static void add_rev_cmdline_list(struct rev_info *revs,
1131                                  struct commit_list *commit_list,
1132                                  int whence,
1133                                  unsigned flags)
1134 {
1135         while (commit_list) {
1136                 struct object *object = &commit_list->item->object;
1137                 add_rev_cmdline(revs, object, sha1_to_hex(object->sha1),
1138                                 whence, flags);
1139                 commit_list = commit_list->next;
1140         }
1141 }
1142
1143 struct all_refs_cb {
1144         int all_flags;
1145         int warned_bad_reflog;
1146         struct rev_info *all_revs;
1147         const char *name_for_errormsg;
1148 };
1149
1150 int ref_excluded(struct string_list *ref_excludes, const char *path)
1151 {
1152         struct string_list_item *item;
1153
1154         if (!ref_excludes)
1155                 return 0;
1156         for_each_string_list_item(item, ref_excludes) {
1157                 if (!wildmatch(item->string, path, 0, NULL))
1158                         return 1;
1159         }
1160         return 0;
1161 }
1162
1163 static int handle_one_ref(const char *path, const struct object_id *oid,
1164                           int flag, void *cb_data)
1165 {
1166         struct all_refs_cb *cb = cb_data;
1167         struct object *object;
1168
1169         if (ref_excluded(cb->all_revs->ref_excludes, path))
1170             return 0;
1171
1172         object = get_reference(cb->all_revs, path, oid->hash, cb->all_flags);
1173         add_rev_cmdline(cb->all_revs, object, path, REV_CMD_REF, cb->all_flags);
1174         add_pending_sha1(cb->all_revs, path, oid->hash, cb->all_flags);
1175         return 0;
1176 }
1177
1178 static void init_all_refs_cb(struct all_refs_cb *cb, struct rev_info *revs,
1179         unsigned flags)
1180 {
1181         cb->all_revs = revs;
1182         cb->all_flags = flags;
1183 }
1184
1185 void clear_ref_exclusion(struct string_list **ref_excludes_p)
1186 {
1187         if (*ref_excludes_p) {
1188                 string_list_clear(*ref_excludes_p, 0);
1189                 free(*ref_excludes_p);
1190         }
1191         *ref_excludes_p = NULL;
1192 }
1193
1194 void add_ref_exclusion(struct string_list **ref_excludes_p, const char *exclude)
1195 {
1196         if (!*ref_excludes_p) {
1197                 *ref_excludes_p = xcalloc(1, sizeof(**ref_excludes_p));
1198                 (*ref_excludes_p)->strdup_strings = 1;
1199         }
1200         string_list_append(*ref_excludes_p, exclude);
1201 }
1202
1203 static void handle_refs(const char *submodule, struct rev_info *revs, unsigned flags,
1204                 int (*for_each)(const char *, each_ref_fn, void *))
1205 {
1206         struct all_refs_cb cb;
1207         init_all_refs_cb(&cb, revs, flags);
1208         for_each(submodule, handle_one_ref, &cb);
1209 }
1210
1211 static void handle_one_reflog_commit(unsigned char *sha1, void *cb_data)
1212 {
1213         struct all_refs_cb *cb = cb_data;
1214         if (!is_null_sha1(sha1)) {
1215                 struct object *o = parse_object(sha1);
1216                 if (o) {
1217                         o->flags |= cb->all_flags;
1218                         /* ??? CMDLINEFLAGS ??? */
1219                         add_pending_object(cb->all_revs, o, "");
1220                 }
1221                 else if (!cb->warned_bad_reflog) {
1222                         warning("reflog of '%s' references pruned commits",
1223                                 cb->name_for_errormsg);
1224                         cb->warned_bad_reflog = 1;
1225                 }
1226         }
1227 }
1228
1229 static int handle_one_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
1230                 const char *email, unsigned long timestamp, int tz,
1231                 const char *message, void *cb_data)
1232 {
1233         handle_one_reflog_commit(osha1, cb_data);
1234         handle_one_reflog_commit(nsha1, cb_data);
1235         return 0;
1236 }
1237
1238 static int handle_one_reflog(const char *path, const struct object_id *oid,
1239                              int flag, void *cb_data)
1240 {
1241         struct all_refs_cb *cb = cb_data;
1242         cb->warned_bad_reflog = 0;
1243         cb->name_for_errormsg = path;
1244         for_each_reflog_ent(path, handle_one_reflog_ent, cb_data);
1245         return 0;
1246 }
1247
1248 void add_reflogs_to_pending(struct rev_info *revs, unsigned flags)
1249 {
1250         struct all_refs_cb cb;
1251
1252         cb.all_revs = revs;
1253         cb.all_flags = flags;
1254         for_each_reflog(handle_one_reflog, &cb);
1255 }
1256
1257 static void add_cache_tree(struct cache_tree *it, struct rev_info *revs,
1258                            struct strbuf *path)
1259 {
1260         size_t baselen = path->len;
1261         int i;
1262
1263         if (it->entry_count >= 0) {
1264                 struct tree *tree = lookup_tree(it->sha1);
1265                 add_pending_object_with_path(revs, &tree->object, "",
1266                                              040000, path->buf);
1267         }
1268
1269         for (i = 0; i < it->subtree_nr; i++) {
1270                 struct cache_tree_sub *sub = it->down[i];
1271                 strbuf_addf(path, "%s%s", baselen ? "/" : "", sub->name);
1272                 add_cache_tree(sub->cache_tree, revs, path);
1273                 strbuf_setlen(path, baselen);
1274         }
1275
1276 }
1277
1278 void add_index_objects_to_pending(struct rev_info *revs, unsigned flags)
1279 {
1280         int i;
1281
1282         read_cache();
1283         for (i = 0; i < active_nr; i++) {
1284                 struct cache_entry *ce = active_cache[i];
1285                 struct blob *blob;
1286
1287                 if (S_ISGITLINK(ce->ce_mode))
1288                         continue;
1289
1290                 blob = lookup_blob(ce->sha1);
1291                 if (!blob)
1292                         die("unable to add index blob to traversal");
1293                 add_pending_object_with_path(revs, &blob->object, "",
1294                                              ce->ce_mode, ce->name);
1295         }
1296
1297         if (active_cache_tree) {
1298                 struct strbuf path = STRBUF_INIT;
1299                 add_cache_tree(active_cache_tree, revs, &path);
1300                 strbuf_release(&path);
1301         }
1302 }
1303
1304 static int add_parents_only(struct rev_info *revs, const char *arg_, int flags)
1305 {
1306         unsigned char sha1[20];
1307         struct object *it;
1308         struct commit *commit;
1309         struct commit_list *parents;
1310         const char *arg = arg_;
1311
1312         if (*arg == '^') {
1313                 flags ^= UNINTERESTING | BOTTOM;
1314                 arg++;
1315         }
1316         if (get_sha1_committish(arg, sha1))
1317                 return 0;
1318         while (1) {
1319                 it = get_reference(revs, arg, sha1, 0);
1320                 if (!it && revs->ignore_missing)
1321                         return 0;
1322                 if (it->type != OBJ_TAG)
1323                         break;
1324                 if (!((struct tag*)it)->tagged)
1325                         return 0;
1326                 hashcpy(sha1, ((struct tag*)it)->tagged->sha1);
1327         }
1328         if (it->type != OBJ_COMMIT)
1329                 return 0;
1330         commit = (struct commit *)it;
1331         for (parents = commit->parents; parents; parents = parents->next) {
1332                 it = &parents->item->object;
1333                 it->flags |= flags;
1334                 add_rev_cmdline(revs, it, arg_, REV_CMD_PARENTS_ONLY, flags);
1335                 add_pending_object(revs, it, arg);
1336         }
1337         return 1;
1338 }
1339
1340 void init_revisions(struct rev_info *revs, const char *prefix)
1341 {
1342         memset(revs, 0, sizeof(*revs));
1343
1344         revs->abbrev = DEFAULT_ABBREV;
1345         revs->ignore_merges = 1;
1346         revs->simplify_history = 1;
1347         DIFF_OPT_SET(&revs->pruning, RECURSIVE);
1348         DIFF_OPT_SET(&revs->pruning, QUICK);
1349         revs->pruning.add_remove = file_add_remove;
1350         revs->pruning.change = file_change;
1351         revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1352         revs->dense = 1;
1353         revs->prefix = prefix;
1354         revs->max_age = -1;
1355         revs->min_age = -1;
1356         revs->skip_count = -1;
1357         revs->max_count = -1;
1358         revs->max_parents = -1;
1359
1360         revs->commit_format = CMIT_FMT_DEFAULT;
1361
1362         init_grep_defaults();
1363         grep_init(&revs->grep_filter, prefix);
1364         revs->grep_filter.status_only = 1;
1365         revs->grep_filter.regflags = REG_NEWLINE;
1366
1367         diff_setup(&revs->diffopt);
1368         if (prefix && !revs->diffopt.prefix) {
1369                 revs->diffopt.prefix = prefix;
1370                 revs->diffopt.prefix_length = strlen(prefix);
1371         }
1372
1373         revs->notes_opt.use_default_notes = -1;
1374 }
1375
1376 static void add_pending_commit_list(struct rev_info *revs,
1377                                     struct commit_list *commit_list,
1378                                     unsigned int flags)
1379 {
1380         while (commit_list) {
1381                 struct object *object = &commit_list->item->object;
1382                 object->flags |= flags;
1383                 add_pending_object(revs, object, sha1_to_hex(object->sha1));
1384                 commit_list = commit_list->next;
1385         }
1386 }
1387
1388 static void prepare_show_merge(struct rev_info *revs)
1389 {
1390         struct commit_list *bases;
1391         struct commit *head, *other;
1392         unsigned char sha1[20];
1393         const char **prune = NULL;
1394         int i, prune_num = 1; /* counting terminating NULL */
1395
1396         if (get_sha1("HEAD", sha1))
1397                 die("--merge without HEAD?");
1398         head = lookup_commit_or_die(sha1, "HEAD");
1399         if (get_sha1("MERGE_HEAD", sha1))
1400                 die("--merge without MERGE_HEAD?");
1401         other = lookup_commit_or_die(sha1, "MERGE_HEAD");
1402         add_pending_object(revs, &head->object, "HEAD");
1403         add_pending_object(revs, &other->object, "MERGE_HEAD");
1404         bases = get_merge_bases(head, other);
1405         add_rev_cmdline_list(revs, bases, REV_CMD_MERGE_BASE, UNINTERESTING | BOTTOM);
1406         add_pending_commit_list(revs, bases, UNINTERESTING | BOTTOM);
1407         free_commit_list(bases);
1408         head->object.flags |= SYMMETRIC_LEFT;
1409
1410         if (!active_nr)
1411                 read_cache();
1412         for (i = 0; i < active_nr; i++) {
1413                 const struct cache_entry *ce = active_cache[i];
1414                 if (!ce_stage(ce))
1415                         continue;
1416                 if (ce_path_match(ce, &revs->prune_data, NULL)) {
1417                         prune_num++;
1418                         REALLOC_ARRAY(prune, prune_num);
1419                         prune[prune_num-2] = ce->name;
1420                         prune[prune_num-1] = NULL;
1421                 }
1422                 while ((i+1 < active_nr) &&
1423                        ce_same_name(ce, active_cache[i+1]))
1424                         i++;
1425         }
1426         free_pathspec(&revs->prune_data);
1427         parse_pathspec(&revs->prune_data, PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
1428                        PATHSPEC_PREFER_FULL | PATHSPEC_LITERAL_PATH, "", prune);
1429         revs->limited = 1;
1430 }
1431
1432 int handle_revision_arg(const char *arg_, struct rev_info *revs, int flags, unsigned revarg_opt)
1433 {
1434         struct object_context oc;
1435         char *dotdot;
1436         struct object *object;
1437         unsigned char sha1[20];
1438         int local_flags;
1439         const char *arg = arg_;
1440         int cant_be_filename = revarg_opt & REVARG_CANNOT_BE_FILENAME;
1441         unsigned get_sha1_flags = 0;
1442
1443         flags = flags & UNINTERESTING ? flags | BOTTOM : flags & ~BOTTOM;
1444
1445         dotdot = strstr(arg, "..");
1446         if (dotdot) {
1447                 unsigned char from_sha1[20];
1448                 const char *next = dotdot + 2;
1449                 const char *this = arg;
1450                 int symmetric = *next == '.';
1451                 unsigned int flags_exclude = flags ^ (UNINTERESTING | BOTTOM);
1452                 static const char head_by_default[] = "HEAD";
1453                 unsigned int a_flags;
1454
1455                 *dotdot = 0;
1456                 next += symmetric;
1457
1458                 if (!*next)
1459                         next = head_by_default;
1460                 if (dotdot == arg)
1461                         this = head_by_default;
1462                 if (this == head_by_default && next == head_by_default &&
1463                     !symmetric) {
1464                         /*
1465                          * Just ".."?  That is not a range but the
1466                          * pathspec for the parent directory.
1467                          */
1468                         if (!cant_be_filename) {
1469                                 *dotdot = '.';
1470                                 return -1;
1471                         }
1472                 }
1473                 if (!get_sha1_committish(this, from_sha1) &&
1474                     !get_sha1_committish(next, sha1)) {
1475                         struct object *a_obj, *b_obj;
1476
1477                         if (!cant_be_filename) {
1478                                 *dotdot = '.';
1479                                 verify_non_filename(revs->prefix, arg);
1480                         }
1481
1482                         a_obj = parse_object(from_sha1);
1483                         b_obj = parse_object(sha1);
1484                         if (!a_obj || !b_obj) {
1485                         missing:
1486                                 if (revs->ignore_missing)
1487                                         return 0;
1488                                 die(symmetric
1489                                     ? "Invalid symmetric difference expression %s"
1490                                     : "Invalid revision range %s", arg);
1491                         }
1492
1493                         if (!symmetric) {
1494                                 /* just A..B */
1495                                 a_flags = flags_exclude;
1496                         } else {
1497                                 /* A...B -- find merge bases between the two */
1498                                 struct commit *a, *b;
1499                                 struct commit_list *exclude;
1500
1501                                 a = (a_obj->type == OBJ_COMMIT
1502                                      ? (struct commit *)a_obj
1503                                      : lookup_commit_reference(a_obj->sha1));
1504                                 b = (b_obj->type == OBJ_COMMIT
1505                                      ? (struct commit *)b_obj
1506                                      : lookup_commit_reference(b_obj->sha1));
1507                                 if (!a || !b)
1508                                         goto missing;
1509                                 exclude = get_merge_bases(a, b);
1510                                 add_rev_cmdline_list(revs, exclude,
1511                                                      REV_CMD_MERGE_BASE,
1512                                                      flags_exclude);
1513                                 add_pending_commit_list(revs, exclude,
1514                                                         flags_exclude);
1515                                 free_commit_list(exclude);
1516
1517                                 a_flags = flags | SYMMETRIC_LEFT;
1518                         }
1519
1520                         a_obj->flags |= a_flags;
1521                         b_obj->flags |= flags;
1522                         add_rev_cmdline(revs, a_obj, this,
1523                                         REV_CMD_LEFT, a_flags);
1524                         add_rev_cmdline(revs, b_obj, next,
1525                                         REV_CMD_RIGHT, flags);
1526                         add_pending_object(revs, a_obj, this);
1527                         add_pending_object(revs, b_obj, next);
1528                         return 0;
1529                 }
1530                 *dotdot = '.';
1531         }
1532         dotdot = strstr(arg, "^@");
1533         if (dotdot && !dotdot[2]) {
1534                 *dotdot = 0;
1535                 if (add_parents_only(revs, arg, flags))
1536                         return 0;
1537                 *dotdot = '^';
1538         }
1539         dotdot = strstr(arg, "^!");
1540         if (dotdot && !dotdot[2]) {
1541                 *dotdot = 0;
1542                 if (!add_parents_only(revs, arg, flags ^ (UNINTERESTING | BOTTOM)))
1543                         *dotdot = '^';
1544         }
1545
1546         local_flags = 0;
1547         if (*arg == '^') {
1548                 local_flags = UNINTERESTING | BOTTOM;
1549                 arg++;
1550         }
1551
1552         if (revarg_opt & REVARG_COMMITTISH)
1553                 get_sha1_flags = GET_SHA1_COMMITTISH;
1554
1555         if (get_sha1_with_context(arg, get_sha1_flags, sha1, &oc))
1556                 return revs->ignore_missing ? 0 : -1;
1557         if (!cant_be_filename)
1558                 verify_non_filename(revs->prefix, arg);
1559         object = get_reference(revs, arg, sha1, flags ^ local_flags);
1560         add_rev_cmdline(revs, object, arg_, REV_CMD_REV, flags ^ local_flags);
1561         add_pending_object_with_mode(revs, object, arg, oc.mode);
1562         return 0;
1563 }
1564
1565 struct cmdline_pathspec {
1566         int alloc;
1567         int nr;
1568         const char **path;
1569 };
1570
1571 static void append_prune_data(struct cmdline_pathspec *prune, const char **av)
1572 {
1573         while (*av) {
1574                 ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1575                 prune->path[prune->nr++] = *(av++);
1576         }
1577 }
1578
1579 static void read_pathspec_from_stdin(struct rev_info *revs, struct strbuf *sb,
1580                                      struct cmdline_pathspec *prune)
1581 {
1582         while (strbuf_getwholeline(sb, stdin, '\n') != EOF) {
1583                 int len = sb->len;
1584                 if (len && sb->buf[len - 1] == '\n')
1585                         sb->buf[--len] = '\0';
1586                 ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1587                 prune->path[prune->nr++] = xstrdup(sb->buf);
1588         }
1589 }
1590
1591 static void read_revisions_from_stdin(struct rev_info *revs,
1592                                       struct cmdline_pathspec *prune)
1593 {
1594         struct strbuf sb;
1595         int seen_dashdash = 0;
1596         int save_warning;
1597
1598         save_warning = warn_on_object_refname_ambiguity;
1599         warn_on_object_refname_ambiguity = 0;
1600
1601         strbuf_init(&sb, 1000);
1602         while (strbuf_getwholeline(&sb, stdin, '\n') != EOF) {
1603                 int len = sb.len;
1604                 if (len && sb.buf[len - 1] == '\n')
1605                         sb.buf[--len] = '\0';
1606                 if (!len)
1607                         break;
1608                 if (sb.buf[0] == '-') {
1609                         if (len == 2 && sb.buf[1] == '-') {
1610                                 seen_dashdash = 1;
1611                                 break;
1612                         }
1613                         die("options not supported in --stdin mode");
1614                 }
1615                 if (handle_revision_arg(sb.buf, revs, 0,
1616                                         REVARG_CANNOT_BE_FILENAME))
1617                         die("bad revision '%s'", sb.buf);
1618         }
1619         if (seen_dashdash)
1620                 read_pathspec_from_stdin(revs, &sb, prune);
1621
1622         strbuf_release(&sb);
1623         warn_on_object_refname_ambiguity = save_warning;
1624 }
1625
1626 static void add_grep(struct rev_info *revs, const char *ptn, enum grep_pat_token what)
1627 {
1628         append_grep_pattern(&revs->grep_filter, ptn, "command line", 0, what);
1629 }
1630
1631 static void add_header_grep(struct rev_info *revs, enum grep_header_field field, const char *pattern)
1632 {
1633         append_header_grep_pattern(&revs->grep_filter, field, pattern);
1634 }
1635
1636 static void add_message_grep(struct rev_info *revs, const char *pattern)
1637 {
1638         add_grep(revs, pattern, GREP_PATTERN_BODY);
1639 }
1640
1641 static int handle_revision_opt(struct rev_info *revs, int argc, const char **argv,
1642                                int *unkc, const char **unkv)
1643 {
1644         const char *arg = argv[0];
1645         const char *optarg;
1646         int argcount;
1647
1648         /* pseudo revision arguments */
1649         if (!strcmp(arg, "--all") || !strcmp(arg, "--branches") ||
1650             !strcmp(arg, "--tags") || !strcmp(arg, "--remotes") ||
1651             !strcmp(arg, "--reflog") || !strcmp(arg, "--not") ||
1652             !strcmp(arg, "--no-walk") || !strcmp(arg, "--do-walk") ||
1653             !strcmp(arg, "--bisect") || starts_with(arg, "--glob=") ||
1654             !strcmp(arg, "--indexed-objects") ||
1655             starts_with(arg, "--exclude=") ||
1656             starts_with(arg, "--branches=") || starts_with(arg, "--tags=") ||
1657             starts_with(arg, "--remotes=") || starts_with(arg, "--no-walk="))
1658         {
1659                 unkv[(*unkc)++] = arg;
1660                 return 1;
1661         }
1662
1663         if ((argcount = parse_long_opt("max-count", argv, &optarg))) {
1664                 revs->max_count = atoi(optarg);
1665                 revs->no_walk = 0;
1666                 return argcount;
1667         } else if ((argcount = parse_long_opt("skip", argv, &optarg))) {
1668                 revs->skip_count = atoi(optarg);
1669                 return argcount;
1670         } else if ((*arg == '-') && isdigit(arg[1])) {
1671                 /* accept -<digit>, like traditional "head" */
1672                 if (strtol_i(arg + 1, 10, &revs->max_count) < 0 ||
1673                     revs->max_count < 0)
1674                         die("'%s': not a non-negative integer", arg + 1);
1675                 revs->no_walk = 0;
1676         } else if (!strcmp(arg, "-n")) {
1677                 if (argc <= 1)
1678                         return error("-n requires an argument");
1679                 revs->max_count = atoi(argv[1]);
1680                 revs->no_walk = 0;
1681                 return 2;
1682         } else if (starts_with(arg, "-n")) {
1683                 revs->max_count = atoi(arg + 2);
1684                 revs->no_walk = 0;
1685         } else if ((argcount = parse_long_opt("max-age", argv, &optarg))) {
1686                 revs->max_age = atoi(optarg);
1687                 return argcount;
1688         } else if ((argcount = parse_long_opt("since", argv, &optarg))) {
1689                 revs->max_age = approxidate(optarg);
1690                 return argcount;
1691         } else if ((argcount = parse_long_opt("after", argv, &optarg))) {
1692                 revs->max_age = approxidate(optarg);
1693                 return argcount;
1694         } else if ((argcount = parse_long_opt("min-age", argv, &optarg))) {
1695                 revs->min_age = atoi(optarg);
1696                 return argcount;
1697         } else if ((argcount = parse_long_opt("before", argv, &optarg))) {
1698                 revs->min_age = approxidate(optarg);
1699                 return argcount;
1700         } else if ((argcount = parse_long_opt("until", argv, &optarg))) {
1701                 revs->min_age = approxidate(optarg);
1702                 return argcount;
1703         } else if (!strcmp(arg, "--first-parent")) {
1704                 revs->first_parent_only = 1;
1705         } else if (!strcmp(arg, "--ancestry-path")) {
1706                 revs->ancestry_path = 1;
1707                 revs->simplify_history = 0;
1708                 revs->limited = 1;
1709         } else if (!strcmp(arg, "-g") || !strcmp(arg, "--walk-reflogs")) {
1710                 init_reflog_walk(&revs->reflog_info);
1711         } else if (!strcmp(arg, "--default")) {
1712                 if (argc <= 1)
1713                         return error("bad --default argument");
1714                 revs->def = argv[1];
1715                 return 2;
1716         } else if (!strcmp(arg, "--merge")) {
1717                 revs->show_merge = 1;
1718         } else if (!strcmp(arg, "--topo-order")) {
1719                 revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1720                 revs->topo_order = 1;
1721         } else if (!strcmp(arg, "--simplify-merges")) {
1722                 revs->simplify_merges = 1;
1723                 revs->topo_order = 1;
1724                 revs->rewrite_parents = 1;
1725                 revs->simplify_history = 0;
1726                 revs->limited = 1;
1727         } else if (!strcmp(arg, "--simplify-by-decoration")) {
1728                 revs->simplify_merges = 1;
1729                 revs->topo_order = 1;
1730                 revs->rewrite_parents = 1;
1731                 revs->simplify_history = 0;
1732                 revs->simplify_by_decoration = 1;
1733                 revs->limited = 1;
1734                 revs->prune = 1;
1735                 load_ref_decorations(DECORATE_SHORT_REFS);
1736         } else if (!strcmp(arg, "--date-order")) {
1737                 revs->sort_order = REV_SORT_BY_COMMIT_DATE;
1738                 revs->topo_order = 1;
1739         } else if (!strcmp(arg, "--author-date-order")) {
1740                 revs->sort_order = REV_SORT_BY_AUTHOR_DATE;
1741                 revs->topo_order = 1;
1742         } else if (starts_with(arg, "--early-output")) {
1743                 int count = 100;
1744                 switch (arg[14]) {
1745                 case '=':
1746                         count = atoi(arg+15);
1747                         /* Fallthrough */
1748                 case 0:
1749                         revs->topo_order = 1;
1750                        revs->early_output = count;
1751                 }
1752         } else if (!strcmp(arg, "--parents")) {
1753                 revs->rewrite_parents = 1;
1754                 revs->print_parents = 1;
1755         } else if (!strcmp(arg, "--dense")) {
1756                 revs->dense = 1;
1757         } else if (!strcmp(arg, "--sparse")) {
1758                 revs->dense = 0;
1759         } else if (!strcmp(arg, "--show-all")) {
1760                 revs->show_all = 1;
1761         } else if (!strcmp(arg, "--remove-empty")) {
1762                 revs->remove_empty_trees = 1;
1763         } else if (!strcmp(arg, "--merges")) {
1764                 revs->min_parents = 2;
1765         } else if (!strcmp(arg, "--no-merges")) {
1766                 revs->max_parents = 1;
1767         } else if (starts_with(arg, "--min-parents=")) {
1768                 revs->min_parents = atoi(arg+14);
1769         } else if (starts_with(arg, "--no-min-parents")) {
1770                 revs->min_parents = 0;
1771         } else if (starts_with(arg, "--max-parents=")) {
1772                 revs->max_parents = atoi(arg+14);
1773         } else if (starts_with(arg, "--no-max-parents")) {
1774                 revs->max_parents = -1;
1775         } else if (!strcmp(arg, "--boundary")) {
1776                 revs->boundary = 1;
1777         } else if (!strcmp(arg, "--left-right")) {
1778                 revs->left_right = 1;
1779         } else if (!strcmp(arg, "--left-only")) {
1780                 if (revs->right_only)
1781                         die("--left-only is incompatible with --right-only"
1782                             " or --cherry");
1783                 revs->left_only = 1;
1784         } else if (!strcmp(arg, "--right-only")) {
1785                 if (revs->left_only)
1786                         die("--right-only is incompatible with --left-only");
1787                 revs->right_only = 1;
1788         } else if (!strcmp(arg, "--cherry")) {
1789                 if (revs->left_only)
1790                         die("--cherry is incompatible with --left-only");
1791                 revs->cherry_mark = 1;
1792                 revs->right_only = 1;
1793                 revs->max_parents = 1;
1794                 revs->limited = 1;
1795         } else if (!strcmp(arg, "--count")) {
1796                 revs->count = 1;
1797         } else if (!strcmp(arg, "--cherry-mark")) {
1798                 if (revs->cherry_pick)
1799                         die("--cherry-mark is incompatible with --cherry-pick");
1800                 revs->cherry_mark = 1;
1801                 revs->limited = 1; /* needs limit_list() */
1802         } else if (!strcmp(arg, "--cherry-pick")) {
1803                 if (revs->cherry_mark)
1804                         die("--cherry-pick is incompatible with --cherry-mark");
1805                 revs->cherry_pick = 1;
1806                 revs->limited = 1;
1807         } else if (!strcmp(arg, "--objects")) {
1808                 revs->tag_objects = 1;
1809                 revs->tree_objects = 1;
1810                 revs->blob_objects = 1;
1811         } else if (!strcmp(arg, "--objects-edge")) {
1812                 revs->tag_objects = 1;
1813                 revs->tree_objects = 1;
1814                 revs->blob_objects = 1;
1815                 revs->edge_hint = 1;
1816         } else if (!strcmp(arg, "--objects-edge-aggressive")) {
1817                 revs->tag_objects = 1;
1818                 revs->tree_objects = 1;
1819                 revs->blob_objects = 1;
1820                 revs->edge_hint = 1;
1821                 revs->edge_hint_aggressive = 1;
1822         } else if (!strcmp(arg, "--verify-objects")) {
1823                 revs->tag_objects = 1;
1824                 revs->tree_objects = 1;
1825                 revs->blob_objects = 1;
1826                 revs->verify_objects = 1;
1827         } else if (!strcmp(arg, "--unpacked")) {
1828                 revs->unpacked = 1;
1829         } else if (starts_with(arg, "--unpacked=")) {
1830                 die("--unpacked=<packfile> no longer supported.");
1831         } else if (!strcmp(arg, "-r")) {
1832                 revs->diff = 1;
1833                 DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1834         } else if (!strcmp(arg, "-t")) {
1835                 revs->diff = 1;
1836                 DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1837                 DIFF_OPT_SET(&revs->diffopt, TREE_IN_RECURSIVE);
1838         } else if (!strcmp(arg, "-m")) {
1839                 revs->ignore_merges = 0;
1840         } else if (!strcmp(arg, "-c")) {
1841                 revs->diff = 1;
1842                 revs->dense_combined_merges = 0;
1843                 revs->combine_merges = 1;
1844         } else if (!strcmp(arg, "--cc")) {
1845                 revs->diff = 1;
1846                 revs->dense_combined_merges = 1;
1847                 revs->combine_merges = 1;
1848         } else if (!strcmp(arg, "-v")) {
1849                 revs->verbose_header = 1;
1850         } else if (!strcmp(arg, "--pretty")) {
1851                 revs->verbose_header = 1;
1852                 revs->pretty_given = 1;
1853                 get_commit_format(NULL, revs);
1854         } else if (starts_with(arg, "--pretty=") || starts_with(arg, "--format=")) {
1855                 /*
1856                  * Detached form ("--pretty X" as opposed to "--pretty=X")
1857                  * not allowed, since the argument is optional.
1858                  */
1859                 revs->verbose_header = 1;
1860                 revs->pretty_given = 1;
1861                 get_commit_format(arg+9, revs);
1862         } else if (!strcmp(arg, "--show-notes") || !strcmp(arg, "--notes")) {
1863                 revs->show_notes = 1;
1864                 revs->show_notes_given = 1;
1865                 revs->notes_opt.use_default_notes = 1;
1866         } else if (!strcmp(arg, "--show-signature")) {
1867                 revs->show_signature = 1;
1868         } else if (!strcmp(arg, "--show-linear-break") ||
1869                    starts_with(arg, "--show-linear-break=")) {
1870                 if (starts_with(arg, "--show-linear-break="))
1871                         revs->break_bar = xstrdup(arg + 20);
1872                 else
1873                         revs->break_bar = "                    ..........";
1874                 revs->track_linear = 1;
1875                 revs->track_first_time = 1;
1876         } else if (starts_with(arg, "--show-notes=") ||
1877                    starts_with(arg, "--notes=")) {
1878                 struct strbuf buf = STRBUF_INIT;
1879                 revs->show_notes = 1;
1880                 revs->show_notes_given = 1;
1881                 if (starts_with(arg, "--show-notes")) {
1882                         if (revs->notes_opt.use_default_notes < 0)
1883                                 revs->notes_opt.use_default_notes = 1;
1884                         strbuf_addstr(&buf, arg+13);
1885                 }
1886                 else
1887                         strbuf_addstr(&buf, arg+8);
1888                 expand_notes_ref(&buf);
1889                 string_list_append(&revs->notes_opt.extra_notes_refs,
1890                                    strbuf_detach(&buf, NULL));
1891         } else if (!strcmp(arg, "--no-notes")) {
1892                 revs->show_notes = 0;
1893                 revs->show_notes_given = 1;
1894                 revs->notes_opt.use_default_notes = -1;
1895                 /* we have been strdup'ing ourselves, so trick
1896                  * string_list into free()ing strings */
1897                 revs->notes_opt.extra_notes_refs.strdup_strings = 1;
1898                 string_list_clear(&revs->notes_opt.extra_notes_refs, 0);
1899                 revs->notes_opt.extra_notes_refs.strdup_strings = 0;
1900         } else if (!strcmp(arg, "--standard-notes")) {
1901                 revs->show_notes_given = 1;
1902                 revs->notes_opt.use_default_notes = 1;
1903         } else if (!strcmp(arg, "--no-standard-notes")) {
1904                 revs->notes_opt.use_default_notes = 0;
1905         } else if (!strcmp(arg, "--oneline")) {
1906                 revs->verbose_header = 1;
1907                 get_commit_format("oneline", revs);
1908                 revs->pretty_given = 1;
1909                 revs->abbrev_commit = 1;
1910         } else if (!strcmp(arg, "--graph")) {
1911                 revs->topo_order = 1;
1912                 revs->rewrite_parents = 1;
1913                 revs->graph = graph_init(revs);
1914         } else if (!strcmp(arg, "--root")) {
1915                 revs->show_root_diff = 1;
1916         } else if (!strcmp(arg, "--no-commit-id")) {
1917                 revs->no_commit_id = 1;
1918         } else if (!strcmp(arg, "--always")) {
1919                 revs->always_show_header = 1;
1920         } else if (!strcmp(arg, "--no-abbrev")) {
1921                 revs->abbrev = 0;
1922         } else if (!strcmp(arg, "--abbrev")) {
1923                 revs->abbrev = DEFAULT_ABBREV;
1924         } else if (starts_with(arg, "--abbrev=")) {
1925                 revs->abbrev = strtoul(arg + 9, NULL, 10);
1926                 if (revs->abbrev < MINIMUM_ABBREV)
1927                         revs->abbrev = MINIMUM_ABBREV;
1928                 else if (revs->abbrev > 40)
1929                         revs->abbrev = 40;
1930         } else if (!strcmp(arg, "--abbrev-commit")) {
1931                 revs->abbrev_commit = 1;
1932                 revs->abbrev_commit_given = 1;
1933         } else if (!strcmp(arg, "--no-abbrev-commit")) {
1934                 revs->abbrev_commit = 0;
1935         } else if (!strcmp(arg, "--full-diff")) {
1936                 revs->diff = 1;
1937                 revs->full_diff = 1;
1938         } else if (!strcmp(arg, "--full-history")) {
1939                 revs->simplify_history = 0;
1940         } else if (!strcmp(arg, "--relative-date")) {
1941                 revs->date_mode.type = DATE_RELATIVE;
1942                 revs->date_mode_explicit = 1;
1943         } else if ((argcount = parse_long_opt("date", argv, &optarg))) {
1944                 parse_date_format(optarg, &revs->date_mode);
1945                 revs->date_mode_explicit = 1;
1946                 return argcount;
1947         } else if (!strcmp(arg, "--log-size")) {
1948                 revs->show_log_size = 1;
1949         }
1950         /*
1951          * Grepping the commit log
1952          */
1953         else if ((argcount = parse_long_opt("author", argv, &optarg))) {
1954                 add_header_grep(revs, GREP_HEADER_AUTHOR, optarg);
1955                 return argcount;
1956         } else if ((argcount = parse_long_opt("committer", argv, &optarg))) {
1957                 add_header_grep(revs, GREP_HEADER_COMMITTER, optarg);
1958                 return argcount;
1959         } else if ((argcount = parse_long_opt("grep-reflog", argv, &optarg))) {
1960                 add_header_grep(revs, GREP_HEADER_REFLOG, optarg);
1961                 return argcount;
1962         } else if ((argcount = parse_long_opt("grep", argv, &optarg))) {
1963                 add_message_grep(revs, optarg);
1964                 return argcount;
1965         } else if (!strcmp(arg, "--grep-debug")) {
1966                 revs->grep_filter.debug = 1;
1967         } else if (!strcmp(arg, "--basic-regexp")) {
1968                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_BRE, &revs->grep_filter);
1969         } else if (!strcmp(arg, "--extended-regexp") || !strcmp(arg, "-E")) {
1970                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, &revs->grep_filter);
1971         } else if (!strcmp(arg, "--regexp-ignore-case") || !strcmp(arg, "-i")) {
1972                 revs->grep_filter.regflags |= REG_ICASE;
1973                 DIFF_OPT_SET(&revs->diffopt, PICKAXE_IGNORE_CASE);
1974         } else if (!strcmp(arg, "--fixed-strings") || !strcmp(arg, "-F")) {
1975                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_FIXED, &revs->grep_filter);
1976         } else if (!strcmp(arg, "--perl-regexp")) {
1977                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_PCRE, &revs->grep_filter);
1978         } else if (!strcmp(arg, "--all-match")) {
1979                 revs->grep_filter.all_match = 1;
1980         } else if (!strcmp(arg, "--invert-grep")) {
1981                 revs->invert_grep = 1;
1982         } else if ((argcount = parse_long_opt("encoding", argv, &optarg))) {
1983                 if (strcmp(optarg, "none"))
1984                         git_log_output_encoding = xstrdup(optarg);
1985                 else
1986                         git_log_output_encoding = "";
1987                 return argcount;
1988         } else if (!strcmp(arg, "--reverse")) {
1989                 revs->reverse ^= 1;
1990         } else if (!strcmp(arg, "--children")) {
1991                 revs->children.name = "children";
1992                 revs->limited = 1;
1993         } else if (!strcmp(arg, "--ignore-missing")) {
1994                 revs->ignore_missing = 1;
1995         } else {
1996                 int opts = diff_opt_parse(&revs->diffopt, argv, argc);
1997                 if (!opts)
1998                         unkv[(*unkc)++] = arg;
1999                 return opts;
2000         }
2001         if (revs->graph && revs->track_linear)
2002                 die("--show-linear-break and --graph are incompatible");
2003
2004         return 1;
2005 }
2006
2007 void parse_revision_opt(struct rev_info *revs, struct parse_opt_ctx_t *ctx,
2008                         const struct option *options,
2009                         const char * const usagestr[])
2010 {
2011         int n = handle_revision_opt(revs, ctx->argc, ctx->argv,
2012                                     &ctx->cpidx, ctx->out);
2013         if (n <= 0) {
2014                 error("unknown option `%s'", ctx->argv[0]);
2015                 usage_with_options(usagestr, options);
2016         }
2017         ctx->argv += n;
2018         ctx->argc -= n;
2019 }
2020
2021 static int for_each_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data, const char *term) {
2022         struct strbuf bisect_refs = STRBUF_INIT;
2023         int status;
2024         strbuf_addf(&bisect_refs, "refs/bisect/%s", term);
2025         status = for_each_ref_in_submodule(submodule, bisect_refs.buf, fn, cb_data);
2026         strbuf_release(&bisect_refs);
2027         return status;
2028 }
2029
2030 static int for_each_bad_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2031 {
2032         return for_each_bisect_ref(submodule, fn, cb_data, term_bad);
2033 }
2034
2035 static int for_each_good_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2036 {
2037         return for_each_bisect_ref(submodule, fn, cb_data, term_good);
2038 }
2039
2040 static int handle_revision_pseudo_opt(const char *submodule,
2041                                 struct rev_info *revs,
2042                                 int argc, const char **argv, int *flags)
2043 {
2044         const char *arg = argv[0];
2045         const char *optarg;
2046         int argcount;
2047
2048         /*
2049          * NOTE!
2050          *
2051          * Commands like "git shortlog" will not accept the options below
2052          * unless parse_revision_opt queues them (as opposed to erroring
2053          * out).
2054          *
2055          * When implementing your new pseudo-option, remember to
2056          * register it in the list at the top of handle_revision_opt.
2057          */
2058         if (!strcmp(arg, "--all")) {
2059                 handle_refs(submodule, revs, *flags, for_each_ref_submodule);
2060                 handle_refs(submodule, revs, *flags, head_ref_submodule);
2061                 clear_ref_exclusion(&revs->ref_excludes);
2062         } else if (!strcmp(arg, "--branches")) {
2063                 handle_refs(submodule, revs, *flags, for_each_branch_ref_submodule);
2064                 clear_ref_exclusion(&revs->ref_excludes);
2065         } else if (!strcmp(arg, "--bisect")) {
2066                 read_bisect_terms(&term_bad, &term_good);
2067                 handle_refs(submodule, revs, *flags, for_each_bad_bisect_ref);
2068                 handle_refs(submodule, revs, *flags ^ (UNINTERESTING | BOTTOM), for_each_good_bisect_ref);
2069                 revs->bisect = 1;
2070         } else if (!strcmp(arg, "--tags")) {
2071                 handle_refs(submodule, revs, *flags, for_each_tag_ref_submodule);
2072                 clear_ref_exclusion(&revs->ref_excludes);
2073         } else if (!strcmp(arg, "--remotes")) {
2074                 handle_refs(submodule, revs, *flags, for_each_remote_ref_submodule);
2075                 clear_ref_exclusion(&revs->ref_excludes);
2076         } else if ((argcount = parse_long_opt("glob", argv, &optarg))) {
2077                 struct all_refs_cb cb;
2078                 init_all_refs_cb(&cb, revs, *flags);
2079                 for_each_glob_ref(handle_one_ref, optarg, &cb);
2080                 clear_ref_exclusion(&revs->ref_excludes);
2081                 return argcount;
2082         } else if ((argcount = parse_long_opt("exclude", argv, &optarg))) {
2083                 add_ref_exclusion(&revs->ref_excludes, optarg);
2084                 return argcount;
2085         } else if (starts_with(arg, "--branches=")) {
2086                 struct all_refs_cb cb;
2087                 init_all_refs_cb(&cb, revs, *flags);
2088                 for_each_glob_ref_in(handle_one_ref, arg + 11, "refs/heads/", &cb);
2089                 clear_ref_exclusion(&revs->ref_excludes);
2090         } else if (starts_with(arg, "--tags=")) {
2091                 struct all_refs_cb cb;
2092                 init_all_refs_cb(&cb, revs, *flags);
2093                 for_each_glob_ref_in(handle_one_ref, arg + 7, "refs/tags/", &cb);
2094                 clear_ref_exclusion(&revs->ref_excludes);
2095         } else if (starts_with(arg, "--remotes=")) {
2096                 struct all_refs_cb cb;
2097                 init_all_refs_cb(&cb, revs, *flags);
2098                 for_each_glob_ref_in(handle_one_ref, arg + 10, "refs/remotes/", &cb);
2099                 clear_ref_exclusion(&revs->ref_excludes);
2100         } else if (!strcmp(arg, "--reflog")) {
2101                 add_reflogs_to_pending(revs, *flags);
2102         } else if (!strcmp(arg, "--indexed-objects")) {
2103                 add_index_objects_to_pending(revs, *flags);
2104         } else if (!strcmp(arg, "--not")) {
2105                 *flags ^= UNINTERESTING | BOTTOM;
2106         } else if (!strcmp(arg, "--no-walk")) {
2107                 revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2108         } else if (starts_with(arg, "--no-walk=")) {
2109                 /*
2110                  * Detached form ("--no-walk X" as opposed to "--no-walk=X")
2111                  * not allowed, since the argument is optional.
2112                  */
2113                 if (!strcmp(arg + 10, "sorted"))
2114                         revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2115                 else if (!strcmp(arg + 10, "unsorted"))
2116                         revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
2117                 else
2118                         return error("invalid argument to --no-walk");
2119         } else if (!strcmp(arg, "--do-walk")) {
2120                 revs->no_walk = 0;
2121         } else {
2122                 return 0;
2123         }
2124
2125         return 1;
2126 }
2127
2128 static void NORETURN diagnose_missing_default(const char *def)
2129 {
2130         unsigned char sha1[20];
2131         int flags;
2132         const char *refname;
2133
2134         refname = resolve_ref_unsafe(def, 0, sha1, &flags);
2135         if (!refname || !(flags & REF_ISSYMREF) || (flags & REF_ISBROKEN))
2136                 die(_("your current branch appears to be broken"));
2137
2138         skip_prefix(refname, "refs/heads/", &refname);
2139         die(_("your current branch '%s' does not have any commits yet"),
2140             refname);
2141 }
2142
2143 /*
2144  * Parse revision information, filling in the "rev_info" structure,
2145  * and removing the used arguments from the argument list.
2146  *
2147  * Returns the number of arguments left that weren't recognized
2148  * (which are also moved to the head of the argument list)
2149  */
2150 int setup_revisions(int argc, const char **argv, struct rev_info *revs, struct setup_revision_opt *opt)
2151 {
2152         int i, flags, left, seen_dashdash, read_from_stdin, got_rev_arg = 0, revarg_opt;
2153         struct cmdline_pathspec prune_data;
2154         const char *submodule = NULL;
2155
2156         memset(&prune_data, 0, sizeof(prune_data));
2157         if (opt)
2158                 submodule = opt->submodule;
2159
2160         /* First, search for "--" */
2161         if (opt && opt->assume_dashdash) {
2162                 seen_dashdash = 1;
2163         } else {
2164                 seen_dashdash = 0;
2165                 for (i = 1; i < argc; i++) {
2166                         const char *arg = argv[i];
2167                         if (strcmp(arg, "--"))
2168                                 continue;
2169                         argv[i] = NULL;
2170                         argc = i;
2171                         if (argv[i + 1])
2172                                 append_prune_data(&prune_data, argv + i + 1);
2173                         seen_dashdash = 1;
2174                         break;
2175                 }
2176         }
2177
2178         /* Second, deal with arguments and options */
2179         flags = 0;
2180         revarg_opt = opt ? opt->revarg_opt : 0;
2181         if (seen_dashdash)
2182                 revarg_opt |= REVARG_CANNOT_BE_FILENAME;
2183         read_from_stdin = 0;
2184         for (left = i = 1; i < argc; i++) {
2185                 const char *arg = argv[i];
2186                 if (*arg == '-') {
2187                         int opts;
2188
2189                         opts = handle_revision_pseudo_opt(submodule,
2190                                                 revs, argc - i, argv + i,
2191                                                 &flags);
2192                         if (opts > 0) {
2193                                 i += opts - 1;
2194                                 continue;
2195                         }
2196
2197                         if (!strcmp(arg, "--stdin")) {
2198                                 if (revs->disable_stdin) {
2199                                         argv[left++] = arg;
2200                                         continue;
2201                                 }
2202                                 if (read_from_stdin++)
2203                                         die("--stdin given twice?");
2204                                 read_revisions_from_stdin(revs, &prune_data);
2205                                 continue;
2206                         }
2207
2208                         opts = handle_revision_opt(revs, argc - i, argv + i, &left, argv);
2209                         if (opts > 0) {
2210                                 i += opts - 1;
2211                                 continue;
2212                         }
2213                         if (opts < 0)
2214                                 exit(128);
2215                         continue;
2216                 }
2217
2218
2219                 if (handle_revision_arg(arg, revs, flags, revarg_opt)) {
2220                         int j;
2221                         if (seen_dashdash || *arg == '^')
2222                                 die("bad revision '%s'", arg);
2223
2224                         /* If we didn't have a "--":
2225                          * (1) all filenames must exist;
2226                          * (2) all rev-args must not be interpretable
2227                          *     as a valid filename.
2228                          * but the latter we have checked in the main loop.
2229                          */
2230                         for (j = i; j < argc; j++)
2231                                 verify_filename(revs->prefix, argv[j], j == i);
2232
2233                         append_prune_data(&prune_data, argv + i);
2234                         break;
2235                 }
2236                 else
2237                         got_rev_arg = 1;
2238         }
2239
2240         if (prune_data.nr) {
2241                 /*
2242                  * If we need to introduce the magic "a lone ':' means no
2243                  * pathspec whatsoever", here is the place to do so.
2244                  *
2245                  * if (prune_data.nr == 1 && !strcmp(prune_data[0], ":")) {
2246                  *      prune_data.nr = 0;
2247                  *      prune_data.alloc = 0;
2248                  *      free(prune_data.path);
2249                  *      prune_data.path = NULL;
2250                  * } else {
2251                  *      terminate prune_data.alloc with NULL and
2252                  *      call init_pathspec() to set revs->prune_data here.
2253                  * }
2254                  */
2255                 ALLOC_GROW(prune_data.path, prune_data.nr + 1, prune_data.alloc);
2256                 prune_data.path[prune_data.nr++] = NULL;
2257                 parse_pathspec(&revs->prune_data, 0, 0,
2258                                revs->prefix, prune_data.path);
2259         }
2260
2261         if (revs->def == NULL)
2262                 revs->def = opt ? opt->def : NULL;
2263         if (opt && opt->tweak)
2264                 opt->tweak(revs, opt);
2265         if (revs->show_merge)
2266                 prepare_show_merge(revs);
2267         if (revs->def && !revs->pending.nr && !got_rev_arg) {
2268                 unsigned char sha1[20];
2269                 struct object *object;
2270                 struct object_context oc;
2271                 if (get_sha1_with_context(revs->def, 0, sha1, &oc))
2272                         diagnose_missing_default(revs->def);
2273                 object = get_reference(revs, revs->def, sha1, 0);
2274                 add_pending_object_with_mode(revs, object, revs->def, oc.mode);
2275         }
2276
2277         /* Did the user ask for any diff output? Run the diff! */
2278         if (revs->diffopt.output_format & ~DIFF_FORMAT_NO_OUTPUT)
2279                 revs->diff = 1;
2280
2281         /* Pickaxe, diff-filter and rename following need diffs */
2282         if (revs->diffopt.pickaxe ||
2283             revs->diffopt.filter ||
2284             DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2285                 revs->diff = 1;
2286
2287         if (revs->topo_order)
2288                 revs->limited = 1;
2289
2290         if (revs->prune_data.nr) {
2291                 copy_pathspec(&revs->pruning.pathspec, &revs->prune_data);
2292                 /* Can't prune commits with rename following: the paths change.. */
2293                 if (!DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2294                         revs->prune = 1;
2295                 if (!revs->full_diff)
2296                         copy_pathspec(&revs->diffopt.pathspec,
2297                                       &revs->prune_data);
2298         }
2299         if (revs->combine_merges)
2300                 revs->ignore_merges = 0;
2301         revs->diffopt.abbrev = revs->abbrev;
2302
2303         if (revs->line_level_traverse) {
2304                 revs->limited = 1;
2305                 revs->topo_order = 1;
2306         }
2307
2308         diff_setup_done(&revs->diffopt);
2309
2310         grep_commit_pattern_type(GREP_PATTERN_TYPE_UNSPECIFIED,
2311                                  &revs->grep_filter);
2312         compile_grep_patterns(&revs->grep_filter);
2313
2314         if (revs->reverse && revs->reflog_info)
2315                 die("cannot combine --reverse with --walk-reflogs");
2316         if (revs->rewrite_parents && revs->children.name)
2317                 die("cannot combine --parents and --children");
2318
2319         /*
2320          * Limitations on the graph functionality
2321          */
2322         if (revs->reverse && revs->graph)
2323                 die("cannot combine --reverse with --graph");
2324
2325         if (revs->reflog_info && revs->graph)
2326                 die("cannot combine --walk-reflogs with --graph");
2327         if (revs->no_walk && revs->graph)
2328                 die("cannot combine --no-walk with --graph");
2329         if (!revs->reflog_info && revs->grep_filter.use_reflog_filter)
2330                 die("cannot use --grep-reflog without --walk-reflogs");
2331
2332         if (revs->first_parent_only && revs->bisect)
2333                 die(_("--first-parent is incompatible with --bisect"));
2334
2335         return left;
2336 }
2337
2338 static void add_child(struct rev_info *revs, struct commit *parent, struct commit *child)
2339 {
2340         struct commit_list *l = xcalloc(1, sizeof(*l));
2341
2342         l->item = child;
2343         l->next = add_decoration(&revs->children, &parent->object, l);
2344 }
2345
2346 static int remove_duplicate_parents(struct rev_info *revs, struct commit *commit)
2347 {
2348         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2349         struct commit_list **pp, *p;
2350         int surviving_parents;
2351
2352         /* Examine existing parents while marking ones we have seen... */
2353         pp = &commit->parents;
2354         surviving_parents = 0;
2355         while ((p = *pp) != NULL) {
2356                 struct commit *parent = p->item;
2357                 if (parent->object.flags & TMP_MARK) {
2358                         *pp = p->next;
2359                         if (ts)
2360                                 compact_treesame(revs, commit, surviving_parents);
2361                         continue;
2362                 }
2363                 parent->object.flags |= TMP_MARK;
2364                 surviving_parents++;
2365                 pp = &p->next;
2366         }
2367         /* clear the temporary mark */
2368         for (p = commit->parents; p; p = p->next) {
2369                 p->item->object.flags &= ~TMP_MARK;
2370         }
2371         /* no update_treesame() - removing duplicates can't affect TREESAME */
2372         return surviving_parents;
2373 }
2374
2375 struct merge_simplify_state {
2376         struct commit *simplified;
2377 };
2378
2379 static struct merge_simplify_state *locate_simplify_state(struct rev_info *revs, struct commit *commit)
2380 {
2381         struct merge_simplify_state *st;
2382
2383         st = lookup_decoration(&revs->merge_simplification, &commit->object);
2384         if (!st) {
2385                 st = xcalloc(1, sizeof(*st));
2386                 add_decoration(&revs->merge_simplification, &commit->object, st);
2387         }
2388         return st;
2389 }
2390
2391 static int mark_redundant_parents(struct rev_info *revs, struct commit *commit)
2392 {
2393         struct commit_list *h = reduce_heads(commit->parents);
2394         int i = 0, marked = 0;
2395         struct commit_list *po, *pn;
2396
2397         /* Want these for sanity-checking only */
2398         int orig_cnt = commit_list_count(commit->parents);
2399         int cnt = commit_list_count(h);
2400
2401         /*
2402          * Not ready to remove items yet, just mark them for now, based
2403          * on the output of reduce_heads(). reduce_heads outputs the reduced
2404          * set in its original order, so this isn't too hard.
2405          */
2406         po = commit->parents;
2407         pn = h;
2408         while (po) {
2409                 if (pn && po->item == pn->item) {
2410                         pn = pn->next;
2411                         i++;
2412                 } else {
2413                         po->item->object.flags |= TMP_MARK;
2414                         marked++;
2415                 }
2416                 po=po->next;
2417         }
2418
2419         if (i != cnt || cnt+marked != orig_cnt)
2420                 die("mark_redundant_parents %d %d %d %d", orig_cnt, cnt, i, marked);
2421
2422         free_commit_list(h);
2423
2424         return marked;
2425 }
2426
2427 static int mark_treesame_root_parents(struct rev_info *revs, struct commit *commit)
2428 {
2429         struct commit_list *p;
2430         int marked = 0;
2431
2432         for (p = commit->parents; p; p = p->next) {
2433                 struct commit *parent = p->item;
2434                 if (!parent->parents && (parent->object.flags & TREESAME)) {
2435                         parent->object.flags |= TMP_MARK;
2436                         marked++;
2437                 }
2438         }
2439
2440         return marked;
2441 }
2442
2443 /*
2444  * Awkward naming - this means one parent we are TREESAME to.
2445  * cf mark_treesame_root_parents: root parents that are TREESAME (to an
2446  * empty tree). Better name suggestions?
2447  */
2448 static int leave_one_treesame_to_parent(struct rev_info *revs, struct commit *commit)
2449 {
2450         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2451         struct commit *unmarked = NULL, *marked = NULL;
2452         struct commit_list *p;
2453         unsigned n;
2454
2455         for (p = commit->parents, n = 0; p; p = p->next, n++) {
2456                 if (ts->treesame[n]) {
2457                         if (p->item->object.flags & TMP_MARK) {
2458                                 if (!marked)
2459                                         marked = p->item;
2460                         } else {
2461                                 if (!unmarked) {
2462                                         unmarked = p->item;
2463                                         break;
2464                                 }
2465                         }
2466                 }
2467         }
2468
2469         /*
2470          * If we are TREESAME to a marked-for-deletion parent, but not to any
2471          * unmarked parents, unmark the first TREESAME parent. This is the
2472          * parent that the default simplify_history==1 scan would have followed,
2473          * and it doesn't make sense to omit that path when asking for a
2474          * simplified full history. Retaining it improves the chances of
2475          * understanding odd missed merges that took an old version of a file.
2476          *
2477          * Example:
2478          *
2479          *   I--------*X       A modified the file, but mainline merge X used
2480          *    \       /        "-s ours", so took the version from I. X is
2481          *     `-*A--'         TREESAME to I and !TREESAME to A.
2482          *
2483          * Default log from X would produce "I". Without this check,
2484          * --full-history --simplify-merges would produce "I-A-X", showing
2485          * the merge commit X and that it changed A, but not making clear that
2486          * it had just taken the I version. With this check, the topology above
2487          * is retained.
2488          *
2489          * Note that it is possible that the simplification chooses a different
2490          * TREESAME parent from the default, in which case this test doesn't
2491          * activate, and we _do_ drop the default parent. Example:
2492          *
2493          *   I------X         A modified the file, but it was reverted in B,
2494          *    \    /          meaning mainline merge X is TREESAME to both
2495          *    *A-*B           parents.
2496          *
2497          * Default log would produce "I" by following the first parent;
2498          * --full-history --simplify-merges will produce "I-A-B". But this is a
2499          * reasonable result - it presents a logical full history leading from
2500          * I to X, and X is not an important merge.
2501          */
2502         if (!unmarked && marked) {
2503                 marked->object.flags &= ~TMP_MARK;
2504                 return 1;
2505         }
2506
2507         return 0;
2508 }
2509
2510 static int remove_marked_parents(struct rev_info *revs, struct commit *commit)
2511 {
2512         struct commit_list **pp, *p;
2513         int nth_parent, removed = 0;
2514
2515         pp = &commit->parents;
2516         nth_parent = 0;
2517         while ((p = *pp) != NULL) {
2518                 struct commit *parent = p->item;
2519                 if (parent->object.flags & TMP_MARK) {
2520                         parent->object.flags &= ~TMP_MARK;
2521                         *pp = p->next;
2522                         free(p);
2523                         removed++;
2524                         compact_treesame(revs, commit, nth_parent);
2525                         continue;
2526                 }
2527                 pp = &p->next;
2528                 nth_parent++;
2529         }
2530
2531         /* Removing parents can only increase TREESAMEness */
2532         if (removed && !(commit->object.flags & TREESAME))
2533                 update_treesame(revs, commit);
2534
2535         return nth_parent;
2536 }
2537
2538 static struct commit_list **simplify_one(struct rev_info *revs, struct commit *commit, struct commit_list **tail)
2539 {
2540         struct commit_list *p;
2541         struct commit *parent;
2542         struct merge_simplify_state *st, *pst;
2543         int cnt;
2544
2545         st = locate_simplify_state(revs, commit);
2546
2547         /*
2548          * Have we handled this one?
2549          */
2550         if (st->simplified)
2551                 return tail;
2552
2553         /*
2554          * An UNINTERESTING commit simplifies to itself, so does a
2555          * root commit.  We do not rewrite parents of such commit
2556          * anyway.
2557          */
2558         if ((commit->object.flags & UNINTERESTING) || !commit->parents) {
2559                 st->simplified = commit;
2560                 return tail;
2561         }
2562
2563         /*
2564          * Do we know what commit all of our parents that matter
2565          * should be rewritten to?  Otherwise we are not ready to
2566          * rewrite this one yet.
2567          */
2568         for (cnt = 0, p = commit->parents; p; p = p->next) {
2569                 pst = locate_simplify_state(revs, p->item);
2570                 if (!pst->simplified) {
2571                         tail = &commit_list_insert(p->item, tail)->next;
2572                         cnt++;
2573                 }
2574                 if (revs->first_parent_only)
2575                         break;
2576         }
2577         if (cnt) {
2578                 tail = &commit_list_insert(commit, tail)->next;
2579                 return tail;
2580         }
2581
2582         /*
2583          * Rewrite our list of parents. Note that this cannot
2584          * affect our TREESAME flags in any way - a commit is
2585          * always TREESAME to its simplification.
2586          */
2587         for (p = commit->parents; p; p = p->next) {
2588                 pst = locate_simplify_state(revs, p->item);
2589                 p->item = pst->simplified;
2590                 if (revs->first_parent_only)
2591                         break;
2592         }
2593
2594         if (revs->first_parent_only)
2595                 cnt = 1;
2596         else
2597                 cnt = remove_duplicate_parents(revs, commit);
2598
2599         /*
2600          * It is possible that we are a merge and one side branch
2601          * does not have any commit that touches the given paths;
2602          * in such a case, the immediate parent from that branch
2603          * will be rewritten to be the merge base.
2604          *
2605          *      o----X          X: the commit we are looking at;
2606          *     /    /           o: a commit that touches the paths;
2607          * ---o----'
2608          *
2609          * Further, a merge of an independent branch that doesn't
2610          * touch the path will reduce to a treesame root parent:
2611          *
2612          *  ----o----X          X: the commit we are looking at;
2613          *          /           o: a commit that touches the paths;
2614          *         r            r: a root commit not touching the paths
2615          *
2616          * Detect and simplify both cases.
2617          */
2618         if (1 < cnt) {
2619                 int marked = mark_redundant_parents(revs, commit);
2620                 marked += mark_treesame_root_parents(revs, commit);
2621                 if (marked)
2622                         marked -= leave_one_treesame_to_parent(revs, commit);
2623                 if (marked)
2624                         cnt = remove_marked_parents(revs, commit);
2625         }
2626
2627         /*
2628          * A commit simplifies to itself if it is a root, if it is
2629          * UNINTERESTING, if it touches the given paths, or if it is a
2630          * merge and its parents don't simplify to one relevant commit
2631          * (the first two cases are already handled at the beginning of
2632          * this function).
2633          *
2634          * Otherwise, it simplifies to what its sole relevant parent
2635          * simplifies to.
2636          */
2637         if (!cnt ||
2638             (commit->object.flags & UNINTERESTING) ||
2639             !(commit->object.flags & TREESAME) ||
2640             (parent = one_relevant_parent(revs, commit->parents)) == NULL)
2641                 st->simplified = commit;
2642         else {
2643                 pst = locate_simplify_state(revs, parent);
2644                 st->simplified = pst->simplified;
2645         }
2646         return tail;
2647 }
2648
2649 static void simplify_merges(struct rev_info *revs)
2650 {
2651         struct commit_list *list, *next;
2652         struct commit_list *yet_to_do, **tail;
2653         struct commit *commit;
2654
2655         if (!revs->prune)
2656                 return;
2657
2658         /* feed the list reversed */
2659         yet_to_do = NULL;
2660         for (list = revs->commits; list; list = next) {
2661                 commit = list->item;
2662                 next = list->next;
2663                 /*
2664                  * Do not free(list) here yet; the original list
2665                  * is used later in this function.
2666                  */
2667                 commit_list_insert(commit, &yet_to_do);
2668         }
2669         while (yet_to_do) {
2670                 list = yet_to_do;
2671                 yet_to_do = NULL;
2672                 tail = &yet_to_do;
2673                 while (list) {
2674                         commit = pop_commit(&list);
2675                         tail = simplify_one(revs, commit, tail);
2676                 }
2677         }
2678
2679         /* clean up the result, removing the simplified ones */
2680         list = revs->commits;
2681         revs->commits = NULL;
2682         tail = &revs->commits;
2683         while (list) {
2684                 struct merge_simplify_state *st;
2685
2686                 commit = pop_commit(&list);
2687                 st = locate_simplify_state(revs, commit);
2688                 if (st->simplified == commit)
2689                         tail = &commit_list_insert(commit, tail)->next;
2690         }
2691 }
2692
2693 static void set_children(struct rev_info *revs)
2694 {
2695         struct commit_list *l;
2696         for (l = revs->commits; l; l = l->next) {
2697                 struct commit *commit = l->item;
2698                 struct commit_list *p;
2699
2700                 for (p = commit->parents; p; p = p->next)
2701                         add_child(revs, p->item, commit);
2702         }
2703 }
2704
2705 void reset_revision_walk(void)
2706 {
2707         clear_object_flags(SEEN | ADDED | SHOWN);
2708 }
2709
2710 int prepare_revision_walk(struct rev_info *revs)
2711 {
2712         int i;
2713         struct object_array old_pending;
2714         struct commit_list **next = &revs->commits;
2715
2716         memcpy(&old_pending, &revs->pending, sizeof(old_pending));
2717         revs->pending.nr = 0;
2718         revs->pending.alloc = 0;
2719         revs->pending.objects = NULL;
2720         for (i = 0; i < old_pending.nr; i++) {
2721                 struct object_array_entry *e = old_pending.objects + i;
2722                 struct commit *commit = handle_commit(revs, e);
2723                 if (commit) {
2724                         if (!(commit->object.flags & SEEN)) {
2725                                 commit->object.flags |= SEEN;
2726                                 next = commit_list_append(commit, next);
2727                         }
2728                 }
2729         }
2730         if (!revs->leak_pending)
2731                 object_array_clear(&old_pending);
2732
2733         /* Signal whether we need per-parent treesame decoration */
2734         if (revs->simplify_merges ||
2735             (revs->limited && limiting_can_increase_treesame(revs)))
2736                 revs->treesame.name = "treesame";
2737
2738         if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
2739                 commit_list_sort_by_date(&revs->commits);
2740         if (revs->no_walk)
2741                 return 0;
2742         if (revs->limited)
2743                 if (limit_list(revs) < 0)
2744                         return -1;
2745         if (revs->topo_order)
2746                 sort_in_topological_order(&revs->commits, revs->sort_order);
2747         if (revs->line_level_traverse)
2748                 line_log_filter(revs);
2749         if (revs->simplify_merges)
2750                 simplify_merges(revs);
2751         if (revs->children.name)
2752                 set_children(revs);
2753         return 0;
2754 }
2755
2756 static enum rewrite_result rewrite_one(struct rev_info *revs, struct commit **pp)
2757 {
2758         struct commit_list *cache = NULL;
2759
2760         for (;;) {
2761                 struct commit *p = *pp;
2762                 if (!revs->limited)
2763                         if (add_parents_to_list(revs, p, &revs->commits, &cache) < 0)
2764                                 return rewrite_one_error;
2765                 if (p->object.flags & UNINTERESTING)
2766                         return rewrite_one_ok;
2767                 if (!(p->object.flags & TREESAME))
2768                         return rewrite_one_ok;
2769                 if (!p->parents)
2770                         return rewrite_one_noparents;
2771                 if ((p = one_relevant_parent(revs, p->parents)) == NULL)
2772                         return rewrite_one_ok;
2773                 *pp = p;
2774         }
2775 }
2776
2777 int rewrite_parents(struct rev_info *revs, struct commit *commit,
2778         rewrite_parent_fn_t rewrite_parent)
2779 {
2780         struct commit_list **pp = &commit->parents;
2781         while (*pp) {
2782                 struct commit_list *parent = *pp;
2783                 switch (rewrite_parent(revs, &parent->item)) {
2784                 case rewrite_one_ok:
2785                         break;
2786                 case rewrite_one_noparents:
2787                         *pp = parent->next;
2788                         continue;
2789                 case rewrite_one_error:
2790                         return -1;
2791                 }
2792                 pp = &parent->next;
2793         }
2794         remove_duplicate_parents(revs, commit);
2795         return 0;
2796 }
2797
2798 static int commit_rewrite_person(struct strbuf *buf, const char *what, struct string_list *mailmap)
2799 {
2800         char *person, *endp;
2801         size_t len, namelen, maillen;
2802         const char *name;
2803         const char *mail;
2804         struct ident_split ident;
2805
2806         person = strstr(buf->buf, what);
2807         if (!person)
2808                 return 0;
2809
2810         person += strlen(what);
2811         endp = strchr(person, '\n');
2812         if (!endp)
2813                 return 0;
2814
2815         len = endp - person;
2816
2817         if (split_ident_line(&ident, person, len))
2818                 return 0;
2819
2820         mail = ident.mail_begin;
2821         maillen = ident.mail_end - ident.mail_begin;
2822         name = ident.name_begin;
2823         namelen = ident.name_end - ident.name_begin;
2824
2825         if (map_user(mailmap, &mail, &maillen, &name, &namelen)) {
2826                 struct strbuf namemail = STRBUF_INIT;
2827
2828                 strbuf_addf(&namemail, "%.*s <%.*s>",
2829                             (int)namelen, name, (int)maillen, mail);
2830
2831                 strbuf_splice(buf, ident.name_begin - buf->buf,
2832                               ident.mail_end - ident.name_begin + 1,
2833                               namemail.buf, namemail.len);
2834
2835                 strbuf_release(&namemail);
2836
2837                 return 1;
2838         }
2839
2840         return 0;
2841 }
2842
2843 static int commit_match(struct commit *commit, struct rev_info *opt)
2844 {
2845         int retval;
2846         const char *encoding;
2847         const char *message;
2848         struct strbuf buf = STRBUF_INIT;
2849
2850         if (!opt->grep_filter.pattern_list && !opt->grep_filter.header_list)
2851                 return 1;
2852
2853         /* Prepend "fake" headers as needed */
2854         if (opt->grep_filter.use_reflog_filter) {
2855                 strbuf_addstr(&buf, "reflog ");
2856                 get_reflog_message(&buf, opt->reflog_info);
2857                 strbuf_addch(&buf, '\n');
2858         }
2859
2860         /*
2861          * We grep in the user's output encoding, under the assumption that it
2862          * is the encoding they are most likely to write their grep pattern
2863          * for. In addition, it means we will match the "notes" encoding below,
2864          * so we will not end up with a buffer that has two different encodings
2865          * in it.
2866          */
2867         encoding = get_log_output_encoding();
2868         message = logmsg_reencode(commit, NULL, encoding);
2869
2870         /* Copy the commit to temporary if we are using "fake" headers */
2871         if (buf.len)
2872                 strbuf_addstr(&buf, message);
2873
2874         if (opt->grep_filter.header_list && opt->mailmap) {
2875                 if (!buf.len)
2876                         strbuf_addstr(&buf, message);
2877
2878                 commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
2879                 commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
2880         }
2881
2882         /* Append "fake" message parts as needed */
2883         if (opt->show_notes) {
2884                 if (!buf.len)
2885                         strbuf_addstr(&buf, message);
2886                 format_display_notes(commit->object.sha1, &buf, encoding, 1);
2887         }
2888
2889         /*
2890          * Find either in the original commit message, or in the temporary.
2891          * Note that we cast away the constness of "message" here. It is
2892          * const because it may come from the cached commit buffer. That's OK,
2893          * because we know that it is modifiable heap memory, and that while
2894          * grep_buffer may modify it for speed, it will restore any
2895          * changes before returning.
2896          */
2897         if (buf.len)
2898                 retval = grep_buffer(&opt->grep_filter, buf.buf, buf.len);
2899         else
2900                 retval = grep_buffer(&opt->grep_filter,
2901                                      (char *)message, strlen(message));
2902         strbuf_release(&buf);
2903         unuse_commit_buffer(commit, message);
2904         return opt->invert_grep ? !retval : retval;
2905 }
2906
2907 static inline int want_ancestry(const struct rev_info *revs)
2908 {
2909         return (revs->rewrite_parents || revs->children.name);
2910 }
2911
2912 enum commit_action get_commit_action(struct rev_info *revs, struct commit *commit)
2913 {
2914         if (commit->object.flags & SHOWN)
2915                 return commit_ignore;
2916         if (revs->unpacked && has_sha1_pack(commit->object.sha1))
2917                 return commit_ignore;
2918         if (revs->show_all)
2919                 return commit_show;
2920         if (commit->object.flags & UNINTERESTING)
2921                 return commit_ignore;
2922         if (revs->min_age != -1 && (commit->date > revs->min_age))
2923                 return commit_ignore;
2924         if (revs->min_parents || (revs->max_parents >= 0)) {
2925                 int n = commit_list_count(commit->parents);
2926                 if ((n < revs->min_parents) ||
2927                     ((revs->max_parents >= 0) && (n > revs->max_parents)))
2928                         return commit_ignore;
2929         }
2930         if (!commit_match(commit, revs))
2931                 return commit_ignore;
2932         if (revs->prune && revs->dense) {
2933                 /* Commit without changes? */
2934                 if (commit->object.flags & TREESAME) {
2935                         int n;
2936                         struct commit_list *p;
2937                         /* drop merges unless we want parenthood */
2938                         if (!want_ancestry(revs))
2939                                 return commit_ignore;
2940                         /*
2941                          * If we want ancestry, then need to keep any merges
2942                          * between relevant commits to tie together topology.
2943                          * For consistency with TREESAME and simplification
2944                          * use "relevant" here rather than just INTERESTING,
2945                          * to treat bottom commit(s) as part of the topology.
2946                          */
2947                         for (n = 0, p = commit->parents; p; p = p->next)
2948                                 if (relevant_commit(p->item))
2949                                         if (++n >= 2)
2950                                                 return commit_show;
2951                         return commit_ignore;
2952                 }
2953         }
2954         return commit_show;
2955 }
2956
2957 define_commit_slab(saved_parents, struct commit_list *);
2958
2959 #define EMPTY_PARENT_LIST ((struct commit_list *)-1)
2960
2961 /*
2962  * You may only call save_parents() once per commit (this is checked
2963  * for non-root commits).
2964  */
2965 static void save_parents(struct rev_info *revs, struct commit *commit)
2966 {
2967         struct commit_list **pp;
2968
2969         if (!revs->saved_parents_slab) {
2970                 revs->saved_parents_slab = xmalloc(sizeof(struct saved_parents));
2971                 init_saved_parents(revs->saved_parents_slab);
2972         }
2973
2974         pp = saved_parents_at(revs->saved_parents_slab, commit);
2975
2976         /*
2977          * When walking with reflogs, we may visit the same commit
2978          * several times: once for each appearance in the reflog.
2979          *
2980          * In this case, save_parents() will be called multiple times.
2981          * We want to keep only the first set of parents.  We need to
2982          * store a sentinel value for an empty (i.e., NULL) parent
2983          * list to distinguish it from a not-yet-saved list, however.
2984          */
2985         if (*pp)
2986                 return;
2987         if (commit->parents)
2988                 *pp = copy_commit_list(commit->parents);
2989         else
2990                 *pp = EMPTY_PARENT_LIST;
2991 }
2992
2993 static void free_saved_parents(struct rev_info *revs)
2994 {
2995         if (revs->saved_parents_slab)
2996                 clear_saved_parents(revs->saved_parents_slab);
2997 }
2998
2999 struct commit_list *get_saved_parents(struct rev_info *revs, const struct commit *commit)
3000 {
3001         struct commit_list *parents;
3002
3003         if (!revs->saved_parents_slab)
3004                 return commit->parents;
3005
3006         parents = *saved_parents_at(revs->saved_parents_slab, commit);
3007         if (parents == EMPTY_PARENT_LIST)
3008                 return NULL;
3009         return parents;
3010 }
3011
3012 enum commit_action simplify_commit(struct rev_info *revs, struct commit *commit)
3013 {
3014         enum commit_action action = get_commit_action(revs, commit);
3015
3016         if (action == commit_show &&
3017             !revs->show_all &&
3018             revs->prune && revs->dense && want_ancestry(revs)) {
3019                 /*
3020                  * --full-diff on simplified parents is no good: it
3021                  * will show spurious changes from the commits that
3022                  * were elided.  So we save the parents on the side
3023                  * when --full-diff is in effect.
3024                  */
3025                 if (revs->full_diff)
3026                         save_parents(revs, commit);
3027                 if (rewrite_parents(revs, commit, rewrite_one) < 0)
3028                         return commit_error;
3029         }
3030         return action;
3031 }
3032
3033 static void track_linear(struct rev_info *revs, struct commit *commit)
3034 {
3035         if (revs->track_first_time) {
3036                 revs->linear = 1;
3037                 revs->track_first_time = 0;
3038         } else {
3039                 struct commit_list *p;
3040                 for (p = revs->previous_parents; p; p = p->next)
3041                         if (p->item == NULL || /* first commit */
3042                             !hashcmp(p->item->object.sha1, commit->object.sha1))
3043                                 break;
3044                 revs->linear = p != NULL;
3045         }
3046         if (revs->reverse) {
3047                 if (revs->linear)
3048                         commit->object.flags |= TRACK_LINEAR;
3049         }
3050         free_commit_list(revs->previous_parents);
3051         revs->previous_parents = copy_commit_list(commit->parents);
3052 }
3053
3054 static struct commit *get_revision_1(struct rev_info *revs)
3055 {
3056         if (!revs->commits)
3057                 return NULL;
3058
3059         do {
3060                 struct commit *commit = pop_commit(&revs->commits);
3061
3062                 if (revs->reflog_info) {
3063                         save_parents(revs, commit);
3064                         fake_reflog_parent(revs->reflog_info, commit);
3065                         commit->object.flags &= ~(ADDED | SEEN | SHOWN);
3066                 }
3067
3068                 /*
3069                  * If we haven't done the list limiting, we need to look at
3070                  * the parents here. We also need to do the date-based limiting
3071                  * that we'd otherwise have done in limit_list().
3072                  */
3073                 if (!revs->limited) {
3074                         if (revs->max_age != -1 &&
3075                             (commit->date < revs->max_age))
3076                                 continue;
3077                         if (add_parents_to_list(revs, commit, &revs->commits, NULL) < 0) {
3078                                 if (!revs->ignore_missing_links)
3079                                         die("Failed to traverse parents of commit %s",
3080                                                 sha1_to_hex(commit->object.sha1));
3081                         }
3082                 }
3083
3084                 switch (simplify_commit(revs, commit)) {
3085                 case commit_ignore:
3086                         continue;
3087                 case commit_error:
3088                         die("Failed to simplify parents of commit %s",
3089                             sha1_to_hex(commit->object.sha1));
3090                 default:
3091                         if (revs->track_linear)
3092                                 track_linear(revs, commit);
3093                         return commit;
3094                 }
3095         } while (revs->commits);
3096         return NULL;
3097 }
3098
3099 /*
3100  * Return true for entries that have not yet been shown.  (This is an
3101  * object_array_each_func_t.)
3102  */
3103 static int entry_unshown(struct object_array_entry *entry, void *cb_data_unused)
3104 {
3105         return !(entry->item->flags & SHOWN);
3106 }
3107
3108 /*
3109  * If array is on the verge of a realloc, garbage-collect any entries
3110  * that have already been shown to try to free up some space.
3111  */
3112 static void gc_boundary(struct object_array *array)
3113 {
3114         if (array->nr == array->alloc)
3115                 object_array_filter(array, entry_unshown, NULL);
3116 }
3117
3118 static void create_boundary_commit_list(struct rev_info *revs)
3119 {
3120         unsigned i;
3121         struct commit *c;
3122         struct object_array *array = &revs->boundary_commits;
3123         struct object_array_entry *objects = array->objects;
3124
3125         /*
3126          * If revs->commits is non-NULL at this point, an error occurred in
3127          * get_revision_1().  Ignore the error and continue printing the
3128          * boundary commits anyway.  (This is what the code has always
3129          * done.)
3130          */
3131         if (revs->commits) {
3132                 free_commit_list(revs->commits);
3133                 revs->commits = NULL;
3134         }
3135
3136         /*
3137          * Put all of the actual boundary commits from revs->boundary_commits
3138          * into revs->commits
3139          */
3140         for (i = 0; i < array->nr; i++) {
3141                 c = (struct commit *)(objects[i].item);
3142                 if (!c)
3143                         continue;
3144                 if (!(c->object.flags & CHILD_SHOWN))
3145                         continue;
3146                 if (c->object.flags & (SHOWN | BOUNDARY))
3147                         continue;
3148                 c->object.flags |= BOUNDARY;
3149                 commit_list_insert(c, &revs->commits);
3150         }
3151
3152         /*
3153          * If revs->topo_order is set, sort the boundary commits
3154          * in topological order
3155          */
3156         sort_in_topological_order(&revs->commits, revs->sort_order);
3157 }
3158
3159 static struct commit *get_revision_internal(struct rev_info *revs)
3160 {
3161         struct commit *c = NULL;
3162         struct commit_list *l;
3163
3164         if (revs->boundary == 2) {
3165                 /*
3166                  * All of the normal commits have already been returned,
3167                  * and we are now returning boundary commits.
3168                  * create_boundary_commit_list() has populated
3169                  * revs->commits with the remaining commits to return.
3170                  */
3171                 c = pop_commit(&revs->commits);
3172                 if (c)
3173                         c->object.flags |= SHOWN;
3174                 return c;
3175         }
3176
3177         /*
3178          * If our max_count counter has reached zero, then we are done. We
3179          * don't simply return NULL because we still might need to show
3180          * boundary commits. But we want to avoid calling get_revision_1, which
3181          * might do a considerable amount of work finding the next commit only
3182          * for us to throw it away.
3183          *
3184          * If it is non-zero, then either we don't have a max_count at all
3185          * (-1), or it is still counting, in which case we decrement.
3186          */
3187         if (revs->max_count) {
3188                 c = get_revision_1(revs);
3189                 if (c) {
3190                         while (revs->skip_count > 0) {
3191                                 revs->skip_count--;
3192                                 c = get_revision_1(revs);
3193                                 if (!c)
3194                                         break;
3195                         }
3196                 }
3197
3198                 if (revs->max_count > 0)
3199                         revs->max_count--;
3200         }
3201
3202         if (c)
3203                 c->object.flags |= SHOWN;
3204
3205         if (!revs->boundary)
3206                 return c;
3207
3208         if (!c) {
3209                 /*
3210                  * get_revision_1() runs out the commits, and
3211                  * we are done computing the boundaries.
3212                  * switch to boundary commits output mode.
3213                  */
3214                 revs->boundary = 2;
3215
3216                 /*
3217                  * Update revs->commits to contain the list of
3218                  * boundary commits.
3219                  */
3220                 create_boundary_commit_list(revs);
3221
3222                 return get_revision_internal(revs);
3223         }
3224
3225         /*
3226          * boundary commits are the commits that are parents of the
3227          * ones we got from get_revision_1() but they themselves are
3228          * not returned from get_revision_1().  Before returning
3229          * 'c', we need to mark its parents that they could be boundaries.
3230          */
3231
3232         for (l = c->parents; l; l = l->next) {
3233                 struct object *p;
3234                 p = &(l->item->object);
3235                 if (p->flags & (CHILD_SHOWN | SHOWN))
3236                         continue;
3237                 p->flags |= CHILD_SHOWN;
3238                 gc_boundary(&revs->boundary_commits);
3239                 add_object_array(p, NULL, &revs->boundary_commits);
3240         }
3241
3242         return c;
3243 }
3244
3245 struct commit *get_revision(struct rev_info *revs)
3246 {
3247         struct commit *c;
3248         struct commit_list *reversed;
3249
3250         if (revs->reverse) {
3251                 reversed = NULL;
3252                 while ((c = get_revision_internal(revs)))
3253                         commit_list_insert(c, &reversed);
3254                 revs->commits = reversed;
3255                 revs->reverse = 0;
3256                 revs->reverse_output_stage = 1;
3257         }
3258
3259         if (revs->reverse_output_stage) {
3260                 c = pop_commit(&revs->commits);
3261                 if (revs->track_linear)
3262                         revs->linear = !!(c && c->object.flags & TRACK_LINEAR);
3263                 return c;
3264         }
3265
3266         c = get_revision_internal(revs);
3267         if (c && revs->graph)
3268                 graph_update(revs->graph, c);
3269         if (!c) {
3270                 free_saved_parents(revs);
3271                 if (revs->previous_parents) {
3272                         free_commit_list(revs->previous_parents);
3273                         revs->previous_parents = NULL;
3274                 }
3275         }
3276         return c;
3277 }
3278
3279 char *get_revision_mark(const struct rev_info *revs, const struct commit *commit)
3280 {
3281         if (commit->object.flags & BOUNDARY)
3282                 return "-";
3283         else if (commit->object.flags & UNINTERESTING)
3284                 return "^";
3285         else if (commit->object.flags & PATCHSAME)
3286                 return "=";
3287         else if (!revs || revs->left_right) {
3288                 if (commit->object.flags & SYMMETRIC_LEFT)
3289                         return "<";
3290                 else
3291                         return ">";
3292         } else if (revs->graph)
3293                 return "*";
3294         else if (revs->cherry_mark)
3295                 return "+";
3296         return "";
3297 }
3298
3299 void put_revision_mark(const struct rev_info *revs, const struct commit *commit)
3300 {
3301         char *mark = get_revision_mark(revs, commit);
3302         if (!strlen(mark))
3303                 return;
3304         fputs(mark, stdout);
3305         putchar(' ');
3306 }