repack: pack objects mentioned by the index
[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
22 volatile show_early_output_fn_t show_early_output;
23
24 char *path_name(const struct name_path *path, const char *name)
25 {
26         const struct name_path *p;
27         char *n, *m;
28         int nlen = strlen(name);
29         int len = nlen + 1;
30
31         for (p = path; p; p = p->up) {
32                 if (p->elem_len)
33                         len += p->elem_len + 1;
34         }
35         n = xmalloc(len);
36         m = n + len - (nlen + 1);
37         strcpy(m, name);
38         for (p = path; p; p = p->up) {
39                 if (p->elem_len) {
40                         m -= p->elem_len + 1;
41                         memcpy(m, p->elem, p->elem_len);
42                         m[p->elem_len] = '/';
43                 }
44         }
45         return n;
46 }
47
48 static int show_path_component_truncated(FILE *out, const char *name, int len)
49 {
50         int cnt;
51         for (cnt = 0; cnt < len; cnt++) {
52                 int ch = name[cnt];
53                 if (!ch || ch == '\n')
54                         return -1;
55                 fputc(ch, out);
56         }
57         return len;
58 }
59
60 static int show_path_truncated(FILE *out, const struct name_path *path)
61 {
62         int emitted, ours;
63
64         if (!path)
65                 return 0;
66         emitted = show_path_truncated(out, path->up);
67         if (emitted < 0)
68                 return emitted;
69         if (emitted)
70                 fputc('/', out);
71         ours = show_path_component_truncated(out, path->elem, path->elem_len);
72         if (ours < 0)
73                 return ours;
74         return ours || emitted;
75 }
76
77 void show_object_with_name(FILE *out, struct object *obj,
78                            const struct name_path *path, const char *component)
79 {
80         struct name_path leaf;
81         leaf.up = (struct name_path *)path;
82         leaf.elem = component;
83         leaf.elem_len = strlen(component);
84
85         fprintf(out, "%s ", sha1_to_hex(obj->sha1));
86         show_path_truncated(out, &leaf);
87         fputc('\n', out);
88 }
89
90 void add_object(struct object *obj,
91                 struct object_array *p,
92                 struct name_path *path,
93                 const char *name)
94 {
95         char *pn = path_name(path, name);
96         add_object_array(obj, pn, p);
97         free(pn);
98 }
99
100 static void mark_blob_uninteresting(struct blob *blob)
101 {
102         if (!blob)
103                 return;
104         if (blob->object.flags & UNINTERESTING)
105                 return;
106         blob->object.flags |= UNINTERESTING;
107 }
108
109 static void mark_tree_contents_uninteresting(struct tree *tree)
110 {
111         struct tree_desc desc;
112         struct name_entry entry;
113         struct object *obj = &tree->object;
114
115         if (!has_sha1_file(obj->sha1))
116                 return;
117         if (parse_tree(tree) < 0)
118                 die("bad tree %s", sha1_to_hex(obj->sha1));
119
120         init_tree_desc(&desc, tree->buffer, tree->size);
121         while (tree_entry(&desc, &entry)) {
122                 switch (object_type(entry.mode)) {
123                 case OBJ_TREE:
124                         mark_tree_uninteresting(lookup_tree(entry.sha1));
125                         break;
126                 case OBJ_BLOB:
127                         mark_blob_uninteresting(lookup_blob(entry.sha1));
128                         break;
129                 default:
130                         /* Subproject commit - not in this repository */
131                         break;
132                 }
133         }
134
135         /*
136          * We don't care about the tree any more
137          * after it has been marked uninteresting.
138          */
139         free_tree_buffer(tree);
140 }
141
142 void mark_tree_uninteresting(struct tree *tree)
143 {
144         struct object *obj = &tree->object;
145
146         if (!tree)
147                 return;
148         if (obj->flags & UNINTERESTING)
149                 return;
150         obj->flags |= UNINTERESTING;
151         mark_tree_contents_uninteresting(tree);
152 }
153
154 void mark_parents_uninteresting(struct commit *commit)
155 {
156         struct commit_list *parents = NULL, *l;
157
158         for (l = commit->parents; l; l = l->next)
159                 commit_list_insert(l->item, &parents);
160
161         while (parents) {
162                 struct commit *commit = parents->item;
163                 l = parents;
164                 parents = parents->next;
165                 free(l);
166
167                 while (commit) {
168                         /*
169                          * A missing commit is ok iff its parent is marked
170                          * uninteresting.
171                          *
172                          * We just mark such a thing parsed, so that when
173                          * it is popped next time around, we won't be trying
174                          * to parse it and get an error.
175                          */
176                         if (!has_sha1_file(commit->object.sha1))
177                                 commit->object.parsed = 1;
178
179                         if (commit->object.flags & UNINTERESTING)
180                                 break;
181
182                         commit->object.flags |= UNINTERESTING;
183
184                         /*
185                          * Normally we haven't parsed the parent
186                          * yet, so we won't have a parent of a parent
187                          * here. However, it may turn out that we've
188                          * reached this commit some other way (where it
189                          * wasn't uninteresting), in which case we need
190                          * to mark its parents recursively too..
191                          */
192                         if (!commit->parents)
193                                 break;
194
195                         for (l = commit->parents->next; l; l = l->next)
196                                 commit_list_insert(l->item, &parents);
197                         commit = commit->parents->item;
198                 }
199         }
200 }
201
202 static void add_pending_object_with_path(struct rev_info *revs,
203                                          struct object *obj,
204                                          const char *name, unsigned mode,
205                                          const char *path)
206 {
207         if (!obj)
208                 return;
209         if (revs->no_walk && (obj->flags & UNINTERESTING))
210                 revs->no_walk = 0;
211         if (revs->reflog_info && obj->type == OBJ_COMMIT) {
212                 struct strbuf buf = STRBUF_INIT;
213                 int len = interpret_branch_name(name, 0, &buf);
214                 int st;
215
216                 if (0 < len && name[len] && buf.len)
217                         strbuf_addstr(&buf, name + len);
218                 st = add_reflog_for_walk(revs->reflog_info,
219                                          (struct commit *)obj,
220                                          buf.buf[0] ? buf.buf: name);
221                 strbuf_release(&buf);
222                 if (st)
223                         return;
224         }
225         add_object_array_with_path(obj, name, &revs->pending, mode, path);
226 }
227
228 static void add_pending_object_with_mode(struct rev_info *revs,
229                                          struct object *obj,
230                                          const char *name, unsigned mode)
231 {
232         add_pending_object_with_path(revs, obj, name, mode, NULL);
233 }
234
235 void add_pending_object(struct rev_info *revs,
236                         struct object *obj, const char *name)
237 {
238         add_pending_object_with_mode(revs, obj, name, S_IFINVALID);
239 }
240
241 void add_head_to_pending(struct rev_info *revs)
242 {
243         unsigned char sha1[20];
244         struct object *obj;
245         if (get_sha1("HEAD", sha1))
246                 return;
247         obj = parse_object(sha1);
248         if (!obj)
249                 return;
250         add_pending_object(revs, obj, "HEAD");
251 }
252
253 static struct object *get_reference(struct rev_info *revs, const char *name,
254                                     const unsigned char *sha1,
255                                     unsigned int flags)
256 {
257         struct object *object;
258
259         object = parse_object(sha1);
260         if (!object) {
261                 if (revs->ignore_missing)
262                         return object;
263                 die("bad object %s", name);
264         }
265         object->flags |= flags;
266         return object;
267 }
268
269 void add_pending_sha1(struct rev_info *revs, const char *name,
270                       const unsigned char *sha1, unsigned int flags)
271 {
272         struct object *object = get_reference(revs, name, sha1, flags);
273         add_pending_object(revs, object, name);
274 }
275
276 static struct commit *handle_commit(struct rev_info *revs,
277                                     struct object_array_entry *entry)
278 {
279         struct object *object = entry->item;
280         const char *name = entry->name;
281         const char *path = entry->path;
282         unsigned int mode = entry->mode;
283         unsigned long flags = object->flags;
284
285         /*
286          * Tag object? Look what it points to..
287          */
288         while (object->type == OBJ_TAG) {
289                 struct tag *tag = (struct tag *) object;
290                 if (revs->tag_objects && !(flags & UNINTERESTING))
291                         add_pending_object(revs, object, tag->tag);
292                 if (!tag->tagged)
293                         die("bad tag");
294                 object = parse_object(tag->tagged->sha1);
295                 if (!object) {
296                         if (flags & UNINTERESTING)
297                                 return NULL;
298                         die("bad object %s", sha1_to_hex(tag->tagged->sha1));
299                 }
300                 object->flags |= flags;
301                 /*
302                  * We'll handle the tagged object by looping or dropping
303                  * through to the non-tag handlers below. Do not
304                  * propagate data from the tag's pending entry.
305                  */
306                 name = "";
307                 path = NULL;
308                 mode = 0;
309         }
310
311         /*
312          * Commit object? Just return it, we'll do all the complex
313          * reachability crud.
314          */
315         if (object->type == OBJ_COMMIT) {
316                 struct commit *commit = (struct commit *)object;
317                 if (parse_commit(commit) < 0)
318                         die("unable to parse commit %s", name);
319                 if (flags & UNINTERESTING) {
320                         mark_parents_uninteresting(commit);
321                         revs->limited = 1;
322                 }
323                 if (revs->show_source && !commit->util)
324                         commit->util = xstrdup(name);
325                 return commit;
326         }
327
328         /*
329          * Tree object? Either mark it uninteresting, or add it
330          * to the list of objects to look at later..
331          */
332         if (object->type == OBJ_TREE) {
333                 struct tree *tree = (struct tree *)object;
334                 if (!revs->tree_objects)
335                         return NULL;
336                 if (flags & UNINTERESTING) {
337                         mark_tree_contents_uninteresting(tree);
338                         return NULL;
339                 }
340                 add_pending_object_with_path(revs, object, name, mode, path);
341                 return NULL;
342         }
343
344         /*
345          * Blob object? You know the drill by now..
346          */
347         if (object->type == OBJ_BLOB) {
348                 if (!revs->blob_objects)
349                         return NULL;
350                 if (flags & UNINTERESTING)
351                         return NULL;
352                 add_pending_object_with_path(revs, object, name, mode, path);
353                 return NULL;
354         }
355         die("%s is unknown object", name);
356 }
357
358 static int everybody_uninteresting(struct commit_list *orig)
359 {
360         struct commit_list *list = orig;
361         while (list) {
362                 struct commit *commit = list->item;
363                 list = list->next;
364                 if (commit->object.flags & UNINTERESTING)
365                         continue;
366                 return 0;
367         }
368         return 1;
369 }
370
371 /*
372  * A definition of "relevant" commit that we can use to simplify limited graphs
373  * by eliminating side branches.
374  *
375  * A "relevant" commit is one that is !UNINTERESTING (ie we are including it
376  * in our list), or that is a specified BOTTOM commit. Then after computing
377  * a limited list, during processing we can generally ignore boundary merges
378  * coming from outside the graph, (ie from irrelevant parents), and treat
379  * those merges as if they were single-parent. TREESAME is defined to consider
380  * only relevant parents, if any. If we are TREESAME to our on-graph parents,
381  * we don't care if we were !TREESAME to non-graph parents.
382  *
383  * Treating bottom commits as relevant ensures that a limited graph's
384  * connection to the actual bottom commit is not viewed as a side branch, but
385  * treated as part of the graph. For example:
386  *
387  *   ....Z...A---X---o---o---B
388  *        .     /
389  *         W---Y
390  *
391  * When computing "A..B", the A-X connection is at least as important as
392  * Y-X, despite A being flagged UNINTERESTING.
393  *
394  * And when computing --ancestry-path "A..B", the A-X connection is more
395  * important than Y-X, despite both A and Y being flagged UNINTERESTING.
396  */
397 static inline int relevant_commit(struct commit *commit)
398 {
399         return (commit->object.flags & (UNINTERESTING | BOTTOM)) != UNINTERESTING;
400 }
401
402 /*
403  * Return a single relevant commit from a parent list. If we are a TREESAME
404  * commit, and this selects one of our parents, then we can safely simplify to
405  * that parent.
406  */
407 static struct commit *one_relevant_parent(const struct rev_info *revs,
408                                           struct commit_list *orig)
409 {
410         struct commit_list *list = orig;
411         struct commit *relevant = NULL;
412
413         if (!orig)
414                 return NULL;
415
416         /*
417          * For 1-parent commits, or if first-parent-only, then return that
418          * first parent (even if not "relevant" by the above definition).
419          * TREESAME will have been set purely on that parent.
420          */
421         if (revs->first_parent_only || !orig->next)
422                 return orig->item;
423
424         /*
425          * For multi-parent commits, identify a sole relevant parent, if any.
426          * If we have only one relevant parent, then TREESAME will be set purely
427          * with regard to that parent, and we can simplify accordingly.
428          *
429          * If we have more than one relevant parent, or no relevant parents
430          * (and multiple irrelevant ones), then we can't select a parent here
431          * and return NULL.
432          */
433         while (list) {
434                 struct commit *commit = list->item;
435                 list = list->next;
436                 if (relevant_commit(commit)) {
437                         if (relevant)
438                                 return NULL;
439                         relevant = commit;
440                 }
441         }
442         return relevant;
443 }
444
445 /*
446  * The goal is to get REV_TREE_NEW as the result only if the
447  * diff consists of all '+' (and no other changes), REV_TREE_OLD
448  * if the whole diff is removal of old data, and otherwise
449  * REV_TREE_DIFFERENT (of course if the trees are the same we
450  * want REV_TREE_SAME).
451  * That means that once we get to REV_TREE_DIFFERENT, we do not
452  * have to look any further.
453  */
454 static int tree_difference = REV_TREE_SAME;
455
456 static void file_add_remove(struct diff_options *options,
457                     int addremove, unsigned mode,
458                     const unsigned char *sha1,
459                     int sha1_valid,
460                     const char *fullpath, unsigned dirty_submodule)
461 {
462         int diff = addremove == '+' ? REV_TREE_NEW : REV_TREE_OLD;
463
464         tree_difference |= diff;
465         if (tree_difference == REV_TREE_DIFFERENT)
466                 DIFF_OPT_SET(options, HAS_CHANGES);
467 }
468
469 static void file_change(struct diff_options *options,
470                  unsigned old_mode, unsigned new_mode,
471                  const unsigned char *old_sha1,
472                  const unsigned char *new_sha1,
473                  int old_sha1_valid, int new_sha1_valid,
474                  const char *fullpath,
475                  unsigned old_dirty_submodule, unsigned new_dirty_submodule)
476 {
477         tree_difference = REV_TREE_DIFFERENT;
478         DIFF_OPT_SET(options, HAS_CHANGES);
479 }
480
481 static int rev_compare_tree(struct rev_info *revs,
482                             struct commit *parent, struct commit *commit)
483 {
484         struct tree *t1 = parent->tree;
485         struct tree *t2 = commit->tree;
486
487         if (!t1)
488                 return REV_TREE_NEW;
489         if (!t2)
490                 return REV_TREE_OLD;
491
492         if (revs->simplify_by_decoration) {
493                 /*
494                  * If we are simplifying by decoration, then the commit
495                  * is worth showing if it has a tag pointing at it.
496                  */
497                 if (get_name_decoration(&commit->object))
498                         return REV_TREE_DIFFERENT;
499                 /*
500                  * A commit that is not pointed by a tag is uninteresting
501                  * if we are not limited by path.  This means that you will
502                  * see the usual "commits that touch the paths" plus any
503                  * tagged commit by specifying both --simplify-by-decoration
504                  * and pathspec.
505                  */
506                 if (!revs->prune_data.nr)
507                         return REV_TREE_SAME;
508         }
509
510         tree_difference = REV_TREE_SAME;
511         DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
512         if (diff_tree_sha1(t1->object.sha1, t2->object.sha1, "",
513                            &revs->pruning) < 0)
514                 return REV_TREE_DIFFERENT;
515         return tree_difference;
516 }
517
518 static int rev_same_tree_as_empty(struct rev_info *revs, struct commit *commit)
519 {
520         int retval;
521         struct tree *t1 = commit->tree;
522
523         if (!t1)
524                 return 0;
525
526         tree_difference = REV_TREE_SAME;
527         DIFF_OPT_CLR(&revs->pruning, HAS_CHANGES);
528         retval = diff_tree_sha1(NULL, t1->object.sha1, "", &revs->pruning);
529
530         return retval >= 0 && (tree_difference == REV_TREE_SAME);
531 }
532
533 struct treesame_state {
534         unsigned int nparents;
535         unsigned char treesame[FLEX_ARRAY];
536 };
537
538 static struct treesame_state *initialise_treesame(struct rev_info *revs, struct commit *commit)
539 {
540         unsigned n = commit_list_count(commit->parents);
541         struct treesame_state *st = xcalloc(1, sizeof(*st) + n);
542         st->nparents = n;
543         add_decoration(&revs->treesame, &commit->object, st);
544         return st;
545 }
546
547 /*
548  * Must be called immediately after removing the nth_parent from a commit's
549  * parent list, if we are maintaining the per-parent treesame[] decoration.
550  * This does not recalculate the master TREESAME flag - update_treesame()
551  * should be called to update it after a sequence of treesame[] modifications
552  * that may have affected it.
553  */
554 static int compact_treesame(struct rev_info *revs, struct commit *commit, unsigned nth_parent)
555 {
556         struct treesame_state *st;
557         int old_same;
558
559         if (!commit->parents) {
560                 /*
561                  * Have just removed the only parent from a non-merge.
562                  * Different handling, as we lack decoration.
563                  */
564                 if (nth_parent != 0)
565                         die("compact_treesame %u", nth_parent);
566                 old_same = !!(commit->object.flags & TREESAME);
567                 if (rev_same_tree_as_empty(revs, commit))
568                         commit->object.flags |= TREESAME;
569                 else
570                         commit->object.flags &= ~TREESAME;
571                 return old_same;
572         }
573
574         st = lookup_decoration(&revs->treesame, &commit->object);
575         if (!st || nth_parent >= st->nparents)
576                 die("compact_treesame %u", nth_parent);
577
578         old_same = st->treesame[nth_parent];
579         memmove(st->treesame + nth_parent,
580                 st->treesame + nth_parent + 1,
581                 st->nparents - nth_parent - 1);
582
583         /*
584          * If we've just become a non-merge commit, update TREESAME
585          * immediately, and remove the no-longer-needed decoration.
586          * If still a merge, defer update until update_treesame().
587          */
588         if (--st->nparents == 1) {
589                 if (commit->parents->next)
590                         die("compact_treesame parents mismatch");
591                 if (st->treesame[0] && revs->dense)
592                         commit->object.flags |= TREESAME;
593                 else
594                         commit->object.flags &= ~TREESAME;
595                 free(add_decoration(&revs->treesame, &commit->object, NULL));
596         }
597
598         return old_same;
599 }
600
601 static unsigned update_treesame(struct rev_info *revs, struct commit *commit)
602 {
603         if (commit->parents && commit->parents->next) {
604                 unsigned n;
605                 struct treesame_state *st;
606                 struct commit_list *p;
607                 unsigned relevant_parents;
608                 unsigned relevant_change, irrelevant_change;
609
610                 st = lookup_decoration(&revs->treesame, &commit->object);
611                 if (!st)
612                         die("update_treesame %s", sha1_to_hex(commit->object.sha1));
613                 relevant_parents = 0;
614                 relevant_change = irrelevant_change = 0;
615                 for (p = commit->parents, n = 0; p; n++, p = p->next) {
616                         if (relevant_commit(p->item)) {
617                                 relevant_change |= !st->treesame[n];
618                                 relevant_parents++;
619                         } else
620                                 irrelevant_change |= !st->treesame[n];
621                 }
622                 if (relevant_parents ? relevant_change : irrelevant_change)
623                         commit->object.flags &= ~TREESAME;
624                 else
625                         commit->object.flags |= TREESAME;
626         }
627
628         return commit->object.flags & TREESAME;
629 }
630
631 static inline int limiting_can_increase_treesame(const struct rev_info *revs)
632 {
633         /*
634          * TREESAME is irrelevant unless prune && dense;
635          * if simplify_history is set, we can't have a mixture of TREESAME and
636          *    !TREESAME INTERESTING parents (and we don't have treesame[]
637          *    decoration anyway);
638          * if first_parent_only is set, then the TREESAME flag is locked
639          *    against the first parent (and again we lack treesame[] decoration).
640          */
641         return revs->prune && revs->dense &&
642                !revs->simplify_history &&
643                !revs->first_parent_only;
644 }
645
646 static void try_to_simplify_commit(struct rev_info *revs, struct commit *commit)
647 {
648         struct commit_list **pp, *parent;
649         struct treesame_state *ts = NULL;
650         int relevant_change = 0, irrelevant_change = 0;
651         int relevant_parents, nth_parent;
652
653         /*
654          * If we don't do pruning, everything is interesting
655          */
656         if (!revs->prune)
657                 return;
658
659         if (!commit->tree)
660                 return;
661
662         if (!commit->parents) {
663                 if (rev_same_tree_as_empty(revs, commit))
664                         commit->object.flags |= TREESAME;
665                 return;
666         }
667
668         /*
669          * Normal non-merge commit? If we don't want to make the
670          * history dense, we consider it always to be a change..
671          */
672         if (!revs->dense && !commit->parents->next)
673                 return;
674
675         for (pp = &commit->parents, nth_parent = 0, relevant_parents = 0;
676              (parent = *pp) != NULL;
677              pp = &parent->next, nth_parent++) {
678                 struct commit *p = parent->item;
679                 if (relevant_commit(p))
680                         relevant_parents++;
681
682                 if (nth_parent == 1) {
683                         /*
684                          * This our second loop iteration - so we now know
685                          * we're dealing with a merge.
686                          *
687                          * Do not compare with later parents when we care only about
688                          * the first parent chain, in order to avoid derailing the
689                          * traversal to follow a side branch that brought everything
690                          * in the path we are limited to by the pathspec.
691                          */
692                         if (revs->first_parent_only)
693                                 break;
694                         /*
695                          * If this will remain a potentially-simplifiable
696                          * merge, remember per-parent treesame if needed.
697                          * Initialise the array with the comparison from our
698                          * first iteration.
699                          */
700                         if (revs->treesame.name &&
701                             !revs->simplify_history &&
702                             !(commit->object.flags & UNINTERESTING)) {
703                                 ts = initialise_treesame(revs, commit);
704                                 if (!(irrelevant_change || relevant_change))
705                                         ts->treesame[0] = 1;
706                         }
707                 }
708                 if (parse_commit(p) < 0)
709                         die("cannot simplify commit %s (because of %s)",
710                             sha1_to_hex(commit->object.sha1),
711                             sha1_to_hex(p->object.sha1));
712                 switch (rev_compare_tree(revs, p, commit)) {
713                 case REV_TREE_SAME:
714                         if (!revs->simplify_history || !relevant_commit(p)) {
715                                 /* Even if a merge with an uninteresting
716                                  * side branch brought the entire change
717                                  * we are interested in, we do not want
718                                  * to lose the other branches of this
719                                  * merge, so we just keep going.
720                                  */
721                                 if (ts)
722                                         ts->treesame[nth_parent] = 1;
723                                 continue;
724                         }
725                         parent->next = NULL;
726                         commit->parents = parent;
727                         commit->object.flags |= TREESAME;
728                         return;
729
730                 case REV_TREE_NEW:
731                         if (revs->remove_empty_trees &&
732                             rev_same_tree_as_empty(revs, p)) {
733                                 /* We are adding all the specified
734                                  * paths from this parent, so the
735                                  * history beyond this parent is not
736                                  * interesting.  Remove its parents
737                                  * (they are grandparents for us).
738                                  * IOW, we pretend this parent is a
739                                  * "root" commit.
740                                  */
741                                 if (parse_commit(p) < 0)
742                                         die("cannot simplify commit %s (invalid %s)",
743                                             sha1_to_hex(commit->object.sha1),
744                                             sha1_to_hex(p->object.sha1));
745                                 p->parents = NULL;
746                         }
747                 /* fallthrough */
748                 case REV_TREE_OLD:
749                 case REV_TREE_DIFFERENT:
750                         if (relevant_commit(p))
751                                 relevant_change = 1;
752                         else
753                                 irrelevant_change = 1;
754                         continue;
755                 }
756                 die("bad tree compare for commit %s", sha1_to_hex(commit->object.sha1));
757         }
758
759         /*
760          * TREESAME is straightforward for single-parent commits. For merge
761          * commits, it is most useful to define it so that "irrelevant"
762          * parents cannot make us !TREESAME - if we have any relevant
763          * parents, then we only consider TREESAMEness with respect to them,
764          * allowing irrelevant merges from uninteresting branches to be
765          * simplified away. Only if we have only irrelevant parents do we
766          * base TREESAME on them. Note that this logic is replicated in
767          * update_treesame, which should be kept in sync.
768          */
769         if (relevant_parents ? !relevant_change : !irrelevant_change)
770                 commit->object.flags |= TREESAME;
771 }
772
773 static void commit_list_insert_by_date_cached(struct commit *p, struct commit_list **head,
774                     struct commit_list *cached_base, struct commit_list **cache)
775 {
776         struct commit_list *new_entry;
777
778         if (cached_base && p->date < cached_base->item->date)
779                 new_entry = commit_list_insert_by_date(p, &cached_base->next);
780         else
781                 new_entry = commit_list_insert_by_date(p, head);
782
783         if (cache && (!*cache || p->date < (*cache)->item->date))
784                 *cache = new_entry;
785 }
786
787 static int add_parents_to_list(struct rev_info *revs, struct commit *commit,
788                     struct commit_list **list, struct commit_list **cache_ptr)
789 {
790         struct commit_list *parent = commit->parents;
791         unsigned left_flag;
792         struct commit_list *cached_base = cache_ptr ? *cache_ptr : NULL;
793
794         if (commit->object.flags & ADDED)
795                 return 0;
796         commit->object.flags |= ADDED;
797
798         if (revs->include_check &&
799             !revs->include_check(commit, revs->include_check_data))
800                 return 0;
801
802         /*
803          * If the commit is uninteresting, don't try to
804          * prune parents - we want the maximal uninteresting
805          * set.
806          *
807          * Normally we haven't parsed the parent
808          * yet, so we won't have a parent of a parent
809          * here. However, it may turn out that we've
810          * reached this commit some other way (where it
811          * wasn't uninteresting), in which case we need
812          * to mark its parents recursively too..
813          */
814         if (commit->object.flags & UNINTERESTING) {
815                 while (parent) {
816                         struct commit *p = parent->item;
817                         parent = parent->next;
818                         if (p)
819                                 p->object.flags |= UNINTERESTING;
820                         if (parse_commit(p) < 0)
821                                 continue;
822                         if (p->parents)
823                                 mark_parents_uninteresting(p);
824                         if (p->object.flags & SEEN)
825                                 continue;
826                         p->object.flags |= SEEN;
827                         commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
828                 }
829                 return 0;
830         }
831
832         /*
833          * Ok, the commit wasn't uninteresting. Try to
834          * simplify the commit history and find the parent
835          * that has no differences in the path set if one exists.
836          */
837         try_to_simplify_commit(revs, commit);
838
839         if (revs->no_walk)
840                 return 0;
841
842         left_flag = (commit->object.flags & SYMMETRIC_LEFT);
843
844         for (parent = commit->parents; parent; parent = parent->next) {
845                 struct commit *p = parent->item;
846
847                 if (parse_commit(p) < 0)
848                         return -1;
849                 if (revs->show_source && !p->util)
850                         p->util = commit->util;
851                 p->object.flags |= left_flag;
852                 if (!(p->object.flags & SEEN)) {
853                         p->object.flags |= SEEN;
854                         commit_list_insert_by_date_cached(p, list, cached_base, cache_ptr);
855                 }
856                 if (revs->first_parent_only)
857                         break;
858         }
859         return 0;
860 }
861
862 static void cherry_pick_list(struct commit_list *list, struct rev_info *revs)
863 {
864         struct commit_list *p;
865         int left_count = 0, right_count = 0;
866         int left_first;
867         struct patch_ids ids;
868         unsigned cherry_flag;
869
870         /* First count the commits on the left and on the right */
871         for (p = list; p; p = p->next) {
872                 struct commit *commit = p->item;
873                 unsigned flags = commit->object.flags;
874                 if (flags & BOUNDARY)
875                         ;
876                 else if (flags & SYMMETRIC_LEFT)
877                         left_count++;
878                 else
879                         right_count++;
880         }
881
882         if (!left_count || !right_count)
883                 return;
884
885         left_first = left_count < right_count;
886         init_patch_ids(&ids);
887         ids.diffopts.pathspec = revs->diffopt.pathspec;
888
889         /* Compute patch-ids for one side */
890         for (p = list; p; p = p->next) {
891                 struct commit *commit = p->item;
892                 unsigned flags = commit->object.flags;
893
894                 if (flags & BOUNDARY)
895                         continue;
896                 /*
897                  * If we have fewer left, left_first is set and we omit
898                  * commits on the right branch in this loop.  If we have
899                  * fewer right, we skip the left ones.
900                  */
901                 if (left_first != !!(flags & SYMMETRIC_LEFT))
902                         continue;
903                 commit->util = add_commit_patch_id(commit, &ids);
904         }
905
906         /* either cherry_mark or cherry_pick are true */
907         cherry_flag = revs->cherry_mark ? PATCHSAME : SHOWN;
908
909         /* Check the other side */
910         for (p = list; p; p = p->next) {
911                 struct commit *commit = p->item;
912                 struct patch_id *id;
913                 unsigned flags = commit->object.flags;
914
915                 if (flags & BOUNDARY)
916                         continue;
917                 /*
918                  * If we have fewer left, left_first is set and we omit
919                  * commits on the left branch in this loop.
920                  */
921                 if (left_first == !!(flags & SYMMETRIC_LEFT))
922                         continue;
923
924                 /*
925                  * Have we seen the same patch id?
926                  */
927                 id = has_commit_patch_id(commit, &ids);
928                 if (!id)
929                         continue;
930                 id->seen = 1;
931                 commit->object.flags |= cherry_flag;
932         }
933
934         /* Now check the original side for seen ones */
935         for (p = list; p; p = p->next) {
936                 struct commit *commit = p->item;
937                 struct patch_id *ent;
938
939                 ent = commit->util;
940                 if (!ent)
941                         continue;
942                 if (ent->seen)
943                         commit->object.flags |= cherry_flag;
944                 commit->util = NULL;
945         }
946
947         free_patch_ids(&ids);
948 }
949
950 /* How many extra uninteresting commits we want to see.. */
951 #define SLOP 5
952
953 static int still_interesting(struct commit_list *src, unsigned long date, int slop)
954 {
955         /*
956          * No source list at all? We're definitely done..
957          */
958         if (!src)
959                 return 0;
960
961         /*
962          * Does the destination list contain entries with a date
963          * before the source list? Definitely _not_ done.
964          */
965         if (date <= src->item->date)
966                 return SLOP;
967
968         /*
969          * Does the source list still have interesting commits in
970          * it? Definitely not done..
971          */
972         if (!everybody_uninteresting(src))
973                 return SLOP;
974
975         /* Ok, we're closing in.. */
976         return slop-1;
977 }
978
979 /*
980  * "rev-list --ancestry-path A..B" computes commits that are ancestors
981  * of B but not ancestors of A but further limits the result to those
982  * that are descendants of A.  This takes the list of bottom commits and
983  * the result of "A..B" without --ancestry-path, and limits the latter
984  * further to the ones that can reach one of the commits in "bottom".
985  */
986 static void limit_to_ancestry(struct commit_list *bottom, struct commit_list *list)
987 {
988         struct commit_list *p;
989         struct commit_list *rlist = NULL;
990         int made_progress;
991
992         /*
993          * Reverse the list so that it will be likely that we would
994          * process parents before children.
995          */
996         for (p = list; p; p = p->next)
997                 commit_list_insert(p->item, &rlist);
998
999         for (p = bottom; p; p = p->next)
1000                 p->item->object.flags |= TMP_MARK;
1001
1002         /*
1003          * Mark the ones that can reach bottom commits in "list",
1004          * in a bottom-up fashion.
1005          */
1006         do {
1007                 made_progress = 0;
1008                 for (p = rlist; p; p = p->next) {
1009                         struct commit *c = p->item;
1010                         struct commit_list *parents;
1011                         if (c->object.flags & (TMP_MARK | UNINTERESTING))
1012                                 continue;
1013                         for (parents = c->parents;
1014                              parents;
1015                              parents = parents->next) {
1016                                 if (!(parents->item->object.flags & TMP_MARK))
1017                                         continue;
1018                                 c->object.flags |= TMP_MARK;
1019                                 made_progress = 1;
1020                                 break;
1021                         }
1022                 }
1023         } while (made_progress);
1024
1025         /*
1026          * NEEDSWORK: decide if we want to remove parents that are
1027          * not marked with TMP_MARK from commit->parents for commits
1028          * in the resulting list.  We may not want to do that, though.
1029          */
1030
1031         /*
1032          * The ones that are not marked with TMP_MARK are uninteresting
1033          */
1034         for (p = list; p; p = p->next) {
1035                 struct commit *c = p->item;
1036                 if (c->object.flags & TMP_MARK)
1037                         continue;
1038                 c->object.flags |= UNINTERESTING;
1039         }
1040
1041         /* We are done with the TMP_MARK */
1042         for (p = list; p; p = p->next)
1043                 p->item->object.flags &= ~TMP_MARK;
1044         for (p = bottom; p; p = p->next)
1045                 p->item->object.flags &= ~TMP_MARK;
1046         free_commit_list(rlist);
1047 }
1048
1049 /*
1050  * Before walking the history, keep the set of "negative" refs the
1051  * caller has asked to exclude.
1052  *
1053  * This is used to compute "rev-list --ancestry-path A..B", as we need
1054  * to filter the result of "A..B" further to the ones that can actually
1055  * reach A.
1056  */
1057 static struct commit_list *collect_bottom_commits(struct commit_list *list)
1058 {
1059         struct commit_list *elem, *bottom = NULL;
1060         for (elem = list; elem; elem = elem->next)
1061                 if (elem->item->object.flags & BOTTOM)
1062                         commit_list_insert(elem->item, &bottom);
1063         return bottom;
1064 }
1065
1066 /* Assumes either left_only or right_only is set */
1067 static void limit_left_right(struct commit_list *list, struct rev_info *revs)
1068 {
1069         struct commit_list *p;
1070
1071         for (p = list; p; p = p->next) {
1072                 struct commit *commit = p->item;
1073
1074                 if (revs->right_only) {
1075                         if (commit->object.flags & SYMMETRIC_LEFT)
1076                                 commit->object.flags |= SHOWN;
1077                 } else  /* revs->left_only is set */
1078                         if (!(commit->object.flags & SYMMETRIC_LEFT))
1079                                 commit->object.flags |= SHOWN;
1080         }
1081 }
1082
1083 static int limit_list(struct rev_info *revs)
1084 {
1085         int slop = SLOP;
1086         unsigned long date = ~0ul;
1087         struct commit_list *list = revs->commits;
1088         struct commit_list *newlist = NULL;
1089         struct commit_list **p = &newlist;
1090         struct commit_list *bottom = NULL;
1091
1092         if (revs->ancestry_path) {
1093                 bottom = collect_bottom_commits(list);
1094                 if (!bottom)
1095                         die("--ancestry-path given but there are no bottom commits");
1096         }
1097
1098         while (list) {
1099                 struct commit_list *entry = list;
1100                 struct commit *commit = list->item;
1101                 struct object *obj = &commit->object;
1102                 show_early_output_fn_t show;
1103
1104                 list = list->next;
1105                 free(entry);
1106
1107                 if (revs->max_age != -1 && (commit->date < revs->max_age))
1108                         obj->flags |= UNINTERESTING;
1109                 if (add_parents_to_list(revs, commit, &list, NULL) < 0)
1110                         return -1;
1111                 if (obj->flags & UNINTERESTING) {
1112                         mark_parents_uninteresting(commit);
1113                         if (revs->show_all)
1114                                 p = &commit_list_insert(commit, p)->next;
1115                         slop = still_interesting(list, date, slop);
1116                         if (slop)
1117                                 continue;
1118                         /* If showing all, add the whole pending list to the end */
1119                         if (revs->show_all)
1120                                 *p = list;
1121                         break;
1122                 }
1123                 if (revs->min_age != -1 && (commit->date > revs->min_age))
1124                         continue;
1125                 date = commit->date;
1126                 p = &commit_list_insert(commit, p)->next;
1127
1128                 show = show_early_output;
1129                 if (!show)
1130                         continue;
1131
1132                 show(revs, newlist);
1133                 show_early_output = NULL;
1134         }
1135         if (revs->cherry_pick || revs->cherry_mark)
1136                 cherry_pick_list(newlist, revs);
1137
1138         if (revs->left_only || revs->right_only)
1139                 limit_left_right(newlist, revs);
1140
1141         if (bottom) {
1142                 limit_to_ancestry(bottom, newlist);
1143                 free_commit_list(bottom);
1144         }
1145
1146         /*
1147          * Check if any commits have become TREESAME by some of their parents
1148          * becoming UNINTERESTING.
1149          */
1150         if (limiting_can_increase_treesame(revs))
1151                 for (list = newlist; list; list = list->next) {
1152                         struct commit *c = list->item;
1153                         if (c->object.flags & (UNINTERESTING | TREESAME))
1154                                 continue;
1155                         update_treesame(revs, c);
1156                 }
1157
1158         revs->commits = newlist;
1159         return 0;
1160 }
1161
1162 /*
1163  * Add an entry to refs->cmdline with the specified information.
1164  * *name is copied.
1165  */
1166 static void add_rev_cmdline(struct rev_info *revs,
1167                             struct object *item,
1168                             const char *name,
1169                             int whence,
1170                             unsigned flags)
1171 {
1172         struct rev_cmdline_info *info = &revs->cmdline;
1173         int nr = info->nr;
1174
1175         ALLOC_GROW(info->rev, nr + 1, info->alloc);
1176         info->rev[nr].item = item;
1177         info->rev[nr].name = xstrdup(name);
1178         info->rev[nr].whence = whence;
1179         info->rev[nr].flags = flags;
1180         info->nr++;
1181 }
1182
1183 static void add_rev_cmdline_list(struct rev_info *revs,
1184                                  struct commit_list *commit_list,
1185                                  int whence,
1186                                  unsigned flags)
1187 {
1188         while (commit_list) {
1189                 struct object *object = &commit_list->item->object;
1190                 add_rev_cmdline(revs, object, sha1_to_hex(object->sha1),
1191                                 whence, flags);
1192                 commit_list = commit_list->next;
1193         }
1194 }
1195
1196 struct all_refs_cb {
1197         int all_flags;
1198         int warned_bad_reflog;
1199         struct rev_info *all_revs;
1200         const char *name_for_errormsg;
1201 };
1202
1203 int ref_excluded(struct string_list *ref_excludes, const char *path)
1204 {
1205         struct string_list_item *item;
1206
1207         if (!ref_excludes)
1208                 return 0;
1209         for_each_string_list_item(item, ref_excludes) {
1210                 if (!wildmatch(item->string, path, 0, NULL))
1211                         return 1;
1212         }
1213         return 0;
1214 }
1215
1216 static int handle_one_ref(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1217 {
1218         struct all_refs_cb *cb = cb_data;
1219         struct object *object;
1220
1221         if (ref_excluded(cb->all_revs->ref_excludes, path))
1222             return 0;
1223
1224         object = get_reference(cb->all_revs, path, sha1, cb->all_flags);
1225         add_rev_cmdline(cb->all_revs, object, path, REV_CMD_REF, cb->all_flags);
1226         add_pending_sha1(cb->all_revs, path, sha1, cb->all_flags);
1227         return 0;
1228 }
1229
1230 static void init_all_refs_cb(struct all_refs_cb *cb, struct rev_info *revs,
1231         unsigned flags)
1232 {
1233         cb->all_revs = revs;
1234         cb->all_flags = flags;
1235 }
1236
1237 void clear_ref_exclusion(struct string_list **ref_excludes_p)
1238 {
1239         if (*ref_excludes_p) {
1240                 string_list_clear(*ref_excludes_p, 0);
1241                 free(*ref_excludes_p);
1242         }
1243         *ref_excludes_p = NULL;
1244 }
1245
1246 void add_ref_exclusion(struct string_list **ref_excludes_p, const char *exclude)
1247 {
1248         if (!*ref_excludes_p) {
1249                 *ref_excludes_p = xcalloc(1, sizeof(**ref_excludes_p));
1250                 (*ref_excludes_p)->strdup_strings = 1;
1251         }
1252         string_list_append(*ref_excludes_p, exclude);
1253 }
1254
1255 static void handle_refs(const char *submodule, struct rev_info *revs, unsigned flags,
1256                 int (*for_each)(const char *, each_ref_fn, void *))
1257 {
1258         struct all_refs_cb cb;
1259         init_all_refs_cb(&cb, revs, flags);
1260         for_each(submodule, handle_one_ref, &cb);
1261 }
1262
1263 static void handle_one_reflog_commit(unsigned char *sha1, void *cb_data)
1264 {
1265         struct all_refs_cb *cb = cb_data;
1266         if (!is_null_sha1(sha1)) {
1267                 struct object *o = parse_object(sha1);
1268                 if (o) {
1269                         o->flags |= cb->all_flags;
1270                         /* ??? CMDLINEFLAGS ??? */
1271                         add_pending_object(cb->all_revs, o, "");
1272                 }
1273                 else if (!cb->warned_bad_reflog) {
1274                         warning("reflog of '%s' references pruned commits",
1275                                 cb->name_for_errormsg);
1276                         cb->warned_bad_reflog = 1;
1277                 }
1278         }
1279 }
1280
1281 static int handle_one_reflog_ent(unsigned char *osha1, unsigned char *nsha1,
1282                 const char *email, unsigned long timestamp, int tz,
1283                 const char *message, void *cb_data)
1284 {
1285         handle_one_reflog_commit(osha1, cb_data);
1286         handle_one_reflog_commit(nsha1, cb_data);
1287         return 0;
1288 }
1289
1290 static int handle_one_reflog(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1291 {
1292         struct all_refs_cb *cb = cb_data;
1293         cb->warned_bad_reflog = 0;
1294         cb->name_for_errormsg = path;
1295         for_each_reflog_ent(path, handle_one_reflog_ent, cb_data);
1296         return 0;
1297 }
1298
1299 void add_reflogs_to_pending(struct rev_info *revs, unsigned flags)
1300 {
1301         struct all_refs_cb cb;
1302         cb.all_revs = revs;
1303         cb.all_flags = flags;
1304         for_each_reflog(handle_one_reflog, &cb);
1305 }
1306
1307 static void add_cache_tree(struct cache_tree *it, struct rev_info *revs,
1308                            struct strbuf *path)
1309 {
1310         size_t baselen = path->len;
1311         int i;
1312
1313         if (it->entry_count >= 0) {
1314                 struct tree *tree = lookup_tree(it->sha1);
1315                 add_pending_object_with_path(revs, &tree->object, "",
1316                                              040000, path->buf);
1317         }
1318
1319         for (i = 0; i < it->subtree_nr; i++) {
1320                 struct cache_tree_sub *sub = it->down[i];
1321                 strbuf_addf(path, "%s%s", baselen ? "/" : "", sub->name);
1322                 add_cache_tree(sub->cache_tree, revs, path);
1323                 strbuf_setlen(path, baselen);
1324         }
1325
1326 }
1327
1328 void add_index_objects_to_pending(struct rev_info *revs, unsigned flags)
1329 {
1330         int i;
1331
1332         read_cache();
1333         for (i = 0; i < active_nr; i++) {
1334                 struct cache_entry *ce = active_cache[i];
1335                 struct blob *blob;
1336
1337                 if (S_ISGITLINK(ce->ce_mode))
1338                         continue;
1339
1340                 blob = lookup_blob(ce->sha1);
1341                 if (!blob)
1342                         die("unable to add index blob to traversal");
1343                 add_pending_object_with_path(revs, &blob->object, "",
1344                                              ce->ce_mode, ce->name);
1345         }
1346
1347         if (active_cache_tree) {
1348                 struct strbuf path = STRBUF_INIT;
1349                 add_cache_tree(active_cache_tree, revs, &path);
1350                 strbuf_release(&path);
1351         }
1352 }
1353
1354 static int add_parents_only(struct rev_info *revs, const char *arg_, int flags)
1355 {
1356         unsigned char sha1[20];
1357         struct object *it;
1358         struct commit *commit;
1359         struct commit_list *parents;
1360         const char *arg = arg_;
1361
1362         if (*arg == '^') {
1363                 flags ^= UNINTERESTING | BOTTOM;
1364                 arg++;
1365         }
1366         if (get_sha1_committish(arg, sha1))
1367                 return 0;
1368         while (1) {
1369                 it = get_reference(revs, arg, sha1, 0);
1370                 if (!it && revs->ignore_missing)
1371                         return 0;
1372                 if (it->type != OBJ_TAG)
1373                         break;
1374                 if (!((struct tag*)it)->tagged)
1375                         return 0;
1376                 hashcpy(sha1, ((struct tag*)it)->tagged->sha1);
1377         }
1378         if (it->type != OBJ_COMMIT)
1379                 return 0;
1380         commit = (struct commit *)it;
1381         for (parents = commit->parents; parents; parents = parents->next) {
1382                 it = &parents->item->object;
1383                 it->flags |= flags;
1384                 add_rev_cmdline(revs, it, arg_, REV_CMD_PARENTS_ONLY, flags);
1385                 add_pending_object(revs, it, arg);
1386         }
1387         return 1;
1388 }
1389
1390 void init_revisions(struct rev_info *revs, const char *prefix)
1391 {
1392         memset(revs, 0, sizeof(*revs));
1393
1394         revs->abbrev = DEFAULT_ABBREV;
1395         revs->ignore_merges = 1;
1396         revs->simplify_history = 1;
1397         DIFF_OPT_SET(&revs->pruning, RECURSIVE);
1398         DIFF_OPT_SET(&revs->pruning, QUICK);
1399         revs->pruning.add_remove = file_add_remove;
1400         revs->pruning.change = file_change;
1401         revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1402         revs->dense = 1;
1403         revs->prefix = prefix;
1404         revs->max_age = -1;
1405         revs->min_age = -1;
1406         revs->skip_count = -1;
1407         revs->max_count = -1;
1408         revs->max_parents = -1;
1409
1410         revs->commit_format = CMIT_FMT_DEFAULT;
1411
1412         init_grep_defaults();
1413         grep_init(&revs->grep_filter, prefix);
1414         revs->grep_filter.status_only = 1;
1415         revs->grep_filter.regflags = REG_NEWLINE;
1416
1417         diff_setup(&revs->diffopt);
1418         if (prefix && !revs->diffopt.prefix) {
1419                 revs->diffopt.prefix = prefix;
1420                 revs->diffopt.prefix_length = strlen(prefix);
1421         }
1422
1423         revs->notes_opt.use_default_notes = -1;
1424 }
1425
1426 static void add_pending_commit_list(struct rev_info *revs,
1427                                     struct commit_list *commit_list,
1428                                     unsigned int flags)
1429 {
1430         while (commit_list) {
1431                 struct object *object = &commit_list->item->object;
1432                 object->flags |= flags;
1433                 add_pending_object(revs, object, sha1_to_hex(object->sha1));
1434                 commit_list = commit_list->next;
1435         }
1436 }
1437
1438 static void prepare_show_merge(struct rev_info *revs)
1439 {
1440         struct commit_list *bases;
1441         struct commit *head, *other;
1442         unsigned char sha1[20];
1443         const char **prune = NULL;
1444         int i, prune_num = 1; /* counting terminating NULL */
1445
1446         if (get_sha1("HEAD", sha1))
1447                 die("--merge without HEAD?");
1448         head = lookup_commit_or_die(sha1, "HEAD");
1449         if (get_sha1("MERGE_HEAD", sha1))
1450                 die("--merge without MERGE_HEAD?");
1451         other = lookup_commit_or_die(sha1, "MERGE_HEAD");
1452         add_pending_object(revs, &head->object, "HEAD");
1453         add_pending_object(revs, &other->object, "MERGE_HEAD");
1454         bases = get_merge_bases(head, other, 1);
1455         add_rev_cmdline_list(revs, bases, REV_CMD_MERGE_BASE, UNINTERESTING | BOTTOM);
1456         add_pending_commit_list(revs, bases, UNINTERESTING | BOTTOM);
1457         free_commit_list(bases);
1458         head->object.flags |= SYMMETRIC_LEFT;
1459
1460         if (!active_nr)
1461                 read_cache();
1462         for (i = 0; i < active_nr; i++) {
1463                 const struct cache_entry *ce = active_cache[i];
1464                 if (!ce_stage(ce))
1465                         continue;
1466                 if (ce_path_match(ce, &revs->prune_data, NULL)) {
1467                         prune_num++;
1468                         REALLOC_ARRAY(prune, prune_num);
1469                         prune[prune_num-2] = ce->name;
1470                         prune[prune_num-1] = NULL;
1471                 }
1472                 while ((i+1 < active_nr) &&
1473                        ce_same_name(ce, active_cache[i+1]))
1474                         i++;
1475         }
1476         free_pathspec(&revs->prune_data);
1477         parse_pathspec(&revs->prune_data, PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
1478                        PATHSPEC_PREFER_FULL | PATHSPEC_LITERAL_PATH, "", prune);
1479         revs->limited = 1;
1480 }
1481
1482 int handle_revision_arg(const char *arg_, struct rev_info *revs, int flags, unsigned revarg_opt)
1483 {
1484         struct object_context oc;
1485         char *dotdot;
1486         struct object *object;
1487         unsigned char sha1[20];
1488         int local_flags;
1489         const char *arg = arg_;
1490         int cant_be_filename = revarg_opt & REVARG_CANNOT_BE_FILENAME;
1491         unsigned get_sha1_flags = 0;
1492
1493         flags = flags & UNINTERESTING ? flags | BOTTOM : flags & ~BOTTOM;
1494
1495         dotdot = strstr(arg, "..");
1496         if (dotdot) {
1497                 unsigned char from_sha1[20];
1498                 const char *next = dotdot + 2;
1499                 const char *this = arg;
1500                 int symmetric = *next == '.';
1501                 unsigned int flags_exclude = flags ^ (UNINTERESTING | BOTTOM);
1502                 static const char head_by_default[] = "HEAD";
1503                 unsigned int a_flags;
1504
1505                 *dotdot = 0;
1506                 next += symmetric;
1507
1508                 if (!*next)
1509                         next = head_by_default;
1510                 if (dotdot == arg)
1511                         this = head_by_default;
1512                 if (this == head_by_default && next == head_by_default &&
1513                     !symmetric) {
1514                         /*
1515                          * Just ".."?  That is not a range but the
1516                          * pathspec for the parent directory.
1517                          */
1518                         if (!cant_be_filename) {
1519                                 *dotdot = '.';
1520                                 return -1;
1521                         }
1522                 }
1523                 if (!get_sha1_committish(this, from_sha1) &&
1524                     !get_sha1_committish(next, sha1)) {
1525                         struct object *a_obj, *b_obj;
1526
1527                         if (!cant_be_filename) {
1528                                 *dotdot = '.';
1529                                 verify_non_filename(revs->prefix, arg);
1530                         }
1531
1532                         a_obj = parse_object(from_sha1);
1533                         b_obj = parse_object(sha1);
1534                         if (!a_obj || !b_obj) {
1535                         missing:
1536                                 if (revs->ignore_missing)
1537                                         return 0;
1538                                 die(symmetric
1539                                     ? "Invalid symmetric difference expression %s"
1540                                     : "Invalid revision range %s", arg);
1541                         }
1542
1543                         if (!symmetric) {
1544                                 /* just A..B */
1545                                 a_flags = flags_exclude;
1546                         } else {
1547                                 /* A...B -- find merge bases between the two */
1548                                 struct commit *a, *b;
1549                                 struct commit_list *exclude;
1550
1551                                 a = (a_obj->type == OBJ_COMMIT
1552                                      ? (struct commit *)a_obj
1553                                      : lookup_commit_reference(a_obj->sha1));
1554                                 b = (b_obj->type == OBJ_COMMIT
1555                                      ? (struct commit *)b_obj
1556                                      : lookup_commit_reference(b_obj->sha1));
1557                                 if (!a || !b)
1558                                         goto missing;
1559                                 exclude = get_merge_bases(a, b, 1);
1560                                 add_rev_cmdline_list(revs, exclude,
1561                                                      REV_CMD_MERGE_BASE,
1562                                                      flags_exclude);
1563                                 add_pending_commit_list(revs, exclude,
1564                                                         flags_exclude);
1565                                 free_commit_list(exclude);
1566
1567                                 a_flags = flags | SYMMETRIC_LEFT;
1568                         }
1569
1570                         a_obj->flags |= a_flags;
1571                         b_obj->flags |= flags;
1572                         add_rev_cmdline(revs, a_obj, this,
1573                                         REV_CMD_LEFT, a_flags);
1574                         add_rev_cmdline(revs, b_obj, next,
1575                                         REV_CMD_RIGHT, flags);
1576                         add_pending_object(revs, a_obj, this);
1577                         add_pending_object(revs, b_obj, next);
1578                         return 0;
1579                 }
1580                 *dotdot = '.';
1581         }
1582         dotdot = strstr(arg, "^@");
1583         if (dotdot && !dotdot[2]) {
1584                 *dotdot = 0;
1585                 if (add_parents_only(revs, arg, flags))
1586                         return 0;
1587                 *dotdot = '^';
1588         }
1589         dotdot = strstr(arg, "^!");
1590         if (dotdot && !dotdot[2]) {
1591                 *dotdot = 0;
1592                 if (!add_parents_only(revs, arg, flags ^ (UNINTERESTING | BOTTOM)))
1593                         *dotdot = '^';
1594         }
1595
1596         local_flags = 0;
1597         if (*arg == '^') {
1598                 local_flags = UNINTERESTING | BOTTOM;
1599                 arg++;
1600         }
1601
1602         if (revarg_opt & REVARG_COMMITTISH)
1603                 get_sha1_flags = GET_SHA1_COMMITTISH;
1604
1605         if (get_sha1_with_context(arg, get_sha1_flags, sha1, &oc))
1606                 return revs->ignore_missing ? 0 : -1;
1607         if (!cant_be_filename)
1608                 verify_non_filename(revs->prefix, arg);
1609         object = get_reference(revs, arg, sha1, flags ^ local_flags);
1610         add_rev_cmdline(revs, object, arg_, REV_CMD_REV, flags ^ local_flags);
1611         add_pending_object_with_mode(revs, object, arg, oc.mode);
1612         return 0;
1613 }
1614
1615 struct cmdline_pathspec {
1616         int alloc;
1617         int nr;
1618         const char **path;
1619 };
1620
1621 static void append_prune_data(struct cmdline_pathspec *prune, const char **av)
1622 {
1623         while (*av) {
1624                 ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1625                 prune->path[prune->nr++] = *(av++);
1626         }
1627 }
1628
1629 static void read_pathspec_from_stdin(struct rev_info *revs, struct strbuf *sb,
1630                                      struct cmdline_pathspec *prune)
1631 {
1632         while (strbuf_getwholeline(sb, stdin, '\n') != EOF) {
1633                 int len = sb->len;
1634                 if (len && sb->buf[len - 1] == '\n')
1635                         sb->buf[--len] = '\0';
1636                 ALLOC_GROW(prune->path, prune->nr + 1, prune->alloc);
1637                 prune->path[prune->nr++] = xstrdup(sb->buf);
1638         }
1639 }
1640
1641 static void read_revisions_from_stdin(struct rev_info *revs,
1642                                       struct cmdline_pathspec *prune)
1643 {
1644         struct strbuf sb;
1645         int seen_dashdash = 0;
1646         int save_warning;
1647
1648         save_warning = warn_on_object_refname_ambiguity;
1649         warn_on_object_refname_ambiguity = 0;
1650
1651         strbuf_init(&sb, 1000);
1652         while (strbuf_getwholeline(&sb, stdin, '\n') != EOF) {
1653                 int len = sb.len;
1654                 if (len && sb.buf[len - 1] == '\n')
1655                         sb.buf[--len] = '\0';
1656                 if (!len)
1657                         break;
1658                 if (sb.buf[0] == '-') {
1659                         if (len == 2 && sb.buf[1] == '-') {
1660                                 seen_dashdash = 1;
1661                                 break;
1662                         }
1663                         die("options not supported in --stdin mode");
1664                 }
1665                 if (handle_revision_arg(sb.buf, revs, 0,
1666                                         REVARG_CANNOT_BE_FILENAME))
1667                         die("bad revision '%s'", sb.buf);
1668         }
1669         if (seen_dashdash)
1670                 read_pathspec_from_stdin(revs, &sb, prune);
1671
1672         strbuf_release(&sb);
1673         warn_on_object_refname_ambiguity = save_warning;
1674 }
1675
1676 static void add_grep(struct rev_info *revs, const char *ptn, enum grep_pat_token what)
1677 {
1678         append_grep_pattern(&revs->grep_filter, ptn, "command line", 0, what);
1679 }
1680
1681 static void add_header_grep(struct rev_info *revs, enum grep_header_field field, const char *pattern)
1682 {
1683         append_header_grep_pattern(&revs->grep_filter, field, pattern);
1684 }
1685
1686 static void add_message_grep(struct rev_info *revs, const char *pattern)
1687 {
1688         add_grep(revs, pattern, GREP_PATTERN_BODY);
1689 }
1690
1691 static int handle_revision_opt(struct rev_info *revs, int argc, const char **argv,
1692                                int *unkc, const char **unkv)
1693 {
1694         const char *arg = argv[0];
1695         const char *optarg;
1696         int argcount;
1697
1698         /* pseudo revision arguments */
1699         if (!strcmp(arg, "--all") || !strcmp(arg, "--branches") ||
1700             !strcmp(arg, "--tags") || !strcmp(arg, "--remotes") ||
1701             !strcmp(arg, "--reflog") || !strcmp(arg, "--not") ||
1702             !strcmp(arg, "--no-walk") || !strcmp(arg, "--do-walk") ||
1703             !strcmp(arg, "--bisect") || starts_with(arg, "--glob=") ||
1704             !strcmp(arg, "--indexed-objects") ||
1705             starts_with(arg, "--exclude=") ||
1706             starts_with(arg, "--branches=") || starts_with(arg, "--tags=") ||
1707             starts_with(arg, "--remotes=") || starts_with(arg, "--no-walk="))
1708         {
1709                 unkv[(*unkc)++] = arg;
1710                 return 1;
1711         }
1712
1713         if ((argcount = parse_long_opt("max-count", argv, &optarg))) {
1714                 revs->max_count = atoi(optarg);
1715                 revs->no_walk = 0;
1716                 return argcount;
1717         } else if ((argcount = parse_long_opt("skip", argv, &optarg))) {
1718                 revs->skip_count = atoi(optarg);
1719                 return argcount;
1720         } else if ((*arg == '-') && isdigit(arg[1])) {
1721                 /* accept -<digit>, like traditional "head" */
1722                 if (strtol_i(arg + 1, 10, &revs->max_count) < 0 ||
1723                     revs->max_count < 0)
1724                         die("'%s': not a non-negative integer", arg + 1);
1725                 revs->no_walk = 0;
1726         } else if (!strcmp(arg, "-n")) {
1727                 if (argc <= 1)
1728                         return error("-n requires an argument");
1729                 revs->max_count = atoi(argv[1]);
1730                 revs->no_walk = 0;
1731                 return 2;
1732         } else if (starts_with(arg, "-n")) {
1733                 revs->max_count = atoi(arg + 2);
1734                 revs->no_walk = 0;
1735         } else if ((argcount = parse_long_opt("max-age", argv, &optarg))) {
1736                 revs->max_age = atoi(optarg);
1737                 return argcount;
1738         } else if ((argcount = parse_long_opt("since", argv, &optarg))) {
1739                 revs->max_age = approxidate(optarg);
1740                 return argcount;
1741         } else if ((argcount = parse_long_opt("after", argv, &optarg))) {
1742                 revs->max_age = approxidate(optarg);
1743                 return argcount;
1744         } else if ((argcount = parse_long_opt("min-age", argv, &optarg))) {
1745                 revs->min_age = atoi(optarg);
1746                 return argcount;
1747         } else if ((argcount = parse_long_opt("before", argv, &optarg))) {
1748                 revs->min_age = approxidate(optarg);
1749                 return argcount;
1750         } else if ((argcount = parse_long_opt("until", argv, &optarg))) {
1751                 revs->min_age = approxidate(optarg);
1752                 return argcount;
1753         } else if (!strcmp(arg, "--first-parent")) {
1754                 revs->first_parent_only = 1;
1755         } else if (!strcmp(arg, "--ancestry-path")) {
1756                 revs->ancestry_path = 1;
1757                 revs->simplify_history = 0;
1758                 revs->limited = 1;
1759         } else if (!strcmp(arg, "-g") || !strcmp(arg, "--walk-reflogs")) {
1760                 init_reflog_walk(&revs->reflog_info);
1761         } else if (!strcmp(arg, "--default")) {
1762                 if (argc <= 1)
1763                         return error("bad --default argument");
1764                 revs->def = argv[1];
1765                 return 2;
1766         } else if (!strcmp(arg, "--merge")) {
1767                 revs->show_merge = 1;
1768         } else if (!strcmp(arg, "--topo-order")) {
1769                 revs->sort_order = REV_SORT_IN_GRAPH_ORDER;
1770                 revs->topo_order = 1;
1771         } else if (!strcmp(arg, "--simplify-merges")) {
1772                 revs->simplify_merges = 1;
1773                 revs->topo_order = 1;
1774                 revs->rewrite_parents = 1;
1775                 revs->simplify_history = 0;
1776                 revs->limited = 1;
1777         } else if (!strcmp(arg, "--simplify-by-decoration")) {
1778                 revs->simplify_merges = 1;
1779                 revs->topo_order = 1;
1780                 revs->rewrite_parents = 1;
1781                 revs->simplify_history = 0;
1782                 revs->simplify_by_decoration = 1;
1783                 revs->limited = 1;
1784                 revs->prune = 1;
1785                 load_ref_decorations(DECORATE_SHORT_REFS);
1786         } else if (!strcmp(arg, "--date-order")) {
1787                 revs->sort_order = REV_SORT_BY_COMMIT_DATE;
1788                 revs->topo_order = 1;
1789         } else if (!strcmp(arg, "--author-date-order")) {
1790                 revs->sort_order = REV_SORT_BY_AUTHOR_DATE;
1791                 revs->topo_order = 1;
1792         } else if (starts_with(arg, "--early-output")) {
1793                 int count = 100;
1794                 switch (arg[14]) {
1795                 case '=':
1796                         count = atoi(arg+15);
1797                         /* Fallthrough */
1798                 case 0:
1799                         revs->topo_order = 1;
1800                        revs->early_output = count;
1801                 }
1802         } else if (!strcmp(arg, "--parents")) {
1803                 revs->rewrite_parents = 1;
1804                 revs->print_parents = 1;
1805         } else if (!strcmp(arg, "--dense")) {
1806                 revs->dense = 1;
1807         } else if (!strcmp(arg, "--sparse")) {
1808                 revs->dense = 0;
1809         } else if (!strcmp(arg, "--show-all")) {
1810                 revs->show_all = 1;
1811         } else if (!strcmp(arg, "--remove-empty")) {
1812                 revs->remove_empty_trees = 1;
1813         } else if (!strcmp(arg, "--merges")) {
1814                 revs->min_parents = 2;
1815         } else if (!strcmp(arg, "--no-merges")) {
1816                 revs->max_parents = 1;
1817         } else if (starts_with(arg, "--min-parents=")) {
1818                 revs->min_parents = atoi(arg+14);
1819         } else if (starts_with(arg, "--no-min-parents")) {
1820                 revs->min_parents = 0;
1821         } else if (starts_with(arg, "--max-parents=")) {
1822                 revs->max_parents = atoi(arg+14);
1823         } else if (starts_with(arg, "--no-max-parents")) {
1824                 revs->max_parents = -1;
1825         } else if (!strcmp(arg, "--boundary")) {
1826                 revs->boundary = 1;
1827         } else if (!strcmp(arg, "--left-right")) {
1828                 revs->left_right = 1;
1829         } else if (!strcmp(arg, "--left-only")) {
1830                 if (revs->right_only)
1831                         die("--left-only is incompatible with --right-only"
1832                             " or --cherry");
1833                 revs->left_only = 1;
1834         } else if (!strcmp(arg, "--right-only")) {
1835                 if (revs->left_only)
1836                         die("--right-only is incompatible with --left-only");
1837                 revs->right_only = 1;
1838         } else if (!strcmp(arg, "--cherry")) {
1839                 if (revs->left_only)
1840                         die("--cherry is incompatible with --left-only");
1841                 revs->cherry_mark = 1;
1842                 revs->right_only = 1;
1843                 revs->max_parents = 1;
1844                 revs->limited = 1;
1845         } else if (!strcmp(arg, "--count")) {
1846                 revs->count = 1;
1847         } else if (!strcmp(arg, "--cherry-mark")) {
1848                 if (revs->cherry_pick)
1849                         die("--cherry-mark is incompatible with --cherry-pick");
1850                 revs->cherry_mark = 1;
1851                 revs->limited = 1; /* needs limit_list() */
1852         } else if (!strcmp(arg, "--cherry-pick")) {
1853                 if (revs->cherry_mark)
1854                         die("--cherry-pick is incompatible with --cherry-mark");
1855                 revs->cherry_pick = 1;
1856                 revs->limited = 1;
1857         } else if (!strcmp(arg, "--objects")) {
1858                 revs->tag_objects = 1;
1859                 revs->tree_objects = 1;
1860                 revs->blob_objects = 1;
1861         } else if (!strcmp(arg, "--objects-edge")) {
1862                 revs->tag_objects = 1;
1863                 revs->tree_objects = 1;
1864                 revs->blob_objects = 1;
1865                 revs->edge_hint = 1;
1866         } else if (!strcmp(arg, "--verify-objects")) {
1867                 revs->tag_objects = 1;
1868                 revs->tree_objects = 1;
1869                 revs->blob_objects = 1;
1870                 revs->verify_objects = 1;
1871         } else if (!strcmp(arg, "--unpacked")) {
1872                 revs->unpacked = 1;
1873         } else if (starts_with(arg, "--unpacked=")) {
1874                 die("--unpacked=<packfile> no longer supported.");
1875         } else if (!strcmp(arg, "-r")) {
1876                 revs->diff = 1;
1877                 DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1878         } else if (!strcmp(arg, "-t")) {
1879                 revs->diff = 1;
1880                 DIFF_OPT_SET(&revs->diffopt, RECURSIVE);
1881                 DIFF_OPT_SET(&revs->diffopt, TREE_IN_RECURSIVE);
1882         } else if (!strcmp(arg, "-m")) {
1883                 revs->ignore_merges = 0;
1884         } else if (!strcmp(arg, "-c")) {
1885                 revs->diff = 1;
1886                 revs->dense_combined_merges = 0;
1887                 revs->combine_merges = 1;
1888         } else if (!strcmp(arg, "--cc")) {
1889                 revs->diff = 1;
1890                 revs->dense_combined_merges = 1;
1891                 revs->combine_merges = 1;
1892         } else if (!strcmp(arg, "-v")) {
1893                 revs->verbose_header = 1;
1894         } else if (!strcmp(arg, "--pretty")) {
1895                 revs->verbose_header = 1;
1896                 revs->pretty_given = 1;
1897                 get_commit_format(NULL, revs);
1898         } else if (starts_with(arg, "--pretty=") || starts_with(arg, "--format=")) {
1899                 /*
1900                  * Detached form ("--pretty X" as opposed to "--pretty=X")
1901                  * not allowed, since the argument is optional.
1902                  */
1903                 revs->verbose_header = 1;
1904                 revs->pretty_given = 1;
1905                 get_commit_format(arg+9, revs);
1906         } else if (!strcmp(arg, "--show-notes") || !strcmp(arg, "--notes")) {
1907                 revs->show_notes = 1;
1908                 revs->show_notes_given = 1;
1909                 revs->notes_opt.use_default_notes = 1;
1910         } else if (!strcmp(arg, "--show-signature")) {
1911                 revs->show_signature = 1;
1912         } else if (!strcmp(arg, "--show-linear-break") ||
1913                    starts_with(arg, "--show-linear-break=")) {
1914                 if (starts_with(arg, "--show-linear-break="))
1915                         revs->break_bar = xstrdup(arg + 20);
1916                 else
1917                         revs->break_bar = "                    ..........";
1918                 revs->track_linear = 1;
1919                 revs->track_first_time = 1;
1920         } else if (starts_with(arg, "--show-notes=") ||
1921                    starts_with(arg, "--notes=")) {
1922                 struct strbuf buf = STRBUF_INIT;
1923                 revs->show_notes = 1;
1924                 revs->show_notes_given = 1;
1925                 if (starts_with(arg, "--show-notes")) {
1926                         if (revs->notes_opt.use_default_notes < 0)
1927                                 revs->notes_opt.use_default_notes = 1;
1928                         strbuf_addstr(&buf, arg+13);
1929                 }
1930                 else
1931                         strbuf_addstr(&buf, arg+8);
1932                 expand_notes_ref(&buf);
1933                 string_list_append(&revs->notes_opt.extra_notes_refs,
1934                                    strbuf_detach(&buf, NULL));
1935         } else if (!strcmp(arg, "--no-notes")) {
1936                 revs->show_notes = 0;
1937                 revs->show_notes_given = 1;
1938                 revs->notes_opt.use_default_notes = -1;
1939                 /* we have been strdup'ing ourselves, so trick
1940                  * string_list into free()ing strings */
1941                 revs->notes_opt.extra_notes_refs.strdup_strings = 1;
1942                 string_list_clear(&revs->notes_opt.extra_notes_refs, 0);
1943                 revs->notes_opt.extra_notes_refs.strdup_strings = 0;
1944         } else if (!strcmp(arg, "--standard-notes")) {
1945                 revs->show_notes_given = 1;
1946                 revs->notes_opt.use_default_notes = 1;
1947         } else if (!strcmp(arg, "--no-standard-notes")) {
1948                 revs->notes_opt.use_default_notes = 0;
1949         } else if (!strcmp(arg, "--oneline")) {
1950                 revs->verbose_header = 1;
1951                 get_commit_format("oneline", revs);
1952                 revs->pretty_given = 1;
1953                 revs->abbrev_commit = 1;
1954         } else if (!strcmp(arg, "--graph")) {
1955                 revs->topo_order = 1;
1956                 revs->rewrite_parents = 1;
1957                 revs->graph = graph_init(revs);
1958         } else if (!strcmp(arg, "--root")) {
1959                 revs->show_root_diff = 1;
1960         } else if (!strcmp(arg, "--no-commit-id")) {
1961                 revs->no_commit_id = 1;
1962         } else if (!strcmp(arg, "--always")) {
1963                 revs->always_show_header = 1;
1964         } else if (!strcmp(arg, "--no-abbrev")) {
1965                 revs->abbrev = 0;
1966         } else if (!strcmp(arg, "--abbrev")) {
1967                 revs->abbrev = DEFAULT_ABBREV;
1968         } else if (starts_with(arg, "--abbrev=")) {
1969                 revs->abbrev = strtoul(arg + 9, NULL, 10);
1970                 if (revs->abbrev < MINIMUM_ABBREV)
1971                         revs->abbrev = MINIMUM_ABBREV;
1972                 else if (revs->abbrev > 40)
1973                         revs->abbrev = 40;
1974         } else if (!strcmp(arg, "--abbrev-commit")) {
1975                 revs->abbrev_commit = 1;
1976                 revs->abbrev_commit_given = 1;
1977         } else if (!strcmp(arg, "--no-abbrev-commit")) {
1978                 revs->abbrev_commit = 0;
1979         } else if (!strcmp(arg, "--full-diff")) {
1980                 revs->diff = 1;
1981                 revs->full_diff = 1;
1982         } else if (!strcmp(arg, "--full-history")) {
1983                 revs->simplify_history = 0;
1984         } else if (!strcmp(arg, "--relative-date")) {
1985                 revs->date_mode = DATE_RELATIVE;
1986                 revs->date_mode_explicit = 1;
1987         } else if ((argcount = parse_long_opt("date", argv, &optarg))) {
1988                 revs->date_mode = parse_date_format(optarg);
1989                 revs->date_mode_explicit = 1;
1990                 return argcount;
1991         } else if (!strcmp(arg, "--log-size")) {
1992                 revs->show_log_size = 1;
1993         }
1994         /*
1995          * Grepping the commit log
1996          */
1997         else if ((argcount = parse_long_opt("author", argv, &optarg))) {
1998                 add_header_grep(revs, GREP_HEADER_AUTHOR, optarg);
1999                 return argcount;
2000         } else if ((argcount = parse_long_opt("committer", argv, &optarg))) {
2001                 add_header_grep(revs, GREP_HEADER_COMMITTER, optarg);
2002                 return argcount;
2003         } else if ((argcount = parse_long_opt("grep-reflog", argv, &optarg))) {
2004                 add_header_grep(revs, GREP_HEADER_REFLOG, optarg);
2005                 return argcount;
2006         } else if ((argcount = parse_long_opt("grep", argv, &optarg))) {
2007                 add_message_grep(revs, optarg);
2008                 return argcount;
2009         } else if (!strcmp(arg, "--grep-debug")) {
2010                 revs->grep_filter.debug = 1;
2011         } else if (!strcmp(arg, "--basic-regexp")) {
2012                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_BRE, &revs->grep_filter);
2013         } else if (!strcmp(arg, "--extended-regexp") || !strcmp(arg, "-E")) {
2014                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_ERE, &revs->grep_filter);
2015         } else if (!strcmp(arg, "--regexp-ignore-case") || !strcmp(arg, "-i")) {
2016                 revs->grep_filter.regflags |= REG_ICASE;
2017                 DIFF_OPT_SET(&revs->diffopt, PICKAXE_IGNORE_CASE);
2018         } else if (!strcmp(arg, "--fixed-strings") || !strcmp(arg, "-F")) {
2019                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_FIXED, &revs->grep_filter);
2020         } else if (!strcmp(arg, "--perl-regexp")) {
2021                 grep_set_pattern_type_option(GREP_PATTERN_TYPE_PCRE, &revs->grep_filter);
2022         } else if (!strcmp(arg, "--all-match")) {
2023                 revs->grep_filter.all_match = 1;
2024         } else if ((argcount = parse_long_opt("encoding", argv, &optarg))) {
2025                 if (strcmp(optarg, "none"))
2026                         git_log_output_encoding = xstrdup(optarg);
2027                 else
2028                         git_log_output_encoding = "";
2029                 return argcount;
2030         } else if (!strcmp(arg, "--reverse")) {
2031                 revs->reverse ^= 1;
2032         } else if (!strcmp(arg, "--children")) {
2033                 revs->children.name = "children";
2034                 revs->limited = 1;
2035         } else if (!strcmp(arg, "--ignore-missing")) {
2036                 revs->ignore_missing = 1;
2037         } else {
2038                 int opts = diff_opt_parse(&revs->diffopt, argv, argc);
2039                 if (!opts)
2040                         unkv[(*unkc)++] = arg;
2041                 return opts;
2042         }
2043         if (revs->graph && revs->track_linear)
2044                 die("--show-linear-break and --graph are incompatible");
2045
2046         return 1;
2047 }
2048
2049 void parse_revision_opt(struct rev_info *revs, struct parse_opt_ctx_t *ctx,
2050                         const struct option *options,
2051                         const char * const usagestr[])
2052 {
2053         int n = handle_revision_opt(revs, ctx->argc, ctx->argv,
2054                                     &ctx->cpidx, ctx->out);
2055         if (n <= 0) {
2056                 error("unknown option `%s'", ctx->argv[0]);
2057                 usage_with_options(usagestr, options);
2058         }
2059         ctx->argv += n;
2060         ctx->argc -= n;
2061 }
2062
2063 static int for_each_bad_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2064 {
2065         return for_each_ref_in_submodule(submodule, "refs/bisect/bad", fn, cb_data);
2066 }
2067
2068 static int for_each_good_bisect_ref(const char *submodule, each_ref_fn fn, void *cb_data)
2069 {
2070         return for_each_ref_in_submodule(submodule, "refs/bisect/good", fn, cb_data);
2071 }
2072
2073 static int handle_revision_pseudo_opt(const char *submodule,
2074                                 struct rev_info *revs,
2075                                 int argc, const char **argv, int *flags)
2076 {
2077         const char *arg = argv[0];
2078         const char *optarg;
2079         int argcount;
2080
2081         /*
2082          * NOTE!
2083          *
2084          * Commands like "git shortlog" will not accept the options below
2085          * unless parse_revision_opt queues them (as opposed to erroring
2086          * out).
2087          *
2088          * When implementing your new pseudo-option, remember to
2089          * register it in the list at the top of handle_revision_opt.
2090          */
2091         if (!strcmp(arg, "--all")) {
2092                 handle_refs(submodule, revs, *flags, for_each_ref_submodule);
2093                 handle_refs(submodule, revs, *flags, head_ref_submodule);
2094                 clear_ref_exclusion(&revs->ref_excludes);
2095         } else if (!strcmp(arg, "--branches")) {
2096                 handle_refs(submodule, revs, *flags, for_each_branch_ref_submodule);
2097                 clear_ref_exclusion(&revs->ref_excludes);
2098         } else if (!strcmp(arg, "--bisect")) {
2099                 handle_refs(submodule, revs, *flags, for_each_bad_bisect_ref);
2100                 handle_refs(submodule, revs, *flags ^ (UNINTERESTING | BOTTOM), for_each_good_bisect_ref);
2101                 revs->bisect = 1;
2102         } else if (!strcmp(arg, "--tags")) {
2103                 handle_refs(submodule, revs, *flags, for_each_tag_ref_submodule);
2104                 clear_ref_exclusion(&revs->ref_excludes);
2105         } else if (!strcmp(arg, "--remotes")) {
2106                 handle_refs(submodule, revs, *flags, for_each_remote_ref_submodule);
2107                 clear_ref_exclusion(&revs->ref_excludes);
2108         } else if ((argcount = parse_long_opt("glob", argv, &optarg))) {
2109                 struct all_refs_cb cb;
2110                 init_all_refs_cb(&cb, revs, *flags);
2111                 for_each_glob_ref(handle_one_ref, optarg, &cb);
2112                 clear_ref_exclusion(&revs->ref_excludes);
2113                 return argcount;
2114         } else if ((argcount = parse_long_opt("exclude", argv, &optarg))) {
2115                 add_ref_exclusion(&revs->ref_excludes, optarg);
2116                 return argcount;
2117         } else if (starts_with(arg, "--branches=")) {
2118                 struct all_refs_cb cb;
2119                 init_all_refs_cb(&cb, revs, *flags);
2120                 for_each_glob_ref_in(handle_one_ref, arg + 11, "refs/heads/", &cb);
2121                 clear_ref_exclusion(&revs->ref_excludes);
2122         } else if (starts_with(arg, "--tags=")) {
2123                 struct all_refs_cb cb;
2124                 init_all_refs_cb(&cb, revs, *flags);
2125                 for_each_glob_ref_in(handle_one_ref, arg + 7, "refs/tags/", &cb);
2126                 clear_ref_exclusion(&revs->ref_excludes);
2127         } else if (starts_with(arg, "--remotes=")) {
2128                 struct all_refs_cb cb;
2129                 init_all_refs_cb(&cb, revs, *flags);
2130                 for_each_glob_ref_in(handle_one_ref, arg + 10, "refs/remotes/", &cb);
2131                 clear_ref_exclusion(&revs->ref_excludes);
2132         } else if (!strcmp(arg, "--reflog")) {
2133                 add_reflogs_to_pending(revs, *flags);
2134         } else if (!strcmp(arg, "--indexed-objects")) {
2135                 add_index_objects_to_pending(revs, *flags);
2136         } else if (!strcmp(arg, "--not")) {
2137                 *flags ^= UNINTERESTING | BOTTOM;
2138         } else if (!strcmp(arg, "--no-walk")) {
2139                 revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2140         } else if (starts_with(arg, "--no-walk=")) {
2141                 /*
2142                  * Detached form ("--no-walk X" as opposed to "--no-walk=X")
2143                  * not allowed, since the argument is optional.
2144                  */
2145                 if (!strcmp(arg + 10, "sorted"))
2146                         revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
2147                 else if (!strcmp(arg + 10, "unsorted"))
2148                         revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
2149                 else
2150                         return error("invalid argument to --no-walk");
2151         } else if (!strcmp(arg, "--do-walk")) {
2152                 revs->no_walk = 0;
2153         } else {
2154                 return 0;
2155         }
2156
2157         return 1;
2158 }
2159
2160 /*
2161  * Parse revision information, filling in the "rev_info" structure,
2162  * and removing the used arguments from the argument list.
2163  *
2164  * Returns the number of arguments left that weren't recognized
2165  * (which are also moved to the head of the argument list)
2166  */
2167 int setup_revisions(int argc, const char **argv, struct rev_info *revs, struct setup_revision_opt *opt)
2168 {
2169         int i, flags, left, seen_dashdash, read_from_stdin, got_rev_arg = 0, revarg_opt;
2170         struct cmdline_pathspec prune_data;
2171         const char *submodule = NULL;
2172
2173         memset(&prune_data, 0, sizeof(prune_data));
2174         if (opt)
2175                 submodule = opt->submodule;
2176
2177         /* First, search for "--" */
2178         if (opt && opt->assume_dashdash) {
2179                 seen_dashdash = 1;
2180         } else {
2181                 seen_dashdash = 0;
2182                 for (i = 1; i < argc; i++) {
2183                         const char *arg = argv[i];
2184                         if (strcmp(arg, "--"))
2185                                 continue;
2186                         argv[i] = NULL;
2187                         argc = i;
2188                         if (argv[i + 1])
2189                                 append_prune_data(&prune_data, argv + i + 1);
2190                         seen_dashdash = 1;
2191                         break;
2192                 }
2193         }
2194
2195         /* Second, deal with arguments and options */
2196         flags = 0;
2197         revarg_opt = opt ? opt->revarg_opt : 0;
2198         if (seen_dashdash)
2199                 revarg_opt |= REVARG_CANNOT_BE_FILENAME;
2200         read_from_stdin = 0;
2201         for (left = i = 1; i < argc; i++) {
2202                 const char *arg = argv[i];
2203                 if (*arg == '-') {
2204                         int opts;
2205
2206                         opts = handle_revision_pseudo_opt(submodule,
2207                                                 revs, argc - i, argv + i,
2208                                                 &flags);
2209                         if (opts > 0) {
2210                                 i += opts - 1;
2211                                 continue;
2212                         }
2213
2214                         if (!strcmp(arg, "--stdin")) {
2215                                 if (revs->disable_stdin) {
2216                                         argv[left++] = arg;
2217                                         continue;
2218                                 }
2219                                 if (read_from_stdin++)
2220                                         die("--stdin given twice?");
2221                                 read_revisions_from_stdin(revs, &prune_data);
2222                                 continue;
2223                         }
2224
2225                         opts = handle_revision_opt(revs, argc - i, argv + i, &left, argv);
2226                         if (opts > 0) {
2227                                 i += opts - 1;
2228                                 continue;
2229                         }
2230                         if (opts < 0)
2231                                 exit(128);
2232                         continue;
2233                 }
2234
2235
2236                 if (handle_revision_arg(arg, revs, flags, revarg_opt)) {
2237                         int j;
2238                         if (seen_dashdash || *arg == '^')
2239                                 die("bad revision '%s'", arg);
2240
2241                         /* If we didn't have a "--":
2242                          * (1) all filenames must exist;
2243                          * (2) all rev-args must not be interpretable
2244                          *     as a valid filename.
2245                          * but the latter we have checked in the main loop.
2246                          */
2247                         for (j = i; j < argc; j++)
2248                                 verify_filename(revs->prefix, argv[j], j == i);
2249
2250                         append_prune_data(&prune_data, argv + i);
2251                         break;
2252                 }
2253                 else
2254                         got_rev_arg = 1;
2255         }
2256
2257         if (prune_data.nr) {
2258                 /*
2259                  * If we need to introduce the magic "a lone ':' means no
2260                  * pathspec whatsoever", here is the place to do so.
2261                  *
2262                  * if (prune_data.nr == 1 && !strcmp(prune_data[0], ":")) {
2263                  *      prune_data.nr = 0;
2264                  *      prune_data.alloc = 0;
2265                  *      free(prune_data.path);
2266                  *      prune_data.path = NULL;
2267                  * } else {
2268                  *      terminate prune_data.alloc with NULL and
2269                  *      call init_pathspec() to set revs->prune_data here.
2270                  * }
2271                  */
2272                 ALLOC_GROW(prune_data.path, prune_data.nr + 1, prune_data.alloc);
2273                 prune_data.path[prune_data.nr++] = NULL;
2274                 parse_pathspec(&revs->prune_data, 0, 0,
2275                                revs->prefix, prune_data.path);
2276         }
2277
2278         if (revs->def == NULL)
2279                 revs->def = opt ? opt->def : NULL;
2280         if (opt && opt->tweak)
2281                 opt->tweak(revs, opt);
2282         if (revs->show_merge)
2283                 prepare_show_merge(revs);
2284         if (revs->def && !revs->pending.nr && !got_rev_arg) {
2285                 unsigned char sha1[20];
2286                 struct object *object;
2287                 struct object_context oc;
2288                 if (get_sha1_with_context(revs->def, 0, sha1, &oc))
2289                         die("bad default revision '%s'", revs->def);
2290                 object = get_reference(revs, revs->def, sha1, 0);
2291                 add_pending_object_with_mode(revs, object, revs->def, oc.mode);
2292         }
2293
2294         /* Did the user ask for any diff output? Run the diff! */
2295         if (revs->diffopt.output_format & ~DIFF_FORMAT_NO_OUTPUT)
2296                 revs->diff = 1;
2297
2298         /* Pickaxe, diff-filter and rename following need diffs */
2299         if (revs->diffopt.pickaxe ||
2300             revs->diffopt.filter ||
2301             DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2302                 revs->diff = 1;
2303
2304         if (revs->topo_order)
2305                 revs->limited = 1;
2306
2307         if (revs->prune_data.nr) {
2308                 copy_pathspec(&revs->pruning.pathspec, &revs->prune_data);
2309                 /* Can't prune commits with rename following: the paths change.. */
2310                 if (!DIFF_OPT_TST(&revs->diffopt, FOLLOW_RENAMES))
2311                         revs->prune = 1;
2312                 if (!revs->full_diff)
2313                         copy_pathspec(&revs->diffopt.pathspec,
2314                                       &revs->prune_data);
2315         }
2316         if (revs->combine_merges)
2317                 revs->ignore_merges = 0;
2318         revs->diffopt.abbrev = revs->abbrev;
2319
2320         if (revs->line_level_traverse) {
2321                 revs->limited = 1;
2322                 revs->topo_order = 1;
2323         }
2324
2325         diff_setup_done(&revs->diffopt);
2326
2327         grep_commit_pattern_type(GREP_PATTERN_TYPE_UNSPECIFIED,
2328                                  &revs->grep_filter);
2329         compile_grep_patterns(&revs->grep_filter);
2330
2331         if (revs->reverse && revs->reflog_info)
2332                 die("cannot combine --reverse with --walk-reflogs");
2333         if (revs->rewrite_parents && revs->children.name)
2334                 die("cannot combine --parents and --children");
2335
2336         /*
2337          * Limitations on the graph functionality
2338          */
2339         if (revs->reverse && revs->graph)
2340                 die("cannot combine --reverse with --graph");
2341
2342         if (revs->reflog_info && revs->graph)
2343                 die("cannot combine --walk-reflogs with --graph");
2344         if (!revs->reflog_info && revs->grep_filter.use_reflog_filter)
2345                 die("cannot use --grep-reflog without --walk-reflogs");
2346
2347         return left;
2348 }
2349
2350 static void add_child(struct rev_info *revs, struct commit *parent, struct commit *child)
2351 {
2352         struct commit_list *l = xcalloc(1, sizeof(*l));
2353
2354         l->item = child;
2355         l->next = add_decoration(&revs->children, &parent->object, l);
2356 }
2357
2358 static int remove_duplicate_parents(struct rev_info *revs, struct commit *commit)
2359 {
2360         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2361         struct commit_list **pp, *p;
2362         int surviving_parents;
2363
2364         /* Examine existing parents while marking ones we have seen... */
2365         pp = &commit->parents;
2366         surviving_parents = 0;
2367         while ((p = *pp) != NULL) {
2368                 struct commit *parent = p->item;
2369                 if (parent->object.flags & TMP_MARK) {
2370                         *pp = p->next;
2371                         if (ts)
2372                                 compact_treesame(revs, commit, surviving_parents);
2373                         continue;
2374                 }
2375                 parent->object.flags |= TMP_MARK;
2376                 surviving_parents++;
2377                 pp = &p->next;
2378         }
2379         /* clear the temporary mark */
2380         for (p = commit->parents; p; p = p->next) {
2381                 p->item->object.flags &= ~TMP_MARK;
2382         }
2383         /* no update_treesame() - removing duplicates can't affect TREESAME */
2384         return surviving_parents;
2385 }
2386
2387 struct merge_simplify_state {
2388         struct commit *simplified;
2389 };
2390
2391 static struct merge_simplify_state *locate_simplify_state(struct rev_info *revs, struct commit *commit)
2392 {
2393         struct merge_simplify_state *st;
2394
2395         st = lookup_decoration(&revs->merge_simplification, &commit->object);
2396         if (!st) {
2397                 st = xcalloc(1, sizeof(*st));
2398                 add_decoration(&revs->merge_simplification, &commit->object, st);
2399         }
2400         return st;
2401 }
2402
2403 static int mark_redundant_parents(struct rev_info *revs, struct commit *commit)
2404 {
2405         struct commit_list *h = reduce_heads(commit->parents);
2406         int i = 0, marked = 0;
2407         struct commit_list *po, *pn;
2408
2409         /* Want these for sanity-checking only */
2410         int orig_cnt = commit_list_count(commit->parents);
2411         int cnt = commit_list_count(h);
2412
2413         /*
2414          * Not ready to remove items yet, just mark them for now, based
2415          * on the output of reduce_heads(). reduce_heads outputs the reduced
2416          * set in its original order, so this isn't too hard.
2417          */
2418         po = commit->parents;
2419         pn = h;
2420         while (po) {
2421                 if (pn && po->item == pn->item) {
2422                         pn = pn->next;
2423                         i++;
2424                 } else {
2425                         po->item->object.flags |= TMP_MARK;
2426                         marked++;
2427                 }
2428                 po=po->next;
2429         }
2430
2431         if (i != cnt || cnt+marked != orig_cnt)
2432                 die("mark_redundant_parents %d %d %d %d", orig_cnt, cnt, i, marked);
2433
2434         free_commit_list(h);
2435
2436         return marked;
2437 }
2438
2439 static int mark_treesame_root_parents(struct rev_info *revs, struct commit *commit)
2440 {
2441         struct commit_list *p;
2442         int marked = 0;
2443
2444         for (p = commit->parents; p; p = p->next) {
2445                 struct commit *parent = p->item;
2446                 if (!parent->parents && (parent->object.flags & TREESAME)) {
2447                         parent->object.flags |= TMP_MARK;
2448                         marked++;
2449                 }
2450         }
2451
2452         return marked;
2453 }
2454
2455 /*
2456  * Awkward naming - this means one parent we are TREESAME to.
2457  * cf mark_treesame_root_parents: root parents that are TREESAME (to an
2458  * empty tree). Better name suggestions?
2459  */
2460 static int leave_one_treesame_to_parent(struct rev_info *revs, struct commit *commit)
2461 {
2462         struct treesame_state *ts = lookup_decoration(&revs->treesame, &commit->object);
2463         struct commit *unmarked = NULL, *marked = NULL;
2464         struct commit_list *p;
2465         unsigned n;
2466
2467         for (p = commit->parents, n = 0; p; p = p->next, n++) {
2468                 if (ts->treesame[n]) {
2469                         if (p->item->object.flags & TMP_MARK) {
2470                                 if (!marked)
2471                                         marked = p->item;
2472                         } else {
2473                                 if (!unmarked) {
2474                                         unmarked = p->item;
2475                                         break;
2476                                 }
2477                         }
2478                 }
2479         }
2480
2481         /*
2482          * If we are TREESAME to a marked-for-deletion parent, but not to any
2483          * unmarked parents, unmark the first TREESAME parent. This is the
2484          * parent that the default simplify_history==1 scan would have followed,
2485          * and it doesn't make sense to omit that path when asking for a
2486          * simplified full history. Retaining it improves the chances of
2487          * understanding odd missed merges that took an old version of a file.
2488          *
2489          * Example:
2490          *
2491          *   I--------*X       A modified the file, but mainline merge X used
2492          *    \       /        "-s ours", so took the version from I. X is
2493          *     `-*A--'         TREESAME to I and !TREESAME to A.
2494          *
2495          * Default log from X would produce "I". Without this check,
2496          * --full-history --simplify-merges would produce "I-A-X", showing
2497          * the merge commit X and that it changed A, but not making clear that
2498          * it had just taken the I version. With this check, the topology above
2499          * is retained.
2500          *
2501          * Note that it is possible that the simplification chooses a different
2502          * TREESAME parent from the default, in which case this test doesn't
2503          * activate, and we _do_ drop the default parent. Example:
2504          *
2505          *   I------X         A modified the file, but it was reverted in B,
2506          *    \    /          meaning mainline merge X is TREESAME to both
2507          *    *A-*B           parents.
2508          *
2509          * Default log would produce "I" by following the first parent;
2510          * --full-history --simplify-merges will produce "I-A-B". But this is a
2511          * reasonable result - it presents a logical full history leading from
2512          * I to X, and X is not an important merge.
2513          */
2514         if (!unmarked && marked) {
2515                 marked->object.flags &= ~TMP_MARK;
2516                 return 1;
2517         }
2518
2519         return 0;
2520 }
2521
2522 static int remove_marked_parents(struct rev_info *revs, struct commit *commit)
2523 {
2524         struct commit_list **pp, *p;
2525         int nth_parent, removed = 0;
2526
2527         pp = &commit->parents;
2528         nth_parent = 0;
2529         while ((p = *pp) != NULL) {
2530                 struct commit *parent = p->item;
2531                 if (parent->object.flags & TMP_MARK) {
2532                         parent->object.flags &= ~TMP_MARK;
2533                         *pp = p->next;
2534                         free(p);
2535                         removed++;
2536                         compact_treesame(revs, commit, nth_parent);
2537                         continue;
2538                 }
2539                 pp = &p->next;
2540                 nth_parent++;
2541         }
2542
2543         /* Removing parents can only increase TREESAMEness */
2544         if (removed && !(commit->object.flags & TREESAME))
2545                 update_treesame(revs, commit);
2546
2547         return nth_parent;
2548 }
2549
2550 static struct commit_list **simplify_one(struct rev_info *revs, struct commit *commit, struct commit_list **tail)
2551 {
2552         struct commit_list *p;
2553         struct commit *parent;
2554         struct merge_simplify_state *st, *pst;
2555         int cnt;
2556
2557         st = locate_simplify_state(revs, commit);
2558
2559         /*
2560          * Have we handled this one?
2561          */
2562         if (st->simplified)
2563                 return tail;
2564
2565         /*
2566          * An UNINTERESTING commit simplifies to itself, so does a
2567          * root commit.  We do not rewrite parents of such commit
2568          * anyway.
2569          */
2570         if ((commit->object.flags & UNINTERESTING) || !commit->parents) {
2571                 st->simplified = commit;
2572                 return tail;
2573         }
2574
2575         /*
2576          * Do we know what commit all of our parents that matter
2577          * should be rewritten to?  Otherwise we are not ready to
2578          * rewrite this one yet.
2579          */
2580         for (cnt = 0, p = commit->parents; p; p = p->next) {
2581                 pst = locate_simplify_state(revs, p->item);
2582                 if (!pst->simplified) {
2583                         tail = &commit_list_insert(p->item, tail)->next;
2584                         cnt++;
2585                 }
2586                 if (revs->first_parent_only)
2587                         break;
2588         }
2589         if (cnt) {
2590                 tail = &commit_list_insert(commit, tail)->next;
2591                 return tail;
2592         }
2593
2594         /*
2595          * Rewrite our list of parents. Note that this cannot
2596          * affect our TREESAME flags in any way - a commit is
2597          * always TREESAME to its simplification.
2598          */
2599         for (p = commit->parents; p; p = p->next) {
2600                 pst = locate_simplify_state(revs, p->item);
2601                 p->item = pst->simplified;
2602                 if (revs->first_parent_only)
2603                         break;
2604         }
2605
2606         if (revs->first_parent_only)
2607                 cnt = 1;
2608         else
2609                 cnt = remove_duplicate_parents(revs, commit);
2610
2611         /*
2612          * It is possible that we are a merge and one side branch
2613          * does not have any commit that touches the given paths;
2614          * in such a case, the immediate parent from that branch
2615          * will be rewritten to be the merge base.
2616          *
2617          *      o----X          X: the commit we are looking at;
2618          *     /    /           o: a commit that touches the paths;
2619          * ---o----'
2620          *
2621          * Further, a merge of an independent branch that doesn't
2622          * touch the path will reduce to a treesame root parent:
2623          *
2624          *  ----o----X          X: the commit we are looking at;
2625          *          /           o: a commit that touches the paths;
2626          *         r            r: a root commit not touching the paths
2627          *
2628          * Detect and simplify both cases.
2629          */
2630         if (1 < cnt) {
2631                 int marked = mark_redundant_parents(revs, commit);
2632                 marked += mark_treesame_root_parents(revs, commit);
2633                 if (marked)
2634                         marked -= leave_one_treesame_to_parent(revs, commit);
2635                 if (marked)
2636                         cnt = remove_marked_parents(revs, commit);
2637         }
2638
2639         /*
2640          * A commit simplifies to itself if it is a root, if it is
2641          * UNINTERESTING, if it touches the given paths, or if it is a
2642          * merge and its parents don't simplify to one relevant commit
2643          * (the first two cases are already handled at the beginning of
2644          * this function).
2645          *
2646          * Otherwise, it simplifies to what its sole relevant parent
2647          * simplifies to.
2648          */
2649         if (!cnt ||
2650             (commit->object.flags & UNINTERESTING) ||
2651             !(commit->object.flags & TREESAME) ||
2652             (parent = one_relevant_parent(revs, commit->parents)) == NULL)
2653                 st->simplified = commit;
2654         else {
2655                 pst = locate_simplify_state(revs, parent);
2656                 st->simplified = pst->simplified;
2657         }
2658         return tail;
2659 }
2660
2661 static void simplify_merges(struct rev_info *revs)
2662 {
2663         struct commit_list *list, *next;
2664         struct commit_list *yet_to_do, **tail;
2665         struct commit *commit;
2666
2667         if (!revs->prune)
2668                 return;
2669
2670         /* feed the list reversed */
2671         yet_to_do = NULL;
2672         for (list = revs->commits; list; list = next) {
2673                 commit = list->item;
2674                 next = list->next;
2675                 /*
2676                  * Do not free(list) here yet; the original list
2677                  * is used later in this function.
2678                  */
2679                 commit_list_insert(commit, &yet_to_do);
2680         }
2681         while (yet_to_do) {
2682                 list = yet_to_do;
2683                 yet_to_do = NULL;
2684                 tail = &yet_to_do;
2685                 while (list) {
2686                         commit = list->item;
2687                         next = list->next;
2688                         free(list);
2689                         list = next;
2690                         tail = simplify_one(revs, commit, tail);
2691                 }
2692         }
2693
2694         /* clean up the result, removing the simplified ones */
2695         list = revs->commits;
2696         revs->commits = NULL;
2697         tail = &revs->commits;
2698         while (list) {
2699                 struct merge_simplify_state *st;
2700
2701                 commit = list->item;
2702                 next = list->next;
2703                 free(list);
2704                 list = next;
2705                 st = locate_simplify_state(revs, commit);
2706                 if (st->simplified == commit)
2707                         tail = &commit_list_insert(commit, tail)->next;
2708         }
2709 }
2710
2711 static void set_children(struct rev_info *revs)
2712 {
2713         struct commit_list *l;
2714         for (l = revs->commits; l; l = l->next) {
2715                 struct commit *commit = l->item;
2716                 struct commit_list *p;
2717
2718                 for (p = commit->parents; p; p = p->next)
2719                         add_child(revs, p->item, commit);
2720         }
2721 }
2722
2723 void reset_revision_walk(void)
2724 {
2725         clear_object_flags(SEEN | ADDED | SHOWN);
2726 }
2727
2728 int prepare_revision_walk(struct rev_info *revs)
2729 {
2730         int i;
2731         struct object_array old_pending;
2732         struct commit_list **next = &revs->commits;
2733
2734         memcpy(&old_pending, &revs->pending, sizeof(old_pending));
2735         revs->pending.nr = 0;
2736         revs->pending.alloc = 0;
2737         revs->pending.objects = NULL;
2738         for (i = 0; i < old_pending.nr; i++) {
2739                 struct object_array_entry *e = old_pending.objects + i;
2740                 struct commit *commit = handle_commit(revs, e);
2741                 if (commit) {
2742                         if (!(commit->object.flags & SEEN)) {
2743                                 commit->object.flags |= SEEN;
2744                                 next = commit_list_append(commit, next);
2745                         }
2746                 }
2747         }
2748         if (!revs->leak_pending)
2749                 object_array_clear(&old_pending);
2750
2751         /* Signal whether we need per-parent treesame decoration */
2752         if (revs->simplify_merges ||
2753             (revs->limited && limiting_can_increase_treesame(revs)))
2754                 revs->treesame.name = "treesame";
2755
2756         if (revs->no_walk != REVISION_WALK_NO_WALK_UNSORTED)
2757                 commit_list_sort_by_date(&revs->commits);
2758         if (revs->no_walk)
2759                 return 0;
2760         if (revs->limited)
2761                 if (limit_list(revs) < 0)
2762                         return -1;
2763         if (revs->topo_order)
2764                 sort_in_topological_order(&revs->commits, revs->sort_order);
2765         if (revs->line_level_traverse)
2766                 line_log_filter(revs);
2767         if (revs->simplify_merges)
2768                 simplify_merges(revs);
2769         if (revs->children.name)
2770                 set_children(revs);
2771         return 0;
2772 }
2773
2774 static enum rewrite_result rewrite_one(struct rev_info *revs, struct commit **pp)
2775 {
2776         struct commit_list *cache = NULL;
2777
2778         for (;;) {
2779                 struct commit *p = *pp;
2780                 if (!revs->limited)
2781                         if (add_parents_to_list(revs, p, &revs->commits, &cache) < 0)
2782                                 return rewrite_one_error;
2783                 if (p->object.flags & UNINTERESTING)
2784                         return rewrite_one_ok;
2785                 if (!(p->object.flags & TREESAME))
2786                         return rewrite_one_ok;
2787                 if (!p->parents)
2788                         return rewrite_one_noparents;
2789                 if ((p = one_relevant_parent(revs, p->parents)) == NULL)
2790                         return rewrite_one_ok;
2791                 *pp = p;
2792         }
2793 }
2794
2795 int rewrite_parents(struct rev_info *revs, struct commit *commit,
2796         rewrite_parent_fn_t rewrite_parent)
2797 {
2798         struct commit_list **pp = &commit->parents;
2799         while (*pp) {
2800                 struct commit_list *parent = *pp;
2801                 switch (rewrite_parent(revs, &parent->item)) {
2802                 case rewrite_one_ok:
2803                         break;
2804                 case rewrite_one_noparents:
2805                         *pp = parent->next;
2806                         continue;
2807                 case rewrite_one_error:
2808                         return -1;
2809                 }
2810                 pp = &parent->next;
2811         }
2812         remove_duplicate_parents(revs, commit);
2813         return 0;
2814 }
2815
2816 static int commit_rewrite_person(struct strbuf *buf, const char *what, struct string_list *mailmap)
2817 {
2818         char *person, *endp;
2819         size_t len, namelen, maillen;
2820         const char *name;
2821         const char *mail;
2822         struct ident_split ident;
2823
2824         person = strstr(buf->buf, what);
2825         if (!person)
2826                 return 0;
2827
2828         person += strlen(what);
2829         endp = strchr(person, '\n');
2830         if (!endp)
2831                 return 0;
2832
2833         len = endp - person;
2834
2835         if (split_ident_line(&ident, person, len))
2836                 return 0;
2837
2838         mail = ident.mail_begin;
2839         maillen = ident.mail_end - ident.mail_begin;
2840         name = ident.name_begin;
2841         namelen = ident.name_end - ident.name_begin;
2842
2843         if (map_user(mailmap, &mail, &maillen, &name, &namelen)) {
2844                 struct strbuf namemail = STRBUF_INIT;
2845
2846                 strbuf_addf(&namemail, "%.*s <%.*s>",
2847                             (int)namelen, name, (int)maillen, mail);
2848
2849                 strbuf_splice(buf, ident.name_begin - buf->buf,
2850                               ident.mail_end - ident.name_begin + 1,
2851                               namemail.buf, namemail.len);
2852
2853                 strbuf_release(&namemail);
2854
2855                 return 1;
2856         }
2857
2858         return 0;
2859 }
2860
2861 static int commit_match(struct commit *commit, struct rev_info *opt)
2862 {
2863         int retval;
2864         const char *encoding;
2865         const char *message;
2866         struct strbuf buf = STRBUF_INIT;
2867
2868         if (!opt->grep_filter.pattern_list && !opt->grep_filter.header_list)
2869                 return 1;
2870
2871         /* Prepend "fake" headers as needed */
2872         if (opt->grep_filter.use_reflog_filter) {
2873                 strbuf_addstr(&buf, "reflog ");
2874                 get_reflog_message(&buf, opt->reflog_info);
2875                 strbuf_addch(&buf, '\n');
2876         }
2877
2878         /*
2879          * We grep in the user's output encoding, under the assumption that it
2880          * is the encoding they are most likely to write their grep pattern
2881          * for. In addition, it means we will match the "notes" encoding below,
2882          * so we will not end up with a buffer that has two different encodings
2883          * in it.
2884          */
2885         encoding = get_log_output_encoding();
2886         message = logmsg_reencode(commit, NULL, encoding);
2887
2888         /* Copy the commit to temporary if we are using "fake" headers */
2889         if (buf.len)
2890                 strbuf_addstr(&buf, message);
2891
2892         if (opt->grep_filter.header_list && opt->mailmap) {
2893                 if (!buf.len)
2894                         strbuf_addstr(&buf, message);
2895
2896                 commit_rewrite_person(&buf, "\nauthor ", opt->mailmap);
2897                 commit_rewrite_person(&buf, "\ncommitter ", opt->mailmap);
2898         }
2899
2900         /* Append "fake" message parts as needed */
2901         if (opt->show_notes) {
2902                 if (!buf.len)
2903                         strbuf_addstr(&buf, message);
2904                 format_display_notes(commit->object.sha1, &buf, encoding, 1);
2905         }
2906
2907         /*
2908          * Find either in the original commit message, or in the temporary.
2909          * Note that we cast away the constness of "message" here. It is
2910          * const because it may come from the cached commit buffer. That's OK,
2911          * because we know that it is modifiable heap memory, and that while
2912          * grep_buffer may modify it for speed, it will restore any
2913          * changes before returning.
2914          */
2915         if (buf.len)
2916                 retval = grep_buffer(&opt->grep_filter, buf.buf, buf.len);
2917         else
2918                 retval = grep_buffer(&opt->grep_filter,
2919                                      (char *)message, strlen(message));
2920         strbuf_release(&buf);
2921         unuse_commit_buffer(commit, message);
2922         return retval;
2923 }
2924
2925 static inline int want_ancestry(const struct rev_info *revs)
2926 {
2927         return (revs->rewrite_parents || revs->children.name);
2928 }
2929
2930 enum commit_action get_commit_action(struct rev_info *revs, struct commit *commit)
2931 {
2932         if (commit->object.flags & SHOWN)
2933                 return commit_ignore;
2934         if (revs->unpacked && has_sha1_pack(commit->object.sha1))
2935                 return commit_ignore;
2936         if (revs->show_all)
2937                 return commit_show;
2938         if (commit->object.flags & UNINTERESTING)
2939                 return commit_ignore;
2940         if (revs->min_age != -1 && (commit->date > revs->min_age))
2941                 return commit_ignore;
2942         if (revs->min_parents || (revs->max_parents >= 0)) {
2943                 int n = commit_list_count(commit->parents);
2944                 if ((n < revs->min_parents) ||
2945                     ((revs->max_parents >= 0) && (n > revs->max_parents)))
2946                         return commit_ignore;
2947         }
2948         if (!commit_match(commit, revs))
2949                 return commit_ignore;
2950         if (revs->prune && revs->dense) {
2951                 /* Commit without changes? */
2952                 if (commit->object.flags & TREESAME) {
2953                         int n;
2954                         struct commit_list *p;
2955                         /* drop merges unless we want parenthood */
2956                         if (!want_ancestry(revs))
2957                                 return commit_ignore;
2958                         /*
2959                          * If we want ancestry, then need to keep any merges
2960                          * between relevant commits to tie together topology.
2961                          * For consistency with TREESAME and simplification
2962                          * use "relevant" here rather than just INTERESTING,
2963                          * to treat bottom commit(s) as part of the topology.
2964                          */
2965                         for (n = 0, p = commit->parents; p; p = p->next)
2966                                 if (relevant_commit(p->item))
2967                                         if (++n >= 2)
2968                                                 return commit_show;
2969                         return commit_ignore;
2970                 }
2971         }
2972         return commit_show;
2973 }
2974
2975 enum commit_action simplify_commit(struct rev_info *revs, struct commit *commit)
2976 {
2977         enum commit_action action = get_commit_action(revs, commit);
2978
2979         if (action == commit_show &&
2980             !revs->show_all &&
2981             revs->prune && revs->dense && want_ancestry(revs)) {
2982                 /*
2983                  * --full-diff on simplified parents is no good: it
2984                  * will show spurious changes from the commits that
2985                  * were elided.  So we save the parents on the side
2986                  * when --full-diff is in effect.
2987                  */
2988                 if (revs->full_diff)
2989                         save_parents(revs, commit);
2990                 if (rewrite_parents(revs, commit, rewrite_one) < 0)
2991                         return commit_error;
2992         }
2993         return action;
2994 }
2995
2996 static void track_linear(struct rev_info *revs, struct commit *commit)
2997 {
2998         if (revs->track_first_time) {
2999                 revs->linear = 1;
3000                 revs->track_first_time = 0;
3001         } else {
3002                 struct commit_list *p;
3003                 for (p = revs->previous_parents; p; p = p->next)
3004                         if (p->item == NULL || /* first commit */
3005                             !hashcmp(p->item->object.sha1, commit->object.sha1))
3006                                 break;
3007                 revs->linear = p != NULL;
3008         }
3009         if (revs->reverse) {
3010                 if (revs->linear)
3011                         commit->object.flags |= TRACK_LINEAR;
3012         }
3013         free_commit_list(revs->previous_parents);
3014         revs->previous_parents = copy_commit_list(commit->parents);
3015 }
3016
3017 static struct commit *get_revision_1(struct rev_info *revs)
3018 {
3019         if (!revs->commits)
3020                 return NULL;
3021
3022         do {
3023                 struct commit_list *entry = revs->commits;
3024                 struct commit *commit = entry->item;
3025
3026                 revs->commits = entry->next;
3027                 free(entry);
3028
3029                 if (revs->reflog_info) {
3030                         save_parents(revs, commit);
3031                         fake_reflog_parent(revs->reflog_info, commit);
3032                         commit->object.flags &= ~(ADDED | SEEN | SHOWN);
3033                 }
3034
3035                 /*
3036                  * If we haven't done the list limiting, we need to look at
3037                  * the parents here. We also need to do the date-based limiting
3038                  * that we'd otherwise have done in limit_list().
3039                  */
3040                 if (!revs->limited) {
3041                         if (revs->max_age != -1 &&
3042                             (commit->date < revs->max_age))
3043                                 continue;
3044                         if (add_parents_to_list(revs, commit, &revs->commits, NULL) < 0) {
3045                                 if (!revs->ignore_missing_links)
3046                                         die("Failed to traverse parents of commit %s",
3047                                                 sha1_to_hex(commit->object.sha1));
3048                         }
3049                 }
3050
3051                 switch (simplify_commit(revs, commit)) {
3052                 case commit_ignore:
3053                         continue;
3054                 case commit_error:
3055                         die("Failed to simplify parents of commit %s",
3056                             sha1_to_hex(commit->object.sha1));
3057                 default:
3058                         if (revs->track_linear)
3059                                 track_linear(revs, commit);
3060                         return commit;
3061                 }
3062         } while (revs->commits);
3063         return NULL;
3064 }
3065
3066 /*
3067  * Return true for entries that have not yet been shown.  (This is an
3068  * object_array_each_func_t.)
3069  */
3070 static int entry_unshown(struct object_array_entry *entry, void *cb_data_unused)
3071 {
3072         return !(entry->item->flags & SHOWN);
3073 }
3074
3075 /*
3076  * If array is on the verge of a realloc, garbage-collect any entries
3077  * that have already been shown to try to free up some space.
3078  */
3079 static void gc_boundary(struct object_array *array)
3080 {
3081         if (array->nr == array->alloc)
3082                 object_array_filter(array, entry_unshown, NULL);
3083 }
3084
3085 static void create_boundary_commit_list(struct rev_info *revs)
3086 {
3087         unsigned i;
3088         struct commit *c;
3089         struct object_array *array = &revs->boundary_commits;
3090         struct object_array_entry *objects = array->objects;
3091
3092         /*
3093          * If revs->commits is non-NULL at this point, an error occurred in
3094          * get_revision_1().  Ignore the error and continue printing the
3095          * boundary commits anyway.  (This is what the code has always
3096          * done.)
3097          */
3098         if (revs->commits) {
3099                 free_commit_list(revs->commits);
3100                 revs->commits = NULL;
3101         }
3102
3103         /*
3104          * Put all of the actual boundary commits from revs->boundary_commits
3105          * into revs->commits
3106          */
3107         for (i = 0; i < array->nr; i++) {
3108                 c = (struct commit *)(objects[i].item);
3109                 if (!c)
3110                         continue;
3111                 if (!(c->object.flags & CHILD_SHOWN))
3112                         continue;
3113                 if (c->object.flags & (SHOWN | BOUNDARY))
3114                         continue;
3115                 c->object.flags |= BOUNDARY;
3116                 commit_list_insert(c, &revs->commits);
3117         }
3118
3119         /*
3120          * If revs->topo_order is set, sort the boundary commits
3121          * in topological order
3122          */
3123         sort_in_topological_order(&revs->commits, revs->sort_order);
3124 }
3125
3126 static struct commit *get_revision_internal(struct rev_info *revs)
3127 {
3128         struct commit *c = NULL;
3129         struct commit_list *l;
3130
3131         if (revs->boundary == 2) {
3132                 /*
3133                  * All of the normal commits have already been returned,
3134                  * and we are now returning boundary commits.
3135                  * create_boundary_commit_list() has populated
3136                  * revs->commits with the remaining commits to return.
3137                  */
3138                 c = pop_commit(&revs->commits);
3139                 if (c)
3140                         c->object.flags |= SHOWN;
3141                 return c;
3142         }
3143
3144         /*
3145          * If our max_count counter has reached zero, then we are done. We
3146          * don't simply return NULL because we still might need to show
3147          * boundary commits. But we want to avoid calling get_revision_1, which
3148          * might do a considerable amount of work finding the next commit only
3149          * for us to throw it away.
3150          *
3151          * If it is non-zero, then either we don't have a max_count at all
3152          * (-1), or it is still counting, in which case we decrement.
3153          */
3154         if (revs->max_count) {
3155                 c = get_revision_1(revs);
3156                 if (c) {
3157                         while (revs->skip_count > 0) {
3158                                 revs->skip_count--;
3159                                 c = get_revision_1(revs);
3160                                 if (!c)
3161                                         break;
3162                         }
3163                 }
3164
3165                 if (revs->max_count > 0)
3166                         revs->max_count--;
3167         }
3168
3169         if (c)
3170                 c->object.flags |= SHOWN;
3171
3172         if (!revs->boundary)
3173                 return c;
3174
3175         if (!c) {
3176                 /*
3177                  * get_revision_1() runs out the commits, and
3178                  * we are done computing the boundaries.
3179                  * switch to boundary commits output mode.
3180                  */
3181                 revs->boundary = 2;
3182
3183                 /*
3184                  * Update revs->commits to contain the list of
3185                  * boundary commits.
3186                  */
3187                 create_boundary_commit_list(revs);
3188
3189                 return get_revision_internal(revs);
3190         }
3191
3192         /*
3193          * boundary commits are the commits that are parents of the
3194          * ones we got from get_revision_1() but they themselves are
3195          * not returned from get_revision_1().  Before returning
3196          * 'c', we need to mark its parents that they could be boundaries.
3197          */
3198
3199         for (l = c->parents; l; l = l->next) {
3200                 struct object *p;
3201                 p = &(l->item->object);
3202                 if (p->flags & (CHILD_SHOWN | SHOWN))
3203                         continue;
3204                 p->flags |= CHILD_SHOWN;
3205                 gc_boundary(&revs->boundary_commits);
3206                 add_object_array(p, NULL, &revs->boundary_commits);
3207         }
3208
3209         return c;
3210 }
3211
3212 struct commit *get_revision(struct rev_info *revs)
3213 {
3214         struct commit *c;
3215         struct commit_list *reversed;
3216
3217         if (revs->reverse) {
3218                 reversed = NULL;
3219                 while ((c = get_revision_internal(revs)))
3220                         commit_list_insert(c, &reversed);
3221                 revs->commits = reversed;
3222                 revs->reverse = 0;
3223                 revs->reverse_output_stage = 1;
3224         }
3225
3226         if (revs->reverse_output_stage) {
3227                 c = pop_commit(&revs->commits);
3228                 if (revs->track_linear)
3229                         revs->linear = !!(c && c->object.flags & TRACK_LINEAR);
3230                 return c;
3231         }
3232
3233         c = get_revision_internal(revs);
3234         if (c && revs->graph)
3235                 graph_update(revs->graph, c);
3236         if (!c) {
3237                 free_saved_parents(revs);
3238                 if (revs->previous_parents) {
3239                         free_commit_list(revs->previous_parents);
3240                         revs->previous_parents = NULL;
3241                 }
3242         }
3243         return c;
3244 }
3245
3246 char *get_revision_mark(const struct rev_info *revs, const struct commit *commit)
3247 {
3248         if (commit->object.flags & BOUNDARY)
3249                 return "-";
3250         else if (commit->object.flags & UNINTERESTING)
3251                 return "^";
3252         else if (commit->object.flags & PATCHSAME)
3253                 return "=";
3254         else if (!revs || revs->left_right) {
3255                 if (commit->object.flags & SYMMETRIC_LEFT)
3256                         return "<";
3257                 else
3258                         return ">";
3259         } else if (revs->graph)
3260                 return "*";
3261         else if (revs->cherry_mark)
3262                 return "+";
3263         return "";
3264 }
3265
3266 void put_revision_mark(const struct rev_info *revs, const struct commit *commit)
3267 {
3268         char *mark = get_revision_mark(revs, commit);
3269         if (!strlen(mark))
3270                 return;
3271         fputs(mark, stdout);
3272         putchar(' ');
3273 }
3274
3275 define_commit_slab(saved_parents, struct commit_list *);
3276
3277 #define EMPTY_PARENT_LIST ((struct commit_list *)-1)
3278
3279 void save_parents(struct rev_info *revs, struct commit *commit)
3280 {
3281         struct commit_list **pp;
3282
3283         if (!revs->saved_parents_slab) {
3284                 revs->saved_parents_slab = xmalloc(sizeof(struct saved_parents));
3285                 init_saved_parents(revs->saved_parents_slab);
3286         }
3287
3288         pp = saved_parents_at(revs->saved_parents_slab, commit);
3289
3290         /*
3291          * When walking with reflogs, we may visit the same commit
3292          * several times: once for each appearance in the reflog.
3293          *
3294          * In this case, save_parents() will be called multiple times.
3295          * We want to keep only the first set of parents.  We need to
3296          * store a sentinel value for an empty (i.e., NULL) parent
3297          * list to distinguish it from a not-yet-saved list, however.
3298          */
3299         if (*pp)
3300                 return;
3301         if (commit->parents)
3302                 *pp = copy_commit_list(commit->parents);
3303         else
3304                 *pp = EMPTY_PARENT_LIST;
3305 }
3306
3307 struct commit_list *get_saved_parents(struct rev_info *revs, const struct commit *commit)
3308 {
3309         struct commit_list *parents;
3310
3311         if (!revs->saved_parents_slab)
3312                 return commit->parents;
3313
3314         parents = *saved_parents_at(revs->saved_parents_slab, commit);
3315         if (parents == EMPTY_PARENT_LIST)
3316                 return NULL;
3317         return parents;
3318 }
3319
3320 void free_saved_parents(struct rev_info *revs)
3321 {
3322         if (revs->saved_parents_slab)
3323                 clear_saved_parents(revs->saved_parents_slab);
3324 }