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