merge-base --fork-point: extract libified function
[git] / commit.c
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "commit-graph.h"
5 #include "repository.h"
6 #include "object-store.h"
7 #include "pkt-line.h"
8 #include "utf8.h"
9 #include "diff.h"
10 #include "revision.h"
11 #include "notes.h"
12 #include "alloc.h"
13 #include "gpg-interface.h"
14 #include "mergesort.h"
15 #include "commit-slab.h"
16 #include "prio-queue.h"
17 #include "sha1-lookup.h"
18 #include "wt-status.h"
19 #include "advice.h"
20 #include "refs.h"
21
22 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
23
24 int save_commit_buffer = 1;
25
26 const char *commit_type = "commit";
27
28 struct commit *lookup_commit_reference_gently(struct repository *r,
29                 const struct object_id *oid, int quiet)
30 {
31         struct object *obj = deref_tag(r,
32                                        parse_object(r, oid),
33                                        NULL, 0);
34
35         if (!obj)
36                 return NULL;
37         return object_as_type(r, obj, OBJ_COMMIT, quiet);
38 }
39
40 struct commit *lookup_commit_reference(struct repository *r, const struct object_id *oid)
41 {
42         return lookup_commit_reference_gently(r, oid, 0);
43 }
44
45 struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
46 {
47         struct commit *c = lookup_commit_reference(the_repository, oid);
48         if (!c)
49                 die(_("could not parse %s"), ref_name);
50         if (oidcmp(oid, &c->object.oid)) {
51                 warning(_("%s %s is not a commit!"),
52                         ref_name, oid_to_hex(oid));
53         }
54         return c;
55 }
56
57 struct commit *lookup_commit(struct repository *r, const struct object_id *oid)
58 {
59         struct object *obj = lookup_object(r, oid->hash);
60         if (!obj)
61                 return create_object(r, oid->hash,
62                                      alloc_commit_node(r));
63         return object_as_type(r, obj, OBJ_COMMIT, 0);
64 }
65
66 struct commit *lookup_commit_reference_by_name(const char *name)
67 {
68         struct object_id oid;
69         struct commit *commit;
70
71         if (get_oid_committish(name, &oid))
72                 return NULL;
73         commit = lookup_commit_reference(the_repository, &oid);
74         if (parse_commit(commit))
75                 return NULL;
76         return commit;
77 }
78
79 static timestamp_t parse_commit_date(const char *buf, const char *tail)
80 {
81         const char *dateptr;
82
83         if (buf + 6 >= tail)
84                 return 0;
85         if (memcmp(buf, "author", 6))
86                 return 0;
87         while (buf < tail && *buf++ != '\n')
88                 /* nada */;
89         if (buf + 9 >= tail)
90                 return 0;
91         if (memcmp(buf, "committer", 9))
92                 return 0;
93         while (buf < tail && *buf++ != '>')
94                 /* nada */;
95         if (buf >= tail)
96                 return 0;
97         dateptr = buf;
98         while (buf < tail && *buf++ != '\n')
99                 /* nada */;
100         if (buf >= tail)
101                 return 0;
102         /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
103         return parse_timestamp(dateptr, NULL, 10);
104 }
105
106 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
107 {
108         struct commit_graft **commit_graft_table = table;
109         return commit_graft_table[index]->oid.hash;
110 }
111
112 static int commit_graft_pos(struct repository *r, const unsigned char *sha1)
113 {
114         return sha1_pos(sha1, r->parsed_objects->grafts,
115                         r->parsed_objects->grafts_nr,
116                         commit_graft_sha1_access);
117 }
118
119 int register_commit_graft(struct repository *r, struct commit_graft *graft,
120                           int ignore_dups)
121 {
122         int pos = commit_graft_pos(r, graft->oid.hash);
123
124         if (0 <= pos) {
125                 if (ignore_dups)
126                         free(graft);
127                 else {
128                         free(r->parsed_objects->grafts[pos]);
129                         r->parsed_objects->grafts[pos] = graft;
130                 }
131                 return 1;
132         }
133         pos = -pos - 1;
134         ALLOC_GROW(r->parsed_objects->grafts,
135                    r->parsed_objects->grafts_nr + 1,
136                    r->parsed_objects->grafts_alloc);
137         r->parsed_objects->grafts_nr++;
138         if (pos < r->parsed_objects->grafts_nr)
139                 memmove(r->parsed_objects->grafts + pos + 1,
140                         r->parsed_objects->grafts + pos,
141                         (r->parsed_objects->grafts_nr - pos - 1) *
142                         sizeof(*r->parsed_objects->grafts));
143         r->parsed_objects->grafts[pos] = graft;
144         return 0;
145 }
146
147 struct commit_graft *read_graft_line(struct strbuf *line)
148 {
149         /* The format is just "Commit Parent1 Parent2 ...\n" */
150         int i, phase;
151         const char *tail = NULL;
152         struct commit_graft *graft = NULL;
153         struct object_id dummy_oid, *oid;
154
155         strbuf_rtrim(line);
156         if (!line->len || line->buf[0] == '#')
157                 return NULL;
158         /*
159          * phase 0 verifies line, counts hashes in line and allocates graft
160          * phase 1 fills graft
161          */
162         for (phase = 0; phase < 2; phase++) {
163                 oid = graft ? &graft->oid : &dummy_oid;
164                 if (parse_oid_hex(line->buf, oid, &tail))
165                         goto bad_graft_data;
166                 for (i = 0; *tail != '\0'; i++) {
167                         oid = graft ? &graft->parent[i] : &dummy_oid;
168                         if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
169                                 goto bad_graft_data;
170                 }
171                 if (!graft) {
172                         graft = xmalloc(st_add(sizeof(*graft),
173                                                st_mult(sizeof(struct object_id), i)));
174                         graft->nr_parent = i;
175                 }
176         }
177         return graft;
178
179 bad_graft_data:
180         error("bad graft data: %s", line->buf);
181         assert(!graft);
182         return NULL;
183 }
184
185 static int read_graft_file(struct repository *r, const char *graft_file)
186 {
187         FILE *fp = fopen_or_warn(graft_file, "r");
188         struct strbuf buf = STRBUF_INIT;
189         if (!fp)
190                 return -1;
191         if (advice_graft_file_deprecated)
192                 advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
193                          "and will be removed in a future Git version.\n"
194                          "\n"
195                          "Please use \"git replace --convert-graft-file\"\n"
196                          "to convert the grafts into replace refs.\n"
197                          "\n"
198                          "Turn this message off by running\n"
199                          "\"git config advice.graftFileDeprecated false\""));
200         while (!strbuf_getwholeline(&buf, fp, '\n')) {
201                 /* The format is just "Commit Parent1 Parent2 ...\n" */
202                 struct commit_graft *graft = read_graft_line(&buf);
203                 if (!graft)
204                         continue;
205                 if (register_commit_graft(r, graft, 1))
206                         error("duplicate graft data: %s", buf.buf);
207         }
208         fclose(fp);
209         strbuf_release(&buf);
210         return 0;
211 }
212
213 static void prepare_commit_graft(struct repository *r)
214 {
215         char *graft_file;
216
217         if (r->parsed_objects->commit_graft_prepared)
218                 return;
219         if (!startup_info->have_repository)
220                 return;
221
222         graft_file = get_graft_file(r);
223         read_graft_file(r, graft_file);
224         /* make sure shallows are read */
225         is_repository_shallow(r);
226         r->parsed_objects->commit_graft_prepared = 1;
227 }
228
229 struct commit_graft *lookup_commit_graft(struct repository *r, const struct object_id *oid)
230 {
231         int pos;
232         prepare_commit_graft(r);
233         pos = commit_graft_pos(r, oid->hash);
234         if (pos < 0)
235                 return NULL;
236         return r->parsed_objects->grafts[pos];
237 }
238
239 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
240 {
241         int i, ret;
242         for (i = ret = 0; i < the_repository->parsed_objects->grafts_nr && !ret; i++)
243                 ret = fn(the_repository->parsed_objects->grafts[i], cb_data);
244         return ret;
245 }
246
247 int unregister_shallow(const struct object_id *oid)
248 {
249         int pos = commit_graft_pos(the_repository, oid->hash);
250         if (pos < 0)
251                 return -1;
252         if (pos + 1 < the_repository->parsed_objects->grafts_nr)
253                 MOVE_ARRAY(the_repository->parsed_objects->grafts + pos,
254                            the_repository->parsed_objects->grafts + pos + 1,
255                            the_repository->parsed_objects->grafts_nr - pos - 1);
256         the_repository->parsed_objects->grafts_nr--;
257         return 0;
258 }
259
260 struct commit_buffer {
261         void *buffer;
262         unsigned long size;
263 };
264 define_commit_slab(buffer_slab, struct commit_buffer);
265
266 struct buffer_slab *allocate_commit_buffer_slab(void)
267 {
268         struct buffer_slab *bs = xmalloc(sizeof(*bs));
269         init_buffer_slab(bs);
270         return bs;
271 }
272
273 void free_commit_buffer_slab(struct buffer_slab *bs)
274 {
275         clear_buffer_slab(bs);
276         free(bs);
277 }
278
279 void set_commit_buffer(struct repository *r, struct commit *commit, void *buffer, unsigned long size)
280 {
281         struct commit_buffer *v = buffer_slab_at(
282                 r->parsed_objects->buffer_slab, commit);
283         v->buffer = buffer;
284         v->size = size;
285 }
286
287 const void *get_cached_commit_buffer(struct repository *r, const struct commit *commit, unsigned long *sizep)
288 {
289         struct commit_buffer *v = buffer_slab_peek(
290                 r->parsed_objects->buffer_slab, commit);
291         if (!v) {
292                 if (sizep)
293                         *sizep = 0;
294                 return NULL;
295         }
296         if (sizep)
297                 *sizep = v->size;
298         return v->buffer;
299 }
300
301 const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
302 {
303         const void *ret = get_cached_commit_buffer(the_repository, commit, sizep);
304         if (!ret) {
305                 enum object_type type;
306                 unsigned long size;
307                 ret = read_object_file(&commit->object.oid, &type, &size);
308                 if (!ret)
309                         die("cannot read commit object %s",
310                             oid_to_hex(&commit->object.oid));
311                 if (type != OBJ_COMMIT)
312                         die("expected commit for %s, got %s",
313                             oid_to_hex(&commit->object.oid), type_name(type));
314                 if (sizep)
315                         *sizep = size;
316         }
317         return ret;
318 }
319
320 void unuse_commit_buffer(const struct commit *commit, const void *buffer)
321 {
322         struct commit_buffer *v = buffer_slab_peek(
323                 the_repository->parsed_objects->buffer_slab, commit);
324         if (!(v && v->buffer == buffer))
325                 free((void *)buffer);
326 }
327
328 void free_commit_buffer(struct commit *commit)
329 {
330         struct commit_buffer *v = buffer_slab_peek(
331                 the_repository->parsed_objects->buffer_slab, commit);
332         if (v) {
333                 FREE_AND_NULL(v->buffer);
334                 v->size = 0;
335         }
336 }
337
338 struct tree *get_commit_tree(const struct commit *commit)
339 {
340         if (commit->maybe_tree || !commit->object.parsed)
341                 return commit->maybe_tree;
342
343         if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
344                 BUG("commit has NULL tree, but was not loaded from commit-graph");
345
346         return get_commit_tree_in_graph(the_repository, commit);
347 }
348
349 struct object_id *get_commit_tree_oid(const struct commit *commit)
350 {
351         return &get_commit_tree(commit)->object.oid;
352 }
353
354 void release_commit_memory(struct commit *c)
355 {
356         c->maybe_tree = NULL;
357         c->index = 0;
358         free_commit_buffer(c);
359         free_commit_list(c->parents);
360         /* TODO: what about commit->util? */
361
362         c->object.parsed = 0;
363 }
364
365 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
366 {
367         struct commit_buffer *v = buffer_slab_peek(
368                 the_repository->parsed_objects->buffer_slab, commit);
369         void *ret;
370
371         if (!v) {
372                 if (sizep)
373                         *sizep = 0;
374                 return NULL;
375         }
376         ret = v->buffer;
377         if (sizep)
378                 *sizep = v->size;
379
380         v->buffer = NULL;
381         v->size = 0;
382         return ret;
383 }
384
385 int parse_commit_buffer(struct repository *r, struct commit *item, const void *buffer, unsigned long size, int check_graph)
386 {
387         const char *tail = buffer;
388         const char *bufptr = buffer;
389         struct object_id parent;
390         struct commit_list **pptr;
391         struct commit_graft *graft;
392         const int tree_entry_len = the_hash_algo->hexsz + 5;
393         const int parent_entry_len = the_hash_algo->hexsz + 7;
394
395         if (item->object.parsed)
396                 return 0;
397         item->object.parsed = 1;
398         tail += size;
399         if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
400                         bufptr[tree_entry_len] != '\n')
401                 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
402         if (get_oid_hex(bufptr + 5, &parent) < 0)
403                 return error("bad tree pointer in commit %s",
404                              oid_to_hex(&item->object.oid));
405         item->maybe_tree = lookup_tree(r, &parent);
406         bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
407         pptr = &item->parents;
408
409         graft = lookup_commit_graft(r, &item->object.oid);
410         while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
411                 struct commit *new_parent;
412
413                 if (tail <= bufptr + parent_entry_len + 1 ||
414                     get_oid_hex(bufptr + 7, &parent) ||
415                     bufptr[parent_entry_len] != '\n')
416                         return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
417                 bufptr += parent_entry_len + 1;
418                 /*
419                  * The clone is shallow if nr_parent < 0, and we must
420                  * not traverse its real parents even when we unhide them.
421                  */
422                 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
423                         continue;
424                 new_parent = lookup_commit(r, &parent);
425                 if (new_parent)
426                         pptr = &commit_list_insert(new_parent, pptr)->next;
427         }
428         if (graft) {
429                 int i;
430                 struct commit *new_parent;
431                 for (i = 0; i < graft->nr_parent; i++) {
432                         new_parent = lookup_commit(r,
433                                                    &graft->parent[i]);
434                         if (!new_parent)
435                                 continue;
436                         pptr = &commit_list_insert(new_parent, pptr)->next;
437                 }
438         }
439         item->date = parse_commit_date(bufptr, tail);
440
441         if (check_graph)
442                 load_commit_graph_info(the_repository, item);
443
444         return 0;
445 }
446
447 int parse_commit_internal(struct commit *item, int quiet_on_missing, int use_commit_graph)
448 {
449         enum object_type type;
450         void *buffer;
451         unsigned long size;
452         int ret;
453
454         if (!item)
455                 return -1;
456         if (item->object.parsed)
457                 return 0;
458         if (use_commit_graph && parse_commit_in_graph(the_repository, item))
459                 return 0;
460         buffer = read_object_file(&item->object.oid, &type, &size);
461         if (!buffer)
462                 return quiet_on_missing ? -1 :
463                         error("Could not read %s",
464                              oid_to_hex(&item->object.oid));
465         if (type != OBJ_COMMIT) {
466                 free(buffer);
467                 return error("Object %s not a commit",
468                              oid_to_hex(&item->object.oid));
469         }
470
471         ret = parse_commit_buffer(the_repository, item, buffer, size, 0);
472         if (save_commit_buffer && !ret) {
473                 set_commit_buffer(the_repository, item, buffer, size);
474                 return 0;
475         }
476         free(buffer);
477         return ret;
478 }
479
480 int parse_commit_gently(struct commit *item, int quiet_on_missing)
481 {
482         return parse_commit_internal(item, quiet_on_missing, 1);
483 }
484
485 void parse_commit_or_die(struct commit *item)
486 {
487         if (parse_commit(item))
488                 die("unable to parse commit %s",
489                     item ? oid_to_hex(&item->object.oid) : "(null)");
490 }
491
492 int find_commit_subject(const char *commit_buffer, const char **subject)
493 {
494         const char *eol;
495         const char *p = commit_buffer;
496
497         while (*p && (*p != '\n' || p[1] != '\n'))
498                 p++;
499         if (*p) {
500                 p = skip_blank_lines(p + 2);
501                 eol = strchrnul(p, '\n');
502         } else
503                 eol = p;
504
505         *subject = p;
506
507         return eol - p;
508 }
509
510 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
511 {
512         struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
513         new_list->item = item;
514         new_list->next = *list_p;
515         *list_p = new_list;
516         return new_list;
517 }
518
519 unsigned commit_list_count(const struct commit_list *l)
520 {
521         unsigned c = 0;
522         for (; l; l = l->next )
523                 c++;
524         return c;
525 }
526
527 struct commit_list *copy_commit_list(struct commit_list *list)
528 {
529         struct commit_list *head = NULL;
530         struct commit_list **pp = &head;
531         while (list) {
532                 pp = commit_list_append(list->item, pp);
533                 list = list->next;
534         }
535         return head;
536 }
537
538 void free_commit_list(struct commit_list *list)
539 {
540         while (list)
541                 pop_commit(&list);
542 }
543
544 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
545 {
546         struct commit_list **pp = list;
547         struct commit_list *p;
548         while ((p = *pp) != NULL) {
549                 if (p->item->date < item->date) {
550                         break;
551                 }
552                 pp = &p->next;
553         }
554         return commit_list_insert(item, pp);
555 }
556
557 static int commit_list_compare_by_date(const void *a, const void *b)
558 {
559         timestamp_t a_date = ((const struct commit_list *)a)->item->date;
560         timestamp_t b_date = ((const struct commit_list *)b)->item->date;
561         if (a_date < b_date)
562                 return 1;
563         if (a_date > b_date)
564                 return -1;
565         return 0;
566 }
567
568 static void *commit_list_get_next(const void *a)
569 {
570         return ((const struct commit_list *)a)->next;
571 }
572
573 static void commit_list_set_next(void *a, void *next)
574 {
575         ((struct commit_list *)a)->next = next;
576 }
577
578 void commit_list_sort_by_date(struct commit_list **list)
579 {
580         *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
581                                 commit_list_compare_by_date);
582 }
583
584 struct commit *pop_most_recent_commit(struct commit_list **list,
585                                       unsigned int mark)
586 {
587         struct commit *ret = pop_commit(list);
588         struct commit_list *parents = ret->parents;
589
590         while (parents) {
591                 struct commit *commit = parents->item;
592                 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
593                         commit->object.flags |= mark;
594                         commit_list_insert_by_date(commit, list);
595                 }
596                 parents = parents->next;
597         }
598         return ret;
599 }
600
601 static void clear_commit_marks_1(struct commit_list **plist,
602                                  struct commit *commit, unsigned int mark)
603 {
604         while (commit) {
605                 struct commit_list *parents;
606
607                 if (!(mark & commit->object.flags))
608                         return;
609
610                 commit->object.flags &= ~mark;
611
612                 parents = commit->parents;
613                 if (!parents)
614                         return;
615
616                 while ((parents = parents->next))
617                         commit_list_insert(parents->item, plist);
618
619                 commit = commit->parents->item;
620         }
621 }
622
623 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
624 {
625         struct commit_list *list = NULL;
626
627         while (nr--) {
628                 clear_commit_marks_1(&list, *commit, mark);
629                 commit++;
630         }
631         while (list)
632                 clear_commit_marks_1(&list, pop_commit(&list), mark);
633 }
634
635 void clear_commit_marks(struct commit *commit, unsigned int mark)
636 {
637         clear_commit_marks_many(1, &commit, mark);
638 }
639
640 struct commit *pop_commit(struct commit_list **stack)
641 {
642         struct commit_list *top = *stack;
643         struct commit *item = top ? top->item : NULL;
644
645         if (top) {
646                 *stack = top->next;
647                 free(top);
648         }
649         return item;
650 }
651
652 /*
653  * Topological sort support
654  */
655
656 /* count number of children that have not been emitted */
657 define_commit_slab(indegree_slab, int);
658
659 /* record author-date for each commit object */
660 define_commit_slab(author_date_slab, unsigned long);
661
662 static void record_author_date(struct author_date_slab *author_date,
663                                struct commit *commit)
664 {
665         const char *buffer = get_commit_buffer(commit, NULL);
666         struct ident_split ident;
667         const char *ident_line;
668         size_t ident_len;
669         char *date_end;
670         timestamp_t date;
671
672         ident_line = find_commit_header(buffer, "author", &ident_len);
673         if (!ident_line)
674                 goto fail_exit; /* no author line */
675         if (split_ident_line(&ident, ident_line, ident_len) ||
676             !ident.date_begin || !ident.date_end)
677                 goto fail_exit; /* malformed "author" line */
678
679         date = parse_timestamp(ident.date_begin, &date_end, 10);
680         if (date_end != ident.date_end)
681                 goto fail_exit; /* malformed date */
682         *(author_date_slab_at(author_date, commit)) = date;
683
684 fail_exit:
685         unuse_commit_buffer(commit, buffer);
686 }
687
688 static int compare_commits_by_author_date(const void *a_, const void *b_,
689                                           void *cb_data)
690 {
691         const struct commit *a = a_, *b = b_;
692         struct author_date_slab *author_date = cb_data;
693         timestamp_t a_date = *(author_date_slab_at(author_date, a));
694         timestamp_t b_date = *(author_date_slab_at(author_date, b));
695
696         /* newer commits with larger date first */
697         if (a_date < b_date)
698                 return 1;
699         else if (a_date > b_date)
700                 return -1;
701         return 0;
702 }
703
704 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
705 {
706         const struct commit *a = a_, *b = b_;
707
708         /* newer commits first */
709         if (a->generation < b->generation)
710                 return 1;
711         else if (a->generation > b->generation)
712                 return -1;
713
714         /* use date as a heuristic when generations are equal */
715         if (a->date < b->date)
716                 return 1;
717         else if (a->date > b->date)
718                 return -1;
719         return 0;
720 }
721
722 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
723 {
724         const struct commit *a = a_, *b = b_;
725         /* newer commits with larger date first */
726         if (a->date < b->date)
727                 return 1;
728         else if (a->date > b->date)
729                 return -1;
730         return 0;
731 }
732
733 /*
734  * Performs an in-place topological sort on the list supplied.
735  */
736 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
737 {
738         struct commit_list *next, *orig = *list;
739         struct commit_list **pptr;
740         struct indegree_slab indegree;
741         struct prio_queue queue;
742         struct commit *commit;
743         struct author_date_slab author_date;
744
745         if (!orig)
746                 return;
747         *list = NULL;
748
749         init_indegree_slab(&indegree);
750         memset(&queue, '\0', sizeof(queue));
751
752         switch (sort_order) {
753         default: /* REV_SORT_IN_GRAPH_ORDER */
754                 queue.compare = NULL;
755                 break;
756         case REV_SORT_BY_COMMIT_DATE:
757                 queue.compare = compare_commits_by_commit_date;
758                 break;
759         case REV_SORT_BY_AUTHOR_DATE:
760                 init_author_date_slab(&author_date);
761                 queue.compare = compare_commits_by_author_date;
762                 queue.cb_data = &author_date;
763                 break;
764         }
765
766         /* Mark them and clear the indegree */
767         for (next = orig; next; next = next->next) {
768                 struct commit *commit = next->item;
769                 *(indegree_slab_at(&indegree, commit)) = 1;
770                 /* also record the author dates, if needed */
771                 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
772                         record_author_date(&author_date, commit);
773         }
774
775         /* update the indegree */
776         for (next = orig; next; next = next->next) {
777                 struct commit_list *parents = next->item->parents;
778                 while (parents) {
779                         struct commit *parent = parents->item;
780                         int *pi = indegree_slab_at(&indegree, parent);
781
782                         if (*pi)
783                                 (*pi)++;
784                         parents = parents->next;
785                 }
786         }
787
788         /*
789          * find the tips
790          *
791          * tips are nodes not reachable from any other node in the list
792          *
793          * the tips serve as a starting set for the work queue.
794          */
795         for (next = orig; next; next = next->next) {
796                 struct commit *commit = next->item;
797
798                 if (*(indegree_slab_at(&indegree, commit)) == 1)
799                         prio_queue_put(&queue, commit);
800         }
801
802         /*
803          * This is unfortunate; the initial tips need to be shown
804          * in the order given from the revision traversal machinery.
805          */
806         if (sort_order == REV_SORT_IN_GRAPH_ORDER)
807                 prio_queue_reverse(&queue);
808
809         /* We no longer need the commit list */
810         free_commit_list(orig);
811
812         pptr = list;
813         *list = NULL;
814         while ((commit = prio_queue_get(&queue)) != NULL) {
815                 struct commit_list *parents;
816
817                 for (parents = commit->parents; parents ; parents = parents->next) {
818                         struct commit *parent = parents->item;
819                         int *pi = indegree_slab_at(&indegree, parent);
820
821                         if (!*pi)
822                                 continue;
823
824                         /*
825                          * parents are only enqueued for emission
826                          * when all their children have been emitted thereby
827                          * guaranteeing topological order.
828                          */
829                         if (--(*pi) == 1)
830                                 prio_queue_put(&queue, parent);
831                 }
832                 /*
833                  * all children of commit have already been
834                  * emitted. we can emit it now.
835                  */
836                 *(indegree_slab_at(&indegree, commit)) = 0;
837
838                 pptr = &commit_list_insert(commit, pptr)->next;
839         }
840
841         clear_indegree_slab(&indegree);
842         clear_prio_queue(&queue);
843         if (sort_order == REV_SORT_BY_AUTHOR_DATE)
844                 clear_author_date_slab(&author_date);
845 }
846
847 /* merge-base stuff */
848
849 /* Remember to update object flag allocation in object.h */
850 #define PARENT1         (1u<<16)
851 #define PARENT2         (1u<<17)
852 #define STALE           (1u<<18)
853 #define RESULT          (1u<<19)
854
855 static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
856
857 static int queue_has_nonstale(struct prio_queue *queue)
858 {
859         int i;
860         for (i = 0; i < queue->nr; i++) {
861                 struct commit *commit = queue->array[i].data;
862                 if (!(commit->object.flags & STALE))
863                         return 1;
864         }
865         return 0;
866 }
867
868 /* all input commits in one and twos[] must have been parsed! */
869 static struct commit_list *paint_down_to_common(struct commit *one, int n,
870                                                 struct commit **twos,
871                                                 int min_generation)
872 {
873         struct prio_queue queue = { compare_commits_by_gen_then_commit_date };
874         struct commit_list *result = NULL;
875         int i;
876         uint32_t last_gen = GENERATION_NUMBER_INFINITY;
877
878         one->object.flags |= PARENT1;
879         if (!n) {
880                 commit_list_append(one, &result);
881                 return result;
882         }
883         prio_queue_put(&queue, one);
884
885         for (i = 0; i < n; i++) {
886                 twos[i]->object.flags |= PARENT2;
887                 prio_queue_put(&queue, twos[i]);
888         }
889
890         while (queue_has_nonstale(&queue)) {
891                 struct commit *commit = prio_queue_get(&queue);
892                 struct commit_list *parents;
893                 int flags;
894
895                 if (commit->generation > last_gen)
896                         BUG("bad generation skip %8x > %8x at %s",
897                             commit->generation, last_gen,
898                             oid_to_hex(&commit->object.oid));
899                 last_gen = commit->generation;
900
901                 if (commit->generation < min_generation)
902                         break;
903
904                 flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
905                 if (flags == (PARENT1 | PARENT2)) {
906                         if (!(commit->object.flags & RESULT)) {
907                                 commit->object.flags |= RESULT;
908                                 commit_list_insert_by_date(commit, &result);
909                         }
910                         /* Mark parents of a found merge stale */
911                         flags |= STALE;
912                 }
913                 parents = commit->parents;
914                 while (parents) {
915                         struct commit *p = parents->item;
916                         parents = parents->next;
917                         if ((p->object.flags & flags) == flags)
918                                 continue;
919                         if (parse_commit(p))
920                                 return NULL;
921                         p->object.flags |= flags;
922                         prio_queue_put(&queue, p);
923                 }
924         }
925
926         clear_prio_queue(&queue);
927         return result;
928 }
929
930 static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
931 {
932         struct commit_list *list = NULL;
933         struct commit_list *result = NULL;
934         int i;
935
936         for (i = 0; i < n; i++) {
937                 if (one == twos[i])
938                         /*
939                          * We do not mark this even with RESULT so we do not
940                          * have to clean it up.
941                          */
942                         return commit_list_insert(one, &result);
943         }
944
945         if (parse_commit(one))
946                 return NULL;
947         for (i = 0; i < n; i++) {
948                 if (parse_commit(twos[i]))
949                         return NULL;
950         }
951
952         list = paint_down_to_common(one, n, twos, 0);
953
954         while (list) {
955                 struct commit *commit = pop_commit(&list);
956                 if (!(commit->object.flags & STALE))
957                         commit_list_insert_by_date(commit, &result);
958         }
959         return result;
960 }
961
962 struct rev_collect {
963         struct commit **commit;
964         int nr;
965         int alloc;
966         unsigned int initial : 1;
967 };
968
969 static void add_one_commit(struct object_id *oid, struct rev_collect *revs)
970 {
971         struct commit *commit;
972
973         if (is_null_oid(oid))
974                 return;
975
976         commit = lookup_commit(the_repository, oid);
977         if (!commit ||
978             (commit->object.flags & TMP_MARK) ||
979             parse_commit(commit))
980                 return;
981
982         ALLOC_GROW(revs->commit, revs->nr + 1, revs->alloc);
983         revs->commit[revs->nr++] = commit;
984         commit->object.flags |= TMP_MARK;
985 }
986
987 static int collect_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
988                                   const char *ident, timestamp_t timestamp,
989                                   int tz, const char *message, void *cbdata)
990 {
991         struct rev_collect *revs = cbdata;
992
993         if (revs->initial) {
994                 revs->initial = 0;
995                 add_one_commit(ooid, revs);
996         }
997         add_one_commit(noid, revs);
998         return 0;
999 }
1000
1001 struct commit *get_fork_point(const char *refname, struct commit *commit)
1002 {
1003         struct object_id oid;
1004         struct rev_collect revs;
1005         struct commit_list *bases;
1006         int i;
1007         struct commit *ret = NULL;
1008
1009         memset(&revs, 0, sizeof(revs));
1010         revs.initial = 1;
1011         for_each_reflog_ent(refname, collect_one_reflog_ent, &revs);
1012
1013         if (!revs.nr && !get_oid(refname, &oid))
1014                 add_one_commit(&oid, &revs);
1015
1016         for (i = 0; i < revs.nr; i++)
1017                 revs.commit[i]->object.flags &= ~TMP_MARK;
1018
1019         bases = get_merge_bases_many(commit, revs.nr, revs.commit);
1020
1021         /*
1022          * There should be one and only one merge base, when we found
1023          * a common ancestor among reflog entries.
1024          */
1025         if (!bases || bases->next)
1026                 goto cleanup_return;
1027
1028         /* And the found one must be one of the reflog entries */
1029         for (i = 0; i < revs.nr; i++)
1030                 if (&bases->item->object == &revs.commit[i]->object)
1031                         break; /* found */
1032         if (revs.nr <= i)
1033                 goto cleanup_return;
1034
1035         ret = bases->item;
1036
1037 cleanup_return:
1038         free_commit_list(bases);
1039         return ret;
1040 }
1041
1042 struct commit_list *get_octopus_merge_bases(struct commit_list *in)
1043 {
1044         struct commit_list *i, *j, *k, *ret = NULL;
1045
1046         if (!in)
1047                 return ret;
1048
1049         commit_list_insert(in->item, &ret);
1050
1051         for (i = in->next; i; i = i->next) {
1052                 struct commit_list *new_commits = NULL, *end = NULL;
1053
1054                 for (j = ret; j; j = j->next) {
1055                         struct commit_list *bases;
1056                         bases = get_merge_bases(i->item, j->item);
1057                         if (!new_commits)
1058                                 new_commits = bases;
1059                         else
1060                                 end->next = bases;
1061                         for (k = bases; k; k = k->next)
1062                                 end = k;
1063                 }
1064                 ret = new_commits;
1065         }
1066         return ret;
1067 }
1068
1069 static int remove_redundant(struct commit **array, int cnt)
1070 {
1071         /*
1072          * Some commit in the array may be an ancestor of
1073          * another commit.  Move such commit to the end of
1074          * the array, and return the number of commits that
1075          * are independent from each other.
1076          */
1077         struct commit **work;
1078         unsigned char *redundant;
1079         int *filled_index;
1080         int i, j, filled;
1081
1082         work = xcalloc(cnt, sizeof(*work));
1083         redundant = xcalloc(cnt, 1);
1084         ALLOC_ARRAY(filled_index, cnt - 1);
1085
1086         for (i = 0; i < cnt; i++)
1087                 parse_commit(array[i]);
1088         for (i = 0; i < cnt; i++) {
1089                 struct commit_list *common;
1090                 uint32_t min_generation = array[i]->generation;
1091
1092                 if (redundant[i])
1093                         continue;
1094                 for (j = filled = 0; j < cnt; j++) {
1095                         if (i == j || redundant[j])
1096                                 continue;
1097                         filled_index[filled] = j;
1098                         work[filled++] = array[j];
1099
1100                         if (array[j]->generation < min_generation)
1101                                 min_generation = array[j]->generation;
1102                 }
1103                 common = paint_down_to_common(array[i], filled, work,
1104                                               min_generation);
1105                 if (array[i]->object.flags & PARENT2)
1106                         redundant[i] = 1;
1107                 for (j = 0; j < filled; j++)
1108                         if (work[j]->object.flags & PARENT1)
1109                                 redundant[filled_index[j]] = 1;
1110                 clear_commit_marks(array[i], all_flags);
1111                 clear_commit_marks_many(filled, work, all_flags);
1112                 free_commit_list(common);
1113         }
1114
1115         /* Now collect the result */
1116         COPY_ARRAY(work, array, cnt);
1117         for (i = filled = 0; i < cnt; i++)
1118                 if (!redundant[i])
1119                         array[filled++] = work[i];
1120         for (j = filled, i = 0; i < cnt; i++)
1121                 if (redundant[i])
1122                         array[j++] = work[i];
1123         free(work);
1124         free(redundant);
1125         free(filled_index);
1126         return filled;
1127 }
1128
1129 static struct commit_list *get_merge_bases_many_0(struct commit *one,
1130                                                   int n,
1131                                                   struct commit **twos,
1132                                                   int cleanup)
1133 {
1134         struct commit_list *list;
1135         struct commit **rslt;
1136         struct commit_list *result;
1137         int cnt, i;
1138
1139         result = merge_bases_many(one, n, twos);
1140         for (i = 0; i < n; i++) {
1141                 if (one == twos[i])
1142                         return result;
1143         }
1144         if (!result || !result->next) {
1145                 if (cleanup) {
1146                         clear_commit_marks(one, all_flags);
1147                         clear_commit_marks_many(n, twos, all_flags);
1148                 }
1149                 return result;
1150         }
1151
1152         /* There are more than one */
1153         cnt = commit_list_count(result);
1154         rslt = xcalloc(cnt, sizeof(*rslt));
1155         for (list = result, i = 0; list; list = list->next)
1156                 rslt[i++] = list->item;
1157         free_commit_list(result);
1158
1159         clear_commit_marks(one, all_flags);
1160         clear_commit_marks_many(n, twos, all_flags);
1161
1162         cnt = remove_redundant(rslt, cnt);
1163         result = NULL;
1164         for (i = 0; i < cnt; i++)
1165                 commit_list_insert_by_date(rslt[i], &result);
1166         free(rslt);
1167         return result;
1168 }
1169
1170 struct commit_list *get_merge_bases_many(struct commit *one,
1171                                          int n,
1172                                          struct commit **twos)
1173 {
1174         return get_merge_bases_many_0(one, n, twos, 1);
1175 }
1176
1177 struct commit_list *get_merge_bases_many_dirty(struct commit *one,
1178                                                int n,
1179                                                struct commit **twos)
1180 {
1181         return get_merge_bases_many_0(one, n, twos, 0);
1182 }
1183
1184 struct commit_list *get_merge_bases(struct commit *one, struct commit *two)
1185 {
1186         return get_merge_bases_many_0(one, 1, &two, 1);
1187 }
1188
1189 /*
1190  * Is "commit" a descendant of one of the elements on the "with_commit" list?
1191  */
1192 int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
1193 {
1194         if (!with_commit)
1195                 return 1;
1196         while (with_commit) {
1197                 struct commit *other;
1198
1199                 other = with_commit->item;
1200                 with_commit = with_commit->next;
1201                 if (in_merge_bases(other, commit))
1202                         return 1;
1203         }
1204         return 0;
1205 }
1206
1207 /*
1208  * Is "commit" an ancestor of one of the "references"?
1209  */
1210 int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
1211 {
1212         struct commit_list *bases;
1213         int ret = 0, i;
1214         uint32_t min_generation = GENERATION_NUMBER_INFINITY;
1215
1216         if (parse_commit(commit))
1217                 return ret;
1218         for (i = 0; i < nr_reference; i++) {
1219                 if (parse_commit(reference[i]))
1220                         return ret;
1221                 if (reference[i]->generation < min_generation)
1222                         min_generation = reference[i]->generation;
1223         }
1224
1225         if (commit->generation > min_generation)
1226                 return ret;
1227
1228         bases = paint_down_to_common(commit, nr_reference, reference, commit->generation);
1229         if (commit->object.flags & PARENT2)
1230                 ret = 1;
1231         clear_commit_marks(commit, all_flags);
1232         clear_commit_marks_many(nr_reference, reference, all_flags);
1233         free_commit_list(bases);
1234         return ret;
1235 }
1236
1237 /*
1238  * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1239  */
1240 int in_merge_bases(struct commit *commit, struct commit *reference)
1241 {
1242         return in_merge_bases_many(commit, 1, &reference);
1243 }
1244
1245 struct commit_list *reduce_heads(struct commit_list *heads)
1246 {
1247         struct commit_list *p;
1248         struct commit_list *result = NULL, **tail = &result;
1249         struct commit **array;
1250         int num_head, i;
1251
1252         if (!heads)
1253                 return NULL;
1254
1255         /* Uniquify */
1256         for (p = heads; p; p = p->next)
1257                 p->item->object.flags &= ~STALE;
1258         for (p = heads, num_head = 0; p; p = p->next) {
1259                 if (p->item->object.flags & STALE)
1260                         continue;
1261                 p->item->object.flags |= STALE;
1262                 num_head++;
1263         }
1264         array = xcalloc(num_head, sizeof(*array));
1265         for (p = heads, i = 0; p; p = p->next) {
1266                 if (p->item->object.flags & STALE) {
1267                         array[i++] = p->item;
1268                         p->item->object.flags &= ~STALE;
1269                 }
1270         }
1271         num_head = remove_redundant(array, num_head);
1272         for (i = 0; i < num_head; i++)
1273                 tail = &commit_list_insert(array[i], tail)->next;
1274         free(array);
1275         return result;
1276 }
1277
1278 void reduce_heads_replace(struct commit_list **heads)
1279 {
1280         struct commit_list *result = reduce_heads(*heads);
1281         free_commit_list(*heads);
1282         *heads = result;
1283 }
1284
1285 static const char gpg_sig_header[] = "gpgsig";
1286 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1287
1288 static int do_sign_commit(struct strbuf *buf, const char *keyid)
1289 {
1290         struct strbuf sig = STRBUF_INIT;
1291         int inspos, copypos;
1292         const char *eoh;
1293
1294         /* find the end of the header */
1295         eoh = strstr(buf->buf, "\n\n");
1296         if (!eoh)
1297                 inspos = buf->len;
1298         else
1299                 inspos = eoh - buf->buf + 1;
1300
1301         if (!keyid || !*keyid)
1302                 keyid = get_signing_key();
1303         if (sign_buffer(buf, &sig, keyid)) {
1304                 strbuf_release(&sig);
1305                 return -1;
1306         }
1307
1308         for (copypos = 0; sig.buf[copypos]; ) {
1309                 const char *bol = sig.buf + copypos;
1310                 const char *eol = strchrnul(bol, '\n');
1311                 int len = (eol - bol) + !!*eol;
1312
1313                 if (!copypos) {
1314                         strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1315                         inspos += gpg_sig_header_len;
1316                 }
1317                 strbuf_insert(buf, inspos++, " ", 1);
1318                 strbuf_insert(buf, inspos, bol, len);
1319                 inspos += len;
1320                 copypos += len;
1321         }
1322         strbuf_release(&sig);
1323         return 0;
1324 }
1325
1326 int parse_signed_commit(const struct commit *commit,
1327                         struct strbuf *payload, struct strbuf *signature)
1328 {
1329
1330         unsigned long size;
1331         const char *buffer = get_commit_buffer(commit, &size);
1332         int in_signature, saw_signature = -1;
1333         const char *line, *tail;
1334
1335         line = buffer;
1336         tail = buffer + size;
1337         in_signature = 0;
1338         saw_signature = 0;
1339         while (line < tail) {
1340                 const char *sig = NULL;
1341                 const char *next = memchr(line, '\n', tail - line);
1342
1343                 next = next ? next + 1 : tail;
1344                 if (in_signature && line[0] == ' ')
1345                         sig = line + 1;
1346                 else if (starts_with(line, gpg_sig_header) &&
1347                          line[gpg_sig_header_len] == ' ')
1348                         sig = line + gpg_sig_header_len + 1;
1349                 if (sig) {
1350                         strbuf_add(signature, sig, next - sig);
1351                         saw_signature = 1;
1352                         in_signature = 1;
1353                 } else {
1354                         if (*line == '\n')
1355                                 /* dump the whole remainder of the buffer */
1356                                 next = tail;
1357                         strbuf_add(payload, line, next - line);
1358                         in_signature = 0;
1359                 }
1360                 line = next;
1361         }
1362         unuse_commit_buffer(commit, buffer);
1363         return saw_signature;
1364 }
1365
1366 int remove_signature(struct strbuf *buf)
1367 {
1368         const char *line = buf->buf;
1369         const char *tail = buf->buf + buf->len;
1370         int in_signature = 0;
1371         const char *sig_start = NULL;
1372         const char *sig_end = NULL;
1373
1374         while (line < tail) {
1375                 const char *next = memchr(line, '\n', tail - line);
1376                 next = next ? next + 1 : tail;
1377
1378                 if (in_signature && line[0] == ' ')
1379                         sig_end = next;
1380                 else if (starts_with(line, gpg_sig_header) &&
1381                          line[gpg_sig_header_len] == ' ') {
1382                         sig_start = line;
1383                         sig_end = next;
1384                         in_signature = 1;
1385                 } else {
1386                         if (*line == '\n')
1387                                 /* dump the whole remainder of the buffer */
1388                                 next = tail;
1389                         in_signature = 0;
1390                 }
1391                 line = next;
1392         }
1393
1394         if (sig_start)
1395                 strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1396
1397         return sig_start != NULL;
1398 }
1399
1400 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1401 {
1402         struct merge_remote_desc *desc;
1403         struct commit_extra_header *mergetag;
1404         char *buf;
1405         unsigned long size, len;
1406         enum object_type type;
1407
1408         desc = merge_remote_util(parent);
1409         if (!desc || !desc->obj)
1410                 return;
1411         buf = read_object_file(&desc->obj->oid, &type, &size);
1412         if (!buf || type != OBJ_TAG)
1413                 goto free_return;
1414         len = parse_signature(buf, size);
1415         if (size == len)
1416                 goto free_return;
1417         /*
1418          * We could verify this signature and either omit the tag when
1419          * it does not validate, but the integrator may not have the
1420          * public key of the signer of the tag he is merging, while a
1421          * later auditor may have it while auditing, so let's not run
1422          * verify-signed-buffer here for now...
1423          *
1424          * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1425          *      warn("warning: signed tag unverified.");
1426          */
1427         mergetag = xcalloc(1, sizeof(*mergetag));
1428         mergetag->key = xstrdup("mergetag");
1429         mergetag->value = buf;
1430         mergetag->len = size;
1431
1432         **tail = mergetag;
1433         *tail = &mergetag->next;
1434         return;
1435
1436 free_return:
1437         free(buf);
1438 }
1439
1440 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1441 {
1442         struct strbuf payload = STRBUF_INIT;
1443         struct strbuf signature = STRBUF_INIT;
1444         int ret = 1;
1445
1446         sigc->result = 'N';
1447
1448         if (parse_signed_commit(commit, &payload, &signature) <= 0)
1449                 goto out;
1450         ret = check_signature(payload.buf, payload.len, signature.buf,
1451                 signature.len, sigc);
1452
1453  out:
1454         strbuf_release(&payload);
1455         strbuf_release(&signature);
1456
1457         return ret;
1458 }
1459
1460
1461
1462 void append_merge_tag_headers(struct commit_list *parents,
1463                               struct commit_extra_header ***tail)
1464 {
1465         while (parents) {
1466                 struct commit *parent = parents->item;
1467                 handle_signed_tag(parent, tail);
1468                 parents = parents->next;
1469         }
1470 }
1471
1472 static void add_extra_header(struct strbuf *buffer,
1473                              struct commit_extra_header *extra)
1474 {
1475         strbuf_addstr(buffer, extra->key);
1476         if (extra->len)
1477                 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1478         else
1479                 strbuf_addch(buffer, '\n');
1480 }
1481
1482 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1483                                                       const char **exclude)
1484 {
1485         struct commit_extra_header *extra = NULL;
1486         unsigned long size;
1487         const char *buffer = get_commit_buffer(commit, &size);
1488         extra = read_commit_extra_header_lines(buffer, size, exclude);
1489         unuse_commit_buffer(commit, buffer);
1490         return extra;
1491 }
1492
1493 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1494 {
1495         struct commit_extra_header *extra, *to_free;
1496         int res = 0;
1497
1498         to_free = read_commit_extra_headers(commit, NULL);
1499         for (extra = to_free; !res && extra; extra = extra->next) {
1500                 if (strcmp(extra->key, "mergetag"))
1501                         continue; /* not a merge tag */
1502                 res = fn(commit, extra, data);
1503         }
1504         free_commit_extra_headers(to_free);
1505         return res;
1506 }
1507
1508 static inline int standard_header_field(const char *field, size_t len)
1509 {
1510         return ((len == 4 && !memcmp(field, "tree", 4)) ||
1511                 (len == 6 && !memcmp(field, "parent", 6)) ||
1512                 (len == 6 && !memcmp(field, "author", 6)) ||
1513                 (len == 9 && !memcmp(field, "committer", 9)) ||
1514                 (len == 8 && !memcmp(field, "encoding", 8)));
1515 }
1516
1517 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1518 {
1519         if (!exclude)
1520                 return 0;
1521
1522         while (*exclude) {
1523                 size_t xlen = strlen(*exclude);
1524                 if (len == xlen && !memcmp(field, *exclude, xlen))
1525                         return 1;
1526                 exclude++;
1527         }
1528         return 0;
1529 }
1530
1531 static struct commit_extra_header *read_commit_extra_header_lines(
1532         const char *buffer, size_t size,
1533         const char **exclude)
1534 {
1535         struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1536         const char *line, *next, *eof, *eob;
1537         struct strbuf buf = STRBUF_INIT;
1538
1539         for (line = buffer, eob = line + size;
1540              line < eob && *line != '\n';
1541              line = next) {
1542                 next = memchr(line, '\n', eob - line);
1543                 next = next ? next + 1 : eob;
1544                 if (*line == ' ') {
1545                         /* continuation */
1546                         if (it)
1547                                 strbuf_add(&buf, line + 1, next - (line + 1));
1548                         continue;
1549                 }
1550                 if (it)
1551                         it->value = strbuf_detach(&buf, &it->len);
1552                 strbuf_reset(&buf);
1553                 it = NULL;
1554
1555                 eof = memchr(line, ' ', next - line);
1556                 if (!eof)
1557                         eof = next;
1558                 else if (standard_header_field(line, eof - line) ||
1559                          excluded_header_field(line, eof - line, exclude))
1560                         continue;
1561
1562                 it = xcalloc(1, sizeof(*it));
1563                 it->key = xmemdupz(line, eof-line);
1564                 *tail = it;
1565                 tail = &it->next;
1566                 if (eof + 1 < next)
1567                         strbuf_add(&buf, eof + 1, next - (eof + 1));
1568         }
1569         if (it)
1570                 it->value = strbuf_detach(&buf, &it->len);
1571         return extra;
1572 }
1573
1574 void free_commit_extra_headers(struct commit_extra_header *extra)
1575 {
1576         while (extra) {
1577                 struct commit_extra_header *next = extra->next;
1578                 free(extra->key);
1579                 free(extra->value);
1580                 free(extra);
1581                 extra = next;
1582         }
1583 }
1584
1585 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1586                 struct commit_list *parents, struct object_id *ret,
1587                 const char *author, const char *sign_commit)
1588 {
1589         struct commit_extra_header *extra = NULL, **tail = &extra;
1590         int result;
1591
1592         append_merge_tag_headers(parents, &tail);
1593         result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1594                                       author, sign_commit, extra);
1595         free_commit_extra_headers(extra);
1596         return result;
1597 }
1598
1599 static int find_invalid_utf8(const char *buf, int len)
1600 {
1601         int offset = 0;
1602         static const unsigned int max_codepoint[] = {
1603                 0x7f, 0x7ff, 0xffff, 0x10ffff
1604         };
1605
1606         while (len) {
1607                 unsigned char c = *buf++;
1608                 int bytes, bad_offset;
1609                 unsigned int codepoint;
1610                 unsigned int min_val, max_val;
1611
1612                 len--;
1613                 offset++;
1614
1615                 /* Simple US-ASCII? No worries. */
1616                 if (c < 0x80)
1617                         continue;
1618
1619                 bad_offset = offset-1;
1620
1621                 /*
1622                  * Count how many more high bits set: that's how
1623                  * many more bytes this sequence should have.
1624                  */
1625                 bytes = 0;
1626                 while (c & 0x40) {
1627                         c <<= 1;
1628                         bytes++;
1629                 }
1630
1631                 /*
1632                  * Must be between 1 and 3 more bytes.  Longer sequences result in
1633                  * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1634                  */
1635                 if (bytes < 1 || 3 < bytes)
1636                         return bad_offset;
1637
1638                 /* Do we *have* that many bytes? */
1639                 if (len < bytes)
1640                         return bad_offset;
1641
1642                 /*
1643                  * Place the encoded bits at the bottom of the value and compute the
1644                  * valid range.
1645                  */
1646                 codepoint = (c & 0x7f) >> bytes;
1647                 min_val = max_codepoint[bytes-1] + 1;
1648                 max_val = max_codepoint[bytes];
1649
1650                 offset += bytes;
1651                 len -= bytes;
1652
1653                 /* And verify that they are good continuation bytes */
1654                 do {
1655                         codepoint <<= 6;
1656                         codepoint |= *buf & 0x3f;
1657                         if ((*buf++ & 0xc0) != 0x80)
1658                                 return bad_offset;
1659                 } while (--bytes);
1660
1661                 /* Reject codepoints that are out of range for the sequence length. */
1662                 if (codepoint < min_val || codepoint > max_val)
1663                         return bad_offset;
1664                 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1665                 if ((codepoint & 0x1ff800) == 0xd800)
1666                         return bad_offset;
1667                 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1668                 if ((codepoint & 0xfffe) == 0xfffe)
1669                         return bad_offset;
1670                 /* So are anything in the range U+FDD0..U+FDEF. */
1671                 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1672                         return bad_offset;
1673         }
1674         return -1;
1675 }
1676
1677 /*
1678  * This verifies that the buffer is in proper utf8 format.
1679  *
1680  * If it isn't, it assumes any non-utf8 characters are Latin1,
1681  * and does the conversion.
1682  */
1683 static int verify_utf8(struct strbuf *buf)
1684 {
1685         int ok = 1;
1686         long pos = 0;
1687
1688         for (;;) {
1689                 int bad;
1690                 unsigned char c;
1691                 unsigned char replace[2];
1692
1693                 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1694                 if (bad < 0)
1695                         return ok;
1696                 pos += bad;
1697                 ok = 0;
1698                 c = buf->buf[pos];
1699                 strbuf_remove(buf, pos, 1);
1700
1701                 /* We know 'c' must be in the range 128-255 */
1702                 replace[0] = 0xc0 + (c >> 6);
1703                 replace[1] = 0x80 + (c & 0x3f);
1704                 strbuf_insert(buf, pos, replace, 2);
1705                 pos += 2;
1706         }
1707 }
1708
1709 static const char commit_utf8_warn[] =
1710 N_("Warning: commit message did not conform to UTF-8.\n"
1711    "You may want to amend it after fixing the message, or set the config\n"
1712    "variable i18n.commitencoding to the encoding your project uses.\n");
1713
1714 int commit_tree_extended(const char *msg, size_t msg_len,
1715                          const struct object_id *tree,
1716                          struct commit_list *parents, struct object_id *ret,
1717                          const char *author, const char *sign_commit,
1718                          struct commit_extra_header *extra)
1719 {
1720         int result;
1721         int encoding_is_utf8;
1722         struct strbuf buffer;
1723
1724         assert_oid_type(tree, OBJ_TREE);
1725
1726         if (memchr(msg, '\0', msg_len))
1727                 return error("a NUL byte in commit log message not allowed.");
1728
1729         /* Not having i18n.commitencoding is the same as having utf-8 */
1730         encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1731
1732         strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1733         strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1734
1735         /*
1736          * NOTE! This ordering means that the same exact tree merged with a
1737          * different order of parents will be a _different_ changeset even
1738          * if everything else stays the same.
1739          */
1740         while (parents) {
1741                 struct commit *parent = pop_commit(&parents);
1742                 strbuf_addf(&buffer, "parent %s\n",
1743                             oid_to_hex(&parent->object.oid));
1744         }
1745
1746         /* Person/date information */
1747         if (!author)
1748                 author = git_author_info(IDENT_STRICT);
1749         strbuf_addf(&buffer, "author %s\n", author);
1750         strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1751         if (!encoding_is_utf8)
1752                 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1753
1754         while (extra) {
1755                 add_extra_header(&buffer, extra);
1756                 extra = extra->next;
1757         }
1758         strbuf_addch(&buffer, '\n');
1759
1760         /* And add the comment */
1761         strbuf_add(&buffer, msg, msg_len);
1762
1763         /* And check the encoding */
1764         if (encoding_is_utf8 && !verify_utf8(&buffer))
1765                 fprintf(stderr, _(commit_utf8_warn));
1766
1767         if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1768                 result = -1;
1769                 goto out;
1770         }
1771
1772         result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1773 out:
1774         strbuf_release(&buffer);
1775         return result;
1776 }
1777
1778 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1779 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1780
1781 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1782 {
1783         return *merge_desc_slab_at(&merge_desc_slab, commit);
1784 }
1785
1786 void set_merge_remote_desc(struct commit *commit,
1787                            const char *name, struct object *obj)
1788 {
1789         struct merge_remote_desc *desc;
1790         FLEX_ALLOC_STR(desc, name, name);
1791         desc->obj = obj;
1792         *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1793 }
1794
1795 struct commit *get_merge_parent(const char *name)
1796 {
1797         struct object *obj;
1798         struct commit *commit;
1799         struct object_id oid;
1800         if (get_oid(name, &oid))
1801                 return NULL;
1802         obj = parse_object(the_repository, &oid);
1803         commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1804         if (commit && !merge_remote_util(commit))
1805                 set_merge_remote_desc(commit, name, obj);
1806         return commit;
1807 }
1808
1809 /*
1810  * Append a commit to the end of the commit_list.
1811  *
1812  * next starts by pointing to the variable that holds the head of an
1813  * empty commit_list, and is updated to point to the "next" field of
1814  * the last item on the list as new commits are appended.
1815  *
1816  * Usage example:
1817  *
1818  *     struct commit_list *list;
1819  *     struct commit_list **next = &list;
1820  *
1821  *     next = commit_list_append(c1, next);
1822  *     next = commit_list_append(c2, next);
1823  *     assert(commit_list_count(list) == 2);
1824  *     return list;
1825  */
1826 struct commit_list **commit_list_append(struct commit *commit,
1827                                         struct commit_list **next)
1828 {
1829         struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1830         new_commit->item = commit;
1831         *next = new_commit;
1832         new_commit->next = NULL;
1833         return &new_commit->next;
1834 }
1835
1836 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1837 {
1838         int key_len = strlen(key);
1839         const char *line = msg;
1840
1841         while (line) {
1842                 const char *eol = strchrnul(line, '\n');
1843
1844                 if (line == eol)
1845                         return NULL;
1846
1847                 if (eol - line > key_len &&
1848                     !strncmp(line, key, key_len) &&
1849                     line[key_len] == ' ') {
1850                         *out_len = eol - line - key_len - 1;
1851                         return line + key_len + 1;
1852                 }
1853                 line = *eol ? eol + 1 : NULL;
1854         }
1855         return NULL;
1856 }
1857
1858 /*
1859  * Inspect the given string and determine the true "end" of the log message, in
1860  * order to find where to put a new Signed-off-by: line.  Ignored are
1861  * trailing comment lines and blank lines.  To support "git commit -s
1862  * --amend" on an existing commit, we also ignore "Conflicts:".  To
1863  * support "git commit -v", we truncate at cut lines.
1864  *
1865  * Returns the number of bytes from the tail to ignore, to be fed as
1866  * the second parameter to append_signoff().
1867  */
1868 int ignore_non_trailer(const char *buf, size_t len)
1869 {
1870         int boc = 0;
1871         int bol = 0;
1872         int in_old_conflicts_block = 0;
1873         size_t cutoff = wt_status_locate_end(buf, len);
1874
1875         while (bol < cutoff) {
1876                 const char *next_line = memchr(buf + bol, '\n', len - bol);
1877
1878                 if (!next_line)
1879                         next_line = buf + len;
1880                 else
1881                         next_line++;
1882
1883                 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1884                         /* is this the first of the run of comments? */
1885                         if (!boc)
1886                                 boc = bol;
1887                         /* otherwise, it is just continuing */
1888                 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1889                         in_old_conflicts_block = 1;
1890                         if (!boc)
1891                                 boc = bol;
1892                 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1893                         ; /* a pathname in the conflicts block */
1894                 } else if (boc) {
1895                         /* the previous was not trailing comment */
1896                         boc = 0;
1897                         in_old_conflicts_block = 0;
1898                 }
1899                 bol = next_line - buf;
1900         }
1901         return boc ? len - boc : len - cutoff;
1902 }