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