Merge branch 'jt/connectivity-check-after-unshallow'
[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(const struct object_id *oid,
28                                               int quiet)
29 {
30         struct object *obj = deref_tag(parse_object(oid), NULL, 0);
31
32         if (!obj)
33                 return NULL;
34         return object_as_type(obj, OBJ_COMMIT, quiet);
35 }
36
37 struct commit *lookup_commit_reference(const struct object_id *oid)
38 {
39         return lookup_commit_reference_gently(oid, 0);
40 }
41
42 struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
43 {
44         struct commit *c = lookup_commit_reference(oid);
45         if (!c)
46                 die(_("could not parse %s"), ref_name);
47         if (oidcmp(oid, &c->object.oid)) {
48                 warning(_("%s %s is not a commit!"),
49                         ref_name, oid_to_hex(oid));
50         }
51         return c;
52 }
53
54 struct commit *lookup_commit(const struct object_id *oid)
55 {
56         struct object *obj = lookup_object(oid->hash);
57         if (!obj)
58                 return create_object(the_repository, oid->hash,
59                                      alloc_commit_node(the_repository));
60         return object_as_type(obj, OBJ_COMMIT, 0);
61 }
62
63 struct commit *lookup_commit_reference_by_name(const char *name)
64 {
65         struct object_id oid;
66         struct commit *commit;
67
68         if (get_oid_committish(name, &oid))
69                 return NULL;
70         commit = lookup_commit_reference(&oid);
71         if (parse_commit(commit))
72                 return NULL;
73         return commit;
74 }
75
76 static timestamp_t parse_commit_date(const char *buf, const char *tail)
77 {
78         const char *dateptr;
79
80         if (buf + 6 >= tail)
81                 return 0;
82         if (memcmp(buf, "author", 6))
83                 return 0;
84         while (buf < tail && *buf++ != '\n')
85                 /* nada */;
86         if (buf + 9 >= tail)
87                 return 0;
88         if (memcmp(buf, "committer", 9))
89                 return 0;
90         while (buf < tail && *buf++ != '>')
91                 /* nada */;
92         if (buf >= tail)
93                 return 0;
94         dateptr = buf;
95         while (buf < tail && *buf++ != '\n')
96                 /* nada */;
97         if (buf >= tail)
98                 return 0;
99         /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
100         return parse_timestamp(dateptr, NULL, 10);
101 }
102
103 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
104 {
105         struct commit_graft **commit_graft_table = table;
106         return commit_graft_table[index]->oid.hash;
107 }
108
109 static int commit_graft_pos(struct repository *r, const unsigned char *sha1)
110 {
111         return sha1_pos(sha1, r->parsed_objects->grafts,
112                         r->parsed_objects->grafts_nr,
113                         commit_graft_sha1_access);
114 }
115
116 int register_commit_graft(struct repository *r, struct commit_graft *graft,
117                           int ignore_dups)
118 {
119         int pos = commit_graft_pos(r, graft->oid.hash);
120
121         if (0 <= pos) {
122                 if (ignore_dups)
123                         free(graft);
124                 else {
125                         free(r->parsed_objects->grafts[pos]);
126                         r->parsed_objects->grafts[pos] = graft;
127                 }
128                 return 1;
129         }
130         pos = -pos - 1;
131         ALLOC_GROW(r->parsed_objects->grafts,
132                    r->parsed_objects->grafts_nr + 1,
133                    r->parsed_objects->grafts_alloc);
134         r->parsed_objects->grafts_nr++;
135         if (pos < r->parsed_objects->grafts_nr)
136                 memmove(r->parsed_objects->grafts + pos + 1,
137                         r->parsed_objects->grafts + pos,
138                         (r->parsed_objects->grafts_nr - pos - 1) *
139                         sizeof(*r->parsed_objects->grafts));
140         r->parsed_objects->grafts[pos] = graft;
141         return 0;
142 }
143
144 struct commit_graft *read_graft_line(struct strbuf *line)
145 {
146         /* The format is just "Commit Parent1 Parent2 ...\n" */
147         int i, phase;
148         const char *tail = NULL;
149         struct commit_graft *graft = NULL;
150         struct object_id dummy_oid, *oid;
151
152         strbuf_rtrim(line);
153         if (!line->len || line->buf[0] == '#')
154                 return NULL;
155         /*
156          * phase 0 verifies line, counts hashes in line and allocates graft
157          * phase 1 fills graft
158          */
159         for (phase = 0; phase < 2; phase++) {
160                 oid = graft ? &graft->oid : &dummy_oid;
161                 if (parse_oid_hex(line->buf, oid, &tail))
162                         goto bad_graft_data;
163                 for (i = 0; *tail != '\0'; i++) {
164                         oid = graft ? &graft->parent[i] : &dummy_oid;
165                         if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
166                                 goto bad_graft_data;
167                 }
168                 if (!graft) {
169                         graft = xmalloc(st_add(sizeof(*graft),
170                                                st_mult(sizeof(struct object_id), i)));
171                         graft->nr_parent = i;
172                 }
173         }
174         return graft;
175
176 bad_graft_data:
177         error("bad graft data: %s", line->buf);
178         assert(!graft);
179         return NULL;
180 }
181
182 static int read_graft_file(struct repository *r, const char *graft_file)
183 {
184         FILE *fp = fopen_or_warn(graft_file, "r");
185         struct strbuf buf = STRBUF_INIT;
186         if (!fp)
187                 return -1;
188         if (advice_graft_file_deprecated)
189                 advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
190                          "and will be removed in a future Git version.\n"
191                          "\n"
192                          "Please use \"git replace --convert-graft-file\"\n"
193                          "to convert the grafts into replace refs.\n"
194                          "\n"
195                          "Turn this message off by running\n"
196                          "\"git config advice.graftFileDeprecated false\""));
197         while (!strbuf_getwholeline(&buf, fp, '\n')) {
198                 /* The format is just "Commit Parent1 Parent2 ...\n" */
199                 struct commit_graft *graft = read_graft_line(&buf);
200                 if (!graft)
201                         continue;
202                 if (register_commit_graft(r, graft, 1))
203                         error("duplicate graft data: %s", buf.buf);
204         }
205         fclose(fp);
206         strbuf_release(&buf);
207         return 0;
208 }
209
210 static void prepare_commit_graft(struct repository *r)
211 {
212         char *graft_file;
213
214         if (r->parsed_objects->commit_graft_prepared)
215                 return;
216         if (!startup_info->have_repository)
217                 return;
218
219         graft_file = get_graft_file(r);
220         read_graft_file(r, graft_file);
221         /* make sure shallows are read */
222         is_repository_shallow(r);
223         r->parsed_objects->commit_graft_prepared = 1;
224 }
225
226 struct commit_graft *lookup_commit_graft(struct repository *r, const struct object_id *oid)
227 {
228         int pos;
229         prepare_commit_graft(r);
230         pos = commit_graft_pos(r, oid->hash);
231         if (pos < 0)
232                 return NULL;
233         return r->parsed_objects->grafts[pos];
234 }
235
236 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
237 {
238         int i, ret;
239         for (i = ret = 0; i < the_repository->parsed_objects->grafts_nr && !ret; i++)
240                 ret = fn(the_repository->parsed_objects->grafts[i], cb_data);
241         return ret;
242 }
243
244 int unregister_shallow(const struct object_id *oid)
245 {
246         int pos = commit_graft_pos(the_repository, oid->hash);
247         if (pos < 0)
248                 return -1;
249         if (pos + 1 < the_repository->parsed_objects->grafts_nr)
250                 MOVE_ARRAY(the_repository->parsed_objects->grafts + pos,
251                            the_repository->parsed_objects->grafts + pos + 1,
252                            the_repository->parsed_objects->grafts_nr - pos - 1);
253         the_repository->parsed_objects->grafts_nr--;
254         return 0;
255 }
256
257 struct commit_buffer {
258         void *buffer;
259         unsigned long size;
260 };
261 define_commit_slab(buffer_slab, struct commit_buffer);
262 static struct buffer_slab buffer_slab = COMMIT_SLAB_INIT(1, buffer_slab);
263
264 void set_commit_buffer(struct commit *commit, void *buffer, unsigned long size)
265 {
266         struct commit_buffer *v = buffer_slab_at(&buffer_slab, commit);
267         v->buffer = buffer;
268         v->size = size;
269 }
270
271 const void *get_cached_commit_buffer(const struct commit *commit, unsigned long *sizep)
272 {
273         struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
274         if (!v) {
275                 if (sizep)
276                         *sizep = 0;
277                 return NULL;
278         }
279         if (sizep)
280                 *sizep = v->size;
281         return v->buffer;
282 }
283
284 const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
285 {
286         const void *ret = get_cached_commit_buffer(commit, sizep);
287         if (!ret) {
288                 enum object_type type;
289                 unsigned long size;
290                 ret = read_object_file(&commit->object.oid, &type, &size);
291                 if (!ret)
292                         die("cannot read commit object %s",
293                             oid_to_hex(&commit->object.oid));
294                 if (type != OBJ_COMMIT)
295                         die("expected commit for %s, got %s",
296                             oid_to_hex(&commit->object.oid), type_name(type));
297                 if (sizep)
298                         *sizep = size;
299         }
300         return ret;
301 }
302
303 void unuse_commit_buffer(const struct commit *commit, const void *buffer)
304 {
305         struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
306         if (!(v && v->buffer == buffer))
307                 free((void *)buffer);
308 }
309
310 void free_commit_buffer(struct commit *commit)
311 {
312         struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
313         if (v) {
314                 FREE_AND_NULL(v->buffer);
315                 v->size = 0;
316         }
317 }
318
319 struct tree *get_commit_tree(const struct commit *commit)
320 {
321         if (commit->maybe_tree || !commit->object.parsed)
322                 return commit->maybe_tree;
323
324         if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
325                 BUG("commit has NULL tree, but was not loaded from commit-graph");
326
327         return get_commit_tree_in_graph(commit);
328 }
329
330 struct object_id *get_commit_tree_oid(const struct commit *commit)
331 {
332         return &get_commit_tree(commit)->object.oid;
333 }
334
335 void release_commit_memory(struct commit *c)
336 {
337         c->maybe_tree = NULL;
338         c->index = 0;
339         free_commit_buffer(c);
340         free_commit_list(c->parents);
341         /* TODO: what about commit->util? */
342
343         c->object.parsed = 0;
344 }
345
346 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
347 {
348         struct commit_buffer *v = buffer_slab_peek(&buffer_slab, commit);
349         void *ret;
350
351         if (!v) {
352                 if (sizep)
353                         *sizep = 0;
354                 return NULL;
355         }
356         ret = v->buffer;
357         if (sizep)
358                 *sizep = v->size;
359
360         v->buffer = NULL;
361         v->size = 0;
362         return ret;
363 }
364
365 int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size, int check_graph)
366 {
367         const char *tail = buffer;
368         const char *bufptr = buffer;
369         struct object_id parent;
370         struct commit_list **pptr;
371         struct commit_graft *graft;
372         const int tree_entry_len = GIT_SHA1_HEXSZ + 5;
373         const int parent_entry_len = GIT_SHA1_HEXSZ + 7;
374
375         if (item->object.parsed)
376                 return 0;
377         item->object.parsed = 1;
378         tail += size;
379         if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
380                         bufptr[tree_entry_len] != '\n')
381                 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
382         if (get_oid_hex(bufptr + 5, &parent) < 0)
383                 return error("bad tree pointer in commit %s",
384                              oid_to_hex(&item->object.oid));
385         item->maybe_tree = lookup_tree(&parent);
386         bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
387         pptr = &item->parents;
388
389         graft = lookup_commit_graft(the_repository, &item->object.oid);
390         while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
391                 struct commit *new_parent;
392
393                 if (tail <= bufptr + parent_entry_len + 1 ||
394                     get_oid_hex(bufptr + 7, &parent) ||
395                     bufptr[parent_entry_len] != '\n')
396                         return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
397                 bufptr += parent_entry_len + 1;
398                 /*
399                  * The clone is shallow if nr_parent < 0, and we must
400                  * not traverse its real parents even when we unhide them.
401                  */
402                 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
403                         continue;
404                 new_parent = lookup_commit(&parent);
405                 if (new_parent)
406                         pptr = &commit_list_insert(new_parent, pptr)->next;
407         }
408         if (graft) {
409                 int i;
410                 struct commit *new_parent;
411                 for (i = 0; i < graft->nr_parent; i++) {
412                         new_parent = lookup_commit(&graft->parent[i]);
413                         if (!new_parent)
414                                 continue;
415                         pptr = &commit_list_insert(new_parent, pptr)->next;
416                 }
417         }
418         item->date = parse_commit_date(bufptr, tail);
419
420         if (check_graph)
421                 load_commit_graph_info(item);
422
423         return 0;
424 }
425
426 int parse_commit_gently(struct commit *item, int quiet_on_missing)
427 {
428         enum object_type type;
429         void *buffer;
430         unsigned long size;
431         int ret;
432
433         if (!item)
434                 return -1;
435         if (item->object.parsed)
436                 return 0;
437         if (parse_commit_in_graph(item))
438                 return 0;
439         buffer = read_object_file(&item->object.oid, &type, &size);
440         if (!buffer)
441                 return quiet_on_missing ? -1 :
442                         error("Could not read %s",
443                              oid_to_hex(&item->object.oid));
444         if (type != OBJ_COMMIT) {
445                 free(buffer);
446                 return error("Object %s not a commit",
447                              oid_to_hex(&item->object.oid));
448         }
449         ret = parse_commit_buffer(item, buffer, size, 0);
450         if (save_commit_buffer && !ret) {
451                 set_commit_buffer(item, buffer, size);
452                 return 0;
453         }
454         free(buffer);
455         return ret;
456 }
457
458 void parse_commit_or_die(struct commit *item)
459 {
460         if (parse_commit(item))
461                 die("unable to parse commit %s",
462                     item ? oid_to_hex(&item->object.oid) : "(null)");
463 }
464
465 int find_commit_subject(const char *commit_buffer, const char **subject)
466 {
467         const char *eol;
468         const char *p = commit_buffer;
469
470         while (*p && (*p != '\n' || p[1] != '\n'))
471                 p++;
472         if (*p) {
473                 p = skip_blank_lines(p + 2);
474                 eol = strchrnul(p, '\n');
475         } else
476                 eol = p;
477
478         *subject = p;
479
480         return eol - p;
481 }
482
483 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
484 {
485         struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
486         new_list->item = item;
487         new_list->next = *list_p;
488         *list_p = new_list;
489         return new_list;
490 }
491
492 unsigned commit_list_count(const struct commit_list *l)
493 {
494         unsigned c = 0;
495         for (; l; l = l->next )
496                 c++;
497         return c;
498 }
499
500 struct commit_list *copy_commit_list(struct commit_list *list)
501 {
502         struct commit_list *head = NULL;
503         struct commit_list **pp = &head;
504         while (list) {
505                 pp = commit_list_append(list->item, pp);
506                 list = list->next;
507         }
508         return head;
509 }
510
511 void free_commit_list(struct commit_list *list)
512 {
513         while (list)
514                 pop_commit(&list);
515 }
516
517 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
518 {
519         struct commit_list **pp = list;
520         struct commit_list *p;
521         while ((p = *pp) != NULL) {
522                 if (p->item->date < item->date) {
523                         break;
524                 }
525                 pp = &p->next;
526         }
527         return commit_list_insert(item, pp);
528 }
529
530 static int commit_list_compare_by_date(const void *a, const void *b)
531 {
532         timestamp_t a_date = ((const struct commit_list *)a)->item->date;
533         timestamp_t b_date = ((const struct commit_list *)b)->item->date;
534         if (a_date < b_date)
535                 return 1;
536         if (a_date > b_date)
537                 return -1;
538         return 0;
539 }
540
541 static void *commit_list_get_next(const void *a)
542 {
543         return ((const struct commit_list *)a)->next;
544 }
545
546 static void commit_list_set_next(void *a, void *next)
547 {
548         ((struct commit_list *)a)->next = next;
549 }
550
551 void commit_list_sort_by_date(struct commit_list **list)
552 {
553         *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
554                                 commit_list_compare_by_date);
555 }
556
557 struct commit *pop_most_recent_commit(struct commit_list **list,
558                                       unsigned int mark)
559 {
560         struct commit *ret = pop_commit(list);
561         struct commit_list *parents = ret->parents;
562
563         while (parents) {
564                 struct commit *commit = parents->item;
565                 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
566                         commit->object.flags |= mark;
567                         commit_list_insert_by_date(commit, list);
568                 }
569                 parents = parents->next;
570         }
571         return ret;
572 }
573
574 static void clear_commit_marks_1(struct commit_list **plist,
575                                  struct commit *commit, unsigned int mark)
576 {
577         while (commit) {
578                 struct commit_list *parents;
579
580                 if (!(mark & commit->object.flags))
581                         return;
582
583                 commit->object.flags &= ~mark;
584
585                 parents = commit->parents;
586                 if (!parents)
587                         return;
588
589                 while ((parents = parents->next))
590                         commit_list_insert(parents->item, plist);
591
592                 commit = commit->parents->item;
593         }
594 }
595
596 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
597 {
598         struct commit_list *list = NULL;
599
600         while (nr--) {
601                 clear_commit_marks_1(&list, *commit, mark);
602                 commit++;
603         }
604         while (list)
605                 clear_commit_marks_1(&list, pop_commit(&list), mark);
606 }
607
608 void clear_commit_marks(struct commit *commit, unsigned int mark)
609 {
610         clear_commit_marks_many(1, &commit, mark);
611 }
612
613 struct commit *pop_commit(struct commit_list **stack)
614 {
615         struct commit_list *top = *stack;
616         struct commit *item = top ? top->item : NULL;
617
618         if (top) {
619                 *stack = top->next;
620                 free(top);
621         }
622         return item;
623 }
624
625 /*
626  * Topological sort support
627  */
628
629 /* count number of children that have not been emitted */
630 define_commit_slab(indegree_slab, int);
631
632 /* record author-date for each commit object */
633 define_commit_slab(author_date_slab, unsigned long);
634
635 static void record_author_date(struct author_date_slab *author_date,
636                                struct commit *commit)
637 {
638         const char *buffer = get_commit_buffer(commit, NULL);
639         struct ident_split ident;
640         const char *ident_line;
641         size_t ident_len;
642         char *date_end;
643         timestamp_t date;
644
645         ident_line = find_commit_header(buffer, "author", &ident_len);
646         if (!ident_line)
647                 goto fail_exit; /* no author line */
648         if (split_ident_line(&ident, ident_line, ident_len) ||
649             !ident.date_begin || !ident.date_end)
650                 goto fail_exit; /* malformed "author" line */
651
652         date = parse_timestamp(ident.date_begin, &date_end, 10);
653         if (date_end != ident.date_end)
654                 goto fail_exit; /* malformed date */
655         *(author_date_slab_at(author_date, commit)) = date;
656
657 fail_exit:
658         unuse_commit_buffer(commit, buffer);
659 }
660
661 static int compare_commits_by_author_date(const void *a_, const void *b_,
662                                           void *cb_data)
663 {
664         const struct commit *a = a_, *b = b_;
665         struct author_date_slab *author_date = cb_data;
666         timestamp_t a_date = *(author_date_slab_at(author_date, a));
667         timestamp_t b_date = *(author_date_slab_at(author_date, b));
668
669         /* newer commits with larger date first */
670         if (a_date < b_date)
671                 return 1;
672         else if (a_date > b_date)
673                 return -1;
674         return 0;
675 }
676
677 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
678 {
679         const struct commit *a = a_, *b = b_;
680
681         /* newer commits first */
682         if (a->generation < b->generation)
683                 return 1;
684         else if (a->generation > b->generation)
685                 return -1;
686
687         /* use date as a heuristic when generations are equal */
688         if (a->date < b->date)
689                 return 1;
690         else if (a->date > b->date)
691                 return -1;
692         return 0;
693 }
694
695 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
696 {
697         const struct commit *a = a_, *b = b_;
698         /* newer commits with larger date first */
699         if (a->date < b->date)
700                 return 1;
701         else if (a->date > b->date)
702                 return -1;
703         return 0;
704 }
705
706 /*
707  * Performs an in-place topological sort on the list supplied.
708  */
709 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
710 {
711         struct commit_list *next, *orig = *list;
712         struct commit_list **pptr;
713         struct indegree_slab indegree;
714         struct prio_queue queue;
715         struct commit *commit;
716         struct author_date_slab author_date;
717
718         if (!orig)
719                 return;
720         *list = NULL;
721
722         init_indegree_slab(&indegree);
723         memset(&queue, '\0', sizeof(queue));
724
725         switch (sort_order) {
726         default: /* REV_SORT_IN_GRAPH_ORDER */
727                 queue.compare = NULL;
728                 break;
729         case REV_SORT_BY_COMMIT_DATE:
730                 queue.compare = compare_commits_by_commit_date;
731                 break;
732         case REV_SORT_BY_AUTHOR_DATE:
733                 init_author_date_slab(&author_date);
734                 queue.compare = compare_commits_by_author_date;
735                 queue.cb_data = &author_date;
736                 break;
737         }
738
739         /* Mark them and clear the indegree */
740         for (next = orig; next; next = next->next) {
741                 struct commit *commit = next->item;
742                 *(indegree_slab_at(&indegree, commit)) = 1;
743                 /* also record the author dates, if needed */
744                 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
745                         record_author_date(&author_date, commit);
746         }
747
748         /* update the indegree */
749         for (next = orig; next; next = next->next) {
750                 struct commit_list *parents = next->item->parents;
751                 while (parents) {
752                         struct commit *parent = parents->item;
753                         int *pi = indegree_slab_at(&indegree, parent);
754
755                         if (*pi)
756                                 (*pi)++;
757                         parents = parents->next;
758                 }
759         }
760
761         /*
762          * find the tips
763          *
764          * tips are nodes not reachable from any other node in the list
765          *
766          * the tips serve as a starting set for the work queue.
767          */
768         for (next = orig; next; next = next->next) {
769                 struct commit *commit = next->item;
770
771                 if (*(indegree_slab_at(&indegree, commit)) == 1)
772                         prio_queue_put(&queue, commit);
773         }
774
775         /*
776          * This is unfortunate; the initial tips need to be shown
777          * in the order given from the revision traversal machinery.
778          */
779         if (sort_order == REV_SORT_IN_GRAPH_ORDER)
780                 prio_queue_reverse(&queue);
781
782         /* We no longer need the commit list */
783         free_commit_list(orig);
784
785         pptr = list;
786         *list = NULL;
787         while ((commit = prio_queue_get(&queue)) != NULL) {
788                 struct commit_list *parents;
789
790                 for (parents = commit->parents; parents ; parents = parents->next) {
791                         struct commit *parent = parents->item;
792                         int *pi = indegree_slab_at(&indegree, parent);
793
794                         if (!*pi)
795                                 continue;
796
797                         /*
798                          * parents are only enqueued for emission
799                          * when all their children have been emitted thereby
800                          * guaranteeing topological order.
801                          */
802                         if (--(*pi) == 1)
803                                 prio_queue_put(&queue, parent);
804                 }
805                 /*
806                  * all children of commit have already been
807                  * emitted. we can emit it now.
808                  */
809                 *(indegree_slab_at(&indegree, commit)) = 0;
810
811                 pptr = &commit_list_insert(commit, pptr)->next;
812         }
813
814         clear_indegree_slab(&indegree);
815         clear_prio_queue(&queue);
816         if (sort_order == REV_SORT_BY_AUTHOR_DATE)
817                 clear_author_date_slab(&author_date);
818 }
819
820 /* merge-base stuff */
821
822 /* Remember to update object flag allocation in object.h */
823 #define PARENT1         (1u<<16)
824 #define PARENT2         (1u<<17)
825 #define STALE           (1u<<18)
826 #define RESULT          (1u<<19)
827
828 static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
829
830 static int queue_has_nonstale(struct prio_queue *queue)
831 {
832         int i;
833         for (i = 0; i < queue->nr; i++) {
834                 struct commit *commit = queue->array[i].data;
835                 if (!(commit->object.flags & STALE))
836                         return 1;
837         }
838         return 0;
839 }
840
841 /* all input commits in one and twos[] must have been parsed! */
842 static struct commit_list *paint_down_to_common(struct commit *one, int n,
843                                                 struct commit **twos,
844                                                 int min_generation)
845 {
846         struct prio_queue queue = { compare_commits_by_gen_then_commit_date };
847         struct commit_list *result = NULL;
848         int i;
849         uint32_t last_gen = GENERATION_NUMBER_INFINITY;
850
851         one->object.flags |= PARENT1;
852         if (!n) {
853                 commit_list_append(one, &result);
854                 return result;
855         }
856         prio_queue_put(&queue, one);
857
858         for (i = 0; i < n; i++) {
859                 twos[i]->object.flags |= PARENT2;
860                 prio_queue_put(&queue, twos[i]);
861         }
862
863         while (queue_has_nonstale(&queue)) {
864                 struct commit *commit = prio_queue_get(&queue);
865                 struct commit_list *parents;
866                 int flags;
867
868                 if (commit->generation > last_gen)
869                         BUG("bad generation skip %8x > %8x at %s",
870                             commit->generation, last_gen,
871                             oid_to_hex(&commit->object.oid));
872                 last_gen = commit->generation;
873
874                 if (commit->generation < min_generation)
875                         break;
876
877                 flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
878                 if (flags == (PARENT1 | PARENT2)) {
879                         if (!(commit->object.flags & RESULT)) {
880                                 commit->object.flags |= RESULT;
881                                 commit_list_insert_by_date(commit, &result);
882                         }
883                         /* Mark parents of a found merge stale */
884                         flags |= STALE;
885                 }
886                 parents = commit->parents;
887                 while (parents) {
888                         struct commit *p = parents->item;
889                         parents = parents->next;
890                         if ((p->object.flags & flags) == flags)
891                                 continue;
892                         if (parse_commit(p))
893                                 return NULL;
894                         p->object.flags |= flags;
895                         prio_queue_put(&queue, p);
896                 }
897         }
898
899         clear_prio_queue(&queue);
900         return result;
901 }
902
903 static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
904 {
905         struct commit_list *list = NULL;
906         struct commit_list *result = NULL;
907         int i;
908
909         for (i = 0; i < n; i++) {
910                 if (one == twos[i])
911                         /*
912                          * We do not mark this even with RESULT so we do not
913                          * have to clean it up.
914                          */
915                         return commit_list_insert(one, &result);
916         }
917
918         if (parse_commit(one))
919                 return NULL;
920         for (i = 0; i < n; i++) {
921                 if (parse_commit(twos[i]))
922                         return NULL;
923         }
924
925         list = paint_down_to_common(one, n, twos, 0);
926
927         while (list) {
928                 struct commit *commit = pop_commit(&list);
929                 if (!(commit->object.flags & STALE))
930                         commit_list_insert_by_date(commit, &result);
931         }
932         return result;
933 }
934
935 struct commit_list *get_octopus_merge_bases(struct commit_list *in)
936 {
937         struct commit_list *i, *j, *k, *ret = NULL;
938
939         if (!in)
940                 return ret;
941
942         commit_list_insert(in->item, &ret);
943
944         for (i = in->next; i; i = i->next) {
945                 struct commit_list *new_commits = NULL, *end = NULL;
946
947                 for (j = ret; j; j = j->next) {
948                         struct commit_list *bases;
949                         bases = get_merge_bases(i->item, j->item);
950                         if (!new_commits)
951                                 new_commits = bases;
952                         else
953                                 end->next = bases;
954                         for (k = bases; k; k = k->next)
955                                 end = k;
956                 }
957                 ret = new_commits;
958         }
959         return ret;
960 }
961
962 static int remove_redundant(struct commit **array, int cnt)
963 {
964         /*
965          * Some commit in the array may be an ancestor of
966          * another commit.  Move such commit to the end of
967          * the array, and return the number of commits that
968          * are independent from each other.
969          */
970         struct commit **work;
971         unsigned char *redundant;
972         int *filled_index;
973         int i, j, filled;
974
975         work = xcalloc(cnt, sizeof(*work));
976         redundant = xcalloc(cnt, 1);
977         ALLOC_ARRAY(filled_index, cnt - 1);
978
979         for (i = 0; i < cnt; i++)
980                 parse_commit(array[i]);
981         for (i = 0; i < cnt; i++) {
982                 struct commit_list *common;
983                 uint32_t min_generation = array[i]->generation;
984
985                 if (redundant[i])
986                         continue;
987                 for (j = filled = 0; j < cnt; j++) {
988                         if (i == j || redundant[j])
989                                 continue;
990                         filled_index[filled] = j;
991                         work[filled++] = array[j];
992
993                         if (array[j]->generation < min_generation)
994                                 min_generation = array[j]->generation;
995                 }
996                 common = paint_down_to_common(array[i], filled, work,
997                                               min_generation);
998                 if (array[i]->object.flags & PARENT2)
999                         redundant[i] = 1;
1000                 for (j = 0; j < filled; j++)
1001                         if (work[j]->object.flags & PARENT1)
1002                                 redundant[filled_index[j]] = 1;
1003                 clear_commit_marks(array[i], all_flags);
1004                 clear_commit_marks_many(filled, work, all_flags);
1005                 free_commit_list(common);
1006         }
1007
1008         /* Now collect the result */
1009         COPY_ARRAY(work, array, cnt);
1010         for (i = filled = 0; i < cnt; i++)
1011                 if (!redundant[i])
1012                         array[filled++] = work[i];
1013         for (j = filled, i = 0; i < cnt; i++)
1014                 if (redundant[i])
1015                         array[j++] = work[i];
1016         free(work);
1017         free(redundant);
1018         free(filled_index);
1019         return filled;
1020 }
1021
1022 static struct commit_list *get_merge_bases_many_0(struct commit *one,
1023                                                   int n,
1024                                                   struct commit **twos,
1025                                                   int cleanup)
1026 {
1027         struct commit_list *list;
1028         struct commit **rslt;
1029         struct commit_list *result;
1030         int cnt, i;
1031
1032         result = merge_bases_many(one, n, twos);
1033         for (i = 0; i < n; i++) {
1034                 if (one == twos[i])
1035                         return result;
1036         }
1037         if (!result || !result->next) {
1038                 if (cleanup) {
1039                         clear_commit_marks(one, all_flags);
1040                         clear_commit_marks_many(n, twos, all_flags);
1041                 }
1042                 return result;
1043         }
1044
1045         /* There are more than one */
1046         cnt = commit_list_count(result);
1047         rslt = xcalloc(cnt, sizeof(*rslt));
1048         for (list = result, i = 0; list; list = list->next)
1049                 rslt[i++] = list->item;
1050         free_commit_list(result);
1051
1052         clear_commit_marks(one, all_flags);
1053         clear_commit_marks_many(n, twos, all_flags);
1054
1055         cnt = remove_redundant(rslt, cnt);
1056         result = NULL;
1057         for (i = 0; i < cnt; i++)
1058                 commit_list_insert_by_date(rslt[i], &result);
1059         free(rslt);
1060         return result;
1061 }
1062
1063 struct commit_list *get_merge_bases_many(struct commit *one,
1064                                          int n,
1065                                          struct commit **twos)
1066 {
1067         return get_merge_bases_many_0(one, n, twos, 1);
1068 }
1069
1070 struct commit_list *get_merge_bases_many_dirty(struct commit *one,
1071                                                int n,
1072                                                struct commit **twos)
1073 {
1074         return get_merge_bases_many_0(one, n, twos, 0);
1075 }
1076
1077 struct commit_list *get_merge_bases(struct commit *one, struct commit *two)
1078 {
1079         return get_merge_bases_many_0(one, 1, &two, 1);
1080 }
1081
1082 /*
1083  * Is "commit" a descendant of one of the elements on the "with_commit" list?
1084  */
1085 int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
1086 {
1087         if (!with_commit)
1088                 return 1;
1089         while (with_commit) {
1090                 struct commit *other;
1091
1092                 other = with_commit->item;
1093                 with_commit = with_commit->next;
1094                 if (in_merge_bases(other, commit))
1095                         return 1;
1096         }
1097         return 0;
1098 }
1099
1100 /*
1101  * Is "commit" an ancestor of one of the "references"?
1102  */
1103 int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
1104 {
1105         struct commit_list *bases;
1106         int ret = 0, i;
1107         uint32_t min_generation = GENERATION_NUMBER_INFINITY;
1108
1109         if (parse_commit(commit))
1110                 return ret;
1111         for (i = 0; i < nr_reference; i++) {
1112                 if (parse_commit(reference[i]))
1113                         return ret;
1114                 if (reference[i]->generation < min_generation)
1115                         min_generation = reference[i]->generation;
1116         }
1117
1118         if (commit->generation > min_generation)
1119                 return ret;
1120
1121         bases = paint_down_to_common(commit, nr_reference, reference, commit->generation);
1122         if (commit->object.flags & PARENT2)
1123                 ret = 1;
1124         clear_commit_marks(commit, all_flags);
1125         clear_commit_marks_many(nr_reference, reference, all_flags);
1126         free_commit_list(bases);
1127         return ret;
1128 }
1129
1130 /*
1131  * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1132  */
1133 int in_merge_bases(struct commit *commit, struct commit *reference)
1134 {
1135         return in_merge_bases_many(commit, 1, &reference);
1136 }
1137
1138 struct commit_list *reduce_heads(struct commit_list *heads)
1139 {
1140         struct commit_list *p;
1141         struct commit_list *result = NULL, **tail = &result;
1142         struct commit **array;
1143         int num_head, i;
1144
1145         if (!heads)
1146                 return NULL;
1147
1148         /* Uniquify */
1149         for (p = heads; p; p = p->next)
1150                 p->item->object.flags &= ~STALE;
1151         for (p = heads, num_head = 0; p; p = p->next) {
1152                 if (p->item->object.flags & STALE)
1153                         continue;
1154                 p->item->object.flags |= STALE;
1155                 num_head++;
1156         }
1157         array = xcalloc(num_head, sizeof(*array));
1158         for (p = heads, i = 0; p; p = p->next) {
1159                 if (p->item->object.flags & STALE) {
1160                         array[i++] = p->item;
1161                         p->item->object.flags &= ~STALE;
1162                 }
1163         }
1164         num_head = remove_redundant(array, num_head);
1165         for (i = 0; i < num_head; i++)
1166                 tail = &commit_list_insert(array[i], tail)->next;
1167         free(array);
1168         return result;
1169 }
1170
1171 void reduce_heads_replace(struct commit_list **heads)
1172 {
1173         struct commit_list *result = reduce_heads(*heads);
1174         free_commit_list(*heads);
1175         *heads = result;
1176 }
1177
1178 static const char gpg_sig_header[] = "gpgsig";
1179 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1180
1181 static int do_sign_commit(struct strbuf *buf, const char *keyid)
1182 {
1183         struct strbuf sig = STRBUF_INIT;
1184         int inspos, copypos;
1185         const char *eoh;
1186
1187         /* find the end of the header */
1188         eoh = strstr(buf->buf, "\n\n");
1189         if (!eoh)
1190                 inspos = buf->len;
1191         else
1192                 inspos = eoh - buf->buf + 1;
1193
1194         if (!keyid || !*keyid)
1195                 keyid = get_signing_key();
1196         if (sign_buffer(buf, &sig, keyid)) {
1197                 strbuf_release(&sig);
1198                 return -1;
1199         }
1200
1201         for (copypos = 0; sig.buf[copypos]; ) {
1202                 const char *bol = sig.buf + copypos;
1203                 const char *eol = strchrnul(bol, '\n');
1204                 int len = (eol - bol) + !!*eol;
1205
1206                 if (!copypos) {
1207                         strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1208                         inspos += gpg_sig_header_len;
1209                 }
1210                 strbuf_insert(buf, inspos++, " ", 1);
1211                 strbuf_insert(buf, inspos, bol, len);
1212                 inspos += len;
1213                 copypos += len;
1214         }
1215         strbuf_release(&sig);
1216         return 0;
1217 }
1218
1219 int parse_signed_commit(const struct commit *commit,
1220                         struct strbuf *payload, struct strbuf *signature)
1221 {
1222
1223         unsigned long size;
1224         const char *buffer = get_commit_buffer(commit, &size);
1225         int in_signature, saw_signature = -1;
1226         const char *line, *tail;
1227
1228         line = buffer;
1229         tail = buffer + size;
1230         in_signature = 0;
1231         saw_signature = 0;
1232         while (line < tail) {
1233                 const char *sig = NULL;
1234                 const char *next = memchr(line, '\n', tail - line);
1235
1236                 next = next ? next + 1 : tail;
1237                 if (in_signature && line[0] == ' ')
1238                         sig = line + 1;
1239                 else if (starts_with(line, gpg_sig_header) &&
1240                          line[gpg_sig_header_len] == ' ')
1241                         sig = line + gpg_sig_header_len + 1;
1242                 if (sig) {
1243                         strbuf_add(signature, sig, next - sig);
1244                         saw_signature = 1;
1245                         in_signature = 1;
1246                 } else {
1247                         if (*line == '\n')
1248                                 /* dump the whole remainder of the buffer */
1249                                 next = tail;
1250                         strbuf_add(payload, line, next - line);
1251                         in_signature = 0;
1252                 }
1253                 line = next;
1254         }
1255         unuse_commit_buffer(commit, buffer);
1256         return saw_signature;
1257 }
1258
1259 int remove_signature(struct strbuf *buf)
1260 {
1261         const char *line = buf->buf;
1262         const char *tail = buf->buf + buf->len;
1263         int in_signature = 0;
1264         const char *sig_start = NULL;
1265         const char *sig_end = NULL;
1266
1267         while (line < tail) {
1268                 const char *next = memchr(line, '\n', tail - line);
1269                 next = next ? next + 1 : tail;
1270
1271                 if (in_signature && line[0] == ' ')
1272                         sig_end = next;
1273                 else if (starts_with(line, gpg_sig_header) &&
1274                          line[gpg_sig_header_len] == ' ') {
1275                         sig_start = line;
1276                         sig_end = next;
1277                         in_signature = 1;
1278                 } else {
1279                         if (*line == '\n')
1280                                 /* dump the whole remainder of the buffer */
1281                                 next = tail;
1282                         in_signature = 0;
1283                 }
1284                 line = next;
1285         }
1286
1287         if (sig_start)
1288                 strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1289
1290         return sig_start != NULL;
1291 }
1292
1293 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1294 {
1295         struct merge_remote_desc *desc;
1296         struct commit_extra_header *mergetag;
1297         char *buf;
1298         unsigned long size, len;
1299         enum object_type type;
1300
1301         desc = merge_remote_util(parent);
1302         if (!desc || !desc->obj)
1303                 return;
1304         buf = read_object_file(&desc->obj->oid, &type, &size);
1305         if (!buf || type != OBJ_TAG)
1306                 goto free_return;
1307         len = parse_signature(buf, size);
1308         if (size == len)
1309                 goto free_return;
1310         /*
1311          * We could verify this signature and either omit the tag when
1312          * it does not validate, but the integrator may not have the
1313          * public key of the signer of the tag he is merging, while a
1314          * later auditor may have it while auditing, so let's not run
1315          * verify-signed-buffer here for now...
1316          *
1317          * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1318          *      warn("warning: signed tag unverified.");
1319          */
1320         mergetag = xcalloc(1, sizeof(*mergetag));
1321         mergetag->key = xstrdup("mergetag");
1322         mergetag->value = buf;
1323         mergetag->len = size;
1324
1325         **tail = mergetag;
1326         *tail = &mergetag->next;
1327         return;
1328
1329 free_return:
1330         free(buf);
1331 }
1332
1333 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1334 {
1335         struct strbuf payload = STRBUF_INIT;
1336         struct strbuf signature = STRBUF_INIT;
1337         int ret = 1;
1338
1339         sigc->result = 'N';
1340
1341         if (parse_signed_commit(commit, &payload, &signature) <= 0)
1342                 goto out;
1343         ret = check_signature(payload.buf, payload.len, signature.buf,
1344                 signature.len, sigc);
1345
1346  out:
1347         strbuf_release(&payload);
1348         strbuf_release(&signature);
1349
1350         return ret;
1351 }
1352
1353
1354
1355 void append_merge_tag_headers(struct commit_list *parents,
1356                               struct commit_extra_header ***tail)
1357 {
1358         while (parents) {
1359                 struct commit *parent = parents->item;
1360                 handle_signed_tag(parent, tail);
1361                 parents = parents->next;
1362         }
1363 }
1364
1365 static void add_extra_header(struct strbuf *buffer,
1366                              struct commit_extra_header *extra)
1367 {
1368         strbuf_addstr(buffer, extra->key);
1369         if (extra->len)
1370                 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1371         else
1372                 strbuf_addch(buffer, '\n');
1373 }
1374
1375 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1376                                                       const char **exclude)
1377 {
1378         struct commit_extra_header *extra = NULL;
1379         unsigned long size;
1380         const char *buffer = get_commit_buffer(commit, &size);
1381         extra = read_commit_extra_header_lines(buffer, size, exclude);
1382         unuse_commit_buffer(commit, buffer);
1383         return extra;
1384 }
1385
1386 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1387 {
1388         struct commit_extra_header *extra, *to_free;
1389         int res = 0;
1390
1391         to_free = read_commit_extra_headers(commit, NULL);
1392         for (extra = to_free; !res && extra; extra = extra->next) {
1393                 if (strcmp(extra->key, "mergetag"))
1394                         continue; /* not a merge tag */
1395                 res = fn(commit, extra, data);
1396         }
1397         free_commit_extra_headers(to_free);
1398         return res;
1399 }
1400
1401 static inline int standard_header_field(const char *field, size_t len)
1402 {
1403         return ((len == 4 && !memcmp(field, "tree", 4)) ||
1404                 (len == 6 && !memcmp(field, "parent", 6)) ||
1405                 (len == 6 && !memcmp(field, "author", 6)) ||
1406                 (len == 9 && !memcmp(field, "committer", 9)) ||
1407                 (len == 8 && !memcmp(field, "encoding", 8)));
1408 }
1409
1410 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1411 {
1412         if (!exclude)
1413                 return 0;
1414
1415         while (*exclude) {
1416                 size_t xlen = strlen(*exclude);
1417                 if (len == xlen && !memcmp(field, *exclude, xlen))
1418                         return 1;
1419                 exclude++;
1420         }
1421         return 0;
1422 }
1423
1424 static struct commit_extra_header *read_commit_extra_header_lines(
1425         const char *buffer, size_t size,
1426         const char **exclude)
1427 {
1428         struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1429         const char *line, *next, *eof, *eob;
1430         struct strbuf buf = STRBUF_INIT;
1431
1432         for (line = buffer, eob = line + size;
1433              line < eob && *line != '\n';
1434              line = next) {
1435                 next = memchr(line, '\n', eob - line);
1436                 next = next ? next + 1 : eob;
1437                 if (*line == ' ') {
1438                         /* continuation */
1439                         if (it)
1440                                 strbuf_add(&buf, line + 1, next - (line + 1));
1441                         continue;
1442                 }
1443                 if (it)
1444                         it->value = strbuf_detach(&buf, &it->len);
1445                 strbuf_reset(&buf);
1446                 it = NULL;
1447
1448                 eof = memchr(line, ' ', next - line);
1449                 if (!eof)
1450                         eof = next;
1451                 else if (standard_header_field(line, eof - line) ||
1452                          excluded_header_field(line, eof - line, exclude))
1453                         continue;
1454
1455                 it = xcalloc(1, sizeof(*it));
1456                 it->key = xmemdupz(line, eof-line);
1457                 *tail = it;
1458                 tail = &it->next;
1459                 if (eof + 1 < next)
1460                         strbuf_add(&buf, eof + 1, next - (eof + 1));
1461         }
1462         if (it)
1463                 it->value = strbuf_detach(&buf, &it->len);
1464         return extra;
1465 }
1466
1467 void free_commit_extra_headers(struct commit_extra_header *extra)
1468 {
1469         while (extra) {
1470                 struct commit_extra_header *next = extra->next;
1471                 free(extra->key);
1472                 free(extra->value);
1473                 free(extra);
1474                 extra = next;
1475         }
1476 }
1477
1478 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1479                 struct commit_list *parents, struct object_id *ret,
1480                 const char *author, const char *sign_commit)
1481 {
1482         struct commit_extra_header *extra = NULL, **tail = &extra;
1483         int result;
1484
1485         append_merge_tag_headers(parents, &tail);
1486         result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1487                                       author, sign_commit, extra);
1488         free_commit_extra_headers(extra);
1489         return result;
1490 }
1491
1492 static int find_invalid_utf8(const char *buf, int len)
1493 {
1494         int offset = 0;
1495         static const unsigned int max_codepoint[] = {
1496                 0x7f, 0x7ff, 0xffff, 0x10ffff
1497         };
1498
1499         while (len) {
1500                 unsigned char c = *buf++;
1501                 int bytes, bad_offset;
1502                 unsigned int codepoint;
1503                 unsigned int min_val, max_val;
1504
1505                 len--;
1506                 offset++;
1507
1508                 /* Simple US-ASCII? No worries. */
1509                 if (c < 0x80)
1510                         continue;
1511
1512                 bad_offset = offset-1;
1513
1514                 /*
1515                  * Count how many more high bits set: that's how
1516                  * many more bytes this sequence should have.
1517                  */
1518                 bytes = 0;
1519                 while (c & 0x40) {
1520                         c <<= 1;
1521                         bytes++;
1522                 }
1523
1524                 /*
1525                  * Must be between 1 and 3 more bytes.  Longer sequences result in
1526                  * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1527                  */
1528                 if (bytes < 1 || 3 < bytes)
1529                         return bad_offset;
1530
1531                 /* Do we *have* that many bytes? */
1532                 if (len < bytes)
1533                         return bad_offset;
1534
1535                 /*
1536                  * Place the encoded bits at the bottom of the value and compute the
1537                  * valid range.
1538                  */
1539                 codepoint = (c & 0x7f) >> bytes;
1540                 min_val = max_codepoint[bytes-1] + 1;
1541                 max_val = max_codepoint[bytes];
1542
1543                 offset += bytes;
1544                 len -= bytes;
1545
1546                 /* And verify that they are good continuation bytes */
1547                 do {
1548                         codepoint <<= 6;
1549                         codepoint |= *buf & 0x3f;
1550                         if ((*buf++ & 0xc0) != 0x80)
1551                                 return bad_offset;
1552                 } while (--bytes);
1553
1554                 /* Reject codepoints that are out of range for the sequence length. */
1555                 if (codepoint < min_val || codepoint > max_val)
1556                         return bad_offset;
1557                 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1558                 if ((codepoint & 0x1ff800) == 0xd800)
1559                         return bad_offset;
1560                 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1561                 if ((codepoint & 0xfffe) == 0xfffe)
1562                         return bad_offset;
1563                 /* So are anything in the range U+FDD0..U+FDEF. */
1564                 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1565                         return bad_offset;
1566         }
1567         return -1;
1568 }
1569
1570 /*
1571  * This verifies that the buffer is in proper utf8 format.
1572  *
1573  * If it isn't, it assumes any non-utf8 characters are Latin1,
1574  * and does the conversion.
1575  */
1576 static int verify_utf8(struct strbuf *buf)
1577 {
1578         int ok = 1;
1579         long pos = 0;
1580
1581         for (;;) {
1582                 int bad;
1583                 unsigned char c;
1584                 unsigned char replace[2];
1585
1586                 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1587                 if (bad < 0)
1588                         return ok;
1589                 pos += bad;
1590                 ok = 0;
1591                 c = buf->buf[pos];
1592                 strbuf_remove(buf, pos, 1);
1593
1594                 /* We know 'c' must be in the range 128-255 */
1595                 replace[0] = 0xc0 + (c >> 6);
1596                 replace[1] = 0x80 + (c & 0x3f);
1597                 strbuf_insert(buf, pos, replace, 2);
1598                 pos += 2;
1599         }
1600 }
1601
1602 static const char commit_utf8_warn[] =
1603 N_("Warning: commit message did not conform to UTF-8.\n"
1604    "You may want to amend it after fixing the message, or set the config\n"
1605    "variable i18n.commitencoding to the encoding your project uses.\n");
1606
1607 int commit_tree_extended(const char *msg, size_t msg_len,
1608                          const struct object_id *tree,
1609                          struct commit_list *parents, struct object_id *ret,
1610                          const char *author, const char *sign_commit,
1611                          struct commit_extra_header *extra)
1612 {
1613         int result;
1614         int encoding_is_utf8;
1615         struct strbuf buffer;
1616
1617         assert_oid_type(tree, OBJ_TREE);
1618
1619         if (memchr(msg, '\0', msg_len))
1620                 return error("a NUL byte in commit log message not allowed.");
1621
1622         /* Not having i18n.commitencoding is the same as having utf-8 */
1623         encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1624
1625         strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1626         strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1627
1628         /*
1629          * NOTE! This ordering means that the same exact tree merged with a
1630          * different order of parents will be a _different_ changeset even
1631          * if everything else stays the same.
1632          */
1633         while (parents) {
1634                 struct commit *parent = pop_commit(&parents);
1635                 strbuf_addf(&buffer, "parent %s\n",
1636                             oid_to_hex(&parent->object.oid));
1637         }
1638
1639         /* Person/date information */
1640         if (!author)
1641                 author = git_author_info(IDENT_STRICT);
1642         strbuf_addf(&buffer, "author %s\n", author);
1643         strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1644         if (!encoding_is_utf8)
1645                 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1646
1647         while (extra) {
1648                 add_extra_header(&buffer, extra);
1649                 extra = extra->next;
1650         }
1651         strbuf_addch(&buffer, '\n');
1652
1653         /* And add the comment */
1654         strbuf_add(&buffer, msg, msg_len);
1655
1656         /* And check the encoding */
1657         if (encoding_is_utf8 && !verify_utf8(&buffer))
1658                 fprintf(stderr, _(commit_utf8_warn));
1659
1660         if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1661                 result = -1;
1662                 goto out;
1663         }
1664
1665         result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1666 out:
1667         strbuf_release(&buffer);
1668         return result;
1669 }
1670
1671 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1672 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1673
1674 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1675 {
1676         return *merge_desc_slab_at(&merge_desc_slab, commit);
1677 }
1678
1679 void set_merge_remote_desc(struct commit *commit,
1680                            const char *name, struct object *obj)
1681 {
1682         struct merge_remote_desc *desc;
1683         FLEX_ALLOC_STR(desc, name, name);
1684         desc->obj = obj;
1685         *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1686 }
1687
1688 struct commit *get_merge_parent(const char *name)
1689 {
1690         struct object *obj;
1691         struct commit *commit;
1692         struct object_id oid;
1693         if (get_oid(name, &oid))
1694                 return NULL;
1695         obj = parse_object(&oid);
1696         commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1697         if (commit && !merge_remote_util(commit))
1698                 set_merge_remote_desc(commit, name, obj);
1699         return commit;
1700 }
1701
1702 /*
1703  * Append a commit to the end of the commit_list.
1704  *
1705  * next starts by pointing to the variable that holds the head of an
1706  * empty commit_list, and is updated to point to the "next" field of
1707  * the last item on the list as new commits are appended.
1708  *
1709  * Usage example:
1710  *
1711  *     struct commit_list *list;
1712  *     struct commit_list **next = &list;
1713  *
1714  *     next = commit_list_append(c1, next);
1715  *     next = commit_list_append(c2, next);
1716  *     assert(commit_list_count(list) == 2);
1717  *     return list;
1718  */
1719 struct commit_list **commit_list_append(struct commit *commit,
1720                                         struct commit_list **next)
1721 {
1722         struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1723         new_commit->item = commit;
1724         *next = new_commit;
1725         new_commit->next = NULL;
1726         return &new_commit->next;
1727 }
1728
1729 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1730 {
1731         int key_len = strlen(key);
1732         const char *line = msg;
1733
1734         while (line) {
1735                 const char *eol = strchrnul(line, '\n');
1736
1737                 if (line == eol)
1738                         return NULL;
1739
1740                 if (eol - line > key_len &&
1741                     !strncmp(line, key, key_len) &&
1742                     line[key_len] == ' ') {
1743                         *out_len = eol - line - key_len - 1;
1744                         return line + key_len + 1;
1745                 }
1746                 line = *eol ? eol + 1 : NULL;
1747         }
1748         return NULL;
1749 }
1750
1751 /*
1752  * Inspect the given string and determine the true "end" of the log message, in
1753  * order to find where to put a new Signed-off-by: line.  Ignored are
1754  * trailing comment lines and blank lines.  To support "git commit -s
1755  * --amend" on an existing commit, we also ignore "Conflicts:".  To
1756  * support "git commit -v", we truncate at cut lines.
1757  *
1758  * Returns the number of bytes from the tail to ignore, to be fed as
1759  * the second parameter to append_signoff().
1760  */
1761 int ignore_non_trailer(const char *buf, size_t len)
1762 {
1763         int boc = 0;
1764         int bol = 0;
1765         int in_old_conflicts_block = 0;
1766         size_t cutoff = wt_status_locate_end(buf, len);
1767
1768         while (bol < cutoff) {
1769                 const char *next_line = memchr(buf + bol, '\n', len - bol);
1770
1771                 if (!next_line)
1772                         next_line = buf + len;
1773                 else
1774                         next_line++;
1775
1776                 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1777                         /* is this the first of the run of comments? */
1778                         if (!boc)
1779                                 boc = bol;
1780                         /* otherwise, it is just continuing */
1781                 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1782                         in_old_conflicts_block = 1;
1783                         if (!boc)
1784                                 boc = bol;
1785                 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1786                         ; /* a pathname in the conflicts block */
1787                 } else if (boc) {
1788                         /* the previous was not trailing comment */
1789                         boc = 0;
1790                         in_old_conflicts_block = 0;
1791                 }
1792                 bol = next_line - buf;
1793         }
1794         return boc ? len - boc : len - cutoff;
1795 }