sequencer: use logmsg_reencode in get_message
[git.git] / commit.c
blobfbdc480ccf17b4d241e0a950e8dd44dfe9d684ce
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "pkt-line.h"
5 #include "utf8.h"
6 #include "diff.h"
7 #include "revision.h"
8 #include "notes.h"
9 #include "gpg-interface.h"
10 #include "mergesort.h"
11 #include "commit-slab.h"
12 #include "prio-queue.h"
13 #include "sha1-lookup.h"
15 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
17 int save_commit_buffer = 1;
19 const char *commit_type = "commit";
21 static struct commit *check_commit(struct object *obj,
22 const unsigned char *sha1,
23 int quiet)
25 if (obj->type != OBJ_COMMIT) {
26 if (!quiet)
27 error("Object %s is a %s, not a commit",
28 sha1_to_hex(sha1), typename(obj->type));
29 return NULL;
31 return (struct commit *) obj;
34 struct commit *lookup_commit_reference_gently(const unsigned char *sha1,
35 int quiet)
37 struct object *obj = deref_tag(parse_object(sha1), NULL, 0);
39 if (!obj)
40 return NULL;
41 return check_commit(obj, sha1, quiet);
44 struct commit *lookup_commit_reference(const unsigned char *sha1)
46 return lookup_commit_reference_gently(sha1, 0);
49 struct commit *lookup_commit_or_die(const unsigned char *sha1, const char *ref_name)
51 struct commit *c = lookup_commit_reference(sha1);
52 if (!c)
53 die(_("could not parse %s"), ref_name);
54 if (hashcmp(sha1, c->object.sha1)) {
55 warning(_("%s %s is not a commit!"),
56 ref_name, sha1_to_hex(sha1));
58 return c;
61 struct commit *lookup_commit(const unsigned char *sha1)
63 struct object *obj = lookup_object(sha1);
64 if (!obj) {
65 struct commit *c = alloc_commit_node();
66 return create_object(sha1, OBJ_COMMIT, c);
68 if (!obj->type)
69 obj->type = OBJ_COMMIT;
70 return check_commit(obj, sha1, 0);
73 struct commit *lookup_commit_reference_by_name(const char *name)
75 unsigned char sha1[20];
76 struct commit *commit;
78 if (get_sha1_committish(name, sha1))
79 return NULL;
80 commit = lookup_commit_reference(sha1);
81 if (parse_commit(commit))
82 return NULL;
83 return commit;
86 static unsigned long parse_commit_date(const char *buf, const char *tail)
88 const char *dateptr;
90 if (buf + 6 >= tail)
91 return 0;
92 if (memcmp(buf, "author", 6))
93 return 0;
94 while (buf < tail && *buf++ != '\n')
95 /* nada */;
96 if (buf + 9 >= tail)
97 return 0;
98 if (memcmp(buf, "committer", 9))
99 return 0;
100 while (buf < tail && *buf++ != '>')
101 /* nada */;
102 if (buf >= tail)
103 return 0;
104 dateptr = buf;
105 while (buf < tail && *buf++ != '\n')
106 /* nada */;
107 if (buf >= tail)
108 return 0;
109 /* dateptr < buf && buf[-1] == '\n', so strtoul will stop at buf-1 */
110 return strtoul(dateptr, NULL, 10);
113 static struct commit_graft **commit_graft;
114 static int commit_graft_alloc, commit_graft_nr;
116 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
118 struct commit_graft **commit_graft_table = table;
119 return commit_graft_table[index]->sha1;
122 static int commit_graft_pos(const unsigned char *sha1)
124 return sha1_pos(sha1, commit_graft, commit_graft_nr,
125 commit_graft_sha1_access);
128 int register_commit_graft(struct commit_graft *graft, int ignore_dups)
130 int pos = commit_graft_pos(graft->sha1);
132 if (0 <= pos) {
133 if (ignore_dups)
134 free(graft);
135 else {
136 free(commit_graft[pos]);
137 commit_graft[pos] = graft;
139 return 1;
141 pos = -pos - 1;
142 ALLOC_GROW(commit_graft, commit_graft_nr + 1, commit_graft_alloc);
143 commit_graft_nr++;
144 if (pos < commit_graft_nr)
145 memmove(commit_graft + pos + 1,
146 commit_graft + pos,
147 (commit_graft_nr - pos - 1) *
148 sizeof(*commit_graft));
149 commit_graft[pos] = graft;
150 return 0;
153 struct commit_graft *read_graft_line(char *buf, int len)
155 /* The format is just "Commit Parent1 Parent2 ...\n" */
156 int i;
157 struct commit_graft *graft = NULL;
159 while (len && isspace(buf[len-1]))
160 buf[--len] = '\0';
161 if (buf[0] == '#' || buf[0] == '\0')
162 return NULL;
163 if ((len + 1) % 41)
164 goto bad_graft_data;
165 i = (len + 1) / 41 - 1;
166 graft = xmalloc(sizeof(*graft) + 20 * i);
167 graft->nr_parent = i;
168 if (get_sha1_hex(buf, graft->sha1))
169 goto bad_graft_data;
170 for (i = 40; i < len; i += 41) {
171 if (buf[i] != ' ')
172 goto bad_graft_data;
173 if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
174 goto bad_graft_data;
176 return graft;
178 bad_graft_data:
179 error("bad graft data: %s", buf);
180 free(graft);
181 return NULL;
184 static int read_graft_file(const char *graft_file)
186 FILE *fp = fopen(graft_file, "r");
187 struct strbuf buf = STRBUF_INIT;
188 if (!fp)
189 return -1;
190 while (!strbuf_getwholeline(&buf, fp, '\n')) {
191 /* The format is just "Commit Parent1 Parent2 ...\n" */
192 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
193 if (!graft)
194 continue;
195 if (register_commit_graft(graft, 1))
196 error("duplicate graft data: %s", buf.buf);
198 fclose(fp);
199 strbuf_release(&buf);
200 return 0;
203 static void prepare_commit_graft(void)
205 static int commit_graft_prepared;
206 char *graft_file;
208 if (commit_graft_prepared)
209 return;
210 graft_file = get_graft_file();
211 read_graft_file(graft_file);
212 /* make sure shallows are read */
213 is_repository_shallow();
214 commit_graft_prepared = 1;
217 struct commit_graft *lookup_commit_graft(const unsigned char *sha1)
219 int pos;
220 prepare_commit_graft();
221 pos = commit_graft_pos(sha1);
222 if (pos < 0)
223 return NULL;
224 return commit_graft[pos];
227 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
229 int i, ret;
230 for (i = ret = 0; i < commit_graft_nr && !ret; i++)
231 ret = fn(commit_graft[i], cb_data);
232 return ret;
235 int unregister_shallow(const unsigned char *sha1)
237 int pos = commit_graft_pos(sha1);
238 if (pos < 0)
239 return -1;
240 if (pos + 1 < commit_graft_nr)
241 memmove(commit_graft + pos, commit_graft + pos + 1,
242 sizeof(struct commit_graft *)
243 * (commit_graft_nr - pos - 1));
244 commit_graft_nr--;
245 return 0;
248 int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size)
250 const char *tail = buffer;
251 const char *bufptr = buffer;
252 unsigned char parent[20];
253 struct commit_list **pptr;
254 struct commit_graft *graft;
256 if (item->object.parsed)
257 return 0;
258 item->object.parsed = 1;
259 tail += size;
260 if (tail <= bufptr + 46 || memcmp(bufptr, "tree ", 5) || bufptr[45] != '\n')
261 return error("bogus commit object %s", sha1_to_hex(item->object.sha1));
262 if (get_sha1_hex(bufptr + 5, parent) < 0)
263 return error("bad tree pointer in commit %s",
264 sha1_to_hex(item->object.sha1));
265 item->tree = lookup_tree(parent);
266 bufptr += 46; /* "tree " + "hex sha1" + "\n" */
267 pptr = &item->parents;
269 graft = lookup_commit_graft(item->object.sha1);
270 while (bufptr + 48 < tail && !memcmp(bufptr, "parent ", 7)) {
271 struct commit *new_parent;
273 if (tail <= bufptr + 48 ||
274 get_sha1_hex(bufptr + 7, parent) ||
275 bufptr[47] != '\n')
276 return error("bad parents in commit %s", sha1_to_hex(item->object.sha1));
277 bufptr += 48;
279 * The clone is shallow if nr_parent < 0, and we must
280 * not traverse its real parents even when we unhide them.
282 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
283 continue;
284 new_parent = lookup_commit(parent);
285 if (new_parent)
286 pptr = &commit_list_insert(new_parent, pptr)->next;
288 if (graft) {
289 int i;
290 struct commit *new_parent;
291 for (i = 0; i < graft->nr_parent; i++) {
292 new_parent = lookup_commit(graft->parent[i]);
293 if (!new_parent)
294 continue;
295 pptr = &commit_list_insert(new_parent, pptr)->next;
298 item->date = parse_commit_date(bufptr, tail);
300 return 0;
303 int parse_commit(struct commit *item)
305 enum object_type type;
306 void *buffer;
307 unsigned long size;
308 int ret;
310 if (!item)
311 return -1;
312 if (item->object.parsed)
313 return 0;
314 buffer = read_sha1_file(item->object.sha1, &type, &size);
315 if (!buffer)
316 return error("Could not read %s",
317 sha1_to_hex(item->object.sha1));
318 if (type != OBJ_COMMIT) {
319 free(buffer);
320 return error("Object %s not a commit",
321 sha1_to_hex(item->object.sha1));
323 ret = parse_commit_buffer(item, buffer, size);
324 if (save_commit_buffer && !ret) {
325 item->buffer = buffer;
326 return 0;
328 free(buffer);
329 return ret;
332 void parse_commit_or_die(struct commit *item)
334 if (parse_commit(item))
335 die("unable to parse commit %s",
336 item ? sha1_to_hex(item->object.sha1) : "(null)");
339 int find_commit_subject(const char *commit_buffer, const char **subject)
341 const char *eol;
342 const char *p = commit_buffer;
344 while (*p && (*p != '\n' || p[1] != '\n'))
345 p++;
346 if (*p) {
347 p += 2;
348 for (eol = p; *eol && *eol != '\n'; eol++)
349 ; /* do nothing */
350 } else
351 eol = p;
353 *subject = p;
355 return eol - p;
358 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
360 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
361 new_list->item = item;
362 new_list->next = *list_p;
363 *list_p = new_list;
364 return new_list;
367 unsigned commit_list_count(const struct commit_list *l)
369 unsigned c = 0;
370 for (; l; l = l->next )
371 c++;
372 return c;
375 struct commit_list *copy_commit_list(struct commit_list *list)
377 struct commit_list *head = NULL;
378 struct commit_list **pp = &head;
379 while (list) {
380 struct commit_list *new;
381 new = xmalloc(sizeof(struct commit_list));
382 new->item = list->item;
383 new->next = NULL;
384 *pp = new;
385 pp = &new->next;
386 list = list->next;
388 return head;
391 void free_commit_list(struct commit_list *list)
393 while (list) {
394 struct commit_list *temp = list;
395 list = temp->next;
396 free(temp);
400 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
402 struct commit_list **pp = list;
403 struct commit_list *p;
404 while ((p = *pp) != NULL) {
405 if (p->item->date < item->date) {
406 break;
408 pp = &p->next;
410 return commit_list_insert(item, pp);
413 static int commit_list_compare_by_date(const void *a, const void *b)
415 unsigned long a_date = ((const struct commit_list *)a)->item->date;
416 unsigned long b_date = ((const struct commit_list *)b)->item->date;
417 if (a_date < b_date)
418 return 1;
419 if (a_date > b_date)
420 return -1;
421 return 0;
424 static void *commit_list_get_next(const void *a)
426 return ((const struct commit_list *)a)->next;
429 static void commit_list_set_next(void *a, void *next)
431 ((struct commit_list *)a)->next = next;
434 void commit_list_sort_by_date(struct commit_list **list)
436 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
437 commit_list_compare_by_date);
440 struct commit *pop_most_recent_commit(struct commit_list **list,
441 unsigned int mark)
443 struct commit *ret = (*list)->item;
444 struct commit_list *parents = ret->parents;
445 struct commit_list *old = *list;
447 *list = (*list)->next;
448 free(old);
450 while (parents) {
451 struct commit *commit = parents->item;
452 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
453 commit->object.flags |= mark;
454 commit_list_insert_by_date(commit, list);
456 parents = parents->next;
458 return ret;
461 static void clear_commit_marks_1(struct commit_list **plist,
462 struct commit *commit, unsigned int mark)
464 while (commit) {
465 struct commit_list *parents;
467 if (!(mark & commit->object.flags))
468 return;
470 commit->object.flags &= ~mark;
472 parents = commit->parents;
473 if (!parents)
474 return;
476 while ((parents = parents->next))
477 commit_list_insert(parents->item, plist);
479 commit = commit->parents->item;
483 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
485 struct commit_list *list = NULL;
487 while (nr--) {
488 commit_list_insert(*commit, &list);
489 commit++;
491 while (list)
492 clear_commit_marks_1(&list, pop_commit(&list), mark);
495 void clear_commit_marks(struct commit *commit, unsigned int mark)
497 clear_commit_marks_many(1, &commit, mark);
500 void clear_commit_marks_for_object_array(struct object_array *a, unsigned mark)
502 struct object *object;
503 struct commit *commit;
504 unsigned int i;
506 for (i = 0; i < a->nr; i++) {
507 object = a->objects[i].item;
508 commit = lookup_commit_reference_gently(object->sha1, 1);
509 if (commit)
510 clear_commit_marks(commit, mark);
514 struct commit *pop_commit(struct commit_list **stack)
516 struct commit_list *top = *stack;
517 struct commit *item = top ? top->item : NULL;
519 if (top) {
520 *stack = top->next;
521 free(top);
523 return item;
527 * Topological sort support
530 /* count number of children that have not been emitted */
531 define_commit_slab(indegree_slab, int);
533 /* record author-date for each commit object */
534 define_commit_slab(author_date_slab, unsigned long);
536 static void record_author_date(struct author_date_slab *author_date,
537 struct commit *commit)
539 const char *buf, *line_end, *ident_line;
540 char *buffer = NULL;
541 struct ident_split ident;
542 char *date_end;
543 unsigned long date;
545 if (!commit->buffer) {
546 unsigned long size;
547 enum object_type type;
548 buffer = read_sha1_file(commit->object.sha1, &type, &size);
549 if (!buffer)
550 return;
553 for (buf = commit->buffer ? commit->buffer : buffer;
554 buf;
555 buf = line_end + 1) {
556 line_end = strchrnul(buf, '\n');
557 ident_line = skip_prefix(buf, "author ");
558 if (!ident_line) {
559 if (!line_end[0] || line_end[1] == '\n')
560 return; /* end of header */
561 continue;
563 if (split_ident_line(&ident,
564 ident_line, line_end - ident_line) ||
565 !ident.date_begin || !ident.date_end)
566 goto fail_exit; /* malformed "author" line */
567 break;
570 date = strtoul(ident.date_begin, &date_end, 10);
571 if (date_end != ident.date_end)
572 goto fail_exit; /* malformed date */
573 *(author_date_slab_at(author_date, commit)) = date;
575 fail_exit:
576 free(buffer);
579 static int compare_commits_by_author_date(const void *a_, const void *b_,
580 void *cb_data)
582 const struct commit *a = a_, *b = b_;
583 struct author_date_slab *author_date = cb_data;
584 unsigned long a_date = *(author_date_slab_at(author_date, a));
585 unsigned long b_date = *(author_date_slab_at(author_date, b));
587 /* newer commits with larger date first */
588 if (a_date < b_date)
589 return 1;
590 else if (a_date > b_date)
591 return -1;
592 return 0;
595 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
597 const struct commit *a = a_, *b = b_;
598 /* newer commits with larger date first */
599 if (a->date < b->date)
600 return 1;
601 else if (a->date > b->date)
602 return -1;
603 return 0;
607 * Performs an in-place topological sort on the list supplied.
609 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
611 struct commit_list *next, *orig = *list;
612 struct commit_list **pptr;
613 struct indegree_slab indegree;
614 struct prio_queue queue;
615 struct commit *commit;
616 struct author_date_slab author_date;
618 if (!orig)
619 return;
620 *list = NULL;
622 init_indegree_slab(&indegree);
623 memset(&queue, '\0', sizeof(queue));
625 switch (sort_order) {
626 default: /* REV_SORT_IN_GRAPH_ORDER */
627 queue.compare = NULL;
628 break;
629 case REV_SORT_BY_COMMIT_DATE:
630 queue.compare = compare_commits_by_commit_date;
631 break;
632 case REV_SORT_BY_AUTHOR_DATE:
633 init_author_date_slab(&author_date);
634 queue.compare = compare_commits_by_author_date;
635 queue.cb_data = &author_date;
636 break;
639 /* Mark them and clear the indegree */
640 for (next = orig; next; next = next->next) {
641 struct commit *commit = next->item;
642 *(indegree_slab_at(&indegree, commit)) = 1;
643 /* also record the author dates, if needed */
644 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
645 record_author_date(&author_date, commit);
648 /* update the indegree */
649 for (next = orig; next; next = next->next) {
650 struct commit_list *parents = next->item->parents;
651 while (parents) {
652 struct commit *parent = parents->item;
653 int *pi = indegree_slab_at(&indegree, parent);
655 if (*pi)
656 (*pi)++;
657 parents = parents->next;
662 * find the tips
664 * tips are nodes not reachable from any other node in the list
666 * the tips serve as a starting set for the work queue.
668 for (next = orig; next; next = next->next) {
669 struct commit *commit = next->item;
671 if (*(indegree_slab_at(&indegree, commit)) == 1)
672 prio_queue_put(&queue, commit);
676 * This is unfortunate; the initial tips need to be shown
677 * in the order given from the revision traversal machinery.
679 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
680 prio_queue_reverse(&queue);
682 /* We no longer need the commit list */
683 free_commit_list(orig);
685 pptr = list;
686 *list = NULL;
687 while ((commit = prio_queue_get(&queue)) != NULL) {
688 struct commit_list *parents;
690 for (parents = commit->parents; parents ; parents = parents->next) {
691 struct commit *parent = parents->item;
692 int *pi = indegree_slab_at(&indegree, parent);
694 if (!*pi)
695 continue;
698 * parents are only enqueued for emission
699 * when all their children have been emitted thereby
700 * guaranteeing topological order.
702 if (--(*pi) == 1)
703 prio_queue_put(&queue, parent);
706 * all children of commit have already been
707 * emitted. we can emit it now.
709 *(indegree_slab_at(&indegree, commit)) = 0;
711 pptr = &commit_list_insert(commit, pptr)->next;
714 clear_indegree_slab(&indegree);
715 clear_prio_queue(&queue);
716 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
717 clear_author_date_slab(&author_date);
720 /* merge-base stuff */
722 /* Remember to update object flag allocation in object.h */
723 #define PARENT1 (1u<<16)
724 #define PARENT2 (1u<<17)
725 #define STALE (1u<<18)
726 #define RESULT (1u<<19)
728 static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
730 static struct commit *interesting(struct commit_list *list)
732 while (list) {
733 struct commit *commit = list->item;
734 list = list->next;
735 if (commit->object.flags & STALE)
736 continue;
737 return commit;
739 return NULL;
742 /* all input commits in one and twos[] must have been parsed! */
743 static struct commit_list *paint_down_to_common(struct commit *one, int n, struct commit **twos)
745 struct commit_list *list = NULL;
746 struct commit_list *result = NULL;
747 int i;
749 one->object.flags |= PARENT1;
750 commit_list_insert_by_date(one, &list);
751 if (!n)
752 return list;
753 for (i = 0; i < n; i++) {
754 twos[i]->object.flags |= PARENT2;
755 commit_list_insert_by_date(twos[i], &list);
758 while (interesting(list)) {
759 struct commit *commit;
760 struct commit_list *parents;
761 struct commit_list *next;
762 int flags;
764 commit = list->item;
765 next = list->next;
766 free(list);
767 list = next;
769 flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
770 if (flags == (PARENT1 | PARENT2)) {
771 if (!(commit->object.flags & RESULT)) {
772 commit->object.flags |= RESULT;
773 commit_list_insert_by_date(commit, &result);
775 /* Mark parents of a found merge stale */
776 flags |= STALE;
778 parents = commit->parents;
779 while (parents) {
780 struct commit *p = parents->item;
781 parents = parents->next;
782 if ((p->object.flags & flags) == flags)
783 continue;
784 if (parse_commit(p))
785 return NULL;
786 p->object.flags |= flags;
787 commit_list_insert_by_date(p, &list);
791 free_commit_list(list);
792 return result;
795 static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
797 struct commit_list *list = NULL;
798 struct commit_list *result = NULL;
799 int i;
801 for (i = 0; i < n; i++) {
802 if (one == twos[i])
804 * We do not mark this even with RESULT so we do not
805 * have to clean it up.
807 return commit_list_insert(one, &result);
810 if (parse_commit(one))
811 return NULL;
812 for (i = 0; i < n; i++) {
813 if (parse_commit(twos[i]))
814 return NULL;
817 list = paint_down_to_common(one, n, twos);
819 while (list) {
820 struct commit_list *next = list->next;
821 if (!(list->item->object.flags & STALE))
822 commit_list_insert_by_date(list->item, &result);
823 free(list);
824 list = next;
826 return result;
829 struct commit_list *get_octopus_merge_bases(struct commit_list *in)
831 struct commit_list *i, *j, *k, *ret = NULL;
833 if (!in)
834 return ret;
836 commit_list_insert(in->item, &ret);
838 for (i = in->next; i; i = i->next) {
839 struct commit_list *new = NULL, *end = NULL;
841 for (j = ret; j; j = j->next) {
842 struct commit_list *bases;
843 bases = get_merge_bases(i->item, j->item, 1);
844 if (!new)
845 new = bases;
846 else
847 end->next = bases;
848 for (k = bases; k; k = k->next)
849 end = k;
851 ret = new;
853 return ret;
856 static int remove_redundant(struct commit **array, int cnt)
859 * Some commit in the array may be an ancestor of
860 * another commit. Move such commit to the end of
861 * the array, and return the number of commits that
862 * are independent from each other.
864 struct commit **work;
865 unsigned char *redundant;
866 int *filled_index;
867 int i, j, filled;
869 work = xcalloc(cnt, sizeof(*work));
870 redundant = xcalloc(cnt, 1);
871 filled_index = xmalloc(sizeof(*filled_index) * (cnt - 1));
873 for (i = 0; i < cnt; i++)
874 parse_commit(array[i]);
875 for (i = 0; i < cnt; i++) {
876 struct commit_list *common;
878 if (redundant[i])
879 continue;
880 for (j = filled = 0; j < cnt; j++) {
881 if (i == j || redundant[j])
882 continue;
883 filled_index[filled] = j;
884 work[filled++] = array[j];
886 common = paint_down_to_common(array[i], filled, work);
887 if (array[i]->object.flags & PARENT2)
888 redundant[i] = 1;
889 for (j = 0; j < filled; j++)
890 if (work[j]->object.flags & PARENT1)
891 redundant[filled_index[j]] = 1;
892 clear_commit_marks(array[i], all_flags);
893 for (j = 0; j < filled; j++)
894 clear_commit_marks(work[j], all_flags);
895 free_commit_list(common);
898 /* Now collect the result */
899 memcpy(work, array, sizeof(*array) * cnt);
900 for (i = filled = 0; i < cnt; i++)
901 if (!redundant[i])
902 array[filled++] = work[i];
903 for (j = filled, i = 0; i < cnt; i++)
904 if (redundant[i])
905 array[j++] = work[i];
906 free(work);
907 free(redundant);
908 free(filled_index);
909 return filled;
912 struct commit_list *get_merge_bases_many(struct commit *one,
913 int n,
914 struct commit **twos,
915 int cleanup)
917 struct commit_list *list;
918 struct commit **rslt;
919 struct commit_list *result;
920 int cnt, i;
922 result = merge_bases_many(one, n, twos);
923 for (i = 0; i < n; i++) {
924 if (one == twos[i])
925 return result;
927 if (!result || !result->next) {
928 if (cleanup) {
929 clear_commit_marks(one, all_flags);
930 clear_commit_marks_many(n, twos, all_flags);
932 return result;
935 /* There are more than one */
936 cnt = 0;
937 list = result;
938 while (list) {
939 list = list->next;
940 cnt++;
942 rslt = xcalloc(cnt, sizeof(*rslt));
943 for (list = result, i = 0; list; list = list->next)
944 rslt[i++] = list->item;
945 free_commit_list(result);
947 clear_commit_marks(one, all_flags);
948 clear_commit_marks_many(n, twos, all_flags);
950 cnt = remove_redundant(rslt, cnt);
951 result = NULL;
952 for (i = 0; i < cnt; i++)
953 commit_list_insert_by_date(rslt[i], &result);
954 free(rslt);
955 return result;
958 struct commit_list *get_merge_bases(struct commit *one, struct commit *two,
959 int cleanup)
961 return get_merge_bases_many(one, 1, &two, cleanup);
965 * Is "commit" a descendant of one of the elements on the "with_commit" list?
967 int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
969 if (!with_commit)
970 return 1;
971 while (with_commit) {
972 struct commit *other;
974 other = with_commit->item;
975 with_commit = with_commit->next;
976 if (in_merge_bases(other, commit))
977 return 1;
979 return 0;
983 * Is "commit" an ancestor of one of the "references"?
985 int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
987 struct commit_list *bases;
988 int ret = 0, i;
990 if (parse_commit(commit))
991 return ret;
992 for (i = 0; i < nr_reference; i++)
993 if (parse_commit(reference[i]))
994 return ret;
996 bases = paint_down_to_common(commit, nr_reference, reference);
997 if (commit->object.flags & PARENT2)
998 ret = 1;
999 clear_commit_marks(commit, all_flags);
1000 clear_commit_marks_many(nr_reference, reference, all_flags);
1001 free_commit_list(bases);
1002 return ret;
1006 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1008 int in_merge_bases(struct commit *commit, struct commit *reference)
1010 return in_merge_bases_many(commit, 1, &reference);
1013 struct commit_list *reduce_heads(struct commit_list *heads)
1015 struct commit_list *p;
1016 struct commit_list *result = NULL, **tail = &result;
1017 struct commit **array;
1018 int num_head, i;
1020 if (!heads)
1021 return NULL;
1023 /* Uniquify */
1024 for (p = heads; p; p = p->next)
1025 p->item->object.flags &= ~STALE;
1026 for (p = heads, num_head = 0; p; p = p->next) {
1027 if (p->item->object.flags & STALE)
1028 continue;
1029 p->item->object.flags |= STALE;
1030 num_head++;
1032 array = xcalloc(sizeof(*array), num_head);
1033 for (p = heads, i = 0; p; p = p->next) {
1034 if (p->item->object.flags & STALE) {
1035 array[i++] = p->item;
1036 p->item->object.flags &= ~STALE;
1039 num_head = remove_redundant(array, num_head);
1040 for (i = 0; i < num_head; i++)
1041 tail = &commit_list_insert(array[i], tail)->next;
1042 return result;
1045 static const char gpg_sig_header[] = "gpgsig";
1046 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1048 static int do_sign_commit(struct strbuf *buf, const char *keyid)
1050 struct strbuf sig = STRBUF_INIT;
1051 int inspos, copypos;
1053 /* find the end of the header */
1054 inspos = strstr(buf->buf, "\n\n") - buf->buf + 1;
1056 if (!keyid || !*keyid)
1057 keyid = get_signing_key();
1058 if (sign_buffer(buf, &sig, keyid)) {
1059 strbuf_release(&sig);
1060 return -1;
1063 for (copypos = 0; sig.buf[copypos]; ) {
1064 const char *bol = sig.buf + copypos;
1065 const char *eol = strchrnul(bol, '\n');
1066 int len = (eol - bol) + !!*eol;
1068 if (!copypos) {
1069 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1070 inspos += gpg_sig_header_len;
1072 strbuf_insert(buf, inspos++, " ", 1);
1073 strbuf_insert(buf, inspos, bol, len);
1074 inspos += len;
1075 copypos += len;
1077 strbuf_release(&sig);
1078 return 0;
1081 int parse_signed_commit(const unsigned char *sha1,
1082 struct strbuf *payload, struct strbuf *signature)
1084 unsigned long size;
1085 enum object_type type;
1086 char *buffer = read_sha1_file(sha1, &type, &size);
1087 int in_signature, saw_signature = -1;
1088 char *line, *tail;
1090 if (!buffer || type != OBJ_COMMIT)
1091 goto cleanup;
1093 line = buffer;
1094 tail = buffer + size;
1095 in_signature = 0;
1096 saw_signature = 0;
1097 while (line < tail) {
1098 const char *sig = NULL;
1099 char *next = memchr(line, '\n', tail - line);
1101 next = next ? next + 1 : tail;
1102 if (in_signature && line[0] == ' ')
1103 sig = line + 1;
1104 else if (starts_with(line, gpg_sig_header) &&
1105 line[gpg_sig_header_len] == ' ')
1106 sig = line + gpg_sig_header_len + 1;
1107 if (sig) {
1108 strbuf_add(signature, sig, next - sig);
1109 saw_signature = 1;
1110 in_signature = 1;
1111 } else {
1112 if (*line == '\n')
1113 /* dump the whole remainder of the buffer */
1114 next = tail;
1115 strbuf_add(payload, line, next - line);
1116 in_signature = 0;
1118 line = next;
1120 cleanup:
1121 free(buffer);
1122 return saw_signature;
1125 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1127 struct merge_remote_desc *desc;
1128 struct commit_extra_header *mergetag;
1129 char *buf;
1130 unsigned long size, len;
1131 enum object_type type;
1133 desc = merge_remote_util(parent);
1134 if (!desc || !desc->obj)
1135 return;
1136 buf = read_sha1_file(desc->obj->sha1, &type, &size);
1137 if (!buf || type != OBJ_TAG)
1138 goto free_return;
1139 len = parse_signature(buf, size);
1140 if (size == len)
1141 goto free_return;
1143 * We could verify this signature and either omit the tag when
1144 * it does not validate, but the integrator may not have the
1145 * public key of the signer of the tag he is merging, while a
1146 * later auditor may have it while auditing, so let's not run
1147 * verify-signed-buffer here for now...
1149 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1150 * warn("warning: signed tag unverified.");
1152 mergetag = xcalloc(1, sizeof(*mergetag));
1153 mergetag->key = xstrdup("mergetag");
1154 mergetag->value = buf;
1155 mergetag->len = size;
1157 **tail = mergetag;
1158 *tail = &mergetag->next;
1159 return;
1161 free_return:
1162 free(buf);
1165 static struct {
1166 char result;
1167 const char *check;
1168 } sigcheck_gpg_status[] = {
1169 { 'G', "\n[GNUPG:] GOODSIG " },
1170 { 'B', "\n[GNUPG:] BADSIG " },
1171 { 'U', "\n[GNUPG:] TRUST_NEVER" },
1172 { 'U', "\n[GNUPG:] TRUST_UNDEFINED" },
1175 static void parse_gpg_output(struct signature_check *sigc)
1177 const char *buf = sigc->gpg_status;
1178 int i;
1180 /* Iterate over all search strings */
1181 for (i = 0; i < ARRAY_SIZE(sigcheck_gpg_status); i++) {
1182 const char *found, *next;
1184 found = skip_prefix(buf, sigcheck_gpg_status[i].check + 1);
1185 if (!found) {
1186 found = strstr(buf, sigcheck_gpg_status[i].check);
1187 if (!found)
1188 continue;
1189 found += strlen(sigcheck_gpg_status[i].check);
1191 sigc->result = sigcheck_gpg_status[i].result;
1192 /* The trust messages are not followed by key/signer information */
1193 if (sigc->result != 'U') {
1194 sigc->key = xmemdupz(found, 16);
1195 found += 17;
1196 next = strchrnul(found, '\n');
1197 sigc->signer = xmemdupz(found, next - found);
1202 void check_commit_signature(const struct commit* commit, struct signature_check *sigc)
1204 struct strbuf payload = STRBUF_INIT;
1205 struct strbuf signature = STRBUF_INIT;
1206 struct strbuf gpg_output = STRBUF_INIT;
1207 struct strbuf gpg_status = STRBUF_INIT;
1208 int status;
1210 sigc->result = 'N';
1212 if (parse_signed_commit(commit->object.sha1,
1213 &payload, &signature) <= 0)
1214 goto out;
1215 status = verify_signed_buffer(payload.buf, payload.len,
1216 signature.buf, signature.len,
1217 &gpg_output, &gpg_status);
1218 if (status && !gpg_output.len)
1219 goto out;
1220 sigc->gpg_output = strbuf_detach(&gpg_output, NULL);
1221 sigc->gpg_status = strbuf_detach(&gpg_status, NULL);
1222 parse_gpg_output(sigc);
1224 out:
1225 strbuf_release(&gpg_status);
1226 strbuf_release(&gpg_output);
1227 strbuf_release(&payload);
1228 strbuf_release(&signature);
1233 void append_merge_tag_headers(struct commit_list *parents,
1234 struct commit_extra_header ***tail)
1236 while (parents) {
1237 struct commit *parent = parents->item;
1238 handle_signed_tag(parent, tail);
1239 parents = parents->next;
1243 static void add_extra_header(struct strbuf *buffer,
1244 struct commit_extra_header *extra)
1246 strbuf_addstr(buffer, extra->key);
1247 if (extra->len)
1248 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1249 else
1250 strbuf_addch(buffer, '\n');
1253 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1254 const char **exclude)
1256 struct commit_extra_header *extra = NULL;
1257 unsigned long size;
1258 enum object_type type;
1259 char *buffer = read_sha1_file(commit->object.sha1, &type, &size);
1260 if (buffer && type == OBJ_COMMIT)
1261 extra = read_commit_extra_header_lines(buffer, size, exclude);
1262 free(buffer);
1263 return extra;
1266 static inline int standard_header_field(const char *field, size_t len)
1268 return ((len == 4 && !memcmp(field, "tree ", 5)) ||
1269 (len == 6 && !memcmp(field, "parent ", 7)) ||
1270 (len == 6 && !memcmp(field, "author ", 7)) ||
1271 (len == 9 && !memcmp(field, "committer ", 10)) ||
1272 (len == 8 && !memcmp(field, "encoding ", 9)));
1275 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1277 if (!exclude)
1278 return 0;
1280 while (*exclude) {
1281 size_t xlen = strlen(*exclude);
1282 if (len == xlen &&
1283 !memcmp(field, *exclude, xlen) && field[xlen] == ' ')
1284 return 1;
1285 exclude++;
1287 return 0;
1290 static struct commit_extra_header *read_commit_extra_header_lines(
1291 const char *buffer, size_t size,
1292 const char **exclude)
1294 struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1295 const char *line, *next, *eof, *eob;
1296 struct strbuf buf = STRBUF_INIT;
1298 for (line = buffer, eob = line + size;
1299 line < eob && *line != '\n';
1300 line = next) {
1301 next = memchr(line, '\n', eob - line);
1302 next = next ? next + 1 : eob;
1303 if (*line == ' ') {
1304 /* continuation */
1305 if (it)
1306 strbuf_add(&buf, line + 1, next - (line + 1));
1307 continue;
1309 if (it)
1310 it->value = strbuf_detach(&buf, &it->len);
1311 strbuf_reset(&buf);
1312 it = NULL;
1314 eof = strchr(line, ' ');
1315 if (next <= eof)
1316 eof = next;
1318 if (standard_header_field(line, eof - line) ||
1319 excluded_header_field(line, eof - line, exclude))
1320 continue;
1322 it = xcalloc(1, sizeof(*it));
1323 it->key = xmemdupz(line, eof-line);
1324 *tail = it;
1325 tail = &it->next;
1326 if (eof + 1 < next)
1327 strbuf_add(&buf, eof + 1, next - (eof + 1));
1329 if (it)
1330 it->value = strbuf_detach(&buf, &it->len);
1331 return extra;
1334 void free_commit_extra_headers(struct commit_extra_header *extra)
1336 while (extra) {
1337 struct commit_extra_header *next = extra->next;
1338 free(extra->key);
1339 free(extra->value);
1340 free(extra);
1341 extra = next;
1345 int commit_tree(const char *msg, size_t msg_len,
1346 const unsigned char *tree,
1347 struct commit_list *parents, unsigned char *ret,
1348 const char *author, const char *sign_commit)
1350 struct commit_extra_header *extra = NULL, **tail = &extra;
1351 int result;
1353 append_merge_tag_headers(parents, &tail);
1354 result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1355 author, sign_commit, extra);
1356 free_commit_extra_headers(extra);
1357 return result;
1360 static int find_invalid_utf8(const char *buf, int len)
1362 int offset = 0;
1363 static const unsigned int max_codepoint[] = {
1364 0x7f, 0x7ff, 0xffff, 0x10ffff
1367 while (len) {
1368 unsigned char c = *buf++;
1369 int bytes, bad_offset;
1370 unsigned int codepoint;
1371 unsigned int min_val, max_val;
1373 len--;
1374 offset++;
1376 /* Simple US-ASCII? No worries. */
1377 if (c < 0x80)
1378 continue;
1380 bad_offset = offset-1;
1383 * Count how many more high bits set: that's how
1384 * many more bytes this sequence should have.
1386 bytes = 0;
1387 while (c & 0x40) {
1388 c <<= 1;
1389 bytes++;
1393 * Must be between 1 and 3 more bytes. Longer sequences result in
1394 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1396 if (bytes < 1 || 3 < bytes)
1397 return bad_offset;
1399 /* Do we *have* that many bytes? */
1400 if (len < bytes)
1401 return bad_offset;
1404 * Place the encoded bits at the bottom of the value and compute the
1405 * valid range.
1407 codepoint = (c & 0x7f) >> bytes;
1408 min_val = max_codepoint[bytes-1] + 1;
1409 max_val = max_codepoint[bytes];
1411 offset += bytes;
1412 len -= bytes;
1414 /* And verify that they are good continuation bytes */
1415 do {
1416 codepoint <<= 6;
1417 codepoint |= *buf & 0x3f;
1418 if ((*buf++ & 0xc0) != 0x80)
1419 return bad_offset;
1420 } while (--bytes);
1422 /* Reject codepoints that are out of range for the sequence length. */
1423 if (codepoint < min_val || codepoint > max_val)
1424 return bad_offset;
1425 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1426 if ((codepoint & 0x1ff800) == 0xd800)
1427 return bad_offset;
1428 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1429 if ((codepoint & 0xfffe) == 0xfffe)
1430 return bad_offset;
1431 /* So are anything in the range U+FDD0..U+FDEF. */
1432 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1433 return bad_offset;
1435 return -1;
1439 * This verifies that the buffer is in proper utf8 format.
1441 * If it isn't, it assumes any non-utf8 characters are Latin1,
1442 * and does the conversion.
1444 static int verify_utf8(struct strbuf *buf)
1446 int ok = 1;
1447 long pos = 0;
1449 for (;;) {
1450 int bad;
1451 unsigned char c;
1452 unsigned char replace[2];
1454 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1455 if (bad < 0)
1456 return ok;
1457 pos += bad;
1458 ok = 0;
1459 c = buf->buf[pos];
1460 strbuf_remove(buf, pos, 1);
1462 /* We know 'c' must be in the range 128-255 */
1463 replace[0] = 0xc0 + (c >> 6);
1464 replace[1] = 0x80 + (c & 0x3f);
1465 strbuf_insert(buf, pos, replace, 2);
1466 pos += 2;
1470 static const char commit_utf8_warn[] =
1471 "Warning: commit message did not conform to UTF-8.\n"
1472 "You may want to amend it after fixing the message, or set the config\n"
1473 "variable i18n.commitencoding to the encoding your project uses.\n";
1475 int commit_tree_extended(const char *msg, size_t msg_len,
1476 const unsigned char *tree,
1477 struct commit_list *parents, unsigned char *ret,
1478 const char *author, const char *sign_commit,
1479 struct commit_extra_header *extra)
1481 int result;
1482 int encoding_is_utf8;
1483 struct strbuf buffer;
1485 assert_sha1_type(tree, OBJ_TREE);
1487 if (memchr(msg, '\0', msg_len))
1488 return error("a NUL byte in commit log message not allowed.");
1490 /* Not having i18n.commitencoding is the same as having utf-8 */
1491 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1493 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1494 strbuf_addf(&buffer, "tree %s\n", sha1_to_hex(tree));
1497 * NOTE! This ordering means that the same exact tree merged with a
1498 * different order of parents will be a _different_ changeset even
1499 * if everything else stays the same.
1501 while (parents) {
1502 struct commit_list *next = parents->next;
1503 struct commit *parent = parents->item;
1505 strbuf_addf(&buffer, "parent %s\n",
1506 sha1_to_hex(parent->object.sha1));
1507 free(parents);
1508 parents = next;
1511 /* Person/date information */
1512 if (!author)
1513 author = git_author_info(IDENT_STRICT);
1514 strbuf_addf(&buffer, "author %s\n", author);
1515 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1516 if (!encoding_is_utf8)
1517 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1519 while (extra) {
1520 add_extra_header(&buffer, extra);
1521 extra = extra->next;
1523 strbuf_addch(&buffer, '\n');
1525 /* And add the comment */
1526 strbuf_add(&buffer, msg, msg_len);
1528 /* And check the encoding */
1529 if (encoding_is_utf8 && !verify_utf8(&buffer))
1530 fprintf(stderr, commit_utf8_warn);
1532 if (sign_commit && do_sign_commit(&buffer, sign_commit))
1533 return -1;
1535 result = write_sha1_file(buffer.buf, buffer.len, commit_type, ret);
1536 strbuf_release(&buffer);
1537 return result;
1540 struct commit *get_merge_parent(const char *name)
1542 struct object *obj;
1543 struct commit *commit;
1544 unsigned char sha1[20];
1545 if (get_sha1(name, sha1))
1546 return NULL;
1547 obj = parse_object(sha1);
1548 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1549 if (commit && !commit->util) {
1550 struct merge_remote_desc *desc;
1551 desc = xmalloc(sizeof(*desc));
1552 desc->obj = obj;
1553 desc->name = strdup(name);
1554 commit->util = desc;
1556 return commit;
1560 * Append a commit to the end of the commit_list.
1562 * next starts by pointing to the variable that holds the head of an
1563 * empty commit_list, and is updated to point to the "next" field of
1564 * the last item on the list as new commits are appended.
1566 * Usage example:
1568 * struct commit_list *list;
1569 * struct commit_list **next = &list;
1571 * next = commit_list_append(c1, next);
1572 * next = commit_list_append(c2, next);
1573 * assert(commit_list_count(list) == 2);
1574 * return list;
1576 struct commit_list **commit_list_append(struct commit *commit,
1577 struct commit_list **next)
1579 struct commit_list *new = xmalloc(sizeof(struct commit_list));
1580 new->item = commit;
1581 *next = new;
1582 new->next = NULL;
1583 return &new->next;
1586 void print_commit_list(struct commit_list *list,
1587 const char *format_cur,
1588 const char *format_last)
1590 for ( ; list; list = list->next) {
1591 const char *format = list->next ? format_cur : format_last;
1592 printf(format, sha1_to_hex(list->item->object.sha1));