fast-import: refactor parsing of spaces
[git.git] / commit.c
blobdfc0eb02415edccba6195b4d2925196caa242a1a
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";
20 static int commit_count;
22 static struct commit *check_commit(struct object *obj,
23 const unsigned char *sha1,
24 int quiet)
26 if (obj->type != OBJ_COMMIT) {
27 if (!quiet)
28 error("Object %s is a %s, not a commit",
29 sha1_to_hex(sha1), typename(obj->type));
30 return NULL;
32 return (struct commit *) obj;
35 struct commit *lookup_commit_reference_gently(const unsigned char *sha1,
36 int quiet)
38 struct object *obj = deref_tag(parse_object(sha1), NULL, 0);
40 if (!obj)
41 return NULL;
42 return check_commit(obj, sha1, quiet);
45 struct commit *lookup_commit_reference(const unsigned char *sha1)
47 return lookup_commit_reference_gently(sha1, 0);
50 struct commit *lookup_commit_or_die(const unsigned char *sha1, const char *ref_name)
52 struct commit *c = lookup_commit_reference(sha1);
53 if (!c)
54 die(_("could not parse %s"), ref_name);
55 if (hashcmp(sha1, c->object.sha1)) {
56 warning(_("%s %s is not a commit!"),
57 ref_name, sha1_to_hex(sha1));
59 return c;
62 struct commit *lookup_commit(const unsigned char *sha1)
64 struct object *obj = lookup_object(sha1);
65 if (!obj) {
66 struct commit *c = alloc_commit_node();
67 c->index = commit_count++;
68 return create_object(sha1, OBJ_COMMIT, c);
70 if (!obj->type)
71 obj->type = OBJ_COMMIT;
72 return check_commit(obj, sha1, 0);
75 struct commit *lookup_commit_reference_by_name(const char *name)
77 unsigned char sha1[20];
78 struct commit *commit;
80 if (get_sha1_committish(name, sha1))
81 return NULL;
82 commit = lookup_commit_reference(sha1);
83 if (parse_commit(commit))
84 return NULL;
85 return commit;
88 static unsigned long parse_commit_date(const char *buf, const char *tail)
90 const char *dateptr;
92 if (buf + 6 >= tail)
93 return 0;
94 if (memcmp(buf, "author", 6))
95 return 0;
96 while (buf < tail && *buf++ != '\n')
97 /* nada */;
98 if (buf + 9 >= tail)
99 return 0;
100 if (memcmp(buf, "committer", 9))
101 return 0;
102 while (buf < tail && *buf++ != '>')
103 /* nada */;
104 if (buf >= tail)
105 return 0;
106 dateptr = buf;
107 while (buf < tail && *buf++ != '\n')
108 /* nada */;
109 if (buf >= tail)
110 return 0;
111 /* dateptr < buf && buf[-1] == '\n', so strtoul will stop at buf-1 */
112 return strtoul(dateptr, NULL, 10);
115 static struct commit_graft **commit_graft;
116 static int commit_graft_alloc, commit_graft_nr;
118 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
120 struct commit_graft **commit_graft_table = table;
121 return commit_graft_table[index]->sha1;
124 static int commit_graft_pos(const unsigned char *sha1)
126 return sha1_pos(sha1, commit_graft, commit_graft_nr,
127 commit_graft_sha1_access);
130 int register_commit_graft(struct commit_graft *graft, int ignore_dups)
132 int pos = commit_graft_pos(graft->sha1);
134 if (0 <= pos) {
135 if (ignore_dups)
136 free(graft);
137 else {
138 free(commit_graft[pos]);
139 commit_graft[pos] = graft;
141 return 1;
143 pos = -pos - 1;
144 ALLOC_GROW(commit_graft, commit_graft_nr + 1, commit_graft_alloc);
145 commit_graft_nr++;
146 if (pos < commit_graft_nr)
147 memmove(commit_graft + pos + 1,
148 commit_graft + pos,
149 (commit_graft_nr - pos - 1) *
150 sizeof(*commit_graft));
151 commit_graft[pos] = graft;
152 return 0;
155 struct commit_graft *read_graft_line(char *buf, int len)
157 /* The format is just "Commit Parent1 Parent2 ...\n" */
158 int i;
159 struct commit_graft *graft = NULL;
161 while (len && isspace(buf[len-1]))
162 buf[--len] = '\0';
163 if (buf[0] == '#' || buf[0] == '\0')
164 return NULL;
165 if ((len + 1) % 41)
166 goto bad_graft_data;
167 i = (len + 1) / 41 - 1;
168 graft = xmalloc(sizeof(*graft) + 20 * i);
169 graft->nr_parent = i;
170 if (get_sha1_hex(buf, graft->sha1))
171 goto bad_graft_data;
172 for (i = 40; i < len; i += 41) {
173 if (buf[i] != ' ')
174 goto bad_graft_data;
175 if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
176 goto bad_graft_data;
178 return graft;
180 bad_graft_data:
181 error("bad graft data: %s", buf);
182 free(graft);
183 return NULL;
186 static int read_graft_file(const char *graft_file)
188 FILE *fp = fopen(graft_file, "r");
189 struct strbuf buf = STRBUF_INIT;
190 if (!fp)
191 return -1;
192 while (!strbuf_getwholeline(&buf, fp, '\n')) {
193 /* The format is just "Commit Parent1 Parent2 ...\n" */
194 struct commit_graft *graft = read_graft_line(buf.buf, buf.len);
195 if (!graft)
196 continue;
197 if (register_commit_graft(graft, 1))
198 error("duplicate graft data: %s", buf.buf);
200 fclose(fp);
201 strbuf_release(&buf);
202 return 0;
205 static void prepare_commit_graft(void)
207 static int commit_graft_prepared;
208 char *graft_file;
210 if (commit_graft_prepared)
211 return;
212 graft_file = get_graft_file();
213 read_graft_file(graft_file);
214 /* make sure shallows are read */
215 is_repository_shallow();
216 commit_graft_prepared = 1;
219 struct commit_graft *lookup_commit_graft(const unsigned char *sha1)
221 int pos;
222 prepare_commit_graft();
223 pos = commit_graft_pos(sha1);
224 if (pos < 0)
225 return NULL;
226 return commit_graft[pos];
229 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
231 int i, ret;
232 for (i = ret = 0; i < commit_graft_nr && !ret; i++)
233 ret = fn(commit_graft[i], cb_data);
234 return ret;
237 int unregister_shallow(const unsigned char *sha1)
239 int pos = commit_graft_pos(sha1);
240 if (pos < 0)
241 return -1;
242 if (pos + 1 < commit_graft_nr)
243 memmove(commit_graft + pos, commit_graft + pos + 1,
244 sizeof(struct commit_graft *)
245 * (commit_graft_nr - pos - 1));
246 commit_graft_nr--;
247 return 0;
250 int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size)
252 const char *tail = buffer;
253 const char *bufptr = buffer;
254 unsigned char parent[20];
255 struct commit_list **pptr;
256 struct commit_graft *graft;
258 if (item->object.parsed)
259 return 0;
260 item->object.parsed = 1;
261 tail += size;
262 if (tail <= bufptr + 46 || memcmp(bufptr, "tree ", 5) || bufptr[45] != '\n')
263 return error("bogus commit object %s", sha1_to_hex(item->object.sha1));
264 if (get_sha1_hex(bufptr + 5, parent) < 0)
265 return error("bad tree pointer in commit %s",
266 sha1_to_hex(item->object.sha1));
267 item->tree = lookup_tree(parent);
268 bufptr += 46; /* "tree " + "hex sha1" + "\n" */
269 pptr = &item->parents;
271 graft = lookup_commit_graft(item->object.sha1);
272 while (bufptr + 48 < tail && !memcmp(bufptr, "parent ", 7)) {
273 struct commit *new_parent;
275 if (tail <= bufptr + 48 ||
276 get_sha1_hex(bufptr + 7, parent) ||
277 bufptr[47] != '\n')
278 return error("bad parents in commit %s", sha1_to_hex(item->object.sha1));
279 bufptr += 48;
281 * The clone is shallow if nr_parent < 0, and we must
282 * not traverse its real parents even when we unhide them.
284 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
285 continue;
286 new_parent = lookup_commit(parent);
287 if (new_parent)
288 pptr = &commit_list_insert(new_parent, pptr)->next;
290 if (graft) {
291 int i;
292 struct commit *new_parent;
293 for (i = 0; i < graft->nr_parent; i++) {
294 new_parent = lookup_commit(graft->parent[i]);
295 if (!new_parent)
296 continue;
297 pptr = &commit_list_insert(new_parent, pptr)->next;
300 item->date = parse_commit_date(bufptr, tail);
302 return 0;
305 int parse_commit(struct commit *item)
307 enum object_type type;
308 void *buffer;
309 unsigned long size;
310 int ret;
312 if (!item)
313 return -1;
314 if (item->object.parsed)
315 return 0;
316 buffer = read_sha1_file(item->object.sha1, &type, &size);
317 if (!buffer)
318 return error("Could not read %s",
319 sha1_to_hex(item->object.sha1));
320 if (type != OBJ_COMMIT) {
321 free(buffer);
322 return error("Object %s not a commit",
323 sha1_to_hex(item->object.sha1));
325 ret = parse_commit_buffer(item, buffer, size);
326 if (save_commit_buffer && !ret) {
327 item->buffer = buffer;
328 return 0;
330 free(buffer);
331 return ret;
334 void parse_commit_or_die(struct commit *item)
336 if (parse_commit(item))
337 die("unable to parse commit %s",
338 item ? sha1_to_hex(item->object.sha1) : "(null)");
341 int find_commit_subject(const char *commit_buffer, const char **subject)
343 const char *eol;
344 const char *p = commit_buffer;
346 while (*p && (*p != '\n' || p[1] != '\n'))
347 p++;
348 if (*p) {
349 p += 2;
350 for (eol = p; *eol && *eol != '\n'; eol++)
351 ; /* do nothing */
352 } else
353 eol = p;
355 *subject = p;
357 return eol - p;
360 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
362 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
363 new_list->item = item;
364 new_list->next = *list_p;
365 *list_p = new_list;
366 return new_list;
369 unsigned commit_list_count(const struct commit_list *l)
371 unsigned c = 0;
372 for (; l; l = l->next )
373 c++;
374 return c;
377 struct commit_list *copy_commit_list(struct commit_list *list)
379 struct commit_list *head = NULL;
380 struct commit_list **pp = &head;
381 while (list) {
382 struct commit_list *new;
383 new = xmalloc(sizeof(struct commit_list));
384 new->item = list->item;
385 new->next = NULL;
386 *pp = new;
387 pp = &new->next;
388 list = list->next;
390 return head;
393 void free_commit_list(struct commit_list *list)
395 while (list) {
396 struct commit_list *temp = list;
397 list = temp->next;
398 free(temp);
402 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
404 struct commit_list **pp = list;
405 struct commit_list *p;
406 while ((p = *pp) != NULL) {
407 if (p->item->date < item->date) {
408 break;
410 pp = &p->next;
412 return commit_list_insert(item, pp);
415 static int commit_list_compare_by_date(const void *a, const void *b)
417 unsigned long a_date = ((const struct commit_list *)a)->item->date;
418 unsigned long b_date = ((const struct commit_list *)b)->item->date;
419 if (a_date < b_date)
420 return 1;
421 if (a_date > b_date)
422 return -1;
423 return 0;
426 static void *commit_list_get_next(const void *a)
428 return ((const struct commit_list *)a)->next;
431 static void commit_list_set_next(void *a, void *next)
433 ((struct commit_list *)a)->next = next;
436 void commit_list_sort_by_date(struct commit_list **list)
438 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
439 commit_list_compare_by_date);
442 struct commit *pop_most_recent_commit(struct commit_list **list,
443 unsigned int mark)
445 struct commit *ret = (*list)->item;
446 struct commit_list *parents = ret->parents;
447 struct commit_list *old = *list;
449 *list = (*list)->next;
450 free(old);
452 while (parents) {
453 struct commit *commit = parents->item;
454 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
455 commit->object.flags |= mark;
456 commit_list_insert_by_date(commit, list);
458 parents = parents->next;
460 return ret;
463 static void clear_commit_marks_1(struct commit_list **plist,
464 struct commit *commit, unsigned int mark)
466 while (commit) {
467 struct commit_list *parents;
469 if (!(mark & commit->object.flags))
470 return;
472 commit->object.flags &= ~mark;
474 parents = commit->parents;
475 if (!parents)
476 return;
478 while ((parents = parents->next))
479 commit_list_insert(parents->item, plist);
481 commit = commit->parents->item;
485 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
487 struct commit_list *list = NULL;
489 while (nr--) {
490 commit_list_insert(*commit, &list);
491 commit++;
493 while (list)
494 clear_commit_marks_1(&list, pop_commit(&list), mark);
497 void clear_commit_marks(struct commit *commit, unsigned int mark)
499 clear_commit_marks_many(1, &commit, mark);
502 void clear_commit_marks_for_object_array(struct object_array *a, unsigned mark)
504 struct object *object;
505 struct commit *commit;
506 unsigned int i;
508 for (i = 0; i < a->nr; i++) {
509 object = a->objects[i].item;
510 commit = lookup_commit_reference_gently(object->sha1, 1);
511 if (commit)
512 clear_commit_marks(commit, mark);
516 struct commit *pop_commit(struct commit_list **stack)
518 struct commit_list *top = *stack;
519 struct commit *item = top ? top->item : NULL;
521 if (top) {
522 *stack = top->next;
523 free(top);
525 return item;
529 * Topological sort support
532 /* count number of children that have not been emitted */
533 define_commit_slab(indegree_slab, int);
535 /* record author-date for each commit object */
536 define_commit_slab(author_date_slab, unsigned long);
538 static void record_author_date(struct author_date_slab *author_date,
539 struct commit *commit)
541 const char *buf, *line_end, *ident_line;
542 char *buffer = NULL;
543 struct ident_split ident;
544 char *date_end;
545 unsigned long date;
547 if (!commit->buffer) {
548 unsigned long size;
549 enum object_type type;
550 buffer = read_sha1_file(commit->object.sha1, &type, &size);
551 if (!buffer)
552 return;
555 for (buf = commit->buffer ? commit->buffer : buffer;
556 buf;
557 buf = line_end + 1) {
558 line_end = strchrnul(buf, '\n');
559 if (!skip_prefix(buf, "author ", &ident_line)) {
560 if (!line_end[0] || line_end[1] == '\n')
561 return; /* end of header */
562 continue;
564 if (split_ident_line(&ident,
565 ident_line, line_end - ident_line) ||
566 !ident.date_begin || !ident.date_end)
567 goto fail_exit; /* malformed "author" line */
568 break;
571 date = strtoul(ident.date_begin, &date_end, 10);
572 if (date_end != ident.date_end)
573 goto fail_exit; /* malformed date */
574 *(author_date_slab_at(author_date, commit)) = date;
576 fail_exit:
577 free(buffer);
580 static int compare_commits_by_author_date(const void *a_, const void *b_,
581 void *cb_data)
583 const struct commit *a = a_, *b = b_;
584 struct author_date_slab *author_date = cb_data;
585 unsigned long a_date = *(author_date_slab_at(author_date, a));
586 unsigned long b_date = *(author_date_slab_at(author_date, b));
588 /* newer commits with larger date first */
589 if (a_date < b_date)
590 return 1;
591 else if (a_date > b_date)
592 return -1;
593 return 0;
596 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
598 const struct commit *a = a_, *b = b_;
599 /* newer commits with larger date first */
600 if (a->date < b->date)
601 return 1;
602 else if (a->date > b->date)
603 return -1;
604 return 0;
608 * Performs an in-place topological sort on the list supplied.
610 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
612 struct commit_list *next, *orig = *list;
613 struct commit_list **pptr;
614 struct indegree_slab indegree;
615 struct prio_queue queue;
616 struct commit *commit;
617 struct author_date_slab author_date;
619 if (!orig)
620 return;
621 *list = NULL;
623 init_indegree_slab(&indegree);
624 memset(&queue, '\0', sizeof(queue));
626 switch (sort_order) {
627 default: /* REV_SORT_IN_GRAPH_ORDER */
628 queue.compare = NULL;
629 break;
630 case REV_SORT_BY_COMMIT_DATE:
631 queue.compare = compare_commits_by_commit_date;
632 break;
633 case REV_SORT_BY_AUTHOR_DATE:
634 init_author_date_slab(&author_date);
635 queue.compare = compare_commits_by_author_date;
636 queue.cb_data = &author_date;
637 break;
640 /* Mark them and clear the indegree */
641 for (next = orig; next; next = next->next) {
642 struct commit *commit = next->item;
643 *(indegree_slab_at(&indegree, commit)) = 1;
644 /* also record the author dates, if needed */
645 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
646 record_author_date(&author_date, commit);
649 /* update the indegree */
650 for (next = orig; next; next = next->next) {
651 struct commit_list *parents = next->item->parents;
652 while (parents) {
653 struct commit *parent = parents->item;
654 int *pi = indegree_slab_at(&indegree, parent);
656 if (*pi)
657 (*pi)++;
658 parents = parents->next;
663 * find the tips
665 * tips are nodes not reachable from any other node in the list
667 * the tips serve as a starting set for the work queue.
669 for (next = orig; next; next = next->next) {
670 struct commit *commit = next->item;
672 if (*(indegree_slab_at(&indegree, commit)) == 1)
673 prio_queue_put(&queue, commit);
677 * This is unfortunate; the initial tips need to be shown
678 * in the order given from the revision traversal machinery.
680 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
681 prio_queue_reverse(&queue);
683 /* We no longer need the commit list */
684 free_commit_list(orig);
686 pptr = list;
687 *list = NULL;
688 while ((commit = prio_queue_get(&queue)) != NULL) {
689 struct commit_list *parents;
691 for (parents = commit->parents; parents ; parents = parents->next) {
692 struct commit *parent = parents->item;
693 int *pi = indegree_slab_at(&indegree, parent);
695 if (!*pi)
696 continue;
699 * parents are only enqueued for emission
700 * when all their children have been emitted thereby
701 * guaranteeing topological order.
703 if (--(*pi) == 1)
704 prio_queue_put(&queue, parent);
707 * all children of commit have already been
708 * emitted. we can emit it now.
710 *(indegree_slab_at(&indegree, commit)) = 0;
712 pptr = &commit_list_insert(commit, pptr)->next;
715 clear_indegree_slab(&indegree);
716 clear_prio_queue(&queue);
717 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
718 clear_author_date_slab(&author_date);
721 /* merge-base stuff */
723 /* Remember to update object flag allocation in object.h */
724 #define PARENT1 (1u<<16)
725 #define PARENT2 (1u<<17)
726 #define STALE (1u<<18)
727 #define RESULT (1u<<19)
729 static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
731 static struct commit *interesting(struct commit_list *list)
733 while (list) {
734 struct commit *commit = list->item;
735 list = list->next;
736 if (commit->object.flags & STALE)
737 continue;
738 return commit;
740 return NULL;
743 /* all input commits in one and twos[] must have been parsed! */
744 static struct commit_list *paint_down_to_common(struct commit *one, int n, struct commit **twos)
746 struct commit_list *list = NULL;
747 struct commit_list *result = NULL;
748 int i;
750 one->object.flags |= PARENT1;
751 commit_list_insert_by_date(one, &list);
752 if (!n)
753 return list;
754 for (i = 0; i < n; i++) {
755 twos[i]->object.flags |= PARENT2;
756 commit_list_insert_by_date(twos[i], &list);
759 while (interesting(list)) {
760 struct commit *commit;
761 struct commit_list *parents;
762 struct commit_list *next;
763 int flags;
765 commit = list->item;
766 next = list->next;
767 free(list);
768 list = next;
770 flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
771 if (flags == (PARENT1 | PARENT2)) {
772 if (!(commit->object.flags & RESULT)) {
773 commit->object.flags |= RESULT;
774 commit_list_insert_by_date(commit, &result);
776 /* Mark parents of a found merge stale */
777 flags |= STALE;
779 parents = commit->parents;
780 while (parents) {
781 struct commit *p = parents->item;
782 parents = parents->next;
783 if ((p->object.flags & flags) == flags)
784 continue;
785 if (parse_commit(p))
786 return NULL;
787 p->object.flags |= flags;
788 commit_list_insert_by_date(p, &list);
792 free_commit_list(list);
793 return result;
796 static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
798 struct commit_list *list = NULL;
799 struct commit_list *result = NULL;
800 int i;
802 for (i = 0; i < n; i++) {
803 if (one == twos[i])
805 * We do not mark this even with RESULT so we do not
806 * have to clean it up.
808 return commit_list_insert(one, &result);
811 if (parse_commit(one))
812 return NULL;
813 for (i = 0; i < n; i++) {
814 if (parse_commit(twos[i]))
815 return NULL;
818 list = paint_down_to_common(one, n, twos);
820 while (list) {
821 struct commit_list *next = list->next;
822 if (!(list->item->object.flags & STALE))
823 commit_list_insert_by_date(list->item, &result);
824 free(list);
825 list = next;
827 return result;
830 struct commit_list *get_octopus_merge_bases(struct commit_list *in)
832 struct commit_list *i, *j, *k, *ret = NULL;
834 if (!in)
835 return ret;
837 commit_list_insert(in->item, &ret);
839 for (i = in->next; i; i = i->next) {
840 struct commit_list *new = NULL, *end = NULL;
842 for (j = ret; j; j = j->next) {
843 struct commit_list *bases;
844 bases = get_merge_bases(i->item, j->item, 1);
845 if (!new)
846 new = bases;
847 else
848 end->next = bases;
849 for (k = bases; k; k = k->next)
850 end = k;
852 ret = new;
854 return ret;
857 static int remove_redundant(struct commit **array, int cnt)
860 * Some commit in the array may be an ancestor of
861 * another commit. Move such commit to the end of
862 * the array, and return the number of commits that
863 * are independent from each other.
865 struct commit **work;
866 unsigned char *redundant;
867 int *filled_index;
868 int i, j, filled;
870 work = xcalloc(cnt, sizeof(*work));
871 redundant = xcalloc(cnt, 1);
872 filled_index = xmalloc(sizeof(*filled_index) * (cnt - 1));
874 for (i = 0; i < cnt; i++)
875 parse_commit(array[i]);
876 for (i = 0; i < cnt; i++) {
877 struct commit_list *common;
879 if (redundant[i])
880 continue;
881 for (j = filled = 0; j < cnt; j++) {
882 if (i == j || redundant[j])
883 continue;
884 filled_index[filled] = j;
885 work[filled++] = array[j];
887 common = paint_down_to_common(array[i], filled, work);
888 if (array[i]->object.flags & PARENT2)
889 redundant[i] = 1;
890 for (j = 0; j < filled; j++)
891 if (work[j]->object.flags & PARENT1)
892 redundant[filled_index[j]] = 1;
893 clear_commit_marks(array[i], all_flags);
894 for (j = 0; j < filled; j++)
895 clear_commit_marks(work[j], all_flags);
896 free_commit_list(common);
899 /* Now collect the result */
900 memcpy(work, array, sizeof(*array) * cnt);
901 for (i = filled = 0; i < cnt; i++)
902 if (!redundant[i])
903 array[filled++] = work[i];
904 for (j = filled, i = 0; i < cnt; i++)
905 if (redundant[i])
906 array[j++] = work[i];
907 free(work);
908 free(redundant);
909 free(filled_index);
910 return filled;
913 struct commit_list *get_merge_bases_many(struct commit *one,
914 int n,
915 struct commit **twos,
916 int cleanup)
918 struct commit_list *list;
919 struct commit **rslt;
920 struct commit_list *result;
921 int cnt, i;
923 result = merge_bases_many(one, n, twos);
924 for (i = 0; i < n; i++) {
925 if (one == twos[i])
926 return result;
928 if (!result || !result->next) {
929 if (cleanup) {
930 clear_commit_marks(one, all_flags);
931 clear_commit_marks_many(n, twos, all_flags);
933 return result;
936 /* There are more than one */
937 cnt = 0;
938 list = result;
939 while (list) {
940 list = list->next;
941 cnt++;
943 rslt = xcalloc(cnt, sizeof(*rslt));
944 for (list = result, i = 0; list; list = list->next)
945 rslt[i++] = list->item;
946 free_commit_list(result);
948 clear_commit_marks(one, all_flags);
949 clear_commit_marks_many(n, twos, all_flags);
951 cnt = remove_redundant(rslt, cnt);
952 result = NULL;
953 for (i = 0; i < cnt; i++)
954 commit_list_insert_by_date(rslt[i], &result);
955 free(rslt);
956 return result;
959 struct commit_list *get_merge_bases(struct commit *one, struct commit *two,
960 int cleanup)
962 return get_merge_bases_many(one, 1, &two, cleanup);
966 * Is "commit" a descendant of one of the elements on the "with_commit" list?
968 int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
970 if (!with_commit)
971 return 1;
972 while (with_commit) {
973 struct commit *other;
975 other = with_commit->item;
976 with_commit = with_commit->next;
977 if (in_merge_bases(other, commit))
978 return 1;
980 return 0;
984 * Is "commit" an ancestor of one of the "references"?
986 int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
988 struct commit_list *bases;
989 int ret = 0, i;
991 if (parse_commit(commit))
992 return ret;
993 for (i = 0; i < nr_reference; i++)
994 if (parse_commit(reference[i]))
995 return ret;
997 bases = paint_down_to_common(commit, nr_reference, reference);
998 if (commit->object.flags & PARENT2)
999 ret = 1;
1000 clear_commit_marks(commit, all_flags);
1001 clear_commit_marks_many(nr_reference, reference, all_flags);
1002 free_commit_list(bases);
1003 return ret;
1007 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1009 int in_merge_bases(struct commit *commit, struct commit *reference)
1011 return in_merge_bases_many(commit, 1, &reference);
1014 struct commit_list *reduce_heads(struct commit_list *heads)
1016 struct commit_list *p;
1017 struct commit_list *result = NULL, **tail = &result;
1018 struct commit **array;
1019 int num_head, i;
1021 if (!heads)
1022 return NULL;
1024 /* Uniquify */
1025 for (p = heads; p; p = p->next)
1026 p->item->object.flags &= ~STALE;
1027 for (p = heads, num_head = 0; p; p = p->next) {
1028 if (p->item->object.flags & STALE)
1029 continue;
1030 p->item->object.flags |= STALE;
1031 num_head++;
1033 array = xcalloc(num_head, sizeof(*array));
1034 for (p = heads, i = 0; p; p = p->next) {
1035 if (p->item->object.flags & STALE) {
1036 array[i++] = p->item;
1037 p->item->object.flags &= ~STALE;
1040 num_head = remove_redundant(array, num_head);
1041 for (i = 0; i < num_head; i++)
1042 tail = &commit_list_insert(array[i], tail)->next;
1043 return result;
1046 static const char gpg_sig_header[] = "gpgsig";
1047 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1049 static int do_sign_commit(struct strbuf *buf, const char *keyid)
1051 struct strbuf sig = STRBUF_INIT;
1052 int inspos, copypos;
1054 /* find the end of the header */
1055 inspos = strstr(buf->buf, "\n\n") - buf->buf + 1;
1057 if (!keyid || !*keyid)
1058 keyid = get_signing_key();
1059 if (sign_buffer(buf, &sig, keyid)) {
1060 strbuf_release(&sig);
1061 return -1;
1064 for (copypos = 0; sig.buf[copypos]; ) {
1065 const char *bol = sig.buf + copypos;
1066 const char *eol = strchrnul(bol, '\n');
1067 int len = (eol - bol) + !!*eol;
1069 if (!copypos) {
1070 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1071 inspos += gpg_sig_header_len;
1073 strbuf_insert(buf, inspos++, " ", 1);
1074 strbuf_insert(buf, inspos, bol, len);
1075 inspos += len;
1076 copypos += len;
1078 strbuf_release(&sig);
1079 return 0;
1082 int parse_signed_commit(const unsigned char *sha1,
1083 struct strbuf *payload, struct strbuf *signature)
1085 unsigned long size;
1086 enum object_type type;
1087 char *buffer = read_sha1_file(sha1, &type, &size);
1088 int in_signature, saw_signature = -1;
1089 char *line, *tail;
1091 if (!buffer || type != OBJ_COMMIT)
1092 goto cleanup;
1094 line = buffer;
1095 tail = buffer + size;
1096 in_signature = 0;
1097 saw_signature = 0;
1098 while (line < tail) {
1099 const char *sig = NULL;
1100 char *next = memchr(line, '\n', tail - line);
1102 next = next ? next + 1 : tail;
1103 if (in_signature && line[0] == ' ')
1104 sig = line + 1;
1105 else if (starts_with(line, gpg_sig_header) &&
1106 line[gpg_sig_header_len] == ' ')
1107 sig = line + gpg_sig_header_len + 1;
1108 if (sig) {
1109 strbuf_add(signature, sig, next - sig);
1110 saw_signature = 1;
1111 in_signature = 1;
1112 } else {
1113 if (*line == '\n')
1114 /* dump the whole remainder of the buffer */
1115 next = tail;
1116 strbuf_add(payload, line, next - line);
1117 in_signature = 0;
1119 line = next;
1121 cleanup:
1122 free(buffer);
1123 return saw_signature;
1126 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1128 struct merge_remote_desc *desc;
1129 struct commit_extra_header *mergetag;
1130 char *buf;
1131 unsigned long size, len;
1132 enum object_type type;
1134 desc = merge_remote_util(parent);
1135 if (!desc || !desc->obj)
1136 return;
1137 buf = read_sha1_file(desc->obj->sha1, &type, &size);
1138 if (!buf || type != OBJ_TAG)
1139 goto free_return;
1140 len = parse_signature(buf, size);
1141 if (size == len)
1142 goto free_return;
1144 * We could verify this signature and either omit the tag when
1145 * it does not validate, but the integrator may not have the
1146 * public key of the signer of the tag he is merging, while a
1147 * later auditor may have it while auditing, so let's not run
1148 * verify-signed-buffer here for now...
1150 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1151 * warn("warning: signed tag unverified.");
1153 mergetag = xcalloc(1, sizeof(*mergetag));
1154 mergetag->key = xstrdup("mergetag");
1155 mergetag->value = buf;
1156 mergetag->len = size;
1158 **tail = mergetag;
1159 *tail = &mergetag->next;
1160 return;
1162 free_return:
1163 free(buf);
1166 static struct {
1167 char result;
1168 const char *check;
1169 } sigcheck_gpg_status[] = {
1170 { 'G', "\n[GNUPG:] GOODSIG " },
1171 { 'B', "\n[GNUPG:] BADSIG " },
1172 { 'U', "\n[GNUPG:] TRUST_NEVER" },
1173 { 'U', "\n[GNUPG:] TRUST_UNDEFINED" },
1176 static void parse_gpg_output(struct signature_check *sigc)
1178 const char *buf = sigc->gpg_status;
1179 int i;
1181 /* Iterate over all search strings */
1182 for (i = 0; i < ARRAY_SIZE(sigcheck_gpg_status); i++) {
1183 const char *found, *next;
1185 if (!skip_prefix(buf, sigcheck_gpg_status[i].check + 1, &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 struct strbuf *msg, const unsigned char *tree,
1346 struct commit_list *parents, unsigned char *ret,
1347 const char *author, const char *sign_commit)
1349 struct commit_extra_header *extra = NULL, **tail = &extra;
1350 int result;
1352 append_merge_tag_headers(parents, &tail);
1353 result = commit_tree_extended(msg, tree, parents, ret,
1354 author, sign_commit, extra);
1355 free_commit_extra_headers(extra);
1356 return result;
1359 static int find_invalid_utf8(const char *buf, int len)
1361 int offset = 0;
1362 static const unsigned int max_codepoint[] = {
1363 0x7f, 0x7ff, 0xffff, 0x10ffff
1366 while (len) {
1367 unsigned char c = *buf++;
1368 int bytes, bad_offset;
1369 unsigned int codepoint;
1370 unsigned int min_val, max_val;
1372 len--;
1373 offset++;
1375 /* Simple US-ASCII? No worries. */
1376 if (c < 0x80)
1377 continue;
1379 bad_offset = offset-1;
1382 * Count how many more high bits set: that's how
1383 * many more bytes this sequence should have.
1385 bytes = 0;
1386 while (c & 0x40) {
1387 c <<= 1;
1388 bytes++;
1392 * Must be between 1 and 3 more bytes. Longer sequences result in
1393 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1395 if (bytes < 1 || 3 < bytes)
1396 return bad_offset;
1398 /* Do we *have* that many bytes? */
1399 if (len < bytes)
1400 return bad_offset;
1403 * Place the encoded bits at the bottom of the value and compute the
1404 * valid range.
1406 codepoint = (c & 0x7f) >> bytes;
1407 min_val = max_codepoint[bytes-1] + 1;
1408 max_val = max_codepoint[bytes];
1410 offset += bytes;
1411 len -= bytes;
1413 /* And verify that they are good continuation bytes */
1414 do {
1415 codepoint <<= 6;
1416 codepoint |= *buf & 0x3f;
1417 if ((*buf++ & 0xc0) != 0x80)
1418 return bad_offset;
1419 } while (--bytes);
1421 /* Reject codepoints that are out of range for the sequence length. */
1422 if (codepoint < min_val || codepoint > max_val)
1423 return bad_offset;
1424 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1425 if ((codepoint & 0x1ff800) == 0xd800)
1426 return bad_offset;
1427 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1428 if ((codepoint & 0xfffe) == 0xfffe)
1429 return bad_offset;
1430 /* So are anything in the range U+FDD0..U+FDEF. */
1431 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1432 return bad_offset;
1434 return -1;
1438 * This verifies that the buffer is in proper utf8 format.
1440 * If it isn't, it assumes any non-utf8 characters are Latin1,
1441 * and does the conversion.
1443 static int verify_utf8(struct strbuf *buf)
1445 int ok = 1;
1446 long pos = 0;
1448 for (;;) {
1449 int bad;
1450 unsigned char c;
1451 unsigned char replace[2];
1453 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1454 if (bad < 0)
1455 return ok;
1456 pos += bad;
1457 ok = 0;
1458 c = buf->buf[pos];
1459 strbuf_remove(buf, pos, 1);
1461 /* We know 'c' must be in the range 128-255 */
1462 replace[0] = 0xc0 + (c >> 6);
1463 replace[1] = 0x80 + (c & 0x3f);
1464 strbuf_insert(buf, pos, replace, 2);
1465 pos += 2;
1469 static const char commit_utf8_warn[] =
1470 "Warning: commit message did not conform to UTF-8.\n"
1471 "You may want to amend it after fixing the message, or set the config\n"
1472 "variable i18n.commitencoding to the encoding your project uses.\n";
1474 int commit_tree_extended(const struct strbuf *msg, const unsigned char *tree,
1475 struct commit_list *parents, unsigned char *ret,
1476 const char *author, const char *sign_commit,
1477 struct commit_extra_header *extra)
1479 int result;
1480 int encoding_is_utf8;
1481 struct strbuf buffer;
1483 assert_sha1_type(tree, OBJ_TREE);
1485 if (memchr(msg->buf, '\0', msg->len))
1486 return error("a NUL byte in commit log message not allowed.");
1488 /* Not having i18n.commitencoding is the same as having utf-8 */
1489 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1491 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1492 strbuf_addf(&buffer, "tree %s\n", sha1_to_hex(tree));
1495 * NOTE! This ordering means that the same exact tree merged with a
1496 * different order of parents will be a _different_ changeset even
1497 * if everything else stays the same.
1499 while (parents) {
1500 struct commit_list *next = parents->next;
1501 struct commit *parent = parents->item;
1503 strbuf_addf(&buffer, "parent %s\n",
1504 sha1_to_hex(parent->object.sha1));
1505 free(parents);
1506 parents = next;
1509 /* Person/date information */
1510 if (!author)
1511 author = git_author_info(IDENT_STRICT);
1512 strbuf_addf(&buffer, "author %s\n", author);
1513 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1514 if (!encoding_is_utf8)
1515 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1517 while (extra) {
1518 add_extra_header(&buffer, extra);
1519 extra = extra->next;
1521 strbuf_addch(&buffer, '\n');
1523 /* And add the comment */
1524 strbuf_addbuf(&buffer, msg);
1526 /* And check the encoding */
1527 if (encoding_is_utf8 && !verify_utf8(&buffer))
1528 fprintf(stderr, commit_utf8_warn);
1530 if (sign_commit && do_sign_commit(&buffer, sign_commit))
1531 return -1;
1533 result = write_sha1_file(buffer.buf, buffer.len, commit_type, ret);
1534 strbuf_release(&buffer);
1535 return result;
1538 struct commit *get_merge_parent(const char *name)
1540 struct object *obj;
1541 struct commit *commit;
1542 unsigned char sha1[20];
1543 if (get_sha1(name, sha1))
1544 return NULL;
1545 obj = parse_object(sha1);
1546 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1547 if (commit && !commit->util) {
1548 struct merge_remote_desc *desc;
1549 desc = xmalloc(sizeof(*desc));
1550 desc->obj = obj;
1551 desc->name = strdup(name);
1552 commit->util = desc;
1554 return commit;
1558 * Append a commit to the end of the commit_list.
1560 * next starts by pointing to the variable that holds the head of an
1561 * empty commit_list, and is updated to point to the "next" field of
1562 * the last item on the list as new commits are appended.
1564 * Usage example:
1566 * struct commit_list *list;
1567 * struct commit_list **next = &list;
1569 * next = commit_list_append(c1, next);
1570 * next = commit_list_append(c2, next);
1571 * assert(commit_list_count(list) == 2);
1572 * return list;
1574 struct commit_list **commit_list_append(struct commit *commit,
1575 struct commit_list **next)
1577 struct commit_list *new = xmalloc(sizeof(struct commit_list));
1578 new->item = commit;
1579 *next = new;
1580 new->next = NULL;
1581 return &new->next;
1584 void print_commit_list(struct commit_list *list,
1585 const char *format_cur,
1586 const char *format_last)
1588 for ( ; list; list = list->next) {
1589 const char *format = list->next ? format_cur : format_last;
1590 printf(format, sha1_to_hex(list->item->object.sha1));