assume parse_commit checks for NULL commit
[git/jrn.git] / commit.c
blob8535e5cf35bfabe4a9dd0dc19e1caf6fec93531a
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"
14 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
16 int save_commit_buffer = 1;
18 const char *commit_type = "commit";
19 static int commit_count;
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 c->index = commit_count++;
67 return create_object(sha1, OBJ_COMMIT, c);
69 if (!obj->type)
70 obj->type = OBJ_COMMIT;
71 return check_commit(obj, sha1, 0);
74 struct commit *lookup_commit_reference_by_name(const char *name)
76 unsigned char sha1[20];
77 struct commit *commit;
79 if (get_sha1_committish(name, sha1))
80 return NULL;
81 commit = lookup_commit_reference(sha1);
82 if (parse_commit(commit))
83 return NULL;
84 return commit;
87 static unsigned long parse_commit_date(const char *buf, const char *tail)
89 const char *dateptr;
91 if (buf + 6 >= tail)
92 return 0;
93 if (memcmp(buf, "author", 6))
94 return 0;
95 while (buf < tail && *buf++ != '\n')
96 /* nada */;
97 if (buf + 9 >= tail)
98 return 0;
99 if (memcmp(buf, "committer", 9))
100 return 0;
101 while (buf < tail && *buf++ != '>')
102 /* nada */;
103 if (buf >= tail)
104 return 0;
105 dateptr = buf;
106 while (buf < tail && *buf++ != '\n')
107 /* nada */;
108 if (buf >= tail)
109 return 0;
110 /* dateptr < buf && buf[-1] == '\n', so strtoul will stop at buf-1 */
111 return strtoul(dateptr, NULL, 10);
114 static struct commit_graft **commit_graft;
115 static int commit_graft_alloc, commit_graft_nr;
117 static int commit_graft_pos(const unsigned char *sha1)
119 int lo, hi;
120 lo = 0;
121 hi = commit_graft_nr;
122 while (lo < hi) {
123 int mi = (lo + hi) / 2;
124 struct commit_graft *graft = commit_graft[mi];
125 int cmp = hashcmp(sha1, graft->sha1);
126 if (!cmp)
127 return mi;
128 if (cmp < 0)
129 hi = mi;
130 else
131 lo = mi + 1;
133 return -lo - 1;
136 int register_commit_graft(struct commit_graft *graft, int ignore_dups)
138 int pos = commit_graft_pos(graft->sha1);
140 if (0 <= pos) {
141 if (ignore_dups)
142 free(graft);
143 else {
144 free(commit_graft[pos]);
145 commit_graft[pos] = graft;
147 return 1;
149 pos = -pos - 1;
150 if (commit_graft_alloc <= ++commit_graft_nr) {
151 commit_graft_alloc = alloc_nr(commit_graft_alloc);
152 commit_graft = xrealloc(commit_graft,
153 sizeof(*commit_graft) *
154 commit_graft_alloc);
156 if (pos < commit_graft_nr)
157 memmove(commit_graft + pos + 1,
158 commit_graft + pos,
159 (commit_graft_nr - pos - 1) *
160 sizeof(*commit_graft));
161 commit_graft[pos] = graft;
162 return 0;
165 struct commit_graft *read_graft_line(char *buf, int len)
167 /* The format is just "Commit Parent1 Parent2 ...\n" */
168 int i;
169 struct commit_graft *graft = NULL;
171 while (len && isspace(buf[len-1]))
172 buf[--len] = '\0';
173 if (buf[0] == '#' || buf[0] == '\0')
174 return NULL;
175 if ((len + 1) % 41)
176 goto bad_graft_data;
177 i = (len + 1) / 41 - 1;
178 graft = xmalloc(sizeof(*graft) + 20 * i);
179 graft->nr_parent = i;
180 if (get_sha1_hex(buf, graft->sha1))
181 goto bad_graft_data;
182 for (i = 40; i < len; i += 41) {
183 if (buf[i] != ' ')
184 goto bad_graft_data;
185 if (get_sha1_hex(buf + i + 1, graft->parent[i/41]))
186 goto bad_graft_data;
188 return graft;
190 bad_graft_data:
191 error("bad graft data: %s", buf);
192 free(graft);
193 return NULL;
196 static int read_graft_file(const char *graft_file)
198 FILE *fp = fopen(graft_file, "r");
199 char buf[1024];
200 if (!fp)
201 return -1;
202 while (fgets(buf, sizeof(buf), fp)) {
203 /* The format is just "Commit Parent1 Parent2 ...\n" */
204 int len = strlen(buf);
205 struct commit_graft *graft = read_graft_line(buf, len);
206 if (!graft)
207 continue;
208 if (register_commit_graft(graft, 1))
209 error("duplicate graft data: %s", buf);
211 fclose(fp);
212 return 0;
215 static void prepare_commit_graft(void)
217 static int commit_graft_prepared;
218 char *graft_file;
220 if (commit_graft_prepared)
221 return;
222 graft_file = get_graft_file();
223 read_graft_file(graft_file);
224 /* make sure shallows are read */
225 is_repository_shallow();
226 commit_graft_prepared = 1;
229 struct commit_graft *lookup_commit_graft(const unsigned char *sha1)
231 int pos;
232 prepare_commit_graft();
233 pos = commit_graft_pos(sha1);
234 if (pos < 0)
235 return NULL;
236 return commit_graft[pos];
239 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
241 int i, ret;
242 for (i = ret = 0; i < commit_graft_nr && !ret; i++)
243 ret = fn(commit_graft[i], cb_data);
244 return ret;
247 int unregister_shallow(const unsigned char *sha1)
249 int pos = commit_graft_pos(sha1);
250 if (pos < 0)
251 return -1;
252 if (pos + 1 < commit_graft_nr)
253 memmove(commit_graft + pos, commit_graft + pos + 1,
254 sizeof(struct commit_graft *)
255 * (commit_graft_nr - pos - 1));
256 commit_graft_nr--;
257 return 0;
260 int parse_commit_buffer(struct commit *item, const void *buffer, unsigned long size)
262 const char *tail = buffer;
263 const char *bufptr = buffer;
264 unsigned char parent[20];
265 struct commit_list **pptr;
266 struct commit_graft *graft;
268 if (item->object.parsed)
269 return 0;
270 item->object.parsed = 1;
271 tail += size;
272 if (tail <= bufptr + 46 || memcmp(bufptr, "tree ", 5) || bufptr[45] != '\n')
273 return error("bogus commit object %s", sha1_to_hex(item->object.sha1));
274 if (get_sha1_hex(bufptr + 5, parent) < 0)
275 return error("bad tree pointer in commit %s",
276 sha1_to_hex(item->object.sha1));
277 item->tree = lookup_tree(parent);
278 bufptr += 46; /* "tree " + "hex sha1" + "\n" */
279 pptr = &item->parents;
281 graft = lookup_commit_graft(item->object.sha1);
282 while (bufptr + 48 < tail && !memcmp(bufptr, "parent ", 7)) {
283 struct commit *new_parent;
285 if (tail <= bufptr + 48 ||
286 get_sha1_hex(bufptr + 7, parent) ||
287 bufptr[47] != '\n')
288 return error("bad parents in commit %s", sha1_to_hex(item->object.sha1));
289 bufptr += 48;
291 * The clone is shallow if nr_parent < 0, and we must
292 * not traverse its real parents even when we unhide them.
294 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
295 continue;
296 new_parent = lookup_commit(parent);
297 if (new_parent)
298 pptr = &commit_list_insert(new_parent, pptr)->next;
300 if (graft) {
301 int i;
302 struct commit *new_parent;
303 for (i = 0; i < graft->nr_parent; i++) {
304 new_parent = lookup_commit(graft->parent[i]);
305 if (!new_parent)
306 continue;
307 pptr = &commit_list_insert(new_parent, pptr)->next;
310 item->date = parse_commit_date(bufptr, tail);
312 return 0;
315 int parse_commit(struct commit *item)
317 enum object_type type;
318 void *buffer;
319 unsigned long size;
320 int ret;
322 if (!item)
323 return -1;
324 if (item->object.parsed)
325 return 0;
326 buffer = read_sha1_file(item->object.sha1, &type, &size);
327 if (!buffer)
328 return error("Could not read %s",
329 sha1_to_hex(item->object.sha1));
330 if (type != OBJ_COMMIT) {
331 free(buffer);
332 return error("Object %s not a commit",
333 sha1_to_hex(item->object.sha1));
335 ret = parse_commit_buffer(item, buffer, size);
336 if (save_commit_buffer && !ret) {
337 item->buffer = buffer;
338 return 0;
340 free(buffer);
341 return ret;
344 void parse_commit_or_die(struct commit *item)
346 if (parse_commit(item))
347 die("unable to parse commit %s",
348 item ? sha1_to_hex(item->object.sha1) : "(null)");
351 int find_commit_subject(const char *commit_buffer, const char **subject)
353 const char *eol;
354 const char *p = commit_buffer;
356 while (*p && (*p != '\n' || p[1] != '\n'))
357 p++;
358 if (*p) {
359 p += 2;
360 for (eol = p; *eol && *eol != '\n'; eol++)
361 ; /* do nothing */
362 } else
363 eol = p;
365 *subject = p;
367 return eol - p;
370 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
372 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
373 new_list->item = item;
374 new_list->next = *list_p;
375 *list_p = new_list;
376 return new_list;
379 unsigned commit_list_count(const struct commit_list *l)
381 unsigned c = 0;
382 for (; l; l = l->next )
383 c++;
384 return c;
387 void free_commit_list(struct commit_list *list)
389 while (list) {
390 struct commit_list *temp = list;
391 list = temp->next;
392 free(temp);
396 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
398 struct commit_list **pp = list;
399 struct commit_list *p;
400 while ((p = *pp) != NULL) {
401 if (p->item->date < item->date) {
402 break;
404 pp = &p->next;
406 return commit_list_insert(item, pp);
409 static int commit_list_compare_by_date(const void *a, const void *b)
411 unsigned long a_date = ((const struct commit_list *)a)->item->date;
412 unsigned long b_date = ((const struct commit_list *)b)->item->date;
413 if (a_date < b_date)
414 return 1;
415 if (a_date > b_date)
416 return -1;
417 return 0;
420 static void *commit_list_get_next(const void *a)
422 return ((const struct commit_list *)a)->next;
425 static void commit_list_set_next(void *a, void *next)
427 ((struct commit_list *)a)->next = next;
430 void commit_list_sort_by_date(struct commit_list **list)
432 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
433 commit_list_compare_by_date);
436 struct commit *pop_most_recent_commit(struct commit_list **list,
437 unsigned int mark)
439 struct commit *ret = (*list)->item;
440 struct commit_list *parents = ret->parents;
441 struct commit_list *old = *list;
443 *list = (*list)->next;
444 free(old);
446 while (parents) {
447 struct commit *commit = parents->item;
448 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
449 commit->object.flags |= mark;
450 commit_list_insert_by_date(commit, list);
452 parents = parents->next;
454 return ret;
457 static void clear_commit_marks_1(struct commit_list **plist,
458 struct commit *commit, unsigned int mark)
460 while (commit) {
461 struct commit_list *parents;
463 if (!(mark & commit->object.flags))
464 return;
466 commit->object.flags &= ~mark;
468 parents = commit->parents;
469 if (!parents)
470 return;
472 while ((parents = parents->next))
473 commit_list_insert(parents->item, plist);
475 commit = commit->parents->item;
479 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
481 struct commit_list *list = NULL;
483 while (nr--) {
484 commit_list_insert(*commit, &list);
485 commit++;
487 while (list)
488 clear_commit_marks_1(&list, pop_commit(&list), mark);
491 void clear_commit_marks(struct commit *commit, unsigned int mark)
493 clear_commit_marks_many(1, &commit, mark);
496 void clear_commit_marks_for_object_array(struct object_array *a, unsigned mark)
498 struct object *object;
499 struct commit *commit;
500 unsigned int i;
502 for (i = 0; i < a->nr; i++) {
503 object = a->objects[i].item;
504 commit = lookup_commit_reference_gently(object->sha1, 1);
505 if (commit)
506 clear_commit_marks(commit, mark);
510 struct commit *pop_commit(struct commit_list **stack)
512 struct commit_list *top = *stack;
513 struct commit *item = top ? top->item : NULL;
515 if (top) {
516 *stack = top->next;
517 free(top);
519 return item;
523 * Topological sort support
526 /* count number of children that have not been emitted */
527 define_commit_slab(indegree_slab, int);
529 /* record author-date for each commit object */
530 define_commit_slab(author_date_slab, unsigned long);
532 static void record_author_date(struct author_date_slab *author_date,
533 struct commit *commit)
535 const char *buf, *line_end;
536 char *buffer = NULL;
537 struct ident_split ident;
538 char *date_end;
539 unsigned long date;
541 if (!commit->buffer) {
542 unsigned long size;
543 enum object_type type;
544 buffer = read_sha1_file(commit->object.sha1, &type, &size);
545 if (!buffer)
546 return;
549 for (buf = commit->buffer ? commit->buffer : buffer;
550 buf;
551 buf = line_end + 1) {
552 line_end = strchrnul(buf, '\n');
553 if (prefixcmp(buf, "author ")) {
554 if (!line_end[0] || line_end[1] == '\n')
555 return; /* end of header */
556 continue;
558 if (split_ident_line(&ident,
559 buf + strlen("author "),
560 line_end - (buf + strlen("author "))) ||
561 !ident.date_begin || !ident.date_end)
562 goto fail_exit; /* malformed "author" line */
563 break;
566 date = strtoul(ident.date_begin, &date_end, 10);
567 if (date_end != ident.date_end)
568 goto fail_exit; /* malformed date */
569 *(author_date_slab_at(author_date, commit)) = date;
571 fail_exit:
572 free(buffer);
575 static int compare_commits_by_author_date(const void *a_, const void *b_,
576 void *cb_data)
578 const struct commit *a = a_, *b = b_;
579 struct author_date_slab *author_date = cb_data;
580 unsigned long a_date = *(author_date_slab_at(author_date, a));
581 unsigned long b_date = *(author_date_slab_at(author_date, b));
583 /* newer commits with larger date first */
584 if (a_date < b_date)
585 return 1;
586 else if (a_date > b_date)
587 return -1;
588 return 0;
591 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
593 const struct commit *a = a_, *b = b_;
594 /* newer commits with larger date first */
595 if (a->date < b->date)
596 return 1;
597 else if (a->date > b->date)
598 return -1;
599 return 0;
603 * Performs an in-place topological sort on the list supplied.
605 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
607 struct commit_list *next, *orig = *list;
608 struct commit_list **pptr;
609 struct indegree_slab indegree;
610 struct prio_queue queue;
611 struct commit *commit;
612 struct author_date_slab author_date;
614 if (!orig)
615 return;
616 *list = NULL;
618 init_indegree_slab(&indegree);
619 memset(&queue, '\0', sizeof(queue));
621 switch (sort_order) {
622 default: /* REV_SORT_IN_GRAPH_ORDER */
623 queue.compare = NULL;
624 break;
625 case REV_SORT_BY_COMMIT_DATE:
626 queue.compare = compare_commits_by_commit_date;
627 break;
628 case REV_SORT_BY_AUTHOR_DATE:
629 init_author_date_slab(&author_date);
630 queue.compare = compare_commits_by_author_date;
631 queue.cb_data = &author_date;
632 break;
635 /* Mark them and clear the indegree */
636 for (next = orig; next; next = next->next) {
637 struct commit *commit = next->item;
638 *(indegree_slab_at(&indegree, commit)) = 1;
639 /* also record the author dates, if needed */
640 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
641 record_author_date(&author_date, commit);
644 /* update the indegree */
645 for (next = orig; next; next = next->next) {
646 struct commit_list *parents = next->item->parents;
647 while (parents) {
648 struct commit *parent = parents->item;
649 int *pi = indegree_slab_at(&indegree, parent);
651 if (*pi)
652 (*pi)++;
653 parents = parents->next;
658 * find the tips
660 * tips are nodes not reachable from any other node in the list
662 * the tips serve as a starting set for the work queue.
664 for (next = orig; next; next = next->next) {
665 struct commit *commit = next->item;
667 if (*(indegree_slab_at(&indegree, commit)) == 1)
668 prio_queue_put(&queue, commit);
672 * This is unfortunate; the initial tips need to be shown
673 * in the order given from the revision traversal machinery.
675 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
676 prio_queue_reverse(&queue);
678 /* We no longer need the commit list */
679 free_commit_list(orig);
681 pptr = list;
682 *list = NULL;
683 while ((commit = prio_queue_get(&queue)) != NULL) {
684 struct commit_list *parents;
686 for (parents = commit->parents; parents ; parents = parents->next) {
687 struct commit *parent = parents->item;
688 int *pi = indegree_slab_at(&indegree, parent);
690 if (!*pi)
691 continue;
694 * parents are only enqueued for emission
695 * when all their children have been emitted thereby
696 * guaranteeing topological order.
698 if (--(*pi) == 1)
699 prio_queue_put(&queue, parent);
702 * all children of commit have already been
703 * emitted. we can emit it now.
705 *(indegree_slab_at(&indegree, commit)) = 0;
707 pptr = &commit_list_insert(commit, pptr)->next;
710 clear_indegree_slab(&indegree);
711 clear_prio_queue(&queue);
712 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
713 clear_author_date_slab(&author_date);
716 /* merge-base stuff */
718 /* bits #0..15 in revision.h */
719 #define PARENT1 (1u<<16)
720 #define PARENT2 (1u<<17)
721 #define STALE (1u<<18)
722 #define RESULT (1u<<19)
724 static const unsigned all_flags = (PARENT1 | PARENT2 | STALE | RESULT);
726 static struct commit *interesting(struct commit_list *list)
728 while (list) {
729 struct commit *commit = list->item;
730 list = list->next;
731 if (commit->object.flags & STALE)
732 continue;
733 return commit;
735 return NULL;
738 /* all input commits in one and twos[] must have been parsed! */
739 static struct commit_list *paint_down_to_common(struct commit *one, int n, struct commit **twos)
741 struct commit_list *list = NULL;
742 struct commit_list *result = NULL;
743 int i;
745 one->object.flags |= PARENT1;
746 commit_list_insert_by_date(one, &list);
747 if (!n)
748 return list;
749 for (i = 0; i < n; i++) {
750 twos[i]->object.flags |= PARENT2;
751 commit_list_insert_by_date(twos[i], &list);
754 while (interesting(list)) {
755 struct commit *commit;
756 struct commit_list *parents;
757 struct commit_list *next;
758 int flags;
760 commit = list->item;
761 next = list->next;
762 free(list);
763 list = next;
765 flags = commit->object.flags & (PARENT1 | PARENT2 | STALE);
766 if (flags == (PARENT1 | PARENT2)) {
767 if (!(commit->object.flags & RESULT)) {
768 commit->object.flags |= RESULT;
769 commit_list_insert_by_date(commit, &result);
771 /* Mark parents of a found merge stale */
772 flags |= STALE;
774 parents = commit->parents;
775 while (parents) {
776 struct commit *p = parents->item;
777 parents = parents->next;
778 if ((p->object.flags & flags) == flags)
779 continue;
780 if (parse_commit(p))
781 return NULL;
782 p->object.flags |= flags;
783 commit_list_insert_by_date(p, &list);
787 free_commit_list(list);
788 return result;
791 static struct commit_list *merge_bases_many(struct commit *one, int n, struct commit **twos)
793 struct commit_list *list = NULL;
794 struct commit_list *result = NULL;
795 int i;
797 for (i = 0; i < n; i++) {
798 if (one == twos[i])
800 * We do not mark this even with RESULT so we do not
801 * have to clean it up.
803 return commit_list_insert(one, &result);
806 if (parse_commit(one))
807 return NULL;
808 for (i = 0; i < n; i++) {
809 if (parse_commit(twos[i]))
810 return NULL;
813 list = paint_down_to_common(one, n, twos);
815 while (list) {
816 struct commit_list *next = list->next;
817 if (!(list->item->object.flags & STALE))
818 commit_list_insert_by_date(list->item, &result);
819 free(list);
820 list = next;
822 return result;
825 struct commit_list *get_octopus_merge_bases(struct commit_list *in)
827 struct commit_list *i, *j, *k, *ret = NULL;
828 struct commit_list **pptr = &ret;
830 for (i = in; i; i = i->next) {
831 if (!ret)
832 pptr = &commit_list_insert(i->item, pptr)->next;
833 else {
834 struct commit_list *new = NULL, *end = NULL;
836 for (j = ret; j; j = j->next) {
837 struct commit_list *bases;
838 bases = get_merge_bases(i->item, j->item, 1);
839 if (!new)
840 new = bases;
841 else
842 end->next = bases;
843 for (k = bases; k; k = k->next)
844 end = k;
846 ret = new;
849 return ret;
852 static int remove_redundant(struct commit **array, int cnt)
855 * Some commit in the array may be an ancestor of
856 * another commit. Move such commit to the end of
857 * the array, and return the number of commits that
858 * are independent from each other.
860 struct commit **work;
861 unsigned char *redundant;
862 int *filled_index;
863 int i, j, filled;
865 work = xcalloc(cnt, sizeof(*work));
866 redundant = xcalloc(cnt, 1);
867 filled_index = xmalloc(sizeof(*filled_index) * (cnt - 1));
869 for (i = 0; i < cnt; i++)
870 parse_commit(array[i]);
871 for (i = 0; i < cnt; i++) {
872 struct commit_list *common;
874 if (redundant[i])
875 continue;
876 for (j = filled = 0; j < cnt; j++) {
877 if (i == j || redundant[j])
878 continue;
879 filled_index[filled] = j;
880 work[filled++] = array[j];
882 common = paint_down_to_common(array[i], filled, work);
883 if (array[i]->object.flags & PARENT2)
884 redundant[i] = 1;
885 for (j = 0; j < filled; j++)
886 if (work[j]->object.flags & PARENT1)
887 redundant[filled_index[j]] = 1;
888 clear_commit_marks(array[i], all_flags);
889 for (j = 0; j < filled; j++)
890 clear_commit_marks(work[j], all_flags);
891 free_commit_list(common);
894 /* Now collect the result */
895 memcpy(work, array, sizeof(*array) * cnt);
896 for (i = filled = 0; i < cnt; i++)
897 if (!redundant[i])
898 array[filled++] = work[i];
899 for (j = filled, i = 0; i < cnt; i++)
900 if (redundant[i])
901 array[j++] = work[i];
902 free(work);
903 free(redundant);
904 free(filled_index);
905 return filled;
908 struct commit_list *get_merge_bases_many(struct commit *one,
909 int n,
910 struct commit **twos,
911 int cleanup)
913 struct commit_list *list;
914 struct commit **rslt;
915 struct commit_list *result;
916 int cnt, i;
918 result = merge_bases_many(one, n, twos);
919 for (i = 0; i < n; i++) {
920 if (one == twos[i])
921 return result;
923 if (!result || !result->next) {
924 if (cleanup) {
925 clear_commit_marks(one, all_flags);
926 clear_commit_marks_many(n, twos, all_flags);
928 return result;
931 /* There are more than one */
932 cnt = 0;
933 list = result;
934 while (list) {
935 list = list->next;
936 cnt++;
938 rslt = xcalloc(cnt, sizeof(*rslt));
939 for (list = result, i = 0; list; list = list->next)
940 rslt[i++] = list->item;
941 free_commit_list(result);
943 clear_commit_marks(one, all_flags);
944 clear_commit_marks_many(n, twos, all_flags);
946 cnt = remove_redundant(rslt, cnt);
947 result = NULL;
948 for (i = 0; i < cnt; i++)
949 commit_list_insert_by_date(rslt[i], &result);
950 free(rslt);
951 return result;
954 struct commit_list *get_merge_bases(struct commit *one, struct commit *two,
955 int cleanup)
957 return get_merge_bases_many(one, 1, &two, cleanup);
961 * Is "commit" a descendant of one of the elements on the "with_commit" list?
963 int is_descendant_of(struct commit *commit, struct commit_list *with_commit)
965 if (!with_commit)
966 return 1;
967 while (with_commit) {
968 struct commit *other;
970 other = with_commit->item;
971 with_commit = with_commit->next;
972 if (in_merge_bases(other, commit))
973 return 1;
975 return 0;
979 * Is "commit" an ancestor of one of the "references"?
981 int in_merge_bases_many(struct commit *commit, int nr_reference, struct commit **reference)
983 struct commit_list *bases;
984 int ret = 0, i;
986 if (parse_commit(commit))
987 return ret;
988 for (i = 0; i < nr_reference; i++)
989 if (parse_commit(reference[i]))
990 return ret;
992 bases = paint_down_to_common(commit, nr_reference, reference);
993 if (commit->object.flags & PARENT2)
994 ret = 1;
995 clear_commit_marks(commit, all_flags);
996 clear_commit_marks_many(nr_reference, reference, all_flags);
997 free_commit_list(bases);
998 return ret;
1002 * Is "commit" an ancestor of (i.e. reachable from) the "reference"?
1004 int in_merge_bases(struct commit *commit, struct commit *reference)
1006 return in_merge_bases_many(commit, 1, &reference);
1009 struct commit_list *reduce_heads(struct commit_list *heads)
1011 struct commit_list *p;
1012 struct commit_list *result = NULL, **tail = &result;
1013 struct commit **array;
1014 int num_head, i;
1016 if (!heads)
1017 return NULL;
1019 /* Uniquify */
1020 for (p = heads; p; p = p->next)
1021 p->item->object.flags &= ~STALE;
1022 for (p = heads, num_head = 0; p; p = p->next) {
1023 if (p->item->object.flags & STALE)
1024 continue;
1025 p->item->object.flags |= STALE;
1026 num_head++;
1028 array = xcalloc(sizeof(*array), num_head);
1029 for (p = heads, i = 0; p; p = p->next) {
1030 if (p->item->object.flags & STALE) {
1031 array[i++] = p->item;
1032 p->item->object.flags &= ~STALE;
1035 num_head = remove_redundant(array, num_head);
1036 for (i = 0; i < num_head; i++)
1037 tail = &commit_list_insert(array[i], tail)->next;
1038 return result;
1041 static const char gpg_sig_header[] = "gpgsig";
1042 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
1044 static int do_sign_commit(struct strbuf *buf, const char *keyid)
1046 struct strbuf sig = STRBUF_INIT;
1047 int inspos, copypos;
1049 /* find the end of the header */
1050 inspos = strstr(buf->buf, "\n\n") - buf->buf + 1;
1052 if (!keyid || !*keyid)
1053 keyid = get_signing_key();
1054 if (sign_buffer(buf, &sig, keyid)) {
1055 strbuf_release(&sig);
1056 return -1;
1059 for (copypos = 0; sig.buf[copypos]; ) {
1060 const char *bol = sig.buf + copypos;
1061 const char *eol = strchrnul(bol, '\n');
1062 int len = (eol - bol) + !!*eol;
1064 if (!copypos) {
1065 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
1066 inspos += gpg_sig_header_len;
1068 strbuf_insert(buf, inspos++, " ", 1);
1069 strbuf_insert(buf, inspos, bol, len);
1070 inspos += len;
1071 copypos += len;
1073 strbuf_release(&sig);
1074 return 0;
1077 int parse_signed_commit(const unsigned char *sha1,
1078 struct strbuf *payload, struct strbuf *signature)
1080 unsigned long size;
1081 enum object_type type;
1082 char *buffer = read_sha1_file(sha1, &type, &size);
1083 int in_signature, saw_signature = -1;
1084 char *line, *tail;
1086 if (!buffer || type != OBJ_COMMIT)
1087 goto cleanup;
1089 line = buffer;
1090 tail = buffer + size;
1091 in_signature = 0;
1092 saw_signature = 0;
1093 while (line < tail) {
1094 const char *sig = NULL;
1095 char *next = memchr(line, '\n', tail - line);
1097 next = next ? next + 1 : tail;
1098 if (in_signature && line[0] == ' ')
1099 sig = line + 1;
1100 else if (!prefixcmp(line, gpg_sig_header) &&
1101 line[gpg_sig_header_len] == ' ')
1102 sig = line + gpg_sig_header_len + 1;
1103 if (sig) {
1104 strbuf_add(signature, sig, next - sig);
1105 saw_signature = 1;
1106 in_signature = 1;
1107 } else {
1108 if (*line == '\n')
1109 /* dump the whole remainder of the buffer */
1110 next = tail;
1111 strbuf_add(payload, line, next - line);
1112 in_signature = 0;
1114 line = next;
1116 cleanup:
1117 free(buffer);
1118 return saw_signature;
1121 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1123 struct merge_remote_desc *desc;
1124 struct commit_extra_header *mergetag;
1125 char *buf;
1126 unsigned long size, len;
1127 enum object_type type;
1129 desc = merge_remote_util(parent);
1130 if (!desc || !desc->obj)
1131 return;
1132 buf = read_sha1_file(desc->obj->sha1, &type, &size);
1133 if (!buf || type != OBJ_TAG)
1134 goto free_return;
1135 len = parse_signature(buf, size);
1136 if (size == len)
1137 goto free_return;
1139 * We could verify this signature and either omit the tag when
1140 * it does not validate, but the integrator may not have the
1141 * public key of the signer of the tag he is merging, while a
1142 * later auditor may have it while auditing, so let's not run
1143 * verify-signed-buffer here for now...
1145 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1146 * warn("warning: signed tag unverified.");
1148 mergetag = xcalloc(1, sizeof(*mergetag));
1149 mergetag->key = xstrdup("mergetag");
1150 mergetag->value = buf;
1151 mergetag->len = size;
1153 **tail = mergetag;
1154 *tail = &mergetag->next;
1155 return;
1157 free_return:
1158 free(buf);
1161 static struct {
1162 char result;
1163 const char *check;
1164 } sigcheck_gpg_status[] = {
1165 { 'G', "\n[GNUPG:] GOODSIG " },
1166 { 'B', "\n[GNUPG:] BADSIG " },
1167 { 'U', "\n[GNUPG:] TRUST_NEVER" },
1168 { 'U', "\n[GNUPG:] TRUST_UNDEFINED" },
1171 static void parse_gpg_output(struct signature_check *sigc)
1173 const char *buf = sigc->gpg_status;
1174 int i;
1176 /* Iterate over all search strings */
1177 for (i = 0; i < ARRAY_SIZE(sigcheck_gpg_status); i++) {
1178 const char *found, *next;
1180 if (!prefixcmp(buf, sigcheck_gpg_status[i].check + 1)) {
1181 /* At the very beginning of the buffer */
1182 found = buf + strlen(sigcheck_gpg_status[i].check + 1);
1183 } else {
1184 found = strstr(buf, sigcheck_gpg_status[i].check);
1185 if (!found)
1186 continue;
1187 found += strlen(sigcheck_gpg_status[i].check);
1189 sigc->result = sigcheck_gpg_status[i].result;
1190 /* The trust messages are not followed by key/signer information */
1191 if (sigc->result != 'U') {
1192 sigc->key = xmemdupz(found, 16);
1193 found += 17;
1194 next = strchrnul(found, '\n');
1195 sigc->signer = xmemdupz(found, next - found);
1200 void check_commit_signature(const struct commit* commit, struct signature_check *sigc)
1202 struct strbuf payload = STRBUF_INIT;
1203 struct strbuf signature = STRBUF_INIT;
1204 struct strbuf gpg_output = STRBUF_INIT;
1205 struct strbuf gpg_status = STRBUF_INIT;
1206 int status;
1208 sigc->result = 'N';
1210 if (parse_signed_commit(commit->object.sha1,
1211 &payload, &signature) <= 0)
1212 goto out;
1213 status = verify_signed_buffer(payload.buf, payload.len,
1214 signature.buf, signature.len,
1215 &gpg_output, &gpg_status);
1216 if (status && !gpg_output.len)
1217 goto out;
1218 sigc->gpg_output = strbuf_detach(&gpg_output, NULL);
1219 sigc->gpg_status = strbuf_detach(&gpg_status, NULL);
1220 parse_gpg_output(sigc);
1222 out:
1223 strbuf_release(&gpg_status);
1224 strbuf_release(&gpg_output);
1225 strbuf_release(&payload);
1226 strbuf_release(&signature);
1231 void append_merge_tag_headers(struct commit_list *parents,
1232 struct commit_extra_header ***tail)
1234 while (parents) {
1235 struct commit *parent = parents->item;
1236 handle_signed_tag(parent, tail);
1237 parents = parents->next;
1241 static void add_extra_header(struct strbuf *buffer,
1242 struct commit_extra_header *extra)
1244 strbuf_addstr(buffer, extra->key);
1245 if (extra->len)
1246 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1247 else
1248 strbuf_addch(buffer, '\n');
1251 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1252 const char **exclude)
1254 struct commit_extra_header *extra = NULL;
1255 unsigned long size;
1256 enum object_type type;
1257 char *buffer = read_sha1_file(commit->object.sha1, &type, &size);
1258 if (buffer && type == OBJ_COMMIT)
1259 extra = read_commit_extra_header_lines(buffer, size, exclude);
1260 free(buffer);
1261 return extra;
1264 static inline int standard_header_field(const char *field, size_t len)
1266 return ((len == 4 && !memcmp(field, "tree ", 5)) ||
1267 (len == 6 && !memcmp(field, "parent ", 7)) ||
1268 (len == 6 && !memcmp(field, "author ", 7)) ||
1269 (len == 9 && !memcmp(field, "committer ", 10)) ||
1270 (len == 8 && !memcmp(field, "encoding ", 9)));
1273 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1275 if (!exclude)
1276 return 0;
1278 while (*exclude) {
1279 size_t xlen = strlen(*exclude);
1280 if (len == xlen &&
1281 !memcmp(field, *exclude, xlen) && field[xlen] == ' ')
1282 return 1;
1283 exclude++;
1285 return 0;
1288 static struct commit_extra_header *read_commit_extra_header_lines(
1289 const char *buffer, size_t size,
1290 const char **exclude)
1292 struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1293 const char *line, *next, *eof, *eob;
1294 struct strbuf buf = STRBUF_INIT;
1296 for (line = buffer, eob = line + size;
1297 line < eob && *line != '\n';
1298 line = next) {
1299 next = memchr(line, '\n', eob - line);
1300 next = next ? next + 1 : eob;
1301 if (*line == ' ') {
1302 /* continuation */
1303 if (it)
1304 strbuf_add(&buf, line + 1, next - (line + 1));
1305 continue;
1307 if (it)
1308 it->value = strbuf_detach(&buf, &it->len);
1309 strbuf_reset(&buf);
1310 it = NULL;
1312 eof = strchr(line, ' ');
1313 if (next <= eof)
1314 eof = next;
1316 if (standard_header_field(line, eof - line) ||
1317 excluded_header_field(line, eof - line, exclude))
1318 continue;
1320 it = xcalloc(1, sizeof(*it));
1321 it->key = xmemdupz(line, eof-line);
1322 *tail = it;
1323 tail = &it->next;
1324 if (eof + 1 < next)
1325 strbuf_add(&buf, eof + 1, next - (eof + 1));
1327 if (it)
1328 it->value = strbuf_detach(&buf, &it->len);
1329 return extra;
1332 void free_commit_extra_headers(struct commit_extra_header *extra)
1334 while (extra) {
1335 struct commit_extra_header *next = extra->next;
1336 free(extra->key);
1337 free(extra->value);
1338 free(extra);
1339 extra = next;
1343 int commit_tree(const struct strbuf *msg, unsigned char *tree,
1344 struct commit_list *parents, unsigned char *ret,
1345 const char *author, const char *sign_commit)
1347 struct commit_extra_header *extra = NULL, **tail = &extra;
1348 int result;
1350 append_merge_tag_headers(parents, &tail);
1351 result = commit_tree_extended(msg, tree, parents, ret,
1352 author, sign_commit, extra);
1353 free_commit_extra_headers(extra);
1354 return result;
1357 static int find_invalid_utf8(const char *buf, int len)
1359 int offset = 0;
1360 static const unsigned int max_codepoint[] = {
1361 0x7f, 0x7ff, 0xffff, 0x10ffff
1364 while (len) {
1365 unsigned char c = *buf++;
1366 int bytes, bad_offset;
1367 unsigned int codepoint;
1368 unsigned int min_val, max_val;
1370 len--;
1371 offset++;
1373 /* Simple US-ASCII? No worries. */
1374 if (c < 0x80)
1375 continue;
1377 bad_offset = offset-1;
1380 * Count how many more high bits set: that's how
1381 * many more bytes this sequence should have.
1383 bytes = 0;
1384 while (c & 0x40) {
1385 c <<= 1;
1386 bytes++;
1390 * Must be between 1 and 3 more bytes. Longer sequences result in
1391 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1393 if (bytes < 1 || 3 < bytes)
1394 return bad_offset;
1396 /* Do we *have* that many bytes? */
1397 if (len < bytes)
1398 return bad_offset;
1401 * Place the encoded bits at the bottom of the value and compute the
1402 * valid range.
1404 codepoint = (c & 0x7f) >> bytes;
1405 min_val = max_codepoint[bytes-1] + 1;
1406 max_val = max_codepoint[bytes];
1408 offset += bytes;
1409 len -= bytes;
1411 /* And verify that they are good continuation bytes */
1412 do {
1413 codepoint <<= 6;
1414 codepoint |= *buf & 0x3f;
1415 if ((*buf++ & 0xc0) != 0x80)
1416 return bad_offset;
1417 } while (--bytes);
1419 /* Reject codepoints that are out of range for the sequence length. */
1420 if (codepoint < min_val || codepoint > max_val)
1421 return bad_offset;
1422 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1423 if ((codepoint & 0x1ff800) == 0xd800)
1424 return bad_offset;
1425 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1426 if ((codepoint & 0xfffe) == 0xfffe)
1427 return bad_offset;
1428 /* So are anything in the range U+FDD0..U+FDEF. */
1429 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1430 return bad_offset;
1432 return -1;
1436 * This verifies that the buffer is in proper utf8 format.
1438 * If it isn't, it assumes any non-utf8 characters are Latin1,
1439 * and does the conversion.
1441 static int verify_utf8(struct strbuf *buf)
1443 int ok = 1;
1444 long pos = 0;
1446 for (;;) {
1447 int bad;
1448 unsigned char c;
1449 unsigned char replace[2];
1451 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1452 if (bad < 0)
1453 return ok;
1454 pos += bad;
1455 ok = 0;
1456 c = buf->buf[pos];
1457 strbuf_remove(buf, pos, 1);
1459 /* We know 'c' must be in the range 128-255 */
1460 replace[0] = 0xc0 + (c >> 6);
1461 replace[1] = 0x80 + (c & 0x3f);
1462 strbuf_insert(buf, pos, replace, 2);
1463 pos += 2;
1467 static const char commit_utf8_warn[] =
1468 "Warning: commit message did not conform to UTF-8.\n"
1469 "You may want to amend it after fixing the message, or set the config\n"
1470 "variable i18n.commitencoding to the encoding your project uses.\n";
1472 int commit_tree_extended(const struct strbuf *msg, unsigned char *tree,
1473 struct commit_list *parents, unsigned char *ret,
1474 const char *author, const char *sign_commit,
1475 struct commit_extra_header *extra)
1477 int result;
1478 int encoding_is_utf8;
1479 struct strbuf buffer;
1481 assert_sha1_type(tree, OBJ_TREE);
1483 if (memchr(msg->buf, '\0', msg->len))
1484 return error("a NUL byte in commit log message not allowed.");
1486 /* Not having i18n.commitencoding is the same as having utf-8 */
1487 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1489 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1490 strbuf_addf(&buffer, "tree %s\n", sha1_to_hex(tree));
1493 * NOTE! This ordering means that the same exact tree merged with a
1494 * different order of parents will be a _different_ changeset even
1495 * if everything else stays the same.
1497 while (parents) {
1498 struct commit_list *next = parents->next;
1499 struct commit *parent = parents->item;
1501 strbuf_addf(&buffer, "parent %s\n",
1502 sha1_to_hex(parent->object.sha1));
1503 free(parents);
1504 parents = next;
1507 /* Person/date information */
1508 if (!author)
1509 author = git_author_info(IDENT_STRICT);
1510 strbuf_addf(&buffer, "author %s\n", author);
1511 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1512 if (!encoding_is_utf8)
1513 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1515 while (extra) {
1516 add_extra_header(&buffer, extra);
1517 extra = extra->next;
1519 strbuf_addch(&buffer, '\n');
1521 /* And add the comment */
1522 strbuf_addbuf(&buffer, msg);
1524 /* And check the encoding */
1525 if (encoding_is_utf8 && !verify_utf8(&buffer))
1526 fprintf(stderr, commit_utf8_warn);
1528 if (sign_commit && do_sign_commit(&buffer, sign_commit))
1529 return -1;
1531 result = write_sha1_file(buffer.buf, buffer.len, commit_type, ret);
1532 strbuf_release(&buffer);
1533 return result;
1536 struct commit *get_merge_parent(const char *name)
1538 struct object *obj;
1539 struct commit *commit;
1540 unsigned char sha1[20];
1541 if (get_sha1(name, sha1))
1542 return NULL;
1543 obj = parse_object(sha1);
1544 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1545 if (commit && !commit->util) {
1546 struct merge_remote_desc *desc;
1547 desc = xmalloc(sizeof(*desc));
1548 desc->obj = obj;
1549 desc->name = strdup(name);
1550 commit->util = desc;
1552 return commit;
1556 * Append a commit to the end of the commit_list.
1558 * next starts by pointing to the variable that holds the head of an
1559 * empty commit_list, and is updated to point to the "next" field of
1560 * the last item on the list as new commits are appended.
1562 * Usage example:
1564 * struct commit_list *list;
1565 * struct commit_list **next = &list;
1567 * next = commit_list_append(c1, next);
1568 * next = commit_list_append(c2, next);
1569 * assert(commit_list_count(list) == 2);
1570 * return list;
1572 struct commit_list **commit_list_append(struct commit *commit,
1573 struct commit_list **next)
1575 struct commit_list *new = xmalloc(sizeof(struct commit_list));
1576 new->item = commit;
1577 *next = new;
1578 new->next = NULL;
1579 return &new->next;
1582 void print_commit_list(struct commit_list *list,
1583 const char *format_cur,
1584 const char *format_last)
1586 for ( ; list; list = list->next) {
1587 const char *format = list->next ? format_cur : format_last;
1588 printf(format, sha1_to_hex(list->item->object.sha1));