ref-filter: add deltabase option
[git.git] / commit.c
blobd566d7e45c17cfa53493f228dbe46f2f92713ac9
1 #include "cache.h"
2 #include "tag.h"
3 #include "commit.h"
4 #include "commit-graph.h"
5 #include "repository.h"
6 #include "object-store.h"
7 #include "pkt-line.h"
8 #include "utf8.h"
9 #include "diff.h"
10 #include "revision.h"
11 #include "notes.h"
12 #include "alloc.h"
13 #include "gpg-interface.h"
14 #include "mergesort.h"
15 #include "commit-slab.h"
16 #include "prio-queue.h"
17 #include "sha1-lookup.h"
18 #include "wt-status.h"
19 #include "advice.h"
20 #include "refs.h"
21 #include "commit-reach.h"
23 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
25 int save_commit_buffer = 1;
27 const char *commit_type = "commit";
29 struct commit *lookup_commit_reference_gently(struct repository *r,
30 const struct object_id *oid, int quiet)
32 struct object *obj = deref_tag(r,
33 parse_object(r, oid),
34 NULL, 0);
36 if (!obj)
37 return NULL;
38 return object_as_type(r, obj, OBJ_COMMIT, quiet);
41 struct commit *lookup_commit_reference(struct repository *r, const struct object_id *oid)
43 return lookup_commit_reference_gently(r, oid, 0);
46 struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
48 struct commit *c = lookup_commit_reference(the_repository, oid);
49 if (!c)
50 die(_("could not parse %s"), ref_name);
51 if (!oideq(oid, &c->object.oid)) {
52 warning(_("%s %s is not a commit!"),
53 ref_name, oid_to_hex(oid));
55 return c;
58 struct commit *lookup_commit(struct repository *r, const struct object_id *oid)
60 struct object *obj = lookup_object(r, oid->hash);
61 if (!obj)
62 return create_object(r, oid->hash,
63 alloc_commit_node(r));
64 return object_as_type(r, obj, OBJ_COMMIT, 0);
67 struct commit *lookup_commit_reference_by_name(const char *name)
69 struct object_id oid;
70 struct commit *commit;
72 if (get_oid_committish(name, &oid))
73 return NULL;
74 commit = lookup_commit_reference(the_repository, &oid);
75 if (parse_commit(commit))
76 return NULL;
77 return commit;
80 static timestamp_t parse_commit_date(const char *buf, const char *tail)
82 const char *dateptr;
84 if (buf + 6 >= tail)
85 return 0;
86 if (memcmp(buf, "author", 6))
87 return 0;
88 while (buf < tail && *buf++ != '\n')
89 /* nada */;
90 if (buf + 9 >= tail)
91 return 0;
92 if (memcmp(buf, "committer", 9))
93 return 0;
94 while (buf < tail && *buf++ != '>')
95 /* nada */;
96 if (buf >= tail)
97 return 0;
98 dateptr = buf;
99 while (buf < tail && *buf++ != '\n')
100 /* nada */;
101 if (buf >= tail)
102 return 0;
103 /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
104 return parse_timestamp(dateptr, NULL, 10);
107 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
109 struct commit_graft **commit_graft_table = table;
110 return commit_graft_table[index]->oid.hash;
113 static int commit_graft_pos(struct repository *r, const unsigned char *sha1)
115 return sha1_pos(sha1, r->parsed_objects->grafts,
116 r->parsed_objects->grafts_nr,
117 commit_graft_sha1_access);
120 int register_commit_graft(struct repository *r, struct commit_graft *graft,
121 int ignore_dups)
123 int pos = commit_graft_pos(r, graft->oid.hash);
125 if (0 <= pos) {
126 if (ignore_dups)
127 free(graft);
128 else {
129 free(r->parsed_objects->grafts[pos]);
130 r->parsed_objects->grafts[pos] = graft;
132 return 1;
134 pos = -pos - 1;
135 ALLOC_GROW(r->parsed_objects->grafts,
136 r->parsed_objects->grafts_nr + 1,
137 r->parsed_objects->grafts_alloc);
138 r->parsed_objects->grafts_nr++;
139 if (pos < r->parsed_objects->grafts_nr)
140 memmove(r->parsed_objects->grafts + pos + 1,
141 r->parsed_objects->grafts + pos,
142 (r->parsed_objects->grafts_nr - pos - 1) *
143 sizeof(*r->parsed_objects->grafts));
144 r->parsed_objects->grafts[pos] = graft;
145 return 0;
148 struct commit_graft *read_graft_line(struct strbuf *line)
150 /* The format is just "Commit Parent1 Parent2 ...\n" */
151 int i, phase;
152 const char *tail = NULL;
153 struct commit_graft *graft = NULL;
154 struct object_id dummy_oid, *oid;
156 strbuf_rtrim(line);
157 if (!line->len || line->buf[0] == '#')
158 return NULL;
160 * phase 0 verifies line, counts hashes in line and allocates graft
161 * phase 1 fills graft
163 for (phase = 0; phase < 2; phase++) {
164 oid = graft ? &graft->oid : &dummy_oid;
165 if (parse_oid_hex(line->buf, oid, &tail))
166 goto bad_graft_data;
167 for (i = 0; *tail != '\0'; i++) {
168 oid = graft ? &graft->parent[i] : &dummy_oid;
169 if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
170 goto bad_graft_data;
172 if (!graft) {
173 graft = xmalloc(st_add(sizeof(*graft),
174 st_mult(sizeof(struct object_id), i)));
175 graft->nr_parent = i;
178 return graft;
180 bad_graft_data:
181 error("bad graft data: %s", line->buf);
182 assert(!graft);
183 return NULL;
186 static int read_graft_file(struct repository *r, const char *graft_file)
188 FILE *fp = fopen_or_warn(graft_file, "r");
189 struct strbuf buf = STRBUF_INIT;
190 if (!fp)
191 return -1;
192 if (advice_graft_file_deprecated)
193 advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
194 "and will be removed in a future Git version.\n"
195 "\n"
196 "Please use \"git replace --convert-graft-file\"\n"
197 "to convert the grafts into replace refs.\n"
198 "\n"
199 "Turn this message off by running\n"
200 "\"git config advice.graftFileDeprecated false\""));
201 while (!strbuf_getwholeline(&buf, fp, '\n')) {
202 /* The format is just "Commit Parent1 Parent2 ...\n" */
203 struct commit_graft *graft = read_graft_line(&buf);
204 if (!graft)
205 continue;
206 if (register_commit_graft(r, graft, 1))
207 error("duplicate graft data: %s", buf.buf);
209 fclose(fp);
210 strbuf_release(&buf);
211 return 0;
214 void prepare_commit_graft(struct repository *r)
216 char *graft_file;
218 if (r->parsed_objects->commit_graft_prepared)
219 return;
220 if (!startup_info->have_repository)
221 return;
223 graft_file = get_graft_file(r);
224 read_graft_file(r, graft_file);
225 /* make sure shallows are read */
226 is_repository_shallow(r);
227 r->parsed_objects->commit_graft_prepared = 1;
230 struct commit_graft *lookup_commit_graft(struct repository *r, const struct object_id *oid)
232 int pos;
233 prepare_commit_graft(r);
234 pos = commit_graft_pos(r, oid->hash);
235 if (pos < 0)
236 return NULL;
237 return r->parsed_objects->grafts[pos];
240 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
242 int i, ret;
243 for (i = ret = 0; i < the_repository->parsed_objects->grafts_nr && !ret; i++)
244 ret = fn(the_repository->parsed_objects->grafts[i], cb_data);
245 return ret;
248 int unregister_shallow(const struct object_id *oid)
250 int pos = commit_graft_pos(the_repository, oid->hash);
251 if (pos < 0)
252 return -1;
253 if (pos + 1 < the_repository->parsed_objects->grafts_nr)
254 MOVE_ARRAY(the_repository->parsed_objects->grafts + pos,
255 the_repository->parsed_objects->grafts + pos + 1,
256 the_repository->parsed_objects->grafts_nr - pos - 1);
257 the_repository->parsed_objects->grafts_nr--;
258 return 0;
261 struct commit_buffer {
262 void *buffer;
263 unsigned long size;
265 define_commit_slab(buffer_slab, struct commit_buffer);
267 struct buffer_slab *allocate_commit_buffer_slab(void)
269 struct buffer_slab *bs = xmalloc(sizeof(*bs));
270 init_buffer_slab(bs);
271 return bs;
274 void free_commit_buffer_slab(struct buffer_slab *bs)
276 clear_buffer_slab(bs);
277 free(bs);
280 void set_commit_buffer(struct repository *r, struct commit *commit, void *buffer, unsigned long size)
282 struct commit_buffer *v = buffer_slab_at(
283 r->parsed_objects->buffer_slab, commit);
284 v->buffer = buffer;
285 v->size = size;
288 const void *get_cached_commit_buffer(struct repository *r, const struct commit *commit, unsigned long *sizep)
290 struct commit_buffer *v = buffer_slab_peek(
291 r->parsed_objects->buffer_slab, commit);
292 if (!v) {
293 if (sizep)
294 *sizep = 0;
295 return NULL;
297 if (sizep)
298 *sizep = v->size;
299 return v->buffer;
302 const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
304 const void *ret = get_cached_commit_buffer(the_repository, commit, sizep);
305 if (!ret) {
306 enum object_type type;
307 unsigned long size;
308 ret = read_object_file(&commit->object.oid, &type, &size);
309 if (!ret)
310 die("cannot read commit object %s",
311 oid_to_hex(&commit->object.oid));
312 if (type != OBJ_COMMIT)
313 die("expected commit for %s, got %s",
314 oid_to_hex(&commit->object.oid), type_name(type));
315 if (sizep)
316 *sizep = size;
318 return ret;
321 void unuse_commit_buffer(const struct commit *commit, const void *buffer)
323 struct commit_buffer *v = buffer_slab_peek(
324 the_repository->parsed_objects->buffer_slab, commit);
325 if (!(v && v->buffer == buffer))
326 free((void *)buffer);
329 void free_commit_buffer(struct commit *commit)
331 struct commit_buffer *v = buffer_slab_peek(
332 the_repository->parsed_objects->buffer_slab, commit);
333 if (v) {
334 FREE_AND_NULL(v->buffer);
335 v->size = 0;
339 struct tree *get_commit_tree(const struct commit *commit)
341 if (commit->maybe_tree || !commit->object.parsed)
342 return commit->maybe_tree;
344 if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
345 BUG("commit has NULL tree, but was not loaded from commit-graph");
347 return get_commit_tree_in_graph(the_repository, commit);
350 struct object_id *get_commit_tree_oid(const struct commit *commit)
352 return &get_commit_tree(commit)->object.oid;
355 void release_commit_memory(struct commit *c)
357 c->maybe_tree = NULL;
358 c->index = 0;
359 free_commit_buffer(c);
360 free_commit_list(c->parents);
361 /* TODO: what about commit->util? */
363 c->object.parsed = 0;
366 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
368 struct commit_buffer *v = buffer_slab_peek(
369 the_repository->parsed_objects->buffer_slab, commit);
370 void *ret;
372 if (!v) {
373 if (sizep)
374 *sizep = 0;
375 return NULL;
377 ret = v->buffer;
378 if (sizep)
379 *sizep = v->size;
381 v->buffer = NULL;
382 v->size = 0;
383 return ret;
386 int parse_commit_buffer(struct repository *r, struct commit *item, const void *buffer, unsigned long size, int check_graph)
388 const char *tail = buffer;
389 const char *bufptr = buffer;
390 struct object_id parent;
391 struct commit_list **pptr;
392 struct commit_graft *graft;
393 const int tree_entry_len = the_hash_algo->hexsz + 5;
394 const int parent_entry_len = the_hash_algo->hexsz + 7;
396 if (item->object.parsed)
397 return 0;
398 item->object.parsed = 1;
399 tail += size;
400 if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
401 bufptr[tree_entry_len] != '\n')
402 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
403 if (get_oid_hex(bufptr + 5, &parent) < 0)
404 return error("bad tree pointer in commit %s",
405 oid_to_hex(&item->object.oid));
406 item->maybe_tree = lookup_tree(r, &parent);
407 bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
408 pptr = &item->parents;
410 graft = lookup_commit_graft(r, &item->object.oid);
411 while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
412 struct commit *new_parent;
414 if (tail <= bufptr + parent_entry_len + 1 ||
415 get_oid_hex(bufptr + 7, &parent) ||
416 bufptr[parent_entry_len] != '\n')
417 return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
418 bufptr += parent_entry_len + 1;
420 * The clone is shallow if nr_parent < 0, and we must
421 * not traverse its real parents even when we unhide them.
423 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
424 continue;
425 new_parent = lookup_commit(r, &parent);
426 if (new_parent)
427 pptr = &commit_list_insert(new_parent, pptr)->next;
429 if (graft) {
430 int i;
431 struct commit *new_parent;
432 for (i = 0; i < graft->nr_parent; i++) {
433 new_parent = lookup_commit(r,
434 &graft->parent[i]);
435 if (!new_parent)
436 continue;
437 pptr = &commit_list_insert(new_parent, pptr)->next;
440 item->date = parse_commit_date(bufptr, tail);
442 if (check_graph)
443 load_commit_graph_info(the_repository, item);
445 return 0;
448 int parse_commit_internal(struct commit *item, int quiet_on_missing, int use_commit_graph)
450 enum object_type type;
451 void *buffer;
452 unsigned long size;
453 int ret;
455 if (!item)
456 return -1;
457 if (item->object.parsed)
458 return 0;
459 if (use_commit_graph && parse_commit_in_graph(the_repository, item))
460 return 0;
461 buffer = read_object_file(&item->object.oid, &type, &size);
462 if (!buffer)
463 return quiet_on_missing ? -1 :
464 error("Could not read %s",
465 oid_to_hex(&item->object.oid));
466 if (type != OBJ_COMMIT) {
467 free(buffer);
468 return error("Object %s not a commit",
469 oid_to_hex(&item->object.oid));
472 ret = parse_commit_buffer(the_repository, item, buffer, size, 0);
473 if (save_commit_buffer && !ret) {
474 set_commit_buffer(the_repository, item, buffer, size);
475 return 0;
477 free(buffer);
478 return ret;
481 int parse_commit_gently(struct commit *item, int quiet_on_missing)
483 return parse_commit_internal(item, quiet_on_missing, 1);
486 void parse_commit_or_die(struct commit *item)
488 if (parse_commit(item))
489 die("unable to parse commit %s",
490 item ? oid_to_hex(&item->object.oid) : "(null)");
493 int find_commit_subject(const char *commit_buffer, const char **subject)
495 const char *eol;
496 const char *p = commit_buffer;
498 while (*p && (*p != '\n' || p[1] != '\n'))
499 p++;
500 if (*p) {
501 p = skip_blank_lines(p + 2);
502 eol = strchrnul(p, '\n');
503 } else
504 eol = p;
506 *subject = p;
508 return eol - p;
511 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
513 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
514 new_list->item = item;
515 new_list->next = *list_p;
516 *list_p = new_list;
517 return new_list;
520 unsigned commit_list_count(const struct commit_list *l)
522 unsigned c = 0;
523 for (; l; l = l->next )
524 c++;
525 return c;
528 struct commit_list *copy_commit_list(struct commit_list *list)
530 struct commit_list *head = NULL;
531 struct commit_list **pp = &head;
532 while (list) {
533 pp = commit_list_append(list->item, pp);
534 list = list->next;
536 return head;
539 void free_commit_list(struct commit_list *list)
541 while (list)
542 pop_commit(&list);
545 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
547 struct commit_list **pp = list;
548 struct commit_list *p;
549 while ((p = *pp) != NULL) {
550 if (p->item->date < item->date) {
551 break;
553 pp = &p->next;
555 return commit_list_insert(item, pp);
558 static int commit_list_compare_by_date(const void *a, const void *b)
560 timestamp_t a_date = ((const struct commit_list *)a)->item->date;
561 timestamp_t b_date = ((const struct commit_list *)b)->item->date;
562 if (a_date < b_date)
563 return 1;
564 if (a_date > b_date)
565 return -1;
566 return 0;
569 static void *commit_list_get_next(const void *a)
571 return ((const struct commit_list *)a)->next;
574 static void commit_list_set_next(void *a, void *next)
576 ((struct commit_list *)a)->next = next;
579 void commit_list_sort_by_date(struct commit_list **list)
581 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
582 commit_list_compare_by_date);
585 struct commit *pop_most_recent_commit(struct commit_list **list,
586 unsigned int mark)
588 struct commit *ret = pop_commit(list);
589 struct commit_list *parents = ret->parents;
591 while (parents) {
592 struct commit *commit = parents->item;
593 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
594 commit->object.flags |= mark;
595 commit_list_insert_by_date(commit, list);
597 parents = parents->next;
599 return ret;
602 static void clear_commit_marks_1(struct commit_list **plist,
603 struct commit *commit, unsigned int mark)
605 while (commit) {
606 struct commit_list *parents;
608 if (!(mark & commit->object.flags))
609 return;
611 commit->object.flags &= ~mark;
613 parents = commit->parents;
614 if (!parents)
615 return;
617 while ((parents = parents->next))
618 commit_list_insert(parents->item, plist);
620 commit = commit->parents->item;
624 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
626 struct commit_list *list = NULL;
628 while (nr--) {
629 clear_commit_marks_1(&list, *commit, mark);
630 commit++;
632 while (list)
633 clear_commit_marks_1(&list, pop_commit(&list), mark);
636 void clear_commit_marks(struct commit *commit, unsigned int mark)
638 clear_commit_marks_many(1, &commit, mark);
641 struct commit *pop_commit(struct commit_list **stack)
643 struct commit_list *top = *stack;
644 struct commit *item = top ? top->item : NULL;
646 if (top) {
647 *stack = top->next;
648 free(top);
650 return item;
654 * Topological sort support
657 /* count number of children that have not been emitted */
658 define_commit_slab(indegree_slab, int);
660 /* record author-date for each commit object */
661 define_commit_slab(author_date_slab, timestamp_t);
663 static void record_author_date(struct author_date_slab *author_date,
664 struct commit *commit)
666 const char *buffer = get_commit_buffer(commit, NULL);
667 struct ident_split ident;
668 const char *ident_line;
669 size_t ident_len;
670 char *date_end;
671 timestamp_t date;
673 ident_line = find_commit_header(buffer, "author", &ident_len);
674 if (!ident_line)
675 goto fail_exit; /* no author line */
676 if (split_ident_line(&ident, ident_line, ident_len) ||
677 !ident.date_begin || !ident.date_end)
678 goto fail_exit; /* malformed "author" line */
680 date = parse_timestamp(ident.date_begin, &date_end, 10);
681 if (date_end != ident.date_end)
682 goto fail_exit; /* malformed date */
683 *(author_date_slab_at(author_date, commit)) = date;
685 fail_exit:
686 unuse_commit_buffer(commit, buffer);
689 static int compare_commits_by_author_date(const void *a_, const void *b_,
690 void *cb_data)
692 const struct commit *a = a_, *b = b_;
693 struct author_date_slab *author_date = cb_data;
694 timestamp_t a_date = *(author_date_slab_at(author_date, a));
695 timestamp_t b_date = *(author_date_slab_at(author_date, b));
697 /* newer commits with larger date first */
698 if (a_date < b_date)
699 return 1;
700 else if (a_date > b_date)
701 return -1;
702 return 0;
705 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
707 const struct commit *a = a_, *b = b_;
709 /* newer commits first */
710 if (a->generation < b->generation)
711 return 1;
712 else if (a->generation > b->generation)
713 return -1;
715 /* use date as a heuristic when generations are equal */
716 if (a->date < b->date)
717 return 1;
718 else if (a->date > b->date)
719 return -1;
720 return 0;
723 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
725 const struct commit *a = a_, *b = b_;
726 /* newer commits with larger date first */
727 if (a->date < b->date)
728 return 1;
729 else if (a->date > b->date)
730 return -1;
731 return 0;
735 * Performs an in-place topological sort on the list supplied.
737 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
739 struct commit_list *next, *orig = *list;
740 struct commit_list **pptr;
741 struct indegree_slab indegree;
742 struct prio_queue queue;
743 struct commit *commit;
744 struct author_date_slab author_date;
746 if (!orig)
747 return;
748 *list = NULL;
750 init_indegree_slab(&indegree);
751 memset(&queue, '\0', sizeof(queue));
753 switch (sort_order) {
754 default: /* REV_SORT_IN_GRAPH_ORDER */
755 queue.compare = NULL;
756 break;
757 case REV_SORT_BY_COMMIT_DATE:
758 queue.compare = compare_commits_by_commit_date;
759 break;
760 case REV_SORT_BY_AUTHOR_DATE:
761 init_author_date_slab(&author_date);
762 queue.compare = compare_commits_by_author_date;
763 queue.cb_data = &author_date;
764 break;
767 /* Mark them and clear the indegree */
768 for (next = orig; next; next = next->next) {
769 struct commit *commit = next->item;
770 *(indegree_slab_at(&indegree, commit)) = 1;
771 /* also record the author dates, if needed */
772 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
773 record_author_date(&author_date, commit);
776 /* update the indegree */
777 for (next = orig; next; next = next->next) {
778 struct commit_list *parents = next->item->parents;
779 while (parents) {
780 struct commit *parent = parents->item;
781 int *pi = indegree_slab_at(&indegree, parent);
783 if (*pi)
784 (*pi)++;
785 parents = parents->next;
790 * find the tips
792 * tips are nodes not reachable from any other node in the list
794 * the tips serve as a starting set for the work queue.
796 for (next = orig; next; next = next->next) {
797 struct commit *commit = next->item;
799 if (*(indegree_slab_at(&indegree, commit)) == 1)
800 prio_queue_put(&queue, commit);
804 * This is unfortunate; the initial tips need to be shown
805 * in the order given from the revision traversal machinery.
807 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
808 prio_queue_reverse(&queue);
810 /* We no longer need the commit list */
811 free_commit_list(orig);
813 pptr = list;
814 *list = NULL;
815 while ((commit = prio_queue_get(&queue)) != NULL) {
816 struct commit_list *parents;
818 for (parents = commit->parents; parents ; parents = parents->next) {
819 struct commit *parent = parents->item;
820 int *pi = indegree_slab_at(&indegree, parent);
822 if (!*pi)
823 continue;
826 * parents are only enqueued for emission
827 * when all their children have been emitted thereby
828 * guaranteeing topological order.
830 if (--(*pi) == 1)
831 prio_queue_put(&queue, parent);
834 * all children of commit have already been
835 * emitted. we can emit it now.
837 *(indegree_slab_at(&indegree, commit)) = 0;
839 pptr = &commit_list_insert(commit, pptr)->next;
842 clear_indegree_slab(&indegree);
843 clear_prio_queue(&queue);
844 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
845 clear_author_date_slab(&author_date);
848 struct rev_collect {
849 struct commit **commit;
850 int nr;
851 int alloc;
852 unsigned int initial : 1;
855 static void add_one_commit(struct object_id *oid, struct rev_collect *revs)
857 struct commit *commit;
859 if (is_null_oid(oid))
860 return;
862 commit = lookup_commit(the_repository, oid);
863 if (!commit ||
864 (commit->object.flags & TMP_MARK) ||
865 parse_commit(commit))
866 return;
868 ALLOC_GROW(revs->commit, revs->nr + 1, revs->alloc);
869 revs->commit[revs->nr++] = commit;
870 commit->object.flags |= TMP_MARK;
873 static int collect_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
874 const char *ident, timestamp_t timestamp,
875 int tz, const char *message, void *cbdata)
877 struct rev_collect *revs = cbdata;
879 if (revs->initial) {
880 revs->initial = 0;
881 add_one_commit(ooid, revs);
883 add_one_commit(noid, revs);
884 return 0;
887 struct commit *get_fork_point(const char *refname, struct commit *commit)
889 struct object_id oid;
890 struct rev_collect revs;
891 struct commit_list *bases;
892 int i;
893 struct commit *ret = NULL;
895 memset(&revs, 0, sizeof(revs));
896 revs.initial = 1;
897 for_each_reflog_ent(refname, collect_one_reflog_ent, &revs);
899 if (!revs.nr && !get_oid(refname, &oid))
900 add_one_commit(&oid, &revs);
902 for (i = 0; i < revs.nr; i++)
903 revs.commit[i]->object.flags &= ~TMP_MARK;
905 bases = get_merge_bases_many(commit, revs.nr, revs.commit);
908 * There should be one and only one merge base, when we found
909 * a common ancestor among reflog entries.
911 if (!bases || bases->next)
912 goto cleanup_return;
914 /* And the found one must be one of the reflog entries */
915 for (i = 0; i < revs.nr; i++)
916 if (&bases->item->object == &revs.commit[i]->object)
917 break; /* found */
918 if (revs.nr <= i)
919 goto cleanup_return;
921 ret = bases->item;
923 cleanup_return:
924 free_commit_list(bases);
925 return ret;
928 static const char gpg_sig_header[] = "gpgsig";
929 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
931 static int do_sign_commit(struct strbuf *buf, const char *keyid)
933 struct strbuf sig = STRBUF_INIT;
934 int inspos, copypos;
935 const char *eoh;
937 /* find the end of the header */
938 eoh = strstr(buf->buf, "\n\n");
939 if (!eoh)
940 inspos = buf->len;
941 else
942 inspos = eoh - buf->buf + 1;
944 if (!keyid || !*keyid)
945 keyid = get_signing_key();
946 if (sign_buffer(buf, &sig, keyid)) {
947 strbuf_release(&sig);
948 return -1;
951 for (copypos = 0; sig.buf[copypos]; ) {
952 const char *bol = sig.buf + copypos;
953 const char *eol = strchrnul(bol, '\n');
954 int len = (eol - bol) + !!*eol;
956 if (!copypos) {
957 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
958 inspos += gpg_sig_header_len;
960 strbuf_insert(buf, inspos++, " ", 1);
961 strbuf_insert(buf, inspos, bol, len);
962 inspos += len;
963 copypos += len;
965 strbuf_release(&sig);
966 return 0;
969 int parse_signed_commit(const struct commit *commit,
970 struct strbuf *payload, struct strbuf *signature)
973 unsigned long size;
974 const char *buffer = get_commit_buffer(commit, &size);
975 int in_signature, saw_signature = -1;
976 const char *line, *tail;
978 line = buffer;
979 tail = buffer + size;
980 in_signature = 0;
981 saw_signature = 0;
982 while (line < tail) {
983 const char *sig = NULL;
984 const char *next = memchr(line, '\n', tail - line);
986 next = next ? next + 1 : tail;
987 if (in_signature && line[0] == ' ')
988 sig = line + 1;
989 else if (starts_with(line, gpg_sig_header) &&
990 line[gpg_sig_header_len] == ' ')
991 sig = line + gpg_sig_header_len + 1;
992 if (sig) {
993 strbuf_add(signature, sig, next - sig);
994 saw_signature = 1;
995 in_signature = 1;
996 } else {
997 if (*line == '\n')
998 /* dump the whole remainder of the buffer */
999 next = tail;
1000 strbuf_add(payload, line, next - line);
1001 in_signature = 0;
1003 line = next;
1005 unuse_commit_buffer(commit, buffer);
1006 return saw_signature;
1009 int remove_signature(struct strbuf *buf)
1011 const char *line = buf->buf;
1012 const char *tail = buf->buf + buf->len;
1013 int in_signature = 0;
1014 const char *sig_start = NULL;
1015 const char *sig_end = NULL;
1017 while (line < tail) {
1018 const char *next = memchr(line, '\n', tail - line);
1019 next = next ? next + 1 : tail;
1021 if (in_signature && line[0] == ' ')
1022 sig_end = next;
1023 else if (starts_with(line, gpg_sig_header) &&
1024 line[gpg_sig_header_len] == ' ') {
1025 sig_start = line;
1026 sig_end = next;
1027 in_signature = 1;
1028 } else {
1029 if (*line == '\n')
1030 /* dump the whole remainder of the buffer */
1031 next = tail;
1032 in_signature = 0;
1034 line = next;
1037 if (sig_start)
1038 strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1040 return sig_start != NULL;
1043 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1045 struct merge_remote_desc *desc;
1046 struct commit_extra_header *mergetag;
1047 char *buf;
1048 unsigned long size, len;
1049 enum object_type type;
1051 desc = merge_remote_util(parent);
1052 if (!desc || !desc->obj)
1053 return;
1054 buf = read_object_file(&desc->obj->oid, &type, &size);
1055 if (!buf || type != OBJ_TAG)
1056 goto free_return;
1057 len = parse_signature(buf, size);
1058 if (size == len)
1059 goto free_return;
1061 * We could verify this signature and either omit the tag when
1062 * it does not validate, but the integrator may not have the
1063 * public key of the signer of the tag he is merging, while a
1064 * later auditor may have it while auditing, so let's not run
1065 * verify-signed-buffer here for now...
1067 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1068 * warn("warning: signed tag unverified.");
1070 mergetag = xcalloc(1, sizeof(*mergetag));
1071 mergetag->key = xstrdup("mergetag");
1072 mergetag->value = buf;
1073 mergetag->len = size;
1075 **tail = mergetag;
1076 *tail = &mergetag->next;
1077 return;
1079 free_return:
1080 free(buf);
1083 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1085 struct strbuf payload = STRBUF_INIT;
1086 struct strbuf signature = STRBUF_INIT;
1087 int ret = 1;
1089 sigc->result = 'N';
1091 if (parse_signed_commit(commit, &payload, &signature) <= 0)
1092 goto out;
1093 ret = check_signature(payload.buf, payload.len, signature.buf,
1094 signature.len, sigc);
1096 out:
1097 strbuf_release(&payload);
1098 strbuf_release(&signature);
1100 return ret;
1105 void append_merge_tag_headers(struct commit_list *parents,
1106 struct commit_extra_header ***tail)
1108 while (parents) {
1109 struct commit *parent = parents->item;
1110 handle_signed_tag(parent, tail);
1111 parents = parents->next;
1115 static void add_extra_header(struct strbuf *buffer,
1116 struct commit_extra_header *extra)
1118 strbuf_addstr(buffer, extra->key);
1119 if (extra->len)
1120 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1121 else
1122 strbuf_addch(buffer, '\n');
1125 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1126 const char **exclude)
1128 struct commit_extra_header *extra = NULL;
1129 unsigned long size;
1130 const char *buffer = get_commit_buffer(commit, &size);
1131 extra = read_commit_extra_header_lines(buffer, size, exclude);
1132 unuse_commit_buffer(commit, buffer);
1133 return extra;
1136 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1138 struct commit_extra_header *extra, *to_free;
1139 int res = 0;
1141 to_free = read_commit_extra_headers(commit, NULL);
1142 for (extra = to_free; !res && extra; extra = extra->next) {
1143 if (strcmp(extra->key, "mergetag"))
1144 continue; /* not a merge tag */
1145 res = fn(commit, extra, data);
1147 free_commit_extra_headers(to_free);
1148 return res;
1151 static inline int standard_header_field(const char *field, size_t len)
1153 return ((len == 4 && !memcmp(field, "tree", 4)) ||
1154 (len == 6 && !memcmp(field, "parent", 6)) ||
1155 (len == 6 && !memcmp(field, "author", 6)) ||
1156 (len == 9 && !memcmp(field, "committer", 9)) ||
1157 (len == 8 && !memcmp(field, "encoding", 8)));
1160 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1162 if (!exclude)
1163 return 0;
1165 while (*exclude) {
1166 size_t xlen = strlen(*exclude);
1167 if (len == xlen && !memcmp(field, *exclude, xlen))
1168 return 1;
1169 exclude++;
1171 return 0;
1174 static struct commit_extra_header *read_commit_extra_header_lines(
1175 const char *buffer, size_t size,
1176 const char **exclude)
1178 struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1179 const char *line, *next, *eof, *eob;
1180 struct strbuf buf = STRBUF_INIT;
1182 for (line = buffer, eob = line + size;
1183 line < eob && *line != '\n';
1184 line = next) {
1185 next = memchr(line, '\n', eob - line);
1186 next = next ? next + 1 : eob;
1187 if (*line == ' ') {
1188 /* continuation */
1189 if (it)
1190 strbuf_add(&buf, line + 1, next - (line + 1));
1191 continue;
1193 if (it)
1194 it->value = strbuf_detach(&buf, &it->len);
1195 strbuf_reset(&buf);
1196 it = NULL;
1198 eof = memchr(line, ' ', next - line);
1199 if (!eof)
1200 eof = next;
1201 else if (standard_header_field(line, eof - line) ||
1202 excluded_header_field(line, eof - line, exclude))
1203 continue;
1205 it = xcalloc(1, sizeof(*it));
1206 it->key = xmemdupz(line, eof-line);
1207 *tail = it;
1208 tail = &it->next;
1209 if (eof + 1 < next)
1210 strbuf_add(&buf, eof + 1, next - (eof + 1));
1212 if (it)
1213 it->value = strbuf_detach(&buf, &it->len);
1214 return extra;
1217 void free_commit_extra_headers(struct commit_extra_header *extra)
1219 while (extra) {
1220 struct commit_extra_header *next = extra->next;
1221 free(extra->key);
1222 free(extra->value);
1223 free(extra);
1224 extra = next;
1228 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1229 struct commit_list *parents, struct object_id *ret,
1230 const char *author, const char *sign_commit)
1232 struct commit_extra_header *extra = NULL, **tail = &extra;
1233 int result;
1235 append_merge_tag_headers(parents, &tail);
1236 result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1237 author, sign_commit, extra);
1238 free_commit_extra_headers(extra);
1239 return result;
1242 static int find_invalid_utf8(const char *buf, int len)
1244 int offset = 0;
1245 static const unsigned int max_codepoint[] = {
1246 0x7f, 0x7ff, 0xffff, 0x10ffff
1249 while (len) {
1250 unsigned char c = *buf++;
1251 int bytes, bad_offset;
1252 unsigned int codepoint;
1253 unsigned int min_val, max_val;
1255 len--;
1256 offset++;
1258 /* Simple US-ASCII? No worries. */
1259 if (c < 0x80)
1260 continue;
1262 bad_offset = offset-1;
1265 * Count how many more high bits set: that's how
1266 * many more bytes this sequence should have.
1268 bytes = 0;
1269 while (c & 0x40) {
1270 c <<= 1;
1271 bytes++;
1275 * Must be between 1 and 3 more bytes. Longer sequences result in
1276 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1278 if (bytes < 1 || 3 < bytes)
1279 return bad_offset;
1281 /* Do we *have* that many bytes? */
1282 if (len < bytes)
1283 return bad_offset;
1286 * Place the encoded bits at the bottom of the value and compute the
1287 * valid range.
1289 codepoint = (c & 0x7f) >> bytes;
1290 min_val = max_codepoint[bytes-1] + 1;
1291 max_val = max_codepoint[bytes];
1293 offset += bytes;
1294 len -= bytes;
1296 /* And verify that they are good continuation bytes */
1297 do {
1298 codepoint <<= 6;
1299 codepoint |= *buf & 0x3f;
1300 if ((*buf++ & 0xc0) != 0x80)
1301 return bad_offset;
1302 } while (--bytes);
1304 /* Reject codepoints that are out of range for the sequence length. */
1305 if (codepoint < min_val || codepoint > max_val)
1306 return bad_offset;
1307 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1308 if ((codepoint & 0x1ff800) == 0xd800)
1309 return bad_offset;
1310 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1311 if ((codepoint & 0xfffe) == 0xfffe)
1312 return bad_offset;
1313 /* So are anything in the range U+FDD0..U+FDEF. */
1314 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1315 return bad_offset;
1317 return -1;
1321 * This verifies that the buffer is in proper utf8 format.
1323 * If it isn't, it assumes any non-utf8 characters are Latin1,
1324 * and does the conversion.
1326 static int verify_utf8(struct strbuf *buf)
1328 int ok = 1;
1329 long pos = 0;
1331 for (;;) {
1332 int bad;
1333 unsigned char c;
1334 unsigned char replace[2];
1336 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1337 if (bad < 0)
1338 return ok;
1339 pos += bad;
1340 ok = 0;
1341 c = buf->buf[pos];
1342 strbuf_remove(buf, pos, 1);
1344 /* We know 'c' must be in the range 128-255 */
1345 replace[0] = 0xc0 + (c >> 6);
1346 replace[1] = 0x80 + (c & 0x3f);
1347 strbuf_insert(buf, pos, replace, 2);
1348 pos += 2;
1352 static const char commit_utf8_warn[] =
1353 N_("Warning: commit message did not conform to UTF-8.\n"
1354 "You may want to amend it after fixing the message, or set the config\n"
1355 "variable i18n.commitencoding to the encoding your project uses.\n");
1357 int commit_tree_extended(const char *msg, size_t msg_len,
1358 const struct object_id *tree,
1359 struct commit_list *parents, struct object_id *ret,
1360 const char *author, const char *sign_commit,
1361 struct commit_extra_header *extra)
1363 int result;
1364 int encoding_is_utf8;
1365 struct strbuf buffer;
1367 assert_oid_type(tree, OBJ_TREE);
1369 if (memchr(msg, '\0', msg_len))
1370 return error("a NUL byte in commit log message not allowed.");
1372 /* Not having i18n.commitencoding is the same as having utf-8 */
1373 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1375 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1376 strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1379 * NOTE! This ordering means that the same exact tree merged with a
1380 * different order of parents will be a _different_ changeset even
1381 * if everything else stays the same.
1383 while (parents) {
1384 struct commit *parent = pop_commit(&parents);
1385 strbuf_addf(&buffer, "parent %s\n",
1386 oid_to_hex(&parent->object.oid));
1389 /* Person/date information */
1390 if (!author)
1391 author = git_author_info(IDENT_STRICT);
1392 strbuf_addf(&buffer, "author %s\n", author);
1393 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1394 if (!encoding_is_utf8)
1395 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1397 while (extra) {
1398 add_extra_header(&buffer, extra);
1399 extra = extra->next;
1401 strbuf_addch(&buffer, '\n');
1403 /* And add the comment */
1404 strbuf_add(&buffer, msg, msg_len);
1406 /* And check the encoding */
1407 if (encoding_is_utf8 && !verify_utf8(&buffer))
1408 fprintf(stderr, _(commit_utf8_warn));
1410 if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1411 result = -1;
1412 goto out;
1415 result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1416 out:
1417 strbuf_release(&buffer);
1418 return result;
1421 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1422 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1424 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1426 return *merge_desc_slab_at(&merge_desc_slab, commit);
1429 void set_merge_remote_desc(struct commit *commit,
1430 const char *name, struct object *obj)
1432 struct merge_remote_desc *desc;
1433 FLEX_ALLOC_STR(desc, name, name);
1434 desc->obj = obj;
1435 *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1438 struct commit *get_merge_parent(const char *name)
1440 struct object *obj;
1441 struct commit *commit;
1442 struct object_id oid;
1443 if (get_oid(name, &oid))
1444 return NULL;
1445 obj = parse_object(the_repository, &oid);
1446 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1447 if (commit && !merge_remote_util(commit))
1448 set_merge_remote_desc(commit, name, obj);
1449 return commit;
1453 * Append a commit to the end of the commit_list.
1455 * next starts by pointing to the variable that holds the head of an
1456 * empty commit_list, and is updated to point to the "next" field of
1457 * the last item on the list as new commits are appended.
1459 * Usage example:
1461 * struct commit_list *list;
1462 * struct commit_list **next = &list;
1464 * next = commit_list_append(c1, next);
1465 * next = commit_list_append(c2, next);
1466 * assert(commit_list_count(list) == 2);
1467 * return list;
1469 struct commit_list **commit_list_append(struct commit *commit,
1470 struct commit_list **next)
1472 struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1473 new_commit->item = commit;
1474 *next = new_commit;
1475 new_commit->next = NULL;
1476 return &new_commit->next;
1479 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1481 int key_len = strlen(key);
1482 const char *line = msg;
1484 while (line) {
1485 const char *eol = strchrnul(line, '\n');
1487 if (line == eol)
1488 return NULL;
1490 if (eol - line > key_len &&
1491 !strncmp(line, key, key_len) &&
1492 line[key_len] == ' ') {
1493 *out_len = eol - line - key_len - 1;
1494 return line + key_len + 1;
1496 line = *eol ? eol + 1 : NULL;
1498 return NULL;
1502 * Inspect the given string and determine the true "end" of the log message, in
1503 * order to find where to put a new Signed-off-by: line. Ignored are
1504 * trailing comment lines and blank lines. To support "git commit -s
1505 * --amend" on an existing commit, we also ignore "Conflicts:". To
1506 * support "git commit -v", we truncate at cut lines.
1508 * Returns the number of bytes from the tail to ignore, to be fed as
1509 * the second parameter to append_signoff().
1511 size_t ignore_non_trailer(const char *buf, size_t len)
1513 size_t boc = 0;
1514 size_t bol = 0;
1515 int in_old_conflicts_block = 0;
1516 size_t cutoff = wt_status_locate_end(buf, len);
1518 while (bol < cutoff) {
1519 const char *next_line = memchr(buf + bol, '\n', len - bol);
1521 if (!next_line)
1522 next_line = buf + len;
1523 else
1524 next_line++;
1526 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1527 /* is this the first of the run of comments? */
1528 if (!boc)
1529 boc = bol;
1530 /* otherwise, it is just continuing */
1531 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1532 in_old_conflicts_block = 1;
1533 if (!boc)
1534 boc = bol;
1535 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1536 ; /* a pathname in the conflicts block */
1537 } else if (boc) {
1538 /* the previous was not trailing comment */
1539 boc = 0;
1540 in_old_conflicts_block = 0;
1542 bol = next_line - buf;
1544 return boc ? len - boc : len - cutoff;