subtree: use commits before rejoins for splits
[git.git] / commit.c
blobd0f199e12286983c3f9f23b6567dc78a2ceeaf79
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"
21 static struct commit_extra_header *read_commit_extra_header_lines(const char *buf, size_t len, const char **);
23 int save_commit_buffer = 1;
25 const char *commit_type = "commit";
27 struct commit *lookup_commit_reference_gently(struct repository *r,
28 const struct object_id *oid, int quiet)
30 struct object *obj = deref_tag(r,
31 parse_object(r, oid),
32 NULL, 0);
34 if (!obj)
35 return NULL;
36 return object_as_type(r, obj, OBJ_COMMIT, quiet);
39 struct commit *lookup_commit_reference(struct repository *r, const struct object_id *oid)
41 return lookup_commit_reference_gently(r, oid, 0);
44 struct commit *lookup_commit_or_die(const struct object_id *oid, const char *ref_name)
46 struct commit *c = lookup_commit_reference(the_repository, oid);
47 if (!c)
48 die(_("could not parse %s"), ref_name);
49 if (!oideq(oid, &c->object.oid)) {
50 warning(_("%s %s is not a commit!"),
51 ref_name, oid_to_hex(oid));
53 return c;
56 struct commit *lookup_commit(struct repository *r, const struct object_id *oid)
58 struct object *obj = lookup_object(r, oid->hash);
59 if (!obj)
60 return create_object(r, oid->hash,
61 alloc_commit_node(r));
62 return object_as_type(r, obj, OBJ_COMMIT, 0);
65 struct commit *lookup_commit_reference_by_name(const char *name)
67 struct object_id oid;
68 struct commit *commit;
70 if (get_oid_committish(name, &oid))
71 return NULL;
72 commit = lookup_commit_reference(the_repository, &oid);
73 if (parse_commit(commit))
74 return NULL;
75 return commit;
78 static timestamp_t parse_commit_date(const char *buf, const char *tail)
80 const char *dateptr;
82 if (buf + 6 >= tail)
83 return 0;
84 if (memcmp(buf, "author", 6))
85 return 0;
86 while (buf < tail && *buf++ != '\n')
87 /* nada */;
88 if (buf + 9 >= tail)
89 return 0;
90 if (memcmp(buf, "committer", 9))
91 return 0;
92 while (buf < tail && *buf++ != '>')
93 /* nada */;
94 if (buf >= tail)
95 return 0;
96 dateptr = buf;
97 while (buf < tail && *buf++ != '\n')
98 /* nada */;
99 if (buf >= tail)
100 return 0;
101 /* dateptr < buf && buf[-1] == '\n', so parsing will stop at buf-1 */
102 return parse_timestamp(dateptr, NULL, 10);
105 static const unsigned char *commit_graft_sha1_access(size_t index, void *table)
107 struct commit_graft **commit_graft_table = table;
108 return commit_graft_table[index]->oid.hash;
111 static int commit_graft_pos(struct repository *r, const unsigned char *sha1)
113 return sha1_pos(sha1, r->parsed_objects->grafts,
114 r->parsed_objects->grafts_nr,
115 commit_graft_sha1_access);
118 int register_commit_graft(struct repository *r, struct commit_graft *graft,
119 int ignore_dups)
121 int pos = commit_graft_pos(r, graft->oid.hash);
123 if (0 <= pos) {
124 if (ignore_dups)
125 free(graft);
126 else {
127 free(r->parsed_objects->grafts[pos]);
128 r->parsed_objects->grafts[pos] = graft;
130 return 1;
132 pos = -pos - 1;
133 ALLOC_GROW(r->parsed_objects->grafts,
134 r->parsed_objects->grafts_nr + 1,
135 r->parsed_objects->grafts_alloc);
136 r->parsed_objects->grafts_nr++;
137 if (pos < r->parsed_objects->grafts_nr)
138 memmove(r->parsed_objects->grafts + pos + 1,
139 r->parsed_objects->grafts + pos,
140 (r->parsed_objects->grafts_nr - pos - 1) *
141 sizeof(*r->parsed_objects->grafts));
142 r->parsed_objects->grafts[pos] = graft;
143 return 0;
146 struct commit_graft *read_graft_line(struct strbuf *line)
148 /* The format is just "Commit Parent1 Parent2 ...\n" */
149 int i, phase;
150 const char *tail = NULL;
151 struct commit_graft *graft = NULL;
152 struct object_id dummy_oid, *oid;
154 strbuf_rtrim(line);
155 if (!line->len || line->buf[0] == '#')
156 return NULL;
158 * phase 0 verifies line, counts hashes in line and allocates graft
159 * phase 1 fills graft
161 for (phase = 0; phase < 2; phase++) {
162 oid = graft ? &graft->oid : &dummy_oid;
163 if (parse_oid_hex(line->buf, oid, &tail))
164 goto bad_graft_data;
165 for (i = 0; *tail != '\0'; i++) {
166 oid = graft ? &graft->parent[i] : &dummy_oid;
167 if (!isspace(*tail++) || parse_oid_hex(tail, oid, &tail))
168 goto bad_graft_data;
170 if (!graft) {
171 graft = xmalloc(st_add(sizeof(*graft),
172 st_mult(sizeof(struct object_id), i)));
173 graft->nr_parent = i;
176 return graft;
178 bad_graft_data:
179 error("bad graft data: %s", line->buf);
180 assert(!graft);
181 return NULL;
184 static int read_graft_file(struct repository *r, const char *graft_file)
186 FILE *fp = fopen_or_warn(graft_file, "r");
187 struct strbuf buf = STRBUF_INIT;
188 if (!fp)
189 return -1;
190 if (advice_graft_file_deprecated)
191 advise(_("Support for <GIT_DIR>/info/grafts is deprecated\n"
192 "and will be removed in a future Git version.\n"
193 "\n"
194 "Please use \"git replace --convert-graft-file\"\n"
195 "to convert the grafts into replace refs.\n"
196 "\n"
197 "Turn this message off by running\n"
198 "\"git config advice.graftFileDeprecated false\""));
199 while (!strbuf_getwholeline(&buf, fp, '\n')) {
200 /* The format is just "Commit Parent1 Parent2 ...\n" */
201 struct commit_graft *graft = read_graft_line(&buf);
202 if (!graft)
203 continue;
204 if (register_commit_graft(r, graft, 1))
205 error("duplicate graft data: %s", buf.buf);
207 fclose(fp);
208 strbuf_release(&buf);
209 return 0;
212 static void prepare_commit_graft(struct repository *r)
214 char *graft_file;
216 if (r->parsed_objects->commit_graft_prepared)
217 return;
218 if (!startup_info->have_repository)
219 return;
221 graft_file = get_graft_file(r);
222 read_graft_file(r, graft_file);
223 /* make sure shallows are read */
224 is_repository_shallow(r);
225 r->parsed_objects->commit_graft_prepared = 1;
228 struct commit_graft *lookup_commit_graft(struct repository *r, const struct object_id *oid)
230 int pos;
231 prepare_commit_graft(r);
232 pos = commit_graft_pos(r, oid->hash);
233 if (pos < 0)
234 return NULL;
235 return r->parsed_objects->grafts[pos];
238 int for_each_commit_graft(each_commit_graft_fn fn, void *cb_data)
240 int i, ret;
241 for (i = ret = 0; i < the_repository->parsed_objects->grafts_nr && !ret; i++)
242 ret = fn(the_repository->parsed_objects->grafts[i], cb_data);
243 return ret;
246 int unregister_shallow(const struct object_id *oid)
248 int pos = commit_graft_pos(the_repository, oid->hash);
249 if (pos < 0)
250 return -1;
251 if (pos + 1 < the_repository->parsed_objects->grafts_nr)
252 MOVE_ARRAY(the_repository->parsed_objects->grafts + pos,
253 the_repository->parsed_objects->grafts + pos + 1,
254 the_repository->parsed_objects->grafts_nr - pos - 1);
255 the_repository->parsed_objects->grafts_nr--;
256 return 0;
259 struct commit_buffer {
260 void *buffer;
261 unsigned long size;
263 define_commit_slab(buffer_slab, struct commit_buffer);
265 struct buffer_slab *allocate_commit_buffer_slab(void)
267 struct buffer_slab *bs = xmalloc(sizeof(*bs));
268 init_buffer_slab(bs);
269 return bs;
272 void free_commit_buffer_slab(struct buffer_slab *bs)
274 clear_buffer_slab(bs);
275 free(bs);
278 void set_commit_buffer(struct repository *r, struct commit *commit, void *buffer, unsigned long size)
280 struct commit_buffer *v = buffer_slab_at(
281 r->parsed_objects->buffer_slab, commit);
282 v->buffer = buffer;
283 v->size = size;
286 const void *get_cached_commit_buffer(struct repository *r, const struct commit *commit, unsigned long *sizep)
288 struct commit_buffer *v = buffer_slab_peek(
289 r->parsed_objects->buffer_slab, commit);
290 if (!v) {
291 if (sizep)
292 *sizep = 0;
293 return NULL;
295 if (sizep)
296 *sizep = v->size;
297 return v->buffer;
300 const void *get_commit_buffer(const struct commit *commit, unsigned long *sizep)
302 const void *ret = get_cached_commit_buffer(the_repository, commit, sizep);
303 if (!ret) {
304 enum object_type type;
305 unsigned long size;
306 ret = read_object_file(&commit->object.oid, &type, &size);
307 if (!ret)
308 die("cannot read commit object %s",
309 oid_to_hex(&commit->object.oid));
310 if (type != OBJ_COMMIT)
311 die("expected commit for %s, got %s",
312 oid_to_hex(&commit->object.oid), type_name(type));
313 if (sizep)
314 *sizep = size;
316 return ret;
319 void unuse_commit_buffer(const struct commit *commit, const void *buffer)
321 struct commit_buffer *v = buffer_slab_peek(
322 the_repository->parsed_objects->buffer_slab, commit);
323 if (!(v && v->buffer == buffer))
324 free((void *)buffer);
327 void free_commit_buffer(struct commit *commit)
329 struct commit_buffer *v = buffer_slab_peek(
330 the_repository->parsed_objects->buffer_slab, commit);
331 if (v) {
332 FREE_AND_NULL(v->buffer);
333 v->size = 0;
337 struct tree *get_commit_tree(const struct commit *commit)
339 if (commit->maybe_tree || !commit->object.parsed)
340 return commit->maybe_tree;
342 if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
343 BUG("commit has NULL tree, but was not loaded from commit-graph");
345 return get_commit_tree_in_graph(the_repository, commit);
348 struct object_id *get_commit_tree_oid(const struct commit *commit)
350 return &get_commit_tree(commit)->object.oid;
353 void release_commit_memory(struct commit *c)
355 c->maybe_tree = NULL;
356 c->index = 0;
357 free_commit_buffer(c);
358 free_commit_list(c->parents);
359 /* TODO: what about commit->util? */
361 c->object.parsed = 0;
364 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
366 struct commit_buffer *v = buffer_slab_peek(
367 the_repository->parsed_objects->buffer_slab, commit);
368 void *ret;
370 if (!v) {
371 if (sizep)
372 *sizep = 0;
373 return NULL;
375 ret = v->buffer;
376 if (sizep)
377 *sizep = v->size;
379 v->buffer = NULL;
380 v->size = 0;
381 return ret;
384 int parse_commit_buffer(struct repository *r, struct commit *item, const void *buffer, unsigned long size, int check_graph)
386 const char *tail = buffer;
387 const char *bufptr = buffer;
388 struct object_id parent;
389 struct commit_list **pptr;
390 struct commit_graft *graft;
391 const int tree_entry_len = the_hash_algo->hexsz + 5;
392 const int parent_entry_len = the_hash_algo->hexsz + 7;
394 if (item->object.parsed)
395 return 0;
396 item->object.parsed = 1;
397 tail += size;
398 if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
399 bufptr[tree_entry_len] != '\n')
400 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
401 if (get_oid_hex(bufptr + 5, &parent) < 0)
402 return error("bad tree pointer in commit %s",
403 oid_to_hex(&item->object.oid));
404 item->maybe_tree = lookup_tree(r, &parent);
405 bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
406 pptr = &item->parents;
408 graft = lookup_commit_graft(r, &item->object.oid);
409 while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
410 struct commit *new_parent;
412 if (tail <= bufptr + parent_entry_len + 1 ||
413 get_oid_hex(bufptr + 7, &parent) ||
414 bufptr[parent_entry_len] != '\n')
415 return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
416 bufptr += parent_entry_len + 1;
418 * The clone is shallow if nr_parent < 0, and we must
419 * not traverse its real parents even when we unhide them.
421 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
422 continue;
423 new_parent = lookup_commit(r, &parent);
424 if (new_parent)
425 pptr = &commit_list_insert(new_parent, pptr)->next;
427 if (graft) {
428 int i;
429 struct commit *new_parent;
430 for (i = 0; i < graft->nr_parent; i++) {
431 new_parent = lookup_commit(r,
432 &graft->parent[i]);
433 if (!new_parent)
434 continue;
435 pptr = &commit_list_insert(new_parent, pptr)->next;
438 item->date = parse_commit_date(bufptr, tail);
440 if (check_graph)
441 load_commit_graph_info(the_repository, item);
443 return 0;
446 int parse_commit_internal(struct commit *item, int quiet_on_missing, int use_commit_graph)
448 enum object_type type;
449 void *buffer;
450 unsigned long size;
451 int ret;
453 if (!item)
454 return -1;
455 if (item->object.parsed)
456 return 0;
457 if (use_commit_graph && parse_commit_in_graph(the_repository, item))
458 return 0;
459 buffer = read_object_file(&item->object.oid, &type, &size);
460 if (!buffer)
461 return quiet_on_missing ? -1 :
462 error("Could not read %s",
463 oid_to_hex(&item->object.oid));
464 if (type != OBJ_COMMIT) {
465 free(buffer);
466 return error("Object %s not a commit",
467 oid_to_hex(&item->object.oid));
470 ret = parse_commit_buffer(the_repository, item, buffer, size, 0);
471 if (save_commit_buffer && !ret) {
472 set_commit_buffer(the_repository, item, buffer, size);
473 return 0;
475 free(buffer);
476 return ret;
479 int parse_commit_gently(struct commit *item, int quiet_on_missing)
481 return parse_commit_internal(item, quiet_on_missing, 1);
484 void parse_commit_or_die(struct commit *item)
486 if (parse_commit(item))
487 die("unable to parse commit %s",
488 item ? oid_to_hex(&item->object.oid) : "(null)");
491 int find_commit_subject(const char *commit_buffer, const char **subject)
493 const char *eol;
494 const char *p = commit_buffer;
496 while (*p && (*p != '\n' || p[1] != '\n'))
497 p++;
498 if (*p) {
499 p = skip_blank_lines(p + 2);
500 eol = strchrnul(p, '\n');
501 } else
502 eol = p;
504 *subject = p;
506 return eol - p;
509 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
511 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
512 new_list->item = item;
513 new_list->next = *list_p;
514 *list_p = new_list;
515 return new_list;
518 unsigned commit_list_count(const struct commit_list *l)
520 unsigned c = 0;
521 for (; l; l = l->next )
522 c++;
523 return c;
526 struct commit_list *copy_commit_list(struct commit_list *list)
528 struct commit_list *head = NULL;
529 struct commit_list **pp = &head;
530 while (list) {
531 pp = commit_list_append(list->item, pp);
532 list = list->next;
534 return head;
537 void free_commit_list(struct commit_list *list)
539 while (list)
540 pop_commit(&list);
543 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
545 struct commit_list **pp = list;
546 struct commit_list *p;
547 while ((p = *pp) != NULL) {
548 if (p->item->date < item->date) {
549 break;
551 pp = &p->next;
553 return commit_list_insert(item, pp);
556 static int commit_list_compare_by_date(const void *a, const void *b)
558 timestamp_t a_date = ((const struct commit_list *)a)->item->date;
559 timestamp_t b_date = ((const struct commit_list *)b)->item->date;
560 if (a_date < b_date)
561 return 1;
562 if (a_date > b_date)
563 return -1;
564 return 0;
567 static void *commit_list_get_next(const void *a)
569 return ((const struct commit_list *)a)->next;
572 static void commit_list_set_next(void *a, void *next)
574 ((struct commit_list *)a)->next = next;
577 void commit_list_sort_by_date(struct commit_list **list)
579 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
580 commit_list_compare_by_date);
583 struct commit *pop_most_recent_commit(struct commit_list **list,
584 unsigned int mark)
586 struct commit *ret = pop_commit(list);
587 struct commit_list *parents = ret->parents;
589 while (parents) {
590 struct commit *commit = parents->item;
591 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
592 commit->object.flags |= mark;
593 commit_list_insert_by_date(commit, list);
595 parents = parents->next;
597 return ret;
600 static void clear_commit_marks_1(struct commit_list **plist,
601 struct commit *commit, unsigned int mark)
603 while (commit) {
604 struct commit_list *parents;
606 if (!(mark & commit->object.flags))
607 return;
609 commit->object.flags &= ~mark;
611 parents = commit->parents;
612 if (!parents)
613 return;
615 while ((parents = parents->next))
616 commit_list_insert(parents->item, plist);
618 commit = commit->parents->item;
622 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
624 struct commit_list *list = NULL;
626 while (nr--) {
627 clear_commit_marks_1(&list, *commit, mark);
628 commit++;
630 while (list)
631 clear_commit_marks_1(&list, pop_commit(&list), mark);
634 void clear_commit_marks(struct commit *commit, unsigned int mark)
636 clear_commit_marks_many(1, &commit, mark);
639 struct commit *pop_commit(struct commit_list **stack)
641 struct commit_list *top = *stack;
642 struct commit *item = top ? top->item : NULL;
644 if (top) {
645 *stack = top->next;
646 free(top);
648 return item;
652 * Topological sort support
655 /* count number of children that have not been emitted */
656 define_commit_slab(indegree_slab, int);
658 /* record author-date for each commit object */
659 define_commit_slab(author_date_slab, timestamp_t);
661 static void record_author_date(struct author_date_slab *author_date,
662 struct commit *commit)
664 const char *buffer = get_commit_buffer(commit, NULL);
665 struct ident_split ident;
666 const char *ident_line;
667 size_t ident_len;
668 char *date_end;
669 timestamp_t date;
671 ident_line = find_commit_header(buffer, "author", &ident_len);
672 if (!ident_line)
673 goto fail_exit; /* no author line */
674 if (split_ident_line(&ident, ident_line, ident_len) ||
675 !ident.date_begin || !ident.date_end)
676 goto fail_exit; /* malformed "author" line */
678 date = parse_timestamp(ident.date_begin, &date_end, 10);
679 if (date_end != ident.date_end)
680 goto fail_exit; /* malformed date */
681 *(author_date_slab_at(author_date, commit)) = date;
683 fail_exit:
684 unuse_commit_buffer(commit, buffer);
687 static int compare_commits_by_author_date(const void *a_, const void *b_,
688 void *cb_data)
690 const struct commit *a = a_, *b = b_;
691 struct author_date_slab *author_date = cb_data;
692 timestamp_t a_date = *(author_date_slab_at(author_date, a));
693 timestamp_t b_date = *(author_date_slab_at(author_date, b));
695 /* newer commits with larger date first */
696 if (a_date < b_date)
697 return 1;
698 else if (a_date > b_date)
699 return -1;
700 return 0;
703 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
705 const struct commit *a = a_, *b = b_;
707 /* newer commits first */
708 if (a->generation < b->generation)
709 return 1;
710 else if (a->generation > b->generation)
711 return -1;
713 /* use date as a heuristic when generations are equal */
714 if (a->date < b->date)
715 return 1;
716 else if (a->date > b->date)
717 return -1;
718 return 0;
721 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
723 const struct commit *a = a_, *b = b_;
724 /* newer commits with larger date first */
725 if (a->date < b->date)
726 return 1;
727 else if (a->date > b->date)
728 return -1;
729 return 0;
733 * Performs an in-place topological sort on the list supplied.
735 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
737 struct commit_list *next, *orig = *list;
738 struct commit_list **pptr;
739 struct indegree_slab indegree;
740 struct prio_queue queue;
741 struct commit *commit;
742 struct author_date_slab author_date;
744 if (!orig)
745 return;
746 *list = NULL;
748 init_indegree_slab(&indegree);
749 memset(&queue, '\0', sizeof(queue));
751 switch (sort_order) {
752 default: /* REV_SORT_IN_GRAPH_ORDER */
753 queue.compare = NULL;
754 break;
755 case REV_SORT_BY_COMMIT_DATE:
756 queue.compare = compare_commits_by_commit_date;
757 break;
758 case REV_SORT_BY_AUTHOR_DATE:
759 init_author_date_slab(&author_date);
760 queue.compare = compare_commits_by_author_date;
761 queue.cb_data = &author_date;
762 break;
765 /* Mark them and clear the indegree */
766 for (next = orig; next; next = next->next) {
767 struct commit *commit = next->item;
768 *(indegree_slab_at(&indegree, commit)) = 1;
769 /* also record the author dates, if needed */
770 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
771 record_author_date(&author_date, commit);
774 /* update the indegree */
775 for (next = orig; next; next = next->next) {
776 struct commit_list *parents = next->item->parents;
777 while (parents) {
778 struct commit *parent = parents->item;
779 int *pi = indegree_slab_at(&indegree, parent);
781 if (*pi)
782 (*pi)++;
783 parents = parents->next;
788 * find the tips
790 * tips are nodes not reachable from any other node in the list
792 * the tips serve as a starting set for the work queue.
794 for (next = orig; next; next = next->next) {
795 struct commit *commit = next->item;
797 if (*(indegree_slab_at(&indegree, commit)) == 1)
798 prio_queue_put(&queue, commit);
802 * This is unfortunate; the initial tips need to be shown
803 * in the order given from the revision traversal machinery.
805 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
806 prio_queue_reverse(&queue);
808 /* We no longer need the commit list */
809 free_commit_list(orig);
811 pptr = list;
812 *list = NULL;
813 while ((commit = prio_queue_get(&queue)) != NULL) {
814 struct commit_list *parents;
816 for (parents = commit->parents; parents ; parents = parents->next) {
817 struct commit *parent = parents->item;
818 int *pi = indegree_slab_at(&indegree, parent);
820 if (!*pi)
821 continue;
824 * parents are only enqueued for emission
825 * when all their children have been emitted thereby
826 * guaranteeing topological order.
828 if (--(*pi) == 1)
829 prio_queue_put(&queue, parent);
832 * all children of commit have already been
833 * emitted. we can emit it now.
835 *(indegree_slab_at(&indegree, commit)) = 0;
837 pptr = &commit_list_insert(commit, pptr)->next;
840 clear_indegree_slab(&indegree);
841 clear_prio_queue(&queue);
842 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
843 clear_author_date_slab(&author_date);
846 static const char gpg_sig_header[] = "gpgsig";
847 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
849 static int do_sign_commit(struct strbuf *buf, const char *keyid)
851 struct strbuf sig = STRBUF_INIT;
852 int inspos, copypos;
853 const char *eoh;
855 /* find the end of the header */
856 eoh = strstr(buf->buf, "\n\n");
857 if (!eoh)
858 inspos = buf->len;
859 else
860 inspos = eoh - buf->buf + 1;
862 if (!keyid || !*keyid)
863 keyid = get_signing_key();
864 if (sign_buffer(buf, &sig, keyid)) {
865 strbuf_release(&sig);
866 return -1;
869 for (copypos = 0; sig.buf[copypos]; ) {
870 const char *bol = sig.buf + copypos;
871 const char *eol = strchrnul(bol, '\n');
872 int len = (eol - bol) + !!*eol;
874 if (!copypos) {
875 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
876 inspos += gpg_sig_header_len;
878 strbuf_insert(buf, inspos++, " ", 1);
879 strbuf_insert(buf, inspos, bol, len);
880 inspos += len;
881 copypos += len;
883 strbuf_release(&sig);
884 return 0;
887 int parse_signed_commit(const struct commit *commit,
888 struct strbuf *payload, struct strbuf *signature)
891 unsigned long size;
892 const char *buffer = get_commit_buffer(commit, &size);
893 int in_signature, saw_signature = -1;
894 const char *line, *tail;
896 line = buffer;
897 tail = buffer + size;
898 in_signature = 0;
899 saw_signature = 0;
900 while (line < tail) {
901 const char *sig = NULL;
902 const char *next = memchr(line, '\n', tail - line);
904 next = next ? next + 1 : tail;
905 if (in_signature && line[0] == ' ')
906 sig = line + 1;
907 else if (starts_with(line, gpg_sig_header) &&
908 line[gpg_sig_header_len] == ' ')
909 sig = line + gpg_sig_header_len + 1;
910 if (sig) {
911 strbuf_add(signature, sig, next - sig);
912 saw_signature = 1;
913 in_signature = 1;
914 } else {
915 if (*line == '\n')
916 /* dump the whole remainder of the buffer */
917 next = tail;
918 strbuf_add(payload, line, next - line);
919 in_signature = 0;
921 line = next;
923 unuse_commit_buffer(commit, buffer);
924 return saw_signature;
927 int remove_signature(struct strbuf *buf)
929 const char *line = buf->buf;
930 const char *tail = buf->buf + buf->len;
931 int in_signature = 0;
932 const char *sig_start = NULL;
933 const char *sig_end = NULL;
935 while (line < tail) {
936 const char *next = memchr(line, '\n', tail - line);
937 next = next ? next + 1 : tail;
939 if (in_signature && line[0] == ' ')
940 sig_end = next;
941 else if (starts_with(line, gpg_sig_header) &&
942 line[gpg_sig_header_len] == ' ') {
943 sig_start = line;
944 sig_end = next;
945 in_signature = 1;
946 } else {
947 if (*line == '\n')
948 /* dump the whole remainder of the buffer */
949 next = tail;
950 in_signature = 0;
952 line = next;
955 if (sig_start)
956 strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
958 return sig_start != NULL;
961 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
963 struct merge_remote_desc *desc;
964 struct commit_extra_header *mergetag;
965 char *buf;
966 unsigned long size, len;
967 enum object_type type;
969 desc = merge_remote_util(parent);
970 if (!desc || !desc->obj)
971 return;
972 buf = read_object_file(&desc->obj->oid, &type, &size);
973 if (!buf || type != OBJ_TAG)
974 goto free_return;
975 len = parse_signature(buf, size);
976 if (size == len)
977 goto free_return;
979 * We could verify this signature and either omit the tag when
980 * it does not validate, but the integrator may not have the
981 * public key of the signer of the tag he is merging, while a
982 * later auditor may have it while auditing, so let's not run
983 * verify-signed-buffer here for now...
985 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
986 * warn("warning: signed tag unverified.");
988 mergetag = xcalloc(1, sizeof(*mergetag));
989 mergetag->key = xstrdup("mergetag");
990 mergetag->value = buf;
991 mergetag->len = size;
993 **tail = mergetag;
994 *tail = &mergetag->next;
995 return;
997 free_return:
998 free(buf);
1001 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1003 struct strbuf payload = STRBUF_INIT;
1004 struct strbuf signature = STRBUF_INIT;
1005 int ret = 1;
1007 sigc->result = 'N';
1009 if (parse_signed_commit(commit, &payload, &signature) <= 0)
1010 goto out;
1011 ret = check_signature(payload.buf, payload.len, signature.buf,
1012 signature.len, sigc);
1014 out:
1015 strbuf_release(&payload);
1016 strbuf_release(&signature);
1018 return ret;
1023 void append_merge_tag_headers(struct commit_list *parents,
1024 struct commit_extra_header ***tail)
1026 while (parents) {
1027 struct commit *parent = parents->item;
1028 handle_signed_tag(parent, tail);
1029 parents = parents->next;
1033 static void add_extra_header(struct strbuf *buffer,
1034 struct commit_extra_header *extra)
1036 strbuf_addstr(buffer, extra->key);
1037 if (extra->len)
1038 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1039 else
1040 strbuf_addch(buffer, '\n');
1043 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1044 const char **exclude)
1046 struct commit_extra_header *extra = NULL;
1047 unsigned long size;
1048 const char *buffer = get_commit_buffer(commit, &size);
1049 extra = read_commit_extra_header_lines(buffer, size, exclude);
1050 unuse_commit_buffer(commit, buffer);
1051 return extra;
1054 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1056 struct commit_extra_header *extra, *to_free;
1057 int res = 0;
1059 to_free = read_commit_extra_headers(commit, NULL);
1060 for (extra = to_free; !res && extra; extra = extra->next) {
1061 if (strcmp(extra->key, "mergetag"))
1062 continue; /* not a merge tag */
1063 res = fn(commit, extra, data);
1065 free_commit_extra_headers(to_free);
1066 return res;
1069 static inline int standard_header_field(const char *field, size_t len)
1071 return ((len == 4 && !memcmp(field, "tree", 4)) ||
1072 (len == 6 && !memcmp(field, "parent", 6)) ||
1073 (len == 6 && !memcmp(field, "author", 6)) ||
1074 (len == 9 && !memcmp(field, "committer", 9)) ||
1075 (len == 8 && !memcmp(field, "encoding", 8)));
1078 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1080 if (!exclude)
1081 return 0;
1083 while (*exclude) {
1084 size_t xlen = strlen(*exclude);
1085 if (len == xlen && !memcmp(field, *exclude, xlen))
1086 return 1;
1087 exclude++;
1089 return 0;
1092 static struct commit_extra_header *read_commit_extra_header_lines(
1093 const char *buffer, size_t size,
1094 const char **exclude)
1096 struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1097 const char *line, *next, *eof, *eob;
1098 struct strbuf buf = STRBUF_INIT;
1100 for (line = buffer, eob = line + size;
1101 line < eob && *line != '\n';
1102 line = next) {
1103 next = memchr(line, '\n', eob - line);
1104 next = next ? next + 1 : eob;
1105 if (*line == ' ') {
1106 /* continuation */
1107 if (it)
1108 strbuf_add(&buf, line + 1, next - (line + 1));
1109 continue;
1111 if (it)
1112 it->value = strbuf_detach(&buf, &it->len);
1113 strbuf_reset(&buf);
1114 it = NULL;
1116 eof = memchr(line, ' ', next - line);
1117 if (!eof)
1118 eof = next;
1119 else if (standard_header_field(line, eof - line) ||
1120 excluded_header_field(line, eof - line, exclude))
1121 continue;
1123 it = xcalloc(1, sizeof(*it));
1124 it->key = xmemdupz(line, eof-line);
1125 *tail = it;
1126 tail = &it->next;
1127 if (eof + 1 < next)
1128 strbuf_add(&buf, eof + 1, next - (eof + 1));
1130 if (it)
1131 it->value = strbuf_detach(&buf, &it->len);
1132 return extra;
1135 void free_commit_extra_headers(struct commit_extra_header *extra)
1137 while (extra) {
1138 struct commit_extra_header *next = extra->next;
1139 free(extra->key);
1140 free(extra->value);
1141 free(extra);
1142 extra = next;
1146 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1147 struct commit_list *parents, struct object_id *ret,
1148 const char *author, const char *sign_commit)
1150 struct commit_extra_header *extra = NULL, **tail = &extra;
1151 int result;
1153 append_merge_tag_headers(parents, &tail);
1154 result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1155 author, sign_commit, extra);
1156 free_commit_extra_headers(extra);
1157 return result;
1160 static int find_invalid_utf8(const char *buf, int len)
1162 int offset = 0;
1163 static const unsigned int max_codepoint[] = {
1164 0x7f, 0x7ff, 0xffff, 0x10ffff
1167 while (len) {
1168 unsigned char c = *buf++;
1169 int bytes, bad_offset;
1170 unsigned int codepoint;
1171 unsigned int min_val, max_val;
1173 len--;
1174 offset++;
1176 /* Simple US-ASCII? No worries. */
1177 if (c < 0x80)
1178 continue;
1180 bad_offset = offset-1;
1183 * Count how many more high bits set: that's how
1184 * many more bytes this sequence should have.
1186 bytes = 0;
1187 while (c & 0x40) {
1188 c <<= 1;
1189 bytes++;
1193 * Must be between 1 and 3 more bytes. Longer sequences result in
1194 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1196 if (bytes < 1 || 3 < bytes)
1197 return bad_offset;
1199 /* Do we *have* that many bytes? */
1200 if (len < bytes)
1201 return bad_offset;
1204 * Place the encoded bits at the bottom of the value and compute the
1205 * valid range.
1207 codepoint = (c & 0x7f) >> bytes;
1208 min_val = max_codepoint[bytes-1] + 1;
1209 max_val = max_codepoint[bytes];
1211 offset += bytes;
1212 len -= bytes;
1214 /* And verify that they are good continuation bytes */
1215 do {
1216 codepoint <<= 6;
1217 codepoint |= *buf & 0x3f;
1218 if ((*buf++ & 0xc0) != 0x80)
1219 return bad_offset;
1220 } while (--bytes);
1222 /* Reject codepoints that are out of range for the sequence length. */
1223 if (codepoint < min_val || codepoint > max_val)
1224 return bad_offset;
1225 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1226 if ((codepoint & 0x1ff800) == 0xd800)
1227 return bad_offset;
1228 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1229 if ((codepoint & 0xfffe) == 0xfffe)
1230 return bad_offset;
1231 /* So are anything in the range U+FDD0..U+FDEF. */
1232 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1233 return bad_offset;
1235 return -1;
1239 * This verifies that the buffer is in proper utf8 format.
1241 * If it isn't, it assumes any non-utf8 characters are Latin1,
1242 * and does the conversion.
1244 static int verify_utf8(struct strbuf *buf)
1246 int ok = 1;
1247 long pos = 0;
1249 for (;;) {
1250 int bad;
1251 unsigned char c;
1252 unsigned char replace[2];
1254 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1255 if (bad < 0)
1256 return ok;
1257 pos += bad;
1258 ok = 0;
1259 c = buf->buf[pos];
1260 strbuf_remove(buf, pos, 1);
1262 /* We know 'c' must be in the range 128-255 */
1263 replace[0] = 0xc0 + (c >> 6);
1264 replace[1] = 0x80 + (c & 0x3f);
1265 strbuf_insert(buf, pos, replace, 2);
1266 pos += 2;
1270 static const char commit_utf8_warn[] =
1271 N_("Warning: commit message did not conform to UTF-8.\n"
1272 "You may want to amend it after fixing the message, or set the config\n"
1273 "variable i18n.commitencoding to the encoding your project uses.\n");
1275 int commit_tree_extended(const char *msg, size_t msg_len,
1276 const struct object_id *tree,
1277 struct commit_list *parents, struct object_id *ret,
1278 const char *author, const char *sign_commit,
1279 struct commit_extra_header *extra)
1281 int result;
1282 int encoding_is_utf8;
1283 struct strbuf buffer;
1285 assert_oid_type(tree, OBJ_TREE);
1287 if (memchr(msg, '\0', msg_len))
1288 return error("a NUL byte in commit log message not allowed.");
1290 /* Not having i18n.commitencoding is the same as having utf-8 */
1291 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1293 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1294 strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1297 * NOTE! This ordering means that the same exact tree merged with a
1298 * different order of parents will be a _different_ changeset even
1299 * if everything else stays the same.
1301 while (parents) {
1302 struct commit *parent = pop_commit(&parents);
1303 strbuf_addf(&buffer, "parent %s\n",
1304 oid_to_hex(&parent->object.oid));
1307 /* Person/date information */
1308 if (!author)
1309 author = git_author_info(IDENT_STRICT);
1310 strbuf_addf(&buffer, "author %s\n", author);
1311 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1312 if (!encoding_is_utf8)
1313 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1315 while (extra) {
1316 add_extra_header(&buffer, extra);
1317 extra = extra->next;
1319 strbuf_addch(&buffer, '\n');
1321 /* And add the comment */
1322 strbuf_add(&buffer, msg, msg_len);
1324 /* And check the encoding */
1325 if (encoding_is_utf8 && !verify_utf8(&buffer))
1326 fprintf(stderr, _(commit_utf8_warn));
1328 if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1329 result = -1;
1330 goto out;
1333 result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1334 out:
1335 strbuf_release(&buffer);
1336 return result;
1339 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1340 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1342 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1344 return *merge_desc_slab_at(&merge_desc_slab, commit);
1347 void set_merge_remote_desc(struct commit *commit,
1348 const char *name, struct object *obj)
1350 struct merge_remote_desc *desc;
1351 FLEX_ALLOC_STR(desc, name, name);
1352 desc->obj = obj;
1353 *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1356 struct commit *get_merge_parent(const char *name)
1358 struct object *obj;
1359 struct commit *commit;
1360 struct object_id oid;
1361 if (get_oid(name, &oid))
1362 return NULL;
1363 obj = parse_object(the_repository, &oid);
1364 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1365 if (commit && !merge_remote_util(commit))
1366 set_merge_remote_desc(commit, name, obj);
1367 return commit;
1371 * Append a commit to the end of the commit_list.
1373 * next starts by pointing to the variable that holds the head of an
1374 * empty commit_list, and is updated to point to the "next" field of
1375 * the last item on the list as new commits are appended.
1377 * Usage example:
1379 * struct commit_list *list;
1380 * struct commit_list **next = &list;
1382 * next = commit_list_append(c1, next);
1383 * next = commit_list_append(c2, next);
1384 * assert(commit_list_count(list) == 2);
1385 * return list;
1387 struct commit_list **commit_list_append(struct commit *commit,
1388 struct commit_list **next)
1390 struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1391 new_commit->item = commit;
1392 *next = new_commit;
1393 new_commit->next = NULL;
1394 return &new_commit->next;
1397 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1399 int key_len = strlen(key);
1400 const char *line = msg;
1402 while (line) {
1403 const char *eol = strchrnul(line, '\n');
1405 if (line == eol)
1406 return NULL;
1408 if (eol - line > key_len &&
1409 !strncmp(line, key, key_len) &&
1410 line[key_len] == ' ') {
1411 *out_len = eol - line - key_len - 1;
1412 return line + key_len + 1;
1414 line = *eol ? eol + 1 : NULL;
1416 return NULL;
1420 * Inspect the given string and determine the true "end" of the log message, in
1421 * order to find where to put a new Signed-off-by: line. Ignored are
1422 * trailing comment lines and blank lines. To support "git commit -s
1423 * --amend" on an existing commit, we also ignore "Conflicts:". To
1424 * support "git commit -v", we truncate at cut lines.
1426 * Returns the number of bytes from the tail to ignore, to be fed as
1427 * the second parameter to append_signoff().
1429 size_t ignore_non_trailer(const char *buf, size_t len)
1431 size_t boc = 0;
1432 size_t bol = 0;
1433 int in_old_conflicts_block = 0;
1434 size_t cutoff = wt_status_locate_end(buf, len);
1436 while (bol < cutoff) {
1437 const char *next_line = memchr(buf + bol, '\n', len - bol);
1439 if (!next_line)
1440 next_line = buf + len;
1441 else
1442 next_line++;
1444 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1445 /* is this the first of the run of comments? */
1446 if (!boc)
1447 boc = bol;
1448 /* otherwise, it is just continuing */
1449 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1450 in_old_conflicts_block = 1;
1451 if (!boc)
1452 boc = bol;
1453 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1454 ; /* a pathname in the conflicts block */
1455 } else if (boc) {
1456 /* the previous was not trailing comment */
1457 boc = 0;
1458 in_old_conflicts_block = 0;
1460 bol = next_line - buf;
1462 return boc ? len - boc : len - cutoff;