trace2: report peak memory usage of the process
[git/debian.git] / commit.c
bloba5333c7ac6c373a13f9298b36be5ff94a90a3e3f
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 *repo_get_commit_buffer(struct repository *r,
303 const struct commit *commit,
304 unsigned long *sizep)
306 const void *ret = get_cached_commit_buffer(r, commit, sizep);
307 if (!ret) {
308 enum object_type type;
309 unsigned long size;
310 ret = repo_read_object_file(r, &commit->object.oid, &type, &size);
311 if (!ret)
312 die("cannot read commit object %s",
313 oid_to_hex(&commit->object.oid));
314 if (type != OBJ_COMMIT)
315 die("expected commit for %s, got %s",
316 oid_to_hex(&commit->object.oid), type_name(type));
317 if (sizep)
318 *sizep = size;
320 return ret;
323 void repo_unuse_commit_buffer(struct repository *r,
324 const struct commit *commit,
325 const void *buffer)
327 struct commit_buffer *v = buffer_slab_peek(
328 r->parsed_objects->buffer_slab, commit);
329 if (!(v && v->buffer == buffer))
330 free((void *)buffer);
333 void free_commit_buffer(struct parsed_object_pool *pool, struct commit *commit)
335 struct commit_buffer *v = buffer_slab_peek(
336 pool->buffer_slab, commit);
337 if (v) {
338 FREE_AND_NULL(v->buffer);
339 v->size = 0;
343 struct tree *get_commit_tree(const struct commit *commit)
345 if (commit->maybe_tree || !commit->object.parsed)
346 return commit->maybe_tree;
348 if (commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
349 BUG("commit has NULL tree, but was not loaded from commit-graph");
351 return get_commit_tree_in_graph(the_repository, commit);
354 struct object_id *get_commit_tree_oid(const struct commit *commit)
356 return &get_commit_tree(commit)->object.oid;
359 void release_commit_memory(struct parsed_object_pool *pool, struct commit *c)
361 c->maybe_tree = NULL;
362 c->index = 0;
363 free_commit_buffer(pool, c);
364 free_commit_list(c->parents);
366 c->object.parsed = 0;
369 const void *detach_commit_buffer(struct commit *commit, unsigned long *sizep)
371 struct commit_buffer *v = buffer_slab_peek(
372 the_repository->parsed_objects->buffer_slab, commit);
373 void *ret;
375 if (!v) {
376 if (sizep)
377 *sizep = 0;
378 return NULL;
380 ret = v->buffer;
381 if (sizep)
382 *sizep = v->size;
384 v->buffer = NULL;
385 v->size = 0;
386 return ret;
389 int parse_commit_buffer(struct repository *r, struct commit *item, const void *buffer, unsigned long size, int check_graph)
391 const char *tail = buffer;
392 const char *bufptr = buffer;
393 struct object_id parent;
394 struct commit_list **pptr;
395 struct commit_graft *graft;
396 const int tree_entry_len = the_hash_algo->hexsz + 5;
397 const int parent_entry_len = the_hash_algo->hexsz + 7;
399 if (item->object.parsed)
400 return 0;
401 item->object.parsed = 1;
402 tail += size;
403 if (tail <= bufptr + tree_entry_len + 1 || memcmp(bufptr, "tree ", 5) ||
404 bufptr[tree_entry_len] != '\n')
405 return error("bogus commit object %s", oid_to_hex(&item->object.oid));
406 if (get_oid_hex(bufptr + 5, &parent) < 0)
407 return error("bad tree pointer in commit %s",
408 oid_to_hex(&item->object.oid));
409 item->maybe_tree = lookup_tree(r, &parent);
410 bufptr += tree_entry_len + 1; /* "tree " + "hex sha1" + "\n" */
411 pptr = &item->parents;
413 graft = lookup_commit_graft(r, &item->object.oid);
414 while (bufptr + parent_entry_len < tail && !memcmp(bufptr, "parent ", 7)) {
415 struct commit *new_parent;
417 if (tail <= bufptr + parent_entry_len + 1 ||
418 get_oid_hex(bufptr + 7, &parent) ||
419 bufptr[parent_entry_len] != '\n')
420 return error("bad parents in commit %s", oid_to_hex(&item->object.oid));
421 bufptr += parent_entry_len + 1;
423 * The clone is shallow if nr_parent < 0, and we must
424 * not traverse its real parents even when we unhide them.
426 if (graft && (graft->nr_parent < 0 || grafts_replace_parents))
427 continue;
428 new_parent = lookup_commit(r, &parent);
429 if (new_parent)
430 pptr = &commit_list_insert(new_parent, pptr)->next;
432 if (graft) {
433 int i;
434 struct commit *new_parent;
435 for (i = 0; i < graft->nr_parent; i++) {
436 new_parent = lookup_commit(r,
437 &graft->parent[i]);
438 if (!new_parent)
439 continue;
440 pptr = &commit_list_insert(new_parent, pptr)->next;
443 item->date = parse_commit_date(bufptr, tail);
445 if (check_graph)
446 load_commit_graph_info(the_repository, item);
448 return 0;
451 int repo_parse_commit_internal(struct repository *r,
452 struct commit *item,
453 int quiet_on_missing,
454 int use_commit_graph)
456 enum object_type type;
457 void *buffer;
458 unsigned long size;
459 int ret;
461 if (!item)
462 return -1;
463 if (item->object.parsed)
464 return 0;
465 if (use_commit_graph && parse_commit_in_graph(r, item))
466 return 0;
467 buffer = repo_read_object_file(r, &item->object.oid, &type, &size);
468 if (!buffer)
469 return quiet_on_missing ? -1 :
470 error("Could not read %s",
471 oid_to_hex(&item->object.oid));
472 if (type != OBJ_COMMIT) {
473 free(buffer);
474 return error("Object %s not a commit",
475 oid_to_hex(&item->object.oid));
478 ret = parse_commit_buffer(r, item, buffer, size, 0);
479 if (save_commit_buffer && !ret) {
480 set_commit_buffer(r, item, buffer, size);
481 return 0;
483 free(buffer);
484 return ret;
487 int repo_parse_commit_gently(struct repository *r,
488 struct commit *item, int quiet_on_missing)
490 return repo_parse_commit_internal(r, item, quiet_on_missing, 1);
493 void parse_commit_or_die(struct commit *item)
495 if (parse_commit(item))
496 die("unable to parse commit %s",
497 item ? oid_to_hex(&item->object.oid) : "(null)");
500 int find_commit_subject(const char *commit_buffer, const char **subject)
502 const char *eol;
503 const char *p = commit_buffer;
505 while (*p && (*p != '\n' || p[1] != '\n'))
506 p++;
507 if (*p) {
508 p = skip_blank_lines(p + 2);
509 eol = strchrnul(p, '\n');
510 } else
511 eol = p;
513 *subject = p;
515 return eol - p;
518 struct commit_list *commit_list_insert(struct commit *item, struct commit_list **list_p)
520 struct commit_list *new_list = xmalloc(sizeof(struct commit_list));
521 new_list->item = item;
522 new_list->next = *list_p;
523 *list_p = new_list;
524 return new_list;
527 unsigned commit_list_count(const struct commit_list *l)
529 unsigned c = 0;
530 for (; l; l = l->next )
531 c++;
532 return c;
535 struct commit_list *copy_commit_list(struct commit_list *list)
537 struct commit_list *head = NULL;
538 struct commit_list **pp = &head;
539 while (list) {
540 pp = commit_list_append(list->item, pp);
541 list = list->next;
543 return head;
546 void free_commit_list(struct commit_list *list)
548 while (list)
549 pop_commit(&list);
552 struct commit_list * commit_list_insert_by_date(struct commit *item, struct commit_list **list)
554 struct commit_list **pp = list;
555 struct commit_list *p;
556 while ((p = *pp) != NULL) {
557 if (p->item->date < item->date) {
558 break;
560 pp = &p->next;
562 return commit_list_insert(item, pp);
565 static int commit_list_compare_by_date(const void *a, const void *b)
567 timestamp_t a_date = ((const struct commit_list *)a)->item->date;
568 timestamp_t b_date = ((const struct commit_list *)b)->item->date;
569 if (a_date < b_date)
570 return 1;
571 if (a_date > b_date)
572 return -1;
573 return 0;
576 static void *commit_list_get_next(const void *a)
578 return ((const struct commit_list *)a)->next;
581 static void commit_list_set_next(void *a, void *next)
583 ((struct commit_list *)a)->next = next;
586 void commit_list_sort_by_date(struct commit_list **list)
588 *list = llist_mergesort(*list, commit_list_get_next, commit_list_set_next,
589 commit_list_compare_by_date);
592 struct commit *pop_most_recent_commit(struct commit_list **list,
593 unsigned int mark)
595 struct commit *ret = pop_commit(list);
596 struct commit_list *parents = ret->parents;
598 while (parents) {
599 struct commit *commit = parents->item;
600 if (!parse_commit(commit) && !(commit->object.flags & mark)) {
601 commit->object.flags |= mark;
602 commit_list_insert_by_date(commit, list);
604 parents = parents->next;
606 return ret;
609 static void clear_commit_marks_1(struct commit_list **plist,
610 struct commit *commit, unsigned int mark)
612 while (commit) {
613 struct commit_list *parents;
615 if (!(mark & commit->object.flags))
616 return;
618 commit->object.flags &= ~mark;
620 parents = commit->parents;
621 if (!parents)
622 return;
624 while ((parents = parents->next))
625 commit_list_insert(parents->item, plist);
627 commit = commit->parents->item;
631 void clear_commit_marks_many(int nr, struct commit **commit, unsigned int mark)
633 struct commit_list *list = NULL;
635 while (nr--) {
636 clear_commit_marks_1(&list, *commit, mark);
637 commit++;
639 while (list)
640 clear_commit_marks_1(&list, pop_commit(&list), mark);
643 void clear_commit_marks(struct commit *commit, unsigned int mark)
645 clear_commit_marks_many(1, &commit, mark);
648 struct commit *pop_commit(struct commit_list **stack)
650 struct commit_list *top = *stack;
651 struct commit *item = top ? top->item : NULL;
653 if (top) {
654 *stack = top->next;
655 free(top);
657 return item;
661 * Topological sort support
664 /* count number of children that have not been emitted */
665 define_commit_slab(indegree_slab, int);
667 define_commit_slab(author_date_slab, timestamp_t);
669 void record_author_date(struct author_date_slab *author_date,
670 struct commit *commit)
672 const char *buffer = get_commit_buffer(commit, NULL);
673 struct ident_split ident;
674 const char *ident_line;
675 size_t ident_len;
676 char *date_end;
677 timestamp_t date;
679 ident_line = find_commit_header(buffer, "author", &ident_len);
680 if (!ident_line)
681 goto fail_exit; /* no author line */
682 if (split_ident_line(&ident, ident_line, ident_len) ||
683 !ident.date_begin || !ident.date_end)
684 goto fail_exit; /* malformed "author" line */
686 date = parse_timestamp(ident.date_begin, &date_end, 10);
687 if (date_end != ident.date_end)
688 goto fail_exit; /* malformed date */
689 *(author_date_slab_at(author_date, commit)) = date;
691 fail_exit:
692 unuse_commit_buffer(commit, buffer);
695 int compare_commits_by_author_date(const void *a_, const void *b_,
696 void *cb_data)
698 const struct commit *a = a_, *b = b_;
699 struct author_date_slab *author_date = cb_data;
700 timestamp_t a_date = *(author_date_slab_at(author_date, a));
701 timestamp_t b_date = *(author_date_slab_at(author_date, b));
703 /* newer commits with larger date first */
704 if (a_date < b_date)
705 return 1;
706 else if (a_date > b_date)
707 return -1;
708 return 0;
711 int compare_commits_by_gen_then_commit_date(const void *a_, const void *b_, void *unused)
713 const struct commit *a = a_, *b = b_;
715 /* newer commits first */
716 if (a->generation < b->generation)
717 return 1;
718 else if (a->generation > b->generation)
719 return -1;
721 /* use date as a heuristic when generations are equal */
722 if (a->date < b->date)
723 return 1;
724 else if (a->date > b->date)
725 return -1;
726 return 0;
729 int compare_commits_by_commit_date(const void *a_, const void *b_, void *unused)
731 const struct commit *a = a_, *b = b_;
732 /* newer commits with larger date first */
733 if (a->date < b->date)
734 return 1;
735 else if (a->date > b->date)
736 return -1;
737 return 0;
741 * Performs an in-place topological sort on the list supplied.
743 void sort_in_topological_order(struct commit_list **list, enum rev_sort_order sort_order)
745 struct commit_list *next, *orig = *list;
746 struct commit_list **pptr;
747 struct indegree_slab indegree;
748 struct prio_queue queue;
749 struct commit *commit;
750 struct author_date_slab author_date;
752 if (!orig)
753 return;
754 *list = NULL;
756 init_indegree_slab(&indegree);
757 memset(&queue, '\0', sizeof(queue));
759 switch (sort_order) {
760 default: /* REV_SORT_IN_GRAPH_ORDER */
761 queue.compare = NULL;
762 break;
763 case REV_SORT_BY_COMMIT_DATE:
764 queue.compare = compare_commits_by_commit_date;
765 break;
766 case REV_SORT_BY_AUTHOR_DATE:
767 init_author_date_slab(&author_date);
768 queue.compare = compare_commits_by_author_date;
769 queue.cb_data = &author_date;
770 break;
773 /* Mark them and clear the indegree */
774 for (next = orig; next; next = next->next) {
775 struct commit *commit = next->item;
776 *(indegree_slab_at(&indegree, commit)) = 1;
777 /* also record the author dates, if needed */
778 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
779 record_author_date(&author_date, commit);
782 /* update the indegree */
783 for (next = orig; next; next = next->next) {
784 struct commit_list *parents = next->item->parents;
785 while (parents) {
786 struct commit *parent = parents->item;
787 int *pi = indegree_slab_at(&indegree, parent);
789 if (*pi)
790 (*pi)++;
791 parents = parents->next;
796 * find the tips
798 * tips are nodes not reachable from any other node in the list
800 * the tips serve as a starting set for the work queue.
802 for (next = orig; next; next = next->next) {
803 struct commit *commit = next->item;
805 if (*(indegree_slab_at(&indegree, commit)) == 1)
806 prio_queue_put(&queue, commit);
810 * This is unfortunate; the initial tips need to be shown
811 * in the order given from the revision traversal machinery.
813 if (sort_order == REV_SORT_IN_GRAPH_ORDER)
814 prio_queue_reverse(&queue);
816 /* We no longer need the commit list */
817 free_commit_list(orig);
819 pptr = list;
820 *list = NULL;
821 while ((commit = prio_queue_get(&queue)) != NULL) {
822 struct commit_list *parents;
824 for (parents = commit->parents; parents ; parents = parents->next) {
825 struct commit *parent = parents->item;
826 int *pi = indegree_slab_at(&indegree, parent);
828 if (!*pi)
829 continue;
832 * parents are only enqueued for emission
833 * when all their children have been emitted thereby
834 * guaranteeing topological order.
836 if (--(*pi) == 1)
837 prio_queue_put(&queue, parent);
840 * all children of commit have already been
841 * emitted. we can emit it now.
843 *(indegree_slab_at(&indegree, commit)) = 0;
845 pptr = &commit_list_insert(commit, pptr)->next;
848 clear_indegree_slab(&indegree);
849 clear_prio_queue(&queue);
850 if (sort_order == REV_SORT_BY_AUTHOR_DATE)
851 clear_author_date_slab(&author_date);
854 struct rev_collect {
855 struct commit **commit;
856 int nr;
857 int alloc;
858 unsigned int initial : 1;
861 static void add_one_commit(struct object_id *oid, struct rev_collect *revs)
863 struct commit *commit;
865 if (is_null_oid(oid))
866 return;
868 commit = lookup_commit(the_repository, oid);
869 if (!commit ||
870 (commit->object.flags & TMP_MARK) ||
871 parse_commit(commit))
872 return;
874 ALLOC_GROW(revs->commit, revs->nr + 1, revs->alloc);
875 revs->commit[revs->nr++] = commit;
876 commit->object.flags |= TMP_MARK;
879 static int collect_one_reflog_ent(struct object_id *ooid, struct object_id *noid,
880 const char *ident, timestamp_t timestamp,
881 int tz, const char *message, void *cbdata)
883 struct rev_collect *revs = cbdata;
885 if (revs->initial) {
886 revs->initial = 0;
887 add_one_commit(ooid, revs);
889 add_one_commit(noid, revs);
890 return 0;
893 struct commit *get_fork_point(const char *refname, struct commit *commit)
895 struct object_id oid;
896 struct rev_collect revs;
897 struct commit_list *bases;
898 int i;
899 struct commit *ret = NULL;
901 memset(&revs, 0, sizeof(revs));
902 revs.initial = 1;
903 for_each_reflog_ent(refname, collect_one_reflog_ent, &revs);
905 if (!revs.nr && !get_oid(refname, &oid))
906 add_one_commit(&oid, &revs);
908 for (i = 0; i < revs.nr; i++)
909 revs.commit[i]->object.flags &= ~TMP_MARK;
911 bases = get_merge_bases_many(commit, revs.nr, revs.commit);
914 * There should be one and only one merge base, when we found
915 * a common ancestor among reflog entries.
917 if (!bases || bases->next)
918 goto cleanup_return;
920 /* And the found one must be one of the reflog entries */
921 for (i = 0; i < revs.nr; i++)
922 if (&bases->item->object == &revs.commit[i]->object)
923 break; /* found */
924 if (revs.nr <= i)
925 goto cleanup_return;
927 ret = bases->item;
929 cleanup_return:
930 free_commit_list(bases);
931 return ret;
934 static const char gpg_sig_header[] = "gpgsig";
935 static const int gpg_sig_header_len = sizeof(gpg_sig_header) - 1;
937 static int do_sign_commit(struct strbuf *buf, const char *keyid)
939 struct strbuf sig = STRBUF_INIT;
940 int inspos, copypos;
941 const char *eoh;
943 /* find the end of the header */
944 eoh = strstr(buf->buf, "\n\n");
945 if (!eoh)
946 inspos = buf->len;
947 else
948 inspos = eoh - buf->buf + 1;
950 if (!keyid || !*keyid)
951 keyid = get_signing_key();
952 if (sign_buffer(buf, &sig, keyid)) {
953 strbuf_release(&sig);
954 return -1;
957 for (copypos = 0; sig.buf[copypos]; ) {
958 const char *bol = sig.buf + copypos;
959 const char *eol = strchrnul(bol, '\n');
960 int len = (eol - bol) + !!*eol;
962 if (!copypos) {
963 strbuf_insert(buf, inspos, gpg_sig_header, gpg_sig_header_len);
964 inspos += gpg_sig_header_len;
966 strbuf_insert(buf, inspos++, " ", 1);
967 strbuf_insert(buf, inspos, bol, len);
968 inspos += len;
969 copypos += len;
971 strbuf_release(&sig);
972 return 0;
975 int parse_signed_commit(const struct commit *commit,
976 struct strbuf *payload, struct strbuf *signature)
979 unsigned long size;
980 const char *buffer = get_commit_buffer(commit, &size);
981 int in_signature, saw_signature = -1;
982 const char *line, *tail;
984 line = buffer;
985 tail = buffer + size;
986 in_signature = 0;
987 saw_signature = 0;
988 while (line < tail) {
989 const char *sig = NULL;
990 const char *next = memchr(line, '\n', tail - line);
992 next = next ? next + 1 : tail;
993 if (in_signature && line[0] == ' ')
994 sig = line + 1;
995 else if (starts_with(line, gpg_sig_header) &&
996 line[gpg_sig_header_len] == ' ')
997 sig = line + gpg_sig_header_len + 1;
998 if (sig) {
999 strbuf_add(signature, sig, next - sig);
1000 saw_signature = 1;
1001 in_signature = 1;
1002 } else {
1003 if (*line == '\n')
1004 /* dump the whole remainder of the buffer */
1005 next = tail;
1006 strbuf_add(payload, line, next - line);
1007 in_signature = 0;
1009 line = next;
1011 unuse_commit_buffer(commit, buffer);
1012 return saw_signature;
1015 int remove_signature(struct strbuf *buf)
1017 const char *line = buf->buf;
1018 const char *tail = buf->buf + buf->len;
1019 int in_signature = 0;
1020 const char *sig_start = NULL;
1021 const char *sig_end = NULL;
1023 while (line < tail) {
1024 const char *next = memchr(line, '\n', tail - line);
1025 next = next ? next + 1 : tail;
1027 if (in_signature && line[0] == ' ')
1028 sig_end = next;
1029 else if (starts_with(line, gpg_sig_header) &&
1030 line[gpg_sig_header_len] == ' ') {
1031 sig_start = line;
1032 sig_end = next;
1033 in_signature = 1;
1034 } else {
1035 if (*line == '\n')
1036 /* dump the whole remainder of the buffer */
1037 next = tail;
1038 in_signature = 0;
1040 line = next;
1043 if (sig_start)
1044 strbuf_remove(buf, sig_start - buf->buf, sig_end - sig_start);
1046 return sig_start != NULL;
1049 static void handle_signed_tag(struct commit *parent, struct commit_extra_header ***tail)
1051 struct merge_remote_desc *desc;
1052 struct commit_extra_header *mergetag;
1053 char *buf;
1054 unsigned long size, len;
1055 enum object_type type;
1057 desc = merge_remote_util(parent);
1058 if (!desc || !desc->obj)
1059 return;
1060 buf = read_object_file(&desc->obj->oid, &type, &size);
1061 if (!buf || type != OBJ_TAG)
1062 goto free_return;
1063 len = parse_signature(buf, size);
1064 if (size == len)
1065 goto free_return;
1067 * We could verify this signature and either omit the tag when
1068 * it does not validate, but the integrator may not have the
1069 * public key of the signer of the tag he is merging, while a
1070 * later auditor may have it while auditing, so let's not run
1071 * verify-signed-buffer here for now...
1073 * if (verify_signed_buffer(buf, len, buf + len, size - len, ...))
1074 * warn("warning: signed tag unverified.");
1076 mergetag = xcalloc(1, sizeof(*mergetag));
1077 mergetag->key = xstrdup("mergetag");
1078 mergetag->value = buf;
1079 mergetag->len = size;
1081 **tail = mergetag;
1082 *tail = &mergetag->next;
1083 return;
1085 free_return:
1086 free(buf);
1089 int check_commit_signature(const struct commit *commit, struct signature_check *sigc)
1091 struct strbuf payload = STRBUF_INIT;
1092 struct strbuf signature = STRBUF_INIT;
1093 int ret = 1;
1095 sigc->result = 'N';
1097 if (parse_signed_commit(commit, &payload, &signature) <= 0)
1098 goto out;
1099 ret = check_signature(payload.buf, payload.len, signature.buf,
1100 signature.len, sigc);
1102 out:
1103 strbuf_release(&payload);
1104 strbuf_release(&signature);
1106 return ret;
1109 void verify_merge_signature(struct commit *commit, int verbosity)
1111 char hex[GIT_MAX_HEXSZ + 1];
1112 struct signature_check signature_check;
1113 memset(&signature_check, 0, sizeof(signature_check));
1115 check_commit_signature(commit, &signature_check);
1117 find_unique_abbrev_r(hex, &commit->object.oid, DEFAULT_ABBREV);
1118 switch (signature_check.result) {
1119 case 'G':
1120 break;
1121 case 'U':
1122 die(_("Commit %s has an untrusted GPG signature, "
1123 "allegedly by %s."), hex, signature_check.signer);
1124 case 'B':
1125 die(_("Commit %s has a bad GPG signature "
1126 "allegedly by %s."), hex, signature_check.signer);
1127 default: /* 'N' */
1128 die(_("Commit %s does not have a GPG signature."), hex);
1130 if (verbosity >= 0 && signature_check.result == 'G')
1131 printf(_("Commit %s has a good GPG signature by %s\n"),
1132 hex, signature_check.signer);
1134 signature_check_clear(&signature_check);
1137 void append_merge_tag_headers(struct commit_list *parents,
1138 struct commit_extra_header ***tail)
1140 while (parents) {
1141 struct commit *parent = parents->item;
1142 handle_signed_tag(parent, tail);
1143 parents = parents->next;
1147 static void add_extra_header(struct strbuf *buffer,
1148 struct commit_extra_header *extra)
1150 strbuf_addstr(buffer, extra->key);
1151 if (extra->len)
1152 strbuf_add_lines(buffer, " ", extra->value, extra->len);
1153 else
1154 strbuf_addch(buffer, '\n');
1157 struct commit_extra_header *read_commit_extra_headers(struct commit *commit,
1158 const char **exclude)
1160 struct commit_extra_header *extra = NULL;
1161 unsigned long size;
1162 const char *buffer = get_commit_buffer(commit, &size);
1163 extra = read_commit_extra_header_lines(buffer, size, exclude);
1164 unuse_commit_buffer(commit, buffer);
1165 return extra;
1168 int for_each_mergetag(each_mergetag_fn fn, struct commit *commit, void *data)
1170 struct commit_extra_header *extra, *to_free;
1171 int res = 0;
1173 to_free = read_commit_extra_headers(commit, NULL);
1174 for (extra = to_free; !res && extra; extra = extra->next) {
1175 if (strcmp(extra->key, "mergetag"))
1176 continue; /* not a merge tag */
1177 res = fn(commit, extra, data);
1179 free_commit_extra_headers(to_free);
1180 return res;
1183 static inline int standard_header_field(const char *field, size_t len)
1185 return ((len == 4 && !memcmp(field, "tree", 4)) ||
1186 (len == 6 && !memcmp(field, "parent", 6)) ||
1187 (len == 6 && !memcmp(field, "author", 6)) ||
1188 (len == 9 && !memcmp(field, "committer", 9)) ||
1189 (len == 8 && !memcmp(field, "encoding", 8)));
1192 static int excluded_header_field(const char *field, size_t len, const char **exclude)
1194 if (!exclude)
1195 return 0;
1197 while (*exclude) {
1198 size_t xlen = strlen(*exclude);
1199 if (len == xlen && !memcmp(field, *exclude, xlen))
1200 return 1;
1201 exclude++;
1203 return 0;
1206 static struct commit_extra_header *read_commit_extra_header_lines(
1207 const char *buffer, size_t size,
1208 const char **exclude)
1210 struct commit_extra_header *extra = NULL, **tail = &extra, *it = NULL;
1211 const char *line, *next, *eof, *eob;
1212 struct strbuf buf = STRBUF_INIT;
1214 for (line = buffer, eob = line + size;
1215 line < eob && *line != '\n';
1216 line = next) {
1217 next = memchr(line, '\n', eob - line);
1218 next = next ? next + 1 : eob;
1219 if (*line == ' ') {
1220 /* continuation */
1221 if (it)
1222 strbuf_add(&buf, line + 1, next - (line + 1));
1223 continue;
1225 if (it)
1226 it->value = strbuf_detach(&buf, &it->len);
1227 strbuf_reset(&buf);
1228 it = NULL;
1230 eof = memchr(line, ' ', next - line);
1231 if (!eof)
1232 eof = next;
1233 else if (standard_header_field(line, eof - line) ||
1234 excluded_header_field(line, eof - line, exclude))
1235 continue;
1237 it = xcalloc(1, sizeof(*it));
1238 it->key = xmemdupz(line, eof-line);
1239 *tail = it;
1240 tail = &it->next;
1241 if (eof + 1 < next)
1242 strbuf_add(&buf, eof + 1, next - (eof + 1));
1244 if (it)
1245 it->value = strbuf_detach(&buf, &it->len);
1246 return extra;
1249 void free_commit_extra_headers(struct commit_extra_header *extra)
1251 while (extra) {
1252 struct commit_extra_header *next = extra->next;
1253 free(extra->key);
1254 free(extra->value);
1255 free(extra);
1256 extra = next;
1260 int commit_tree(const char *msg, size_t msg_len, const struct object_id *tree,
1261 struct commit_list *parents, struct object_id *ret,
1262 const char *author, const char *sign_commit)
1264 struct commit_extra_header *extra = NULL, **tail = &extra;
1265 int result;
1267 append_merge_tag_headers(parents, &tail);
1268 result = commit_tree_extended(msg, msg_len, tree, parents, ret,
1269 author, sign_commit, extra);
1270 free_commit_extra_headers(extra);
1271 return result;
1274 static int find_invalid_utf8(const char *buf, int len)
1276 int offset = 0;
1277 static const unsigned int max_codepoint[] = {
1278 0x7f, 0x7ff, 0xffff, 0x10ffff
1281 while (len) {
1282 unsigned char c = *buf++;
1283 int bytes, bad_offset;
1284 unsigned int codepoint;
1285 unsigned int min_val, max_val;
1287 len--;
1288 offset++;
1290 /* Simple US-ASCII? No worries. */
1291 if (c < 0x80)
1292 continue;
1294 bad_offset = offset-1;
1297 * Count how many more high bits set: that's how
1298 * many more bytes this sequence should have.
1300 bytes = 0;
1301 while (c & 0x40) {
1302 c <<= 1;
1303 bytes++;
1307 * Must be between 1 and 3 more bytes. Longer sequences result in
1308 * codepoints beyond U+10FFFF, which are guaranteed never to exist.
1310 if (bytes < 1 || 3 < bytes)
1311 return bad_offset;
1313 /* Do we *have* that many bytes? */
1314 if (len < bytes)
1315 return bad_offset;
1318 * Place the encoded bits at the bottom of the value and compute the
1319 * valid range.
1321 codepoint = (c & 0x7f) >> bytes;
1322 min_val = max_codepoint[bytes-1] + 1;
1323 max_val = max_codepoint[bytes];
1325 offset += bytes;
1326 len -= bytes;
1328 /* And verify that they are good continuation bytes */
1329 do {
1330 codepoint <<= 6;
1331 codepoint |= *buf & 0x3f;
1332 if ((*buf++ & 0xc0) != 0x80)
1333 return bad_offset;
1334 } while (--bytes);
1336 /* Reject codepoints that are out of range for the sequence length. */
1337 if (codepoint < min_val || codepoint > max_val)
1338 return bad_offset;
1339 /* Surrogates are only for UTF-16 and cannot be encoded in UTF-8. */
1340 if ((codepoint & 0x1ff800) == 0xd800)
1341 return bad_offset;
1342 /* U+xxFFFE and U+xxFFFF are guaranteed non-characters. */
1343 if ((codepoint & 0xfffe) == 0xfffe)
1344 return bad_offset;
1345 /* So are anything in the range U+FDD0..U+FDEF. */
1346 if (codepoint >= 0xfdd0 && codepoint <= 0xfdef)
1347 return bad_offset;
1349 return -1;
1353 * This verifies that the buffer is in proper utf8 format.
1355 * If it isn't, it assumes any non-utf8 characters are Latin1,
1356 * and does the conversion.
1358 static int verify_utf8(struct strbuf *buf)
1360 int ok = 1;
1361 long pos = 0;
1363 for (;;) {
1364 int bad;
1365 unsigned char c;
1366 unsigned char replace[2];
1368 bad = find_invalid_utf8(buf->buf + pos, buf->len - pos);
1369 if (bad < 0)
1370 return ok;
1371 pos += bad;
1372 ok = 0;
1373 c = buf->buf[pos];
1374 strbuf_remove(buf, pos, 1);
1376 /* We know 'c' must be in the range 128-255 */
1377 replace[0] = 0xc0 + (c >> 6);
1378 replace[1] = 0x80 + (c & 0x3f);
1379 strbuf_insert(buf, pos, replace, 2);
1380 pos += 2;
1384 static const char commit_utf8_warn[] =
1385 N_("Warning: commit message did not conform to UTF-8.\n"
1386 "You may want to amend it after fixing the message, or set the config\n"
1387 "variable i18n.commitencoding to the encoding your project uses.\n");
1389 int commit_tree_extended(const char *msg, size_t msg_len,
1390 const struct object_id *tree,
1391 struct commit_list *parents, struct object_id *ret,
1392 const char *author, const char *sign_commit,
1393 struct commit_extra_header *extra)
1395 int result;
1396 int encoding_is_utf8;
1397 struct strbuf buffer;
1399 assert_oid_type(tree, OBJ_TREE);
1401 if (memchr(msg, '\0', msg_len))
1402 return error("a NUL byte in commit log message not allowed.");
1404 /* Not having i18n.commitencoding is the same as having utf-8 */
1405 encoding_is_utf8 = is_encoding_utf8(git_commit_encoding);
1407 strbuf_init(&buffer, 8192); /* should avoid reallocs for the headers */
1408 strbuf_addf(&buffer, "tree %s\n", oid_to_hex(tree));
1411 * NOTE! This ordering means that the same exact tree merged with a
1412 * different order of parents will be a _different_ changeset even
1413 * if everything else stays the same.
1415 while (parents) {
1416 struct commit *parent = pop_commit(&parents);
1417 strbuf_addf(&buffer, "parent %s\n",
1418 oid_to_hex(&parent->object.oid));
1421 /* Person/date information */
1422 if (!author)
1423 author = git_author_info(IDENT_STRICT);
1424 strbuf_addf(&buffer, "author %s\n", author);
1425 strbuf_addf(&buffer, "committer %s\n", git_committer_info(IDENT_STRICT));
1426 if (!encoding_is_utf8)
1427 strbuf_addf(&buffer, "encoding %s\n", git_commit_encoding);
1429 while (extra) {
1430 add_extra_header(&buffer, extra);
1431 extra = extra->next;
1433 strbuf_addch(&buffer, '\n');
1435 /* And add the comment */
1436 strbuf_add(&buffer, msg, msg_len);
1438 /* And check the encoding */
1439 if (encoding_is_utf8 && !verify_utf8(&buffer))
1440 fprintf(stderr, _(commit_utf8_warn));
1442 if (sign_commit && do_sign_commit(&buffer, sign_commit)) {
1443 result = -1;
1444 goto out;
1447 result = write_object_file(buffer.buf, buffer.len, commit_type, ret);
1448 out:
1449 strbuf_release(&buffer);
1450 return result;
1453 define_commit_slab(merge_desc_slab, struct merge_remote_desc *);
1454 static struct merge_desc_slab merge_desc_slab = COMMIT_SLAB_INIT(1, merge_desc_slab);
1456 struct merge_remote_desc *merge_remote_util(struct commit *commit)
1458 return *merge_desc_slab_at(&merge_desc_slab, commit);
1461 void set_merge_remote_desc(struct commit *commit,
1462 const char *name, struct object *obj)
1464 struct merge_remote_desc *desc;
1465 FLEX_ALLOC_STR(desc, name, name);
1466 desc->obj = obj;
1467 *merge_desc_slab_at(&merge_desc_slab, commit) = desc;
1470 struct commit *get_merge_parent(const char *name)
1472 struct object *obj;
1473 struct commit *commit;
1474 struct object_id oid;
1475 if (get_oid(name, &oid))
1476 return NULL;
1477 obj = parse_object(the_repository, &oid);
1478 commit = (struct commit *)peel_to_type(name, 0, obj, OBJ_COMMIT);
1479 if (commit && !merge_remote_util(commit))
1480 set_merge_remote_desc(commit, name, obj);
1481 return commit;
1485 * Append a commit to the end of the commit_list.
1487 * next starts by pointing to the variable that holds the head of an
1488 * empty commit_list, and is updated to point to the "next" field of
1489 * the last item on the list as new commits are appended.
1491 * Usage example:
1493 * struct commit_list *list;
1494 * struct commit_list **next = &list;
1496 * next = commit_list_append(c1, next);
1497 * next = commit_list_append(c2, next);
1498 * assert(commit_list_count(list) == 2);
1499 * return list;
1501 struct commit_list **commit_list_append(struct commit *commit,
1502 struct commit_list **next)
1504 struct commit_list *new_commit = xmalloc(sizeof(struct commit_list));
1505 new_commit->item = commit;
1506 *next = new_commit;
1507 new_commit->next = NULL;
1508 return &new_commit->next;
1511 const char *find_commit_header(const char *msg, const char *key, size_t *out_len)
1513 int key_len = strlen(key);
1514 const char *line = msg;
1516 while (line) {
1517 const char *eol = strchrnul(line, '\n');
1519 if (line == eol)
1520 return NULL;
1522 if (eol - line > key_len &&
1523 !strncmp(line, key, key_len) &&
1524 line[key_len] == ' ') {
1525 *out_len = eol - line - key_len - 1;
1526 return line + key_len + 1;
1528 line = *eol ? eol + 1 : NULL;
1530 return NULL;
1534 * Inspect the given string and determine the true "end" of the log message, in
1535 * order to find where to put a new Signed-off-by: line. Ignored are
1536 * trailing comment lines and blank lines. To support "git commit -s
1537 * --amend" on an existing commit, we also ignore "Conflicts:". To
1538 * support "git commit -v", we truncate at cut lines.
1540 * Returns the number of bytes from the tail to ignore, to be fed as
1541 * the second parameter to append_signoff().
1543 size_t ignore_non_trailer(const char *buf, size_t len)
1545 size_t boc = 0;
1546 size_t bol = 0;
1547 int in_old_conflicts_block = 0;
1548 size_t cutoff = wt_status_locate_end(buf, len);
1550 while (bol < cutoff) {
1551 const char *next_line = memchr(buf + bol, '\n', len - bol);
1553 if (!next_line)
1554 next_line = buf + len;
1555 else
1556 next_line++;
1558 if (buf[bol] == comment_line_char || buf[bol] == '\n') {
1559 /* is this the first of the run of comments? */
1560 if (!boc)
1561 boc = bol;
1562 /* otherwise, it is just continuing */
1563 } else if (starts_with(buf + bol, "Conflicts:\n")) {
1564 in_old_conflicts_block = 1;
1565 if (!boc)
1566 boc = bol;
1567 } else if (in_old_conflicts_block && buf[bol] == '\t') {
1568 ; /* a pathname in the conflicts block */
1569 } else if (boc) {
1570 /* the previous was not trailing comment */
1571 boc = 0;
1572 in_old_conflicts_block = 0;
1574 bol = next_line - buf;
1576 return boc ? len - boc : len - cutoff;