stash -p: (partially) fix bug concerning split hunks
[git/debian.git] / commit-graph.c
blobf013a84e294b13b552b62257bbf2fa7e1c353a82
1 #include "cache.h"
2 #include "config.h"
3 #include "dir.h"
4 #include "git-compat-util.h"
5 #include "lockfile.h"
6 #include "pack.h"
7 #include "packfile.h"
8 #include "commit.h"
9 #include "object.h"
10 #include "refs.h"
11 #include "revision.h"
12 #include "sha1-lookup.h"
13 #include "commit-graph.h"
14 #include "object-store.h"
15 #include "alloc.h"
16 #include "hashmap.h"
17 #include "replace-object.h"
18 #include "progress.h"
20 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
21 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
22 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
23 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
24 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
25 #define GRAPH_CHUNKID_BASE 0x42415345 /* "BASE" */
27 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
29 #define GRAPH_VERSION_1 0x1
30 #define GRAPH_VERSION GRAPH_VERSION_1
32 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
33 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
34 #define GRAPH_PARENT_NONE 0x70000000
36 #define GRAPH_LAST_EDGE 0x80000000
38 #define GRAPH_HEADER_SIZE 8
39 #define GRAPH_FANOUT_SIZE (4 * 256)
40 #define GRAPH_CHUNKLOOKUP_WIDTH 12
41 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
42 + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
44 /* Remember to update object flag allocation in object.h */
45 #define REACHABLE (1u<<15)
47 char *get_commit_graph_filename(struct object_directory *odb)
49 return xstrfmt("%s/info/commit-graph", odb->path);
52 static char *get_split_graph_filename(struct object_directory *odb,
53 const char *oid_hex)
55 return xstrfmt("%s/info/commit-graphs/graph-%s.graph", odb->path,
56 oid_hex);
59 static char *get_chain_filename(struct object_directory *odb)
61 return xstrfmt("%s/info/commit-graphs/commit-graph-chain", odb->path);
64 static uint8_t oid_version(void)
66 return 1;
69 static struct commit_graph *alloc_commit_graph(void)
71 struct commit_graph *g = xcalloc(1, sizeof(*g));
72 g->graph_fd = -1;
74 return g;
77 extern int read_replace_refs;
79 static int commit_graph_compatible(struct repository *r)
81 if (!r->gitdir)
82 return 0;
84 if (read_replace_refs) {
85 prepare_replace_object(r);
86 if (hashmap_get_size(&r->objects->replace_map->map))
87 return 0;
90 prepare_commit_graft(r);
91 if (r->parsed_objects && r->parsed_objects->grafts_nr)
92 return 0;
93 if (is_repository_shallow(r))
94 return 0;
96 return 1;
99 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
101 *fd = git_open(graph_file);
102 if (*fd < 0)
103 return 0;
104 if (fstat(*fd, st)) {
105 close(*fd);
106 return 0;
108 return 1;
111 struct commit_graph *load_commit_graph_one_fd_st(int fd, struct stat *st,
112 struct object_directory *odb)
114 void *graph_map;
115 size_t graph_size;
116 struct commit_graph *ret;
118 graph_size = xsize_t(st->st_size);
120 if (graph_size < GRAPH_MIN_SIZE) {
121 close(fd);
122 error(_("commit-graph file is too small"));
123 return NULL;
125 graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
126 ret = parse_commit_graph(graph_map, fd, graph_size);
128 if (ret)
129 ret->odb = odb;
130 else {
131 munmap(graph_map, graph_size);
132 close(fd);
135 return ret;
138 static int verify_commit_graph_lite(struct commit_graph *g)
141 * Basic validation shared between parse_commit_graph()
142 * which'll be called every time the graph is used, and the
143 * much more expensive verify_commit_graph() used by
144 * "commit-graph verify".
146 * There should only be very basic checks here to ensure that
147 * we don't e.g. segfault in fill_commit_in_graph(), but
148 * because this is a very hot codepath nothing that e.g. loops
149 * over g->num_commits, or runs a checksum on the commit-graph
150 * itself.
152 if (!g->chunk_oid_fanout) {
153 error("commit-graph is missing the OID Fanout chunk");
154 return 1;
156 if (!g->chunk_oid_lookup) {
157 error("commit-graph is missing the OID Lookup chunk");
158 return 1;
160 if (!g->chunk_commit_data) {
161 error("commit-graph is missing the Commit Data chunk");
162 return 1;
165 return 0;
168 struct commit_graph *parse_commit_graph(void *graph_map, int fd,
169 size_t graph_size)
171 const unsigned char *data, *chunk_lookup;
172 uint32_t i;
173 struct commit_graph *graph;
174 uint64_t last_chunk_offset;
175 uint32_t last_chunk_id;
176 uint32_t graph_signature;
177 unsigned char graph_version, hash_version;
179 if (!graph_map)
180 return NULL;
182 if (graph_size < GRAPH_MIN_SIZE)
183 return NULL;
185 data = (const unsigned char *)graph_map;
187 graph_signature = get_be32(data);
188 if (graph_signature != GRAPH_SIGNATURE) {
189 error(_("commit-graph signature %X does not match signature %X"),
190 graph_signature, GRAPH_SIGNATURE);
191 return NULL;
194 graph_version = *(unsigned char*)(data + 4);
195 if (graph_version != GRAPH_VERSION) {
196 error(_("commit-graph version %X does not match version %X"),
197 graph_version, GRAPH_VERSION);
198 return NULL;
201 hash_version = *(unsigned char*)(data + 5);
202 if (hash_version != oid_version()) {
203 error(_("commit-graph hash version %X does not match version %X"),
204 hash_version, oid_version());
205 return NULL;
208 graph = alloc_commit_graph();
210 graph->hash_len = the_hash_algo->rawsz;
211 graph->num_chunks = *(unsigned char*)(data + 6);
212 graph->graph_fd = fd;
213 graph->data = graph_map;
214 graph->data_len = graph_size;
216 last_chunk_id = 0;
217 last_chunk_offset = 8;
218 chunk_lookup = data + 8;
219 for (i = 0; i < graph->num_chunks; i++) {
220 uint32_t chunk_id;
221 uint64_t chunk_offset;
222 int chunk_repeated = 0;
224 if (data + graph_size - chunk_lookup <
225 GRAPH_CHUNKLOOKUP_WIDTH) {
226 error(_("commit-graph chunk lookup table entry missing; file may be incomplete"));
227 free(graph);
228 return NULL;
231 chunk_id = get_be32(chunk_lookup + 0);
232 chunk_offset = get_be64(chunk_lookup + 4);
234 chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
236 if (chunk_offset > graph_size - the_hash_algo->rawsz) {
237 error(_("commit-graph improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
238 (uint32_t)chunk_offset);
239 free(graph);
240 return NULL;
243 switch (chunk_id) {
244 case GRAPH_CHUNKID_OIDFANOUT:
245 if (graph->chunk_oid_fanout)
246 chunk_repeated = 1;
247 else
248 graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
249 break;
251 case GRAPH_CHUNKID_OIDLOOKUP:
252 if (graph->chunk_oid_lookup)
253 chunk_repeated = 1;
254 else
255 graph->chunk_oid_lookup = data + chunk_offset;
256 break;
258 case GRAPH_CHUNKID_DATA:
259 if (graph->chunk_commit_data)
260 chunk_repeated = 1;
261 else
262 graph->chunk_commit_data = data + chunk_offset;
263 break;
265 case GRAPH_CHUNKID_EXTRAEDGES:
266 if (graph->chunk_extra_edges)
267 chunk_repeated = 1;
268 else
269 graph->chunk_extra_edges = data + chunk_offset;
270 break;
272 case GRAPH_CHUNKID_BASE:
273 if (graph->chunk_base_graphs)
274 chunk_repeated = 1;
275 else
276 graph->chunk_base_graphs = data + chunk_offset;
279 if (chunk_repeated) {
280 error(_("commit-graph chunk id %08x appears multiple times"), chunk_id);
281 free(graph);
282 return NULL;
285 if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
287 graph->num_commits = (chunk_offset - last_chunk_offset)
288 / graph->hash_len;
291 last_chunk_id = chunk_id;
292 last_chunk_offset = chunk_offset;
295 hashcpy(graph->oid.hash, graph->data + graph->data_len - graph->hash_len);
297 if (verify_commit_graph_lite(graph)) {
298 free(graph);
299 return NULL;
302 return graph;
305 static struct commit_graph *load_commit_graph_one(const char *graph_file,
306 struct object_directory *odb)
309 struct stat st;
310 int fd;
311 struct commit_graph *g;
312 int open_ok = open_commit_graph(graph_file, &fd, &st);
314 if (!open_ok)
315 return NULL;
317 g = load_commit_graph_one_fd_st(fd, &st, odb);
319 if (g)
320 g->filename = xstrdup(graph_file);
322 return g;
325 static struct commit_graph *load_commit_graph_v1(struct repository *r,
326 struct object_directory *odb)
328 char *graph_name = get_commit_graph_filename(odb);
329 struct commit_graph *g = load_commit_graph_one(graph_name, odb);
330 free(graph_name);
332 return g;
335 static int add_graph_to_chain(struct commit_graph *g,
336 struct commit_graph *chain,
337 struct object_id *oids,
338 int n)
340 struct commit_graph *cur_g = chain;
342 if (n && !g->chunk_base_graphs) {
343 warning(_("commit-graph has no base graphs chunk"));
344 return 0;
347 while (n) {
348 n--;
350 if (!cur_g ||
351 !oideq(&oids[n], &cur_g->oid) ||
352 !hasheq(oids[n].hash, g->chunk_base_graphs + g->hash_len * n)) {
353 warning(_("commit-graph chain does not match"));
354 return 0;
357 cur_g = cur_g->base_graph;
360 g->base_graph = chain;
362 if (chain)
363 g->num_commits_in_base = chain->num_commits + chain->num_commits_in_base;
365 return 1;
368 static struct commit_graph *load_commit_graph_chain(struct repository *r,
369 struct object_directory *odb)
371 struct commit_graph *graph_chain = NULL;
372 struct strbuf line = STRBUF_INIT;
373 struct stat st;
374 struct object_id *oids;
375 int i = 0, valid = 1, count;
376 char *chain_name = get_chain_filename(odb);
377 FILE *fp;
378 int stat_res;
380 fp = fopen(chain_name, "r");
381 stat_res = stat(chain_name, &st);
382 free(chain_name);
384 if (!fp ||
385 stat_res ||
386 st.st_size <= the_hash_algo->hexsz)
387 return NULL;
389 count = st.st_size / (the_hash_algo->hexsz + 1);
390 oids = xcalloc(count, sizeof(struct object_id));
392 prepare_alt_odb(r);
394 for (i = 0; i < count; i++) {
395 struct object_directory *odb;
397 if (strbuf_getline_lf(&line, fp) == EOF)
398 break;
400 if (get_oid_hex(line.buf, &oids[i])) {
401 warning(_("invalid commit-graph chain: line '%s' not a hash"),
402 line.buf);
403 valid = 0;
404 break;
407 valid = 0;
408 for (odb = r->objects->odb; odb; odb = odb->next) {
409 char *graph_name = get_split_graph_filename(odb, line.buf);
410 struct commit_graph *g = load_commit_graph_one(graph_name, odb);
412 free(graph_name);
414 if (g) {
415 if (add_graph_to_chain(g, graph_chain, oids, i)) {
416 graph_chain = g;
417 valid = 1;
420 break;
424 if (!valid) {
425 warning(_("unable to find all commit-graph files"));
426 break;
430 free(oids);
431 fclose(fp);
432 strbuf_release(&line);
434 return graph_chain;
437 struct commit_graph *read_commit_graph_one(struct repository *r,
438 struct object_directory *odb)
440 struct commit_graph *g = load_commit_graph_v1(r, odb);
442 if (!g)
443 g = load_commit_graph_chain(r, odb);
445 return g;
448 static void prepare_commit_graph_one(struct repository *r,
449 struct object_directory *odb)
452 if (r->objects->commit_graph)
453 return;
455 r->objects->commit_graph = read_commit_graph_one(r, odb);
459 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
461 * On the first invocation, this function attempts to load the commit
462 * graph if the_repository is configured to have one.
464 static int prepare_commit_graph(struct repository *r)
466 struct object_directory *odb;
469 * This must come before the "already attempted?" check below, because
470 * we want to disable even an already-loaded graph file.
472 if (r->commit_graph_disabled)
473 return 0;
475 if (r->objects->commit_graph_attempted)
476 return !!r->objects->commit_graph;
477 r->objects->commit_graph_attempted = 1;
479 if (git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD, 0))
480 die("dying as requested by the '%s' variable on commit-graph load!",
481 GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD);
483 prepare_repo_settings(r);
485 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
486 r->settings.core_commit_graph != 1)
488 * This repository is not configured to use commit graphs, so
489 * do not load one. (But report commit_graph_attempted anyway
490 * so that commit graph loading is not attempted again for this
491 * repository.)
493 return 0;
495 if (!commit_graph_compatible(r))
496 return 0;
498 prepare_alt_odb(r);
499 for (odb = r->objects->odb;
500 !r->objects->commit_graph && odb;
501 odb = odb->next)
502 prepare_commit_graph_one(r, odb);
503 return !!r->objects->commit_graph;
506 int generation_numbers_enabled(struct repository *r)
508 uint32_t first_generation;
509 struct commit_graph *g;
510 if (!prepare_commit_graph(r))
511 return 0;
513 g = r->objects->commit_graph;
515 if (!g->num_commits)
516 return 0;
518 first_generation = get_be32(g->chunk_commit_data +
519 g->hash_len + 8) >> 2;
521 return !!first_generation;
524 static void close_commit_graph_one(struct commit_graph *g)
526 if (!g)
527 return;
529 close_commit_graph_one(g->base_graph);
530 free_commit_graph(g);
533 void close_commit_graph(struct raw_object_store *o)
535 close_commit_graph_one(o->commit_graph);
536 o->commit_graph = NULL;
539 static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
541 return bsearch_hash(oid->hash, g->chunk_oid_fanout,
542 g->chunk_oid_lookup, g->hash_len, pos);
545 static void load_oid_from_graph(struct commit_graph *g,
546 uint32_t pos,
547 struct object_id *oid)
549 uint32_t lex_index;
551 while (g && pos < g->num_commits_in_base)
552 g = g->base_graph;
554 if (!g)
555 BUG("NULL commit-graph");
557 if (pos >= g->num_commits + g->num_commits_in_base)
558 die(_("invalid commit position. commit-graph is likely corrupt"));
560 lex_index = pos - g->num_commits_in_base;
562 hashcpy(oid->hash, g->chunk_oid_lookup + g->hash_len * lex_index);
565 static struct commit_list **insert_parent_or_die(struct repository *r,
566 struct commit_graph *g,
567 uint32_t pos,
568 struct commit_list **pptr)
570 struct commit *c;
571 struct object_id oid;
573 if (pos >= g->num_commits + g->num_commits_in_base)
574 die("invalid parent position %"PRIu32, pos);
576 load_oid_from_graph(g, pos, &oid);
577 c = lookup_commit(r, &oid);
578 if (!c)
579 die(_("could not find commit %s"), oid_to_hex(&oid));
580 c->graph_pos = pos;
581 return &commit_list_insert(c, pptr)->next;
584 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
586 const unsigned char *commit_data;
587 uint32_t lex_index;
589 while (pos < g->num_commits_in_base)
590 g = g->base_graph;
592 lex_index = pos - g->num_commits_in_base;
593 commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * lex_index;
594 item->graph_pos = pos;
595 item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
598 static inline void set_commit_tree(struct commit *c, struct tree *t)
600 c->maybe_tree = t;
603 static int fill_commit_in_graph(struct repository *r,
604 struct commit *item,
605 struct commit_graph *g, uint32_t pos)
607 uint32_t edge_value;
608 uint32_t *parent_data_ptr;
609 uint64_t date_low, date_high;
610 struct commit_list **pptr;
611 const unsigned char *commit_data;
612 uint32_t lex_index;
614 while (pos < g->num_commits_in_base)
615 g = g->base_graph;
617 if (pos >= g->num_commits + g->num_commits_in_base)
618 die(_("invalid commit position. commit-graph is likely corrupt"));
621 * Store the "full" position, but then use the
622 * "local" position for the rest of the calculation.
624 item->graph_pos = pos;
625 lex_index = pos - g->num_commits_in_base;
627 commit_data = g->chunk_commit_data + (g->hash_len + 16) * lex_index;
629 item->object.parsed = 1;
631 set_commit_tree(item, NULL);
633 date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
634 date_low = get_be32(commit_data + g->hash_len + 12);
635 item->date = (timestamp_t)((date_high << 32) | date_low);
637 item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
639 pptr = &item->parents;
641 edge_value = get_be32(commit_data + g->hash_len);
642 if (edge_value == GRAPH_PARENT_NONE)
643 return 1;
644 pptr = insert_parent_or_die(r, g, edge_value, pptr);
646 edge_value = get_be32(commit_data + g->hash_len + 4);
647 if (edge_value == GRAPH_PARENT_NONE)
648 return 1;
649 if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
650 pptr = insert_parent_or_die(r, g, edge_value, pptr);
651 return 1;
654 parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
655 4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
656 do {
657 edge_value = get_be32(parent_data_ptr);
658 pptr = insert_parent_or_die(r, g,
659 edge_value & GRAPH_EDGE_LAST_MASK,
660 pptr);
661 parent_data_ptr++;
662 } while (!(edge_value & GRAPH_LAST_EDGE));
664 return 1;
667 static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
669 if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
670 *pos = item->graph_pos;
671 return 1;
672 } else {
673 struct commit_graph *cur_g = g;
674 uint32_t lex_index;
676 while (cur_g && !bsearch_graph(cur_g, &(item->object.oid), &lex_index))
677 cur_g = cur_g->base_graph;
679 if (cur_g) {
680 *pos = lex_index + cur_g->num_commits_in_base;
681 return 1;
684 return 0;
688 static int parse_commit_in_graph_one(struct repository *r,
689 struct commit_graph *g,
690 struct commit *item)
692 uint32_t pos;
694 if (item->object.parsed)
695 return 1;
697 if (find_commit_in_graph(item, g, &pos))
698 return fill_commit_in_graph(r, item, g, pos);
700 return 0;
703 int parse_commit_in_graph(struct repository *r, struct commit *item)
705 if (!prepare_commit_graph(r))
706 return 0;
707 return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
710 void load_commit_graph_info(struct repository *r, struct commit *item)
712 uint32_t pos;
713 if (!prepare_commit_graph(r))
714 return;
715 if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
716 fill_commit_graph_info(item, r->objects->commit_graph, pos);
719 static struct tree *load_tree_for_commit(struct repository *r,
720 struct commit_graph *g,
721 struct commit *c)
723 struct object_id oid;
724 const unsigned char *commit_data;
726 while (c->graph_pos < g->num_commits_in_base)
727 g = g->base_graph;
729 commit_data = g->chunk_commit_data +
730 GRAPH_DATA_WIDTH * (c->graph_pos - g->num_commits_in_base);
732 hashcpy(oid.hash, commit_data);
733 set_commit_tree(c, lookup_tree(r, &oid));
735 return c->maybe_tree;
738 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
739 struct commit_graph *g,
740 const struct commit *c)
742 if (c->maybe_tree)
743 return c->maybe_tree;
744 if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
745 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
747 return load_tree_for_commit(r, g, (struct commit *)c);
750 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
752 return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
755 struct packed_commit_list {
756 struct commit **list;
757 int nr;
758 int alloc;
761 struct packed_oid_list {
762 struct object_id *list;
763 int nr;
764 int alloc;
767 struct write_commit_graph_context {
768 struct repository *r;
769 struct object_directory *odb;
770 char *graph_name;
771 struct packed_oid_list oids;
772 struct packed_commit_list commits;
773 int num_extra_edges;
774 unsigned long approx_nr_objects;
775 struct progress *progress;
776 int progress_done;
777 uint64_t progress_cnt;
779 char *base_graph_name;
780 int num_commit_graphs_before;
781 int num_commit_graphs_after;
782 char **commit_graph_filenames_before;
783 char **commit_graph_filenames_after;
784 char **commit_graph_hash_after;
785 uint32_t new_num_commits_in_base;
786 struct commit_graph *new_base_graph;
788 unsigned append:1,
789 report_progress:1,
790 split:1,
791 check_oids:1;
793 const struct split_commit_graph_opts *split_opts;
796 static void write_graph_chunk_fanout(struct hashfile *f,
797 struct write_commit_graph_context *ctx)
799 int i, count = 0;
800 struct commit **list = ctx->commits.list;
803 * Write the first-level table (the list is sorted,
804 * but we use a 256-entry lookup to be able to avoid
805 * having to do eight extra binary search iterations).
807 for (i = 0; i < 256; i++) {
808 while (count < ctx->commits.nr) {
809 if ((*list)->object.oid.hash[0] != i)
810 break;
811 display_progress(ctx->progress, ++ctx->progress_cnt);
812 count++;
813 list++;
816 hashwrite_be32(f, count);
820 static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
821 struct write_commit_graph_context *ctx)
823 struct commit **list = ctx->commits.list;
824 int count;
825 for (count = 0; count < ctx->commits.nr; count++, list++) {
826 display_progress(ctx->progress, ++ctx->progress_cnt);
827 hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
831 static const unsigned char *commit_to_sha1(size_t index, void *table)
833 struct commit **commits = table;
834 return commits[index]->object.oid.hash;
837 static void write_graph_chunk_data(struct hashfile *f, int hash_len,
838 struct write_commit_graph_context *ctx)
840 struct commit **list = ctx->commits.list;
841 struct commit **last = ctx->commits.list + ctx->commits.nr;
842 uint32_t num_extra_edges = 0;
844 while (list < last) {
845 struct commit_list *parent;
846 struct object_id *tree;
847 int edge_value;
848 uint32_t packedDate[2];
849 display_progress(ctx->progress, ++ctx->progress_cnt);
851 if (parse_commit_no_graph(*list))
852 die(_("unable to parse commit %s"),
853 oid_to_hex(&(*list)->object.oid));
854 tree = get_commit_tree_oid(*list);
855 hashwrite(f, tree->hash, hash_len);
857 parent = (*list)->parents;
859 if (!parent)
860 edge_value = GRAPH_PARENT_NONE;
861 else {
862 edge_value = sha1_pos(parent->item->object.oid.hash,
863 ctx->commits.list,
864 ctx->commits.nr,
865 commit_to_sha1);
867 if (edge_value >= 0)
868 edge_value += ctx->new_num_commits_in_base;
869 else {
870 uint32_t pos;
871 if (find_commit_in_graph(parent->item,
872 ctx->new_base_graph,
873 &pos))
874 edge_value = pos;
877 if (edge_value < 0)
878 BUG("missing parent %s for commit %s",
879 oid_to_hex(&parent->item->object.oid),
880 oid_to_hex(&(*list)->object.oid));
883 hashwrite_be32(f, edge_value);
885 if (parent)
886 parent = parent->next;
888 if (!parent)
889 edge_value = GRAPH_PARENT_NONE;
890 else if (parent->next)
891 edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
892 else {
893 edge_value = sha1_pos(parent->item->object.oid.hash,
894 ctx->commits.list,
895 ctx->commits.nr,
896 commit_to_sha1);
898 if (edge_value >= 0)
899 edge_value += ctx->new_num_commits_in_base;
900 else {
901 uint32_t pos;
902 if (find_commit_in_graph(parent->item,
903 ctx->new_base_graph,
904 &pos))
905 edge_value = pos;
908 if (edge_value < 0)
909 BUG("missing parent %s for commit %s",
910 oid_to_hex(&parent->item->object.oid),
911 oid_to_hex(&(*list)->object.oid));
914 hashwrite_be32(f, edge_value);
916 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
917 do {
918 num_extra_edges++;
919 parent = parent->next;
920 } while (parent);
923 if (sizeof((*list)->date) > 4)
924 packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
925 else
926 packedDate[0] = 0;
928 packedDate[0] |= htonl((*list)->generation << 2);
930 packedDate[1] = htonl((*list)->date);
931 hashwrite(f, packedDate, 8);
933 list++;
937 static void write_graph_chunk_extra_edges(struct hashfile *f,
938 struct write_commit_graph_context *ctx)
940 struct commit **list = ctx->commits.list;
941 struct commit **last = ctx->commits.list + ctx->commits.nr;
942 struct commit_list *parent;
944 while (list < last) {
945 int num_parents = 0;
947 display_progress(ctx->progress, ++ctx->progress_cnt);
949 for (parent = (*list)->parents; num_parents < 3 && parent;
950 parent = parent->next)
951 num_parents++;
953 if (num_parents <= 2) {
954 list++;
955 continue;
958 /* Since num_parents > 2, this initializer is safe. */
959 for (parent = (*list)->parents->next; parent; parent = parent->next) {
960 int edge_value = sha1_pos(parent->item->object.oid.hash,
961 ctx->commits.list,
962 ctx->commits.nr,
963 commit_to_sha1);
965 if (edge_value >= 0)
966 edge_value += ctx->new_num_commits_in_base;
967 else {
968 uint32_t pos;
969 if (find_commit_in_graph(parent->item,
970 ctx->new_base_graph,
971 &pos))
972 edge_value = pos;
975 if (edge_value < 0)
976 BUG("missing parent %s for commit %s",
977 oid_to_hex(&parent->item->object.oid),
978 oid_to_hex(&(*list)->object.oid));
979 else if (!parent->next)
980 edge_value |= GRAPH_LAST_EDGE;
982 hashwrite_be32(f, edge_value);
985 list++;
989 static int oid_compare(const void *_a, const void *_b)
991 const struct object_id *a = (const struct object_id *)_a;
992 const struct object_id *b = (const struct object_id *)_b;
993 return oidcmp(a, b);
996 static int add_packed_commits(const struct object_id *oid,
997 struct packed_git *pack,
998 uint32_t pos,
999 void *data)
1001 struct write_commit_graph_context *ctx = (struct write_commit_graph_context*)data;
1002 enum object_type type;
1003 off_t offset = nth_packed_object_offset(pack, pos);
1004 struct object_info oi = OBJECT_INFO_INIT;
1006 if (ctx->progress)
1007 display_progress(ctx->progress, ++ctx->progress_done);
1009 oi.typep = &type;
1010 if (packed_object_info(ctx->r, pack, offset, &oi) < 0)
1011 die(_("unable to get type of object %s"), oid_to_hex(oid));
1013 if (type != OBJ_COMMIT)
1014 return 0;
1016 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1017 oidcpy(&(ctx->oids.list[ctx->oids.nr]), oid);
1018 ctx->oids.nr++;
1020 return 0;
1023 static void add_missing_parents(struct write_commit_graph_context *ctx, struct commit *commit)
1025 struct commit_list *parent;
1026 for (parent = commit->parents; parent; parent = parent->next) {
1027 if (!(parent->item->object.flags & REACHABLE)) {
1028 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1029 oidcpy(&ctx->oids.list[ctx->oids.nr], &(parent->item->object.oid));
1030 ctx->oids.nr++;
1031 parent->item->object.flags |= REACHABLE;
1036 static void close_reachable(struct write_commit_graph_context *ctx)
1038 int i;
1039 struct commit *commit;
1041 if (ctx->report_progress)
1042 ctx->progress = start_delayed_progress(
1043 _("Loading known commits in commit graph"),
1044 ctx->oids.nr);
1045 for (i = 0; i < ctx->oids.nr; i++) {
1046 display_progress(ctx->progress, i + 1);
1047 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1048 if (commit)
1049 commit->object.flags |= REACHABLE;
1051 stop_progress(&ctx->progress);
1054 * As this loop runs, ctx->oids.nr may grow, but not more
1055 * than the number of missing commits in the reachable
1056 * closure.
1058 if (ctx->report_progress)
1059 ctx->progress = start_delayed_progress(
1060 _("Expanding reachable commits in commit graph"),
1062 for (i = 0; i < ctx->oids.nr; i++) {
1063 display_progress(ctx->progress, i + 1);
1064 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1066 if (!commit)
1067 continue;
1068 if (ctx->split) {
1069 if (!parse_commit(commit) &&
1070 commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
1071 add_missing_parents(ctx, commit);
1072 } else if (!parse_commit_no_graph(commit))
1073 add_missing_parents(ctx, commit);
1075 stop_progress(&ctx->progress);
1077 if (ctx->report_progress)
1078 ctx->progress = start_delayed_progress(
1079 _("Clearing commit marks in commit graph"),
1080 ctx->oids.nr);
1081 for (i = 0; i < ctx->oids.nr; i++) {
1082 display_progress(ctx->progress, i + 1);
1083 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1085 if (commit)
1086 commit->object.flags &= ~REACHABLE;
1088 stop_progress(&ctx->progress);
1091 static void compute_generation_numbers(struct write_commit_graph_context *ctx)
1093 int i;
1094 struct commit_list *list = NULL;
1096 if (ctx->report_progress)
1097 ctx->progress = start_delayed_progress(
1098 _("Computing commit graph generation numbers"),
1099 ctx->commits.nr);
1100 for (i = 0; i < ctx->commits.nr; i++) {
1101 display_progress(ctx->progress, i + 1);
1102 if (ctx->commits.list[i]->generation != GENERATION_NUMBER_INFINITY &&
1103 ctx->commits.list[i]->generation != GENERATION_NUMBER_ZERO)
1104 continue;
1106 commit_list_insert(ctx->commits.list[i], &list);
1107 while (list) {
1108 struct commit *current = list->item;
1109 struct commit_list *parent;
1110 int all_parents_computed = 1;
1111 uint32_t max_generation = 0;
1113 for (parent = current->parents; parent; parent = parent->next) {
1114 if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
1115 parent->item->generation == GENERATION_NUMBER_ZERO) {
1116 all_parents_computed = 0;
1117 commit_list_insert(parent->item, &list);
1118 break;
1119 } else if (parent->item->generation > max_generation) {
1120 max_generation = parent->item->generation;
1124 if (all_parents_computed) {
1125 current->generation = max_generation + 1;
1126 pop_commit(&list);
1128 if (current->generation > GENERATION_NUMBER_MAX)
1129 current->generation = GENERATION_NUMBER_MAX;
1133 stop_progress(&ctx->progress);
1136 static int add_ref_to_list(const char *refname,
1137 const struct object_id *oid,
1138 int flags, void *cb_data)
1140 struct string_list *list = (struct string_list *)cb_data;
1142 string_list_append(list, oid_to_hex(oid));
1143 return 0;
1146 int write_commit_graph_reachable(struct object_directory *odb,
1147 enum commit_graph_write_flags flags,
1148 const struct split_commit_graph_opts *split_opts)
1150 struct string_list list = STRING_LIST_INIT_DUP;
1151 int result;
1153 for_each_ref(add_ref_to_list, &list);
1154 result = write_commit_graph(odb, NULL, &list,
1155 flags, split_opts);
1157 string_list_clear(&list, 0);
1158 return result;
1161 static int fill_oids_from_packs(struct write_commit_graph_context *ctx,
1162 struct string_list *pack_indexes)
1164 uint32_t i;
1165 struct strbuf progress_title = STRBUF_INIT;
1166 struct strbuf packname = STRBUF_INIT;
1167 int dirlen;
1169 strbuf_addf(&packname, "%s/pack/", ctx->odb->path);
1170 dirlen = packname.len;
1171 if (ctx->report_progress) {
1172 strbuf_addf(&progress_title,
1173 Q_("Finding commits for commit graph in %d pack",
1174 "Finding commits for commit graph in %d packs",
1175 pack_indexes->nr),
1176 pack_indexes->nr);
1177 ctx->progress = start_delayed_progress(progress_title.buf, 0);
1178 ctx->progress_done = 0;
1180 for (i = 0; i < pack_indexes->nr; i++) {
1181 struct packed_git *p;
1182 strbuf_setlen(&packname, dirlen);
1183 strbuf_addstr(&packname, pack_indexes->items[i].string);
1184 p = add_packed_git(packname.buf, packname.len, 1);
1185 if (!p) {
1186 error(_("error adding pack %s"), packname.buf);
1187 return -1;
1189 if (open_pack_index(p)) {
1190 error(_("error opening index for %s"), packname.buf);
1191 return -1;
1193 for_each_object_in_pack(p, add_packed_commits, ctx,
1194 FOR_EACH_OBJECT_PACK_ORDER);
1195 close_pack(p);
1196 free(p);
1199 stop_progress(&ctx->progress);
1200 strbuf_release(&progress_title);
1201 strbuf_release(&packname);
1203 return 0;
1206 static int fill_oids_from_commit_hex(struct write_commit_graph_context *ctx,
1207 struct string_list *commit_hex)
1209 uint32_t i;
1210 struct strbuf progress_title = STRBUF_INIT;
1212 if (ctx->report_progress) {
1213 strbuf_addf(&progress_title,
1214 Q_("Finding commits for commit graph from %d ref",
1215 "Finding commits for commit graph from %d refs",
1216 commit_hex->nr),
1217 commit_hex->nr);
1218 ctx->progress = start_delayed_progress(
1219 progress_title.buf,
1220 commit_hex->nr);
1222 for (i = 0; i < commit_hex->nr; i++) {
1223 const char *end;
1224 struct object_id oid;
1225 struct commit *result;
1227 display_progress(ctx->progress, i + 1);
1228 if (!parse_oid_hex(commit_hex->items[i].string, &oid, &end) &&
1229 (result = lookup_commit_reference_gently(ctx->r, &oid, 1))) {
1230 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1231 oidcpy(&ctx->oids.list[ctx->oids.nr], &(result->object.oid));
1232 ctx->oids.nr++;
1233 } else if (ctx->check_oids) {
1234 error(_("invalid commit object id: %s"),
1235 commit_hex->items[i].string);
1236 return -1;
1239 stop_progress(&ctx->progress);
1240 strbuf_release(&progress_title);
1242 return 0;
1245 static void fill_oids_from_all_packs(struct write_commit_graph_context *ctx)
1247 if (ctx->report_progress)
1248 ctx->progress = start_delayed_progress(
1249 _("Finding commits for commit graph among packed objects"),
1250 ctx->approx_nr_objects);
1251 for_each_packed_object(add_packed_commits, ctx,
1252 FOR_EACH_OBJECT_PACK_ORDER);
1253 if (ctx->progress_done < ctx->approx_nr_objects)
1254 display_progress(ctx->progress, ctx->approx_nr_objects);
1255 stop_progress(&ctx->progress);
1258 static uint32_t count_distinct_commits(struct write_commit_graph_context *ctx)
1260 uint32_t i, count_distinct = 1;
1262 if (ctx->report_progress)
1263 ctx->progress = start_delayed_progress(
1264 _("Counting distinct commits in commit graph"),
1265 ctx->oids.nr);
1266 display_progress(ctx->progress, 0); /* TODO: Measure QSORT() progress */
1267 QSORT(ctx->oids.list, ctx->oids.nr, oid_compare);
1269 for (i = 1; i < ctx->oids.nr; i++) {
1270 display_progress(ctx->progress, i + 1);
1271 if (!oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i])) {
1272 if (ctx->split) {
1273 struct commit *c = lookup_commit(ctx->r, &ctx->oids.list[i]);
1275 if (!c || c->graph_pos != COMMIT_NOT_FROM_GRAPH)
1276 continue;
1279 count_distinct++;
1282 stop_progress(&ctx->progress);
1284 return count_distinct;
1287 static void copy_oids_to_commits(struct write_commit_graph_context *ctx)
1289 uint32_t i;
1291 ctx->num_extra_edges = 0;
1292 if (ctx->report_progress)
1293 ctx->progress = start_delayed_progress(
1294 _("Finding extra edges in commit graph"),
1295 ctx->oids.nr);
1296 for (i = 0; i < ctx->oids.nr; i++) {
1297 unsigned int num_parents;
1299 display_progress(ctx->progress, i + 1);
1300 if (i > 0 && oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i]))
1301 continue;
1303 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + 1, ctx->commits.alloc);
1304 ctx->commits.list[ctx->commits.nr] = lookup_commit(ctx->r, &ctx->oids.list[i]);
1306 if (ctx->split &&
1307 ctx->commits.list[ctx->commits.nr]->graph_pos != COMMIT_NOT_FROM_GRAPH)
1308 continue;
1310 parse_commit_no_graph(ctx->commits.list[ctx->commits.nr]);
1312 num_parents = commit_list_count(ctx->commits.list[ctx->commits.nr]->parents);
1313 if (num_parents > 2)
1314 ctx->num_extra_edges += num_parents - 1;
1316 ctx->commits.nr++;
1318 stop_progress(&ctx->progress);
1321 static int write_graph_chunk_base_1(struct hashfile *f,
1322 struct commit_graph *g)
1324 int num = 0;
1326 if (!g)
1327 return 0;
1329 num = write_graph_chunk_base_1(f, g->base_graph);
1330 hashwrite(f, g->oid.hash, the_hash_algo->rawsz);
1331 return num + 1;
1334 static int write_graph_chunk_base(struct hashfile *f,
1335 struct write_commit_graph_context *ctx)
1337 int num = write_graph_chunk_base_1(f, ctx->new_base_graph);
1339 if (num != ctx->num_commit_graphs_after - 1) {
1340 error(_("failed to write correct number of base graph ids"));
1341 return -1;
1344 return 0;
1347 static int write_commit_graph_file(struct write_commit_graph_context *ctx)
1349 uint32_t i;
1350 int fd;
1351 struct hashfile *f;
1352 struct lock_file lk = LOCK_INIT;
1353 uint32_t chunk_ids[6];
1354 uint64_t chunk_offsets[6];
1355 const unsigned hashsz = the_hash_algo->rawsz;
1356 struct strbuf progress_title = STRBUF_INIT;
1357 int num_chunks = 3;
1358 struct object_id file_hash;
1360 if (ctx->split) {
1361 struct strbuf tmp_file = STRBUF_INIT;
1363 strbuf_addf(&tmp_file,
1364 "%s/info/commit-graphs/tmp_graph_XXXXXX",
1365 ctx->odb->path);
1366 ctx->graph_name = strbuf_detach(&tmp_file, NULL);
1367 } else {
1368 ctx->graph_name = get_commit_graph_filename(ctx->odb);
1371 if (safe_create_leading_directories(ctx->graph_name)) {
1372 UNLEAK(ctx->graph_name);
1373 error(_("unable to create leading directories of %s"),
1374 ctx->graph_name);
1375 return -1;
1378 if (ctx->split) {
1379 char *lock_name = get_chain_filename(ctx->odb);
1381 hold_lock_file_for_update(&lk, lock_name, LOCK_DIE_ON_ERROR);
1383 fd = git_mkstemp_mode(ctx->graph_name, 0444);
1384 if (fd < 0) {
1385 error(_("unable to create '%s'"), ctx->graph_name);
1386 return -1;
1389 f = hashfd(fd, ctx->graph_name);
1390 } else {
1391 hold_lock_file_for_update(&lk, ctx->graph_name, LOCK_DIE_ON_ERROR);
1392 fd = lk.tempfile->fd;
1393 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
1396 chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
1397 chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
1398 chunk_ids[2] = GRAPH_CHUNKID_DATA;
1399 if (ctx->num_extra_edges) {
1400 chunk_ids[num_chunks] = GRAPH_CHUNKID_EXTRAEDGES;
1401 num_chunks++;
1403 if (ctx->num_commit_graphs_after > 1) {
1404 chunk_ids[num_chunks] = GRAPH_CHUNKID_BASE;
1405 num_chunks++;
1408 chunk_ids[num_chunks] = 0;
1410 chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
1411 chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
1412 chunk_offsets[2] = chunk_offsets[1] + hashsz * ctx->commits.nr;
1413 chunk_offsets[3] = chunk_offsets[2] + (hashsz + 16) * ctx->commits.nr;
1415 num_chunks = 3;
1416 if (ctx->num_extra_edges) {
1417 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1418 4 * ctx->num_extra_edges;
1419 num_chunks++;
1421 if (ctx->num_commit_graphs_after > 1) {
1422 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1423 hashsz * (ctx->num_commit_graphs_after - 1);
1424 num_chunks++;
1427 hashwrite_be32(f, GRAPH_SIGNATURE);
1429 hashwrite_u8(f, GRAPH_VERSION);
1430 hashwrite_u8(f, oid_version());
1431 hashwrite_u8(f, num_chunks);
1432 hashwrite_u8(f, ctx->num_commit_graphs_after - 1);
1434 for (i = 0; i <= num_chunks; i++) {
1435 uint32_t chunk_write[3];
1437 chunk_write[0] = htonl(chunk_ids[i]);
1438 chunk_write[1] = htonl(chunk_offsets[i] >> 32);
1439 chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
1440 hashwrite(f, chunk_write, 12);
1443 if (ctx->report_progress) {
1444 strbuf_addf(&progress_title,
1445 Q_("Writing out commit graph in %d pass",
1446 "Writing out commit graph in %d passes",
1447 num_chunks),
1448 num_chunks);
1449 ctx->progress = start_delayed_progress(
1450 progress_title.buf,
1451 num_chunks * ctx->commits.nr);
1453 write_graph_chunk_fanout(f, ctx);
1454 write_graph_chunk_oids(f, hashsz, ctx);
1455 write_graph_chunk_data(f, hashsz, ctx);
1456 if (ctx->num_extra_edges)
1457 write_graph_chunk_extra_edges(f, ctx);
1458 if (ctx->num_commit_graphs_after > 1 &&
1459 write_graph_chunk_base(f, ctx)) {
1460 return -1;
1462 stop_progress(&ctx->progress);
1463 strbuf_release(&progress_title);
1465 if (ctx->split && ctx->base_graph_name && ctx->num_commit_graphs_after > 1) {
1466 char *new_base_hash = xstrdup(oid_to_hex(&ctx->new_base_graph->oid));
1467 char *new_base_name = get_split_graph_filename(ctx->new_base_graph->odb, new_base_hash);
1469 free(ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2]);
1470 free(ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2]);
1471 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2] = new_base_name;
1472 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2] = new_base_hash;
1475 close_commit_graph(ctx->r->objects);
1476 finalize_hashfile(f, file_hash.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
1478 if (ctx->split) {
1479 FILE *chainf = fdopen_lock_file(&lk, "w");
1480 char *final_graph_name;
1481 int result;
1483 close(fd);
1485 if (!chainf) {
1486 error(_("unable to open commit-graph chain file"));
1487 return -1;
1490 if (ctx->base_graph_name) {
1491 const char *dest = ctx->commit_graph_filenames_after[
1492 ctx->num_commit_graphs_after - 2];
1494 if (strcmp(ctx->base_graph_name, dest)) {
1495 result = rename(ctx->base_graph_name, dest);
1497 if (result) {
1498 error(_("failed to rename base commit-graph file"));
1499 return -1;
1502 } else {
1503 char *graph_name = get_commit_graph_filename(ctx->odb);
1504 unlink(graph_name);
1507 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1] = xstrdup(oid_to_hex(&file_hash));
1508 final_graph_name = get_split_graph_filename(ctx->odb,
1509 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1]);
1510 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 1] = final_graph_name;
1512 result = rename(ctx->graph_name, final_graph_name);
1514 for (i = 0; i < ctx->num_commit_graphs_after; i++)
1515 fprintf(lk.tempfile->fp, "%s\n", ctx->commit_graph_hash_after[i]);
1517 if (result) {
1518 error(_("failed to rename temporary commit-graph file"));
1519 return -1;
1523 commit_lock_file(&lk);
1525 return 0;
1528 static void split_graph_merge_strategy(struct write_commit_graph_context *ctx)
1530 struct commit_graph *g;
1531 uint32_t num_commits;
1532 uint32_t i;
1534 int max_commits = 0;
1535 int size_mult = 2;
1537 if (ctx->split_opts) {
1538 max_commits = ctx->split_opts->max_commits;
1540 if (ctx->split_opts->size_multiple)
1541 size_mult = ctx->split_opts->size_multiple;
1544 g = ctx->r->objects->commit_graph;
1545 num_commits = ctx->commits.nr;
1546 ctx->num_commit_graphs_after = ctx->num_commit_graphs_before + 1;
1548 while (g && (g->num_commits <= size_mult * num_commits ||
1549 (max_commits && num_commits > max_commits))) {
1550 if (g->odb != ctx->odb)
1551 break;
1553 num_commits += g->num_commits;
1554 g = g->base_graph;
1556 ctx->num_commit_graphs_after--;
1559 ctx->new_base_graph = g;
1561 if (ctx->num_commit_graphs_after == 2) {
1562 char *old_graph_name = get_commit_graph_filename(g->odb);
1564 if (!strcmp(g->filename, old_graph_name) &&
1565 g->odb != ctx->odb) {
1566 ctx->num_commit_graphs_after = 1;
1567 ctx->new_base_graph = NULL;
1570 free(old_graph_name);
1573 ALLOC_ARRAY(ctx->commit_graph_filenames_after, ctx->num_commit_graphs_after);
1574 ALLOC_ARRAY(ctx->commit_graph_hash_after, ctx->num_commit_graphs_after);
1576 for (i = 0; i < ctx->num_commit_graphs_after &&
1577 i < ctx->num_commit_graphs_before; i++)
1578 ctx->commit_graph_filenames_after[i] = xstrdup(ctx->commit_graph_filenames_before[i]);
1580 i = ctx->num_commit_graphs_before - 1;
1581 g = ctx->r->objects->commit_graph;
1583 while (g) {
1584 if (i < ctx->num_commit_graphs_after)
1585 ctx->commit_graph_hash_after[i] = xstrdup(oid_to_hex(&g->oid));
1587 i--;
1588 g = g->base_graph;
1592 static void merge_commit_graph(struct write_commit_graph_context *ctx,
1593 struct commit_graph *g)
1595 uint32_t i;
1596 uint32_t offset = g->num_commits_in_base;
1598 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + g->num_commits, ctx->commits.alloc);
1600 for (i = 0; i < g->num_commits; i++) {
1601 struct object_id oid;
1602 struct commit *result;
1604 display_progress(ctx->progress, i + 1);
1606 load_oid_from_graph(g, i + offset, &oid);
1608 /* only add commits if they still exist in the repo */
1609 result = lookup_commit_reference_gently(ctx->r, &oid, 1);
1611 if (result) {
1612 ctx->commits.list[ctx->commits.nr] = result;
1613 ctx->commits.nr++;
1618 static int commit_compare(const void *_a, const void *_b)
1620 const struct commit *a = *(const struct commit **)_a;
1621 const struct commit *b = *(const struct commit **)_b;
1622 return oidcmp(&a->object.oid, &b->object.oid);
1625 static void sort_and_scan_merged_commits(struct write_commit_graph_context *ctx)
1627 uint32_t i;
1629 if (ctx->report_progress)
1630 ctx->progress = start_delayed_progress(
1631 _("Scanning merged commits"),
1632 ctx->commits.nr);
1634 QSORT(ctx->commits.list, ctx->commits.nr, commit_compare);
1636 ctx->num_extra_edges = 0;
1637 for (i = 0; i < ctx->commits.nr; i++) {
1638 display_progress(ctx->progress, i);
1640 if (i && oideq(&ctx->commits.list[i - 1]->object.oid,
1641 &ctx->commits.list[i]->object.oid)) {
1642 die(_("unexpected duplicate commit id %s"),
1643 oid_to_hex(&ctx->commits.list[i]->object.oid));
1644 } else {
1645 unsigned int num_parents;
1647 num_parents = commit_list_count(ctx->commits.list[i]->parents);
1648 if (num_parents > 2)
1649 ctx->num_extra_edges += num_parents - 1;
1653 stop_progress(&ctx->progress);
1656 static void merge_commit_graphs(struct write_commit_graph_context *ctx)
1658 struct commit_graph *g = ctx->r->objects->commit_graph;
1659 uint32_t current_graph_number = ctx->num_commit_graphs_before;
1661 while (g && current_graph_number >= ctx->num_commit_graphs_after) {
1662 current_graph_number--;
1664 if (ctx->report_progress)
1665 ctx->progress = start_delayed_progress(_("Merging commit-graph"), 0);
1667 merge_commit_graph(ctx, g);
1668 stop_progress(&ctx->progress);
1670 g = g->base_graph;
1673 if (g) {
1674 ctx->new_base_graph = g;
1675 ctx->new_num_commits_in_base = g->num_commits + g->num_commits_in_base;
1678 if (ctx->new_base_graph)
1679 ctx->base_graph_name = xstrdup(ctx->new_base_graph->filename);
1681 sort_and_scan_merged_commits(ctx);
1684 static void mark_commit_graphs(struct write_commit_graph_context *ctx)
1686 uint32_t i;
1687 time_t now = time(NULL);
1689 for (i = ctx->num_commit_graphs_after - 1; i < ctx->num_commit_graphs_before; i++) {
1690 struct stat st;
1691 struct utimbuf updated_time;
1693 stat(ctx->commit_graph_filenames_before[i], &st);
1695 updated_time.actime = st.st_atime;
1696 updated_time.modtime = now;
1697 utime(ctx->commit_graph_filenames_before[i], &updated_time);
1701 static void expire_commit_graphs(struct write_commit_graph_context *ctx)
1703 struct strbuf path = STRBUF_INIT;
1704 DIR *dir;
1705 struct dirent *de;
1706 size_t dirnamelen;
1707 timestamp_t expire_time = time(NULL);
1709 if (ctx->split_opts && ctx->split_opts->expire_time)
1710 expire_time -= ctx->split_opts->expire_time;
1711 if (!ctx->split) {
1712 char *chain_file_name = get_chain_filename(ctx->odb);
1713 unlink(chain_file_name);
1714 free(chain_file_name);
1715 ctx->num_commit_graphs_after = 0;
1718 strbuf_addstr(&path, ctx->odb->path);
1719 strbuf_addstr(&path, "/info/commit-graphs");
1720 dir = opendir(path.buf);
1722 if (!dir)
1723 goto out;
1725 strbuf_addch(&path, '/');
1726 dirnamelen = path.len;
1727 while ((de = readdir(dir)) != NULL) {
1728 struct stat st;
1729 uint32_t i, found = 0;
1731 strbuf_setlen(&path, dirnamelen);
1732 strbuf_addstr(&path, de->d_name);
1734 stat(path.buf, &st);
1736 if (st.st_mtime > expire_time)
1737 continue;
1738 if (path.len < 6 || strcmp(path.buf + path.len - 6, ".graph"))
1739 continue;
1741 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
1742 if (!strcmp(ctx->commit_graph_filenames_after[i],
1743 path.buf)) {
1744 found = 1;
1745 break;
1749 if (!found)
1750 unlink(path.buf);
1753 out:
1754 strbuf_release(&path);
1757 int write_commit_graph(struct object_directory *odb,
1758 struct string_list *pack_indexes,
1759 struct string_list *commit_hex,
1760 enum commit_graph_write_flags flags,
1761 const struct split_commit_graph_opts *split_opts)
1763 struct write_commit_graph_context *ctx;
1764 uint32_t i, count_distinct = 0;
1765 int res = 0;
1767 if (!commit_graph_compatible(the_repository))
1768 return 0;
1770 ctx = xcalloc(1, sizeof(struct write_commit_graph_context));
1771 ctx->r = the_repository;
1772 ctx->odb = odb;
1773 ctx->append = flags & COMMIT_GRAPH_WRITE_APPEND ? 1 : 0;
1774 ctx->report_progress = flags & COMMIT_GRAPH_WRITE_PROGRESS ? 1 : 0;
1775 ctx->split = flags & COMMIT_GRAPH_WRITE_SPLIT ? 1 : 0;
1776 ctx->check_oids = flags & COMMIT_GRAPH_WRITE_CHECK_OIDS ? 1 : 0;
1777 ctx->split_opts = split_opts;
1779 if (ctx->split) {
1780 struct commit_graph *g;
1781 prepare_commit_graph(ctx->r);
1783 g = ctx->r->objects->commit_graph;
1785 while (g) {
1786 ctx->num_commit_graphs_before++;
1787 g = g->base_graph;
1790 if (ctx->num_commit_graphs_before) {
1791 ALLOC_ARRAY(ctx->commit_graph_filenames_before, ctx->num_commit_graphs_before);
1792 i = ctx->num_commit_graphs_before;
1793 g = ctx->r->objects->commit_graph;
1795 while (g) {
1796 ctx->commit_graph_filenames_before[--i] = xstrdup(g->filename);
1797 g = g->base_graph;
1802 ctx->approx_nr_objects = approximate_object_count();
1803 ctx->oids.alloc = ctx->approx_nr_objects / 32;
1805 if (ctx->split && split_opts && ctx->oids.alloc > split_opts->max_commits)
1806 ctx->oids.alloc = split_opts->max_commits;
1808 if (ctx->append) {
1809 prepare_commit_graph_one(ctx->r, ctx->odb);
1810 if (ctx->r->objects->commit_graph)
1811 ctx->oids.alloc += ctx->r->objects->commit_graph->num_commits;
1814 if (ctx->oids.alloc < 1024)
1815 ctx->oids.alloc = 1024;
1816 ALLOC_ARRAY(ctx->oids.list, ctx->oids.alloc);
1818 if (ctx->append && ctx->r->objects->commit_graph) {
1819 struct commit_graph *g = ctx->r->objects->commit_graph;
1820 for (i = 0; i < g->num_commits; i++) {
1821 const unsigned char *hash = g->chunk_oid_lookup + g->hash_len * i;
1822 hashcpy(ctx->oids.list[ctx->oids.nr++].hash, hash);
1826 if (pack_indexes) {
1827 if ((res = fill_oids_from_packs(ctx, pack_indexes)))
1828 goto cleanup;
1831 if (commit_hex) {
1832 if ((res = fill_oids_from_commit_hex(ctx, commit_hex)))
1833 goto cleanup;
1836 if (!pack_indexes && !commit_hex)
1837 fill_oids_from_all_packs(ctx);
1839 close_reachable(ctx);
1841 count_distinct = count_distinct_commits(ctx);
1843 if (count_distinct >= GRAPH_EDGE_LAST_MASK) {
1844 error(_("the commit graph format cannot write %d commits"), count_distinct);
1845 res = -1;
1846 goto cleanup;
1849 ctx->commits.alloc = count_distinct;
1850 ALLOC_ARRAY(ctx->commits.list, ctx->commits.alloc);
1852 copy_oids_to_commits(ctx);
1854 if (ctx->commits.nr >= GRAPH_EDGE_LAST_MASK) {
1855 error(_("too many commits to write graph"));
1856 res = -1;
1857 goto cleanup;
1860 if (!ctx->commits.nr)
1861 goto cleanup;
1863 if (ctx->split) {
1864 split_graph_merge_strategy(ctx);
1866 merge_commit_graphs(ctx);
1867 } else
1868 ctx->num_commit_graphs_after = 1;
1870 compute_generation_numbers(ctx);
1872 res = write_commit_graph_file(ctx);
1874 if (ctx->split)
1875 mark_commit_graphs(ctx);
1877 expire_commit_graphs(ctx);
1879 cleanup:
1880 free(ctx->graph_name);
1881 free(ctx->commits.list);
1882 free(ctx->oids.list);
1884 if (ctx->commit_graph_filenames_after) {
1885 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
1886 free(ctx->commit_graph_filenames_after[i]);
1887 free(ctx->commit_graph_hash_after[i]);
1890 for (i = 0; i < ctx->num_commit_graphs_before; i++)
1891 free(ctx->commit_graph_filenames_before[i]);
1893 free(ctx->commit_graph_filenames_after);
1894 free(ctx->commit_graph_filenames_before);
1895 free(ctx->commit_graph_hash_after);
1898 free(ctx);
1900 return res;
1903 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
1904 static int verify_commit_graph_error;
1906 static void graph_report(const char *fmt, ...)
1908 va_list ap;
1910 verify_commit_graph_error = 1;
1911 va_start(ap, fmt);
1912 vfprintf(stderr, fmt, ap);
1913 fprintf(stderr, "\n");
1914 va_end(ap);
1917 #define GENERATION_ZERO_EXISTS 1
1918 #define GENERATION_NUMBER_EXISTS 2
1920 int verify_commit_graph(struct repository *r, struct commit_graph *g, int flags)
1922 uint32_t i, cur_fanout_pos = 0;
1923 struct object_id prev_oid, cur_oid, checksum;
1924 int generation_zero = 0;
1925 struct hashfile *f;
1926 int devnull;
1927 struct progress *progress = NULL;
1928 int local_error = 0;
1930 if (!g) {
1931 graph_report("no commit-graph file loaded");
1932 return 1;
1935 verify_commit_graph_error = verify_commit_graph_lite(g);
1936 if (verify_commit_graph_error)
1937 return verify_commit_graph_error;
1939 devnull = open("/dev/null", O_WRONLY);
1940 f = hashfd(devnull, NULL);
1941 hashwrite(f, g->data, g->data_len - g->hash_len);
1942 finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
1943 if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
1944 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
1945 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
1948 for (i = 0; i < g->num_commits; i++) {
1949 struct commit *graph_commit;
1951 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1953 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
1954 graph_report(_("commit-graph has incorrect OID order: %s then %s"),
1955 oid_to_hex(&prev_oid),
1956 oid_to_hex(&cur_oid));
1958 oidcpy(&prev_oid, &cur_oid);
1960 while (cur_oid.hash[0] > cur_fanout_pos) {
1961 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1963 if (i != fanout_value)
1964 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
1965 cur_fanout_pos, fanout_value, i);
1966 cur_fanout_pos++;
1969 graph_commit = lookup_commit(r, &cur_oid);
1970 if (!parse_commit_in_graph_one(r, g, graph_commit))
1971 graph_report(_("failed to parse commit %s from commit-graph"),
1972 oid_to_hex(&cur_oid));
1975 while (cur_fanout_pos < 256) {
1976 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1978 if (g->num_commits != fanout_value)
1979 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
1980 cur_fanout_pos, fanout_value, i);
1982 cur_fanout_pos++;
1985 if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
1986 return verify_commit_graph_error;
1988 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
1989 progress = start_progress(_("Verifying commits in commit graph"),
1990 g->num_commits);
1992 for (i = 0; i < g->num_commits; i++) {
1993 struct commit *graph_commit, *odb_commit;
1994 struct commit_list *graph_parents, *odb_parents;
1995 uint32_t max_generation = 0;
1997 display_progress(progress, i + 1);
1998 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
2000 graph_commit = lookup_commit(r, &cur_oid);
2001 odb_commit = (struct commit *)create_object(r, &cur_oid, alloc_commit_node(r));
2002 if (parse_commit_internal(odb_commit, 0, 0)) {
2003 graph_report(_("failed to parse commit %s from object database for commit-graph"),
2004 oid_to_hex(&cur_oid));
2005 continue;
2008 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
2009 get_commit_tree_oid(odb_commit)))
2010 graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
2011 oid_to_hex(&cur_oid),
2012 oid_to_hex(get_commit_tree_oid(graph_commit)),
2013 oid_to_hex(get_commit_tree_oid(odb_commit)));
2015 graph_parents = graph_commit->parents;
2016 odb_parents = odb_commit->parents;
2018 while (graph_parents) {
2019 if (odb_parents == NULL) {
2020 graph_report(_("commit-graph parent list for commit %s is too long"),
2021 oid_to_hex(&cur_oid));
2022 break;
2025 /* parse parent in case it is in a base graph */
2026 parse_commit_in_graph_one(r, g, graph_parents->item);
2028 if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
2029 graph_report(_("commit-graph parent for %s is %s != %s"),
2030 oid_to_hex(&cur_oid),
2031 oid_to_hex(&graph_parents->item->object.oid),
2032 oid_to_hex(&odb_parents->item->object.oid));
2034 if (graph_parents->item->generation > max_generation)
2035 max_generation = graph_parents->item->generation;
2037 graph_parents = graph_parents->next;
2038 odb_parents = odb_parents->next;
2041 if (odb_parents != NULL)
2042 graph_report(_("commit-graph parent list for commit %s terminates early"),
2043 oid_to_hex(&cur_oid));
2045 if (!graph_commit->generation) {
2046 if (generation_zero == GENERATION_NUMBER_EXISTS)
2047 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
2048 oid_to_hex(&cur_oid));
2049 generation_zero = GENERATION_ZERO_EXISTS;
2050 } else if (generation_zero == GENERATION_ZERO_EXISTS)
2051 graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
2052 oid_to_hex(&cur_oid));
2054 if (generation_zero == GENERATION_ZERO_EXISTS)
2055 continue;
2058 * If one of our parents has generation GENERATION_NUMBER_MAX, then
2059 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
2060 * extra logic in the following condition.
2062 if (max_generation == GENERATION_NUMBER_MAX)
2063 max_generation--;
2065 if (graph_commit->generation != max_generation + 1)
2066 graph_report(_("commit-graph generation for commit %s is %u != %u"),
2067 oid_to_hex(&cur_oid),
2068 graph_commit->generation,
2069 max_generation + 1);
2071 if (graph_commit->date != odb_commit->date)
2072 graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
2073 oid_to_hex(&cur_oid),
2074 graph_commit->date,
2075 odb_commit->date);
2077 stop_progress(&progress);
2079 local_error = verify_commit_graph_error;
2081 if (!(flags & COMMIT_GRAPH_VERIFY_SHALLOW) && g->base_graph)
2082 local_error |= verify_commit_graph(r, g->base_graph, flags);
2084 return local_error;
2087 void free_commit_graph(struct commit_graph *g)
2089 if (!g)
2090 return;
2091 if (g->graph_fd >= 0) {
2092 munmap((void *)g->data, g->data_len);
2093 g->data = NULL;
2094 close(g->graph_fd);
2096 free(g->filename);
2097 free(g);
2100 void disable_commit_graph(struct repository *r)
2102 r->commit_graph_disabled = 1;