commit-graph: examine changed-path objects in pack order
[git/raj.git] / commit-graph.c
blob31b06f878ced3ebbf843f3bf3352aa6d3a28840c
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"
19 #include "bloom.h"
20 #include "commit-slab.h"
22 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
23 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
24 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
25 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
26 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
27 #define GRAPH_CHUNKID_BASE 0x42415345 /* "BASE" */
28 #define MAX_NUM_CHUNKS 5
30 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
32 #define GRAPH_VERSION_1 0x1
33 #define GRAPH_VERSION GRAPH_VERSION_1
35 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
36 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
37 #define GRAPH_PARENT_NONE 0x70000000
39 #define GRAPH_LAST_EDGE 0x80000000
41 #define GRAPH_HEADER_SIZE 8
42 #define GRAPH_FANOUT_SIZE (4 * 256)
43 #define GRAPH_CHUNKLOOKUP_WIDTH 12
44 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
45 + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
47 /* Remember to update object flag allocation in object.h */
48 #define REACHABLE (1u<<15)
50 /* Keep track of the order in which commits are added to our list. */
51 define_commit_slab(commit_pos, int);
52 static struct commit_pos commit_pos = COMMIT_SLAB_INIT(1, commit_pos);
54 static void set_commit_pos(struct repository *r, const struct object_id *oid)
56 static int32_t max_pos;
57 struct commit *commit = lookup_commit(r, oid);
59 if (!commit)
60 return; /* should never happen, but be lenient */
62 *commit_pos_at(&commit_pos, commit) = max_pos++;
65 static int commit_pos_cmp(const void *va, const void *vb)
67 const struct commit *a = *(const struct commit **)va;
68 const struct commit *b = *(const struct commit **)vb;
69 return commit_pos_at(&commit_pos, a) -
70 commit_pos_at(&commit_pos, b);
73 char *get_commit_graph_filename(struct object_directory *obj_dir)
75 return xstrfmt("%s/info/commit-graph", obj_dir->path);
78 static char *get_split_graph_filename(struct object_directory *odb,
79 const char *oid_hex)
81 return xstrfmt("%s/info/commit-graphs/graph-%s.graph", odb->path,
82 oid_hex);
85 static char *get_chain_filename(struct object_directory *odb)
87 return xstrfmt("%s/info/commit-graphs/commit-graph-chain", odb->path);
90 static uint8_t oid_version(void)
92 return 1;
95 static struct commit_graph *alloc_commit_graph(void)
97 struct commit_graph *g = xcalloc(1, sizeof(*g));
98 g->graph_fd = -1;
100 return g;
103 extern int read_replace_refs;
105 static int commit_graph_compatible(struct repository *r)
107 if (!r->gitdir)
108 return 0;
110 if (read_replace_refs) {
111 prepare_replace_object(r);
112 if (hashmap_get_size(&r->objects->replace_map->map))
113 return 0;
116 prepare_commit_graft(r);
117 if (r->parsed_objects && r->parsed_objects->grafts_nr)
118 return 0;
119 if (is_repository_shallow(r))
120 return 0;
122 return 1;
125 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
127 *fd = git_open(graph_file);
128 if (*fd < 0)
129 return 0;
130 if (fstat(*fd, st)) {
131 close(*fd);
132 return 0;
134 return 1;
137 struct commit_graph *load_commit_graph_one_fd_st(int fd, struct stat *st,
138 struct object_directory *odb)
140 void *graph_map;
141 size_t graph_size;
142 struct commit_graph *ret;
144 graph_size = xsize_t(st->st_size);
146 if (graph_size < GRAPH_MIN_SIZE) {
147 close(fd);
148 error(_("commit-graph file is too small"));
149 return NULL;
151 graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
152 ret = parse_commit_graph(graph_map, fd, graph_size);
154 if (ret)
155 ret->odb = odb;
156 else {
157 munmap(graph_map, graph_size);
158 close(fd);
161 return ret;
164 static int verify_commit_graph_lite(struct commit_graph *g)
167 * Basic validation shared between parse_commit_graph()
168 * which'll be called every time the graph is used, and the
169 * much more expensive verify_commit_graph() used by
170 * "commit-graph verify".
172 * There should only be very basic checks here to ensure that
173 * we don't e.g. segfault in fill_commit_in_graph(), but
174 * because this is a very hot codepath nothing that e.g. loops
175 * over g->num_commits, or runs a checksum on the commit-graph
176 * itself.
178 if (!g->chunk_oid_fanout) {
179 error("commit-graph is missing the OID Fanout chunk");
180 return 1;
182 if (!g->chunk_oid_lookup) {
183 error("commit-graph is missing the OID Lookup chunk");
184 return 1;
186 if (!g->chunk_commit_data) {
187 error("commit-graph is missing the Commit Data chunk");
188 return 1;
191 return 0;
194 struct commit_graph *parse_commit_graph(void *graph_map, int fd,
195 size_t graph_size)
197 const unsigned char *data, *chunk_lookup;
198 uint32_t i;
199 struct commit_graph *graph;
200 uint64_t last_chunk_offset;
201 uint32_t last_chunk_id;
202 uint32_t graph_signature;
203 unsigned char graph_version, hash_version;
205 if (!graph_map)
206 return NULL;
208 if (graph_size < GRAPH_MIN_SIZE)
209 return NULL;
211 data = (const unsigned char *)graph_map;
213 graph_signature = get_be32(data);
214 if (graph_signature != GRAPH_SIGNATURE) {
215 error(_("commit-graph signature %X does not match signature %X"),
216 graph_signature, GRAPH_SIGNATURE);
217 return NULL;
220 graph_version = *(unsigned char*)(data + 4);
221 if (graph_version != GRAPH_VERSION) {
222 error(_("commit-graph version %X does not match version %X"),
223 graph_version, GRAPH_VERSION);
224 return NULL;
227 hash_version = *(unsigned char*)(data + 5);
228 if (hash_version != oid_version()) {
229 error(_("commit-graph hash version %X does not match version %X"),
230 hash_version, oid_version());
231 return NULL;
234 graph = alloc_commit_graph();
236 graph->hash_len = the_hash_algo->rawsz;
237 graph->num_chunks = *(unsigned char*)(data + 6);
238 graph->graph_fd = fd;
239 graph->data = graph_map;
240 graph->data_len = graph_size;
242 last_chunk_id = 0;
243 last_chunk_offset = 8;
244 chunk_lookup = data + 8;
245 for (i = 0; i < graph->num_chunks; i++) {
246 uint32_t chunk_id;
247 uint64_t chunk_offset;
248 int chunk_repeated = 0;
250 if (data + graph_size - chunk_lookup <
251 GRAPH_CHUNKLOOKUP_WIDTH) {
252 error(_("commit-graph chunk lookup table entry missing; file may be incomplete"));
253 free(graph);
254 return NULL;
257 chunk_id = get_be32(chunk_lookup + 0);
258 chunk_offset = get_be64(chunk_lookup + 4);
260 chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
262 if (chunk_offset > graph_size - the_hash_algo->rawsz) {
263 error(_("commit-graph improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
264 (uint32_t)chunk_offset);
265 free(graph);
266 return NULL;
269 switch (chunk_id) {
270 case GRAPH_CHUNKID_OIDFANOUT:
271 if (graph->chunk_oid_fanout)
272 chunk_repeated = 1;
273 else
274 graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
275 break;
277 case GRAPH_CHUNKID_OIDLOOKUP:
278 if (graph->chunk_oid_lookup)
279 chunk_repeated = 1;
280 else
281 graph->chunk_oid_lookup = data + chunk_offset;
282 break;
284 case GRAPH_CHUNKID_DATA:
285 if (graph->chunk_commit_data)
286 chunk_repeated = 1;
287 else
288 graph->chunk_commit_data = data + chunk_offset;
289 break;
291 case GRAPH_CHUNKID_EXTRAEDGES:
292 if (graph->chunk_extra_edges)
293 chunk_repeated = 1;
294 else
295 graph->chunk_extra_edges = data + chunk_offset;
296 break;
298 case GRAPH_CHUNKID_BASE:
299 if (graph->chunk_base_graphs)
300 chunk_repeated = 1;
301 else
302 graph->chunk_base_graphs = data + chunk_offset;
305 if (chunk_repeated) {
306 error(_("commit-graph chunk id %08x appears multiple times"), chunk_id);
307 free(graph);
308 return NULL;
311 if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
313 graph->num_commits = (chunk_offset - last_chunk_offset)
314 / graph->hash_len;
317 last_chunk_id = chunk_id;
318 last_chunk_offset = chunk_offset;
321 hashcpy(graph->oid.hash, graph->data + graph->data_len - graph->hash_len);
323 if (verify_commit_graph_lite(graph)) {
324 free(graph);
325 return NULL;
328 return graph;
331 static struct commit_graph *load_commit_graph_one(const char *graph_file,
332 struct object_directory *odb)
335 struct stat st;
336 int fd;
337 struct commit_graph *g;
338 int open_ok = open_commit_graph(graph_file, &fd, &st);
340 if (!open_ok)
341 return NULL;
343 g = load_commit_graph_one_fd_st(fd, &st, odb);
345 if (g)
346 g->filename = xstrdup(graph_file);
348 return g;
351 static struct commit_graph *load_commit_graph_v1(struct repository *r,
352 struct object_directory *odb)
354 char *graph_name = get_commit_graph_filename(odb);
355 struct commit_graph *g = load_commit_graph_one(graph_name, odb);
356 free(graph_name);
358 return g;
361 static int add_graph_to_chain(struct commit_graph *g,
362 struct commit_graph *chain,
363 struct object_id *oids,
364 int n)
366 struct commit_graph *cur_g = chain;
368 if (n && !g->chunk_base_graphs) {
369 warning(_("commit-graph has no base graphs chunk"));
370 return 0;
373 while (n) {
374 n--;
376 if (!cur_g ||
377 !oideq(&oids[n], &cur_g->oid) ||
378 !hasheq(oids[n].hash, g->chunk_base_graphs + g->hash_len * n)) {
379 warning(_("commit-graph chain does not match"));
380 return 0;
383 cur_g = cur_g->base_graph;
386 g->base_graph = chain;
388 if (chain)
389 g->num_commits_in_base = chain->num_commits + chain->num_commits_in_base;
391 return 1;
394 static struct commit_graph *load_commit_graph_chain(struct repository *r,
395 struct object_directory *odb)
397 struct commit_graph *graph_chain = NULL;
398 struct strbuf line = STRBUF_INIT;
399 struct stat st;
400 struct object_id *oids;
401 int i = 0, valid = 1, count;
402 char *chain_name = get_chain_filename(odb);
403 FILE *fp;
404 int stat_res;
406 fp = fopen(chain_name, "r");
407 stat_res = stat(chain_name, &st);
408 free(chain_name);
410 if (!fp ||
411 stat_res ||
412 st.st_size <= the_hash_algo->hexsz)
413 return NULL;
415 count = st.st_size / (the_hash_algo->hexsz + 1);
416 oids = xcalloc(count, sizeof(struct object_id));
418 prepare_alt_odb(r);
420 for (i = 0; i < count; i++) {
421 struct object_directory *odb;
423 if (strbuf_getline_lf(&line, fp) == EOF)
424 break;
426 if (get_oid_hex(line.buf, &oids[i])) {
427 warning(_("invalid commit-graph chain: line '%s' not a hash"),
428 line.buf);
429 valid = 0;
430 break;
433 valid = 0;
434 for (odb = r->objects->odb; odb; odb = odb->next) {
435 char *graph_name = get_split_graph_filename(odb, line.buf);
436 struct commit_graph *g = load_commit_graph_one(graph_name, odb);
438 free(graph_name);
440 if (g) {
441 if (add_graph_to_chain(g, graph_chain, oids, i)) {
442 graph_chain = g;
443 valid = 1;
446 break;
450 if (!valid) {
451 warning(_("unable to find all commit-graph files"));
452 break;
456 free(oids);
457 fclose(fp);
458 strbuf_release(&line);
460 return graph_chain;
463 struct commit_graph *read_commit_graph_one(struct repository *r,
464 struct object_directory *odb)
466 struct commit_graph *g = load_commit_graph_v1(r, odb);
468 if (!g)
469 g = load_commit_graph_chain(r, odb);
471 return g;
474 static void prepare_commit_graph_one(struct repository *r,
475 struct object_directory *odb)
478 if (r->objects->commit_graph)
479 return;
481 r->objects->commit_graph = read_commit_graph_one(r, odb);
485 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
487 * On the first invocation, this function attempts to load the commit
488 * graph if the_repository is configured to have one.
490 static int prepare_commit_graph(struct repository *r)
492 struct object_directory *odb;
495 * This must come before the "already attempted?" check below, because
496 * we want to disable even an already-loaded graph file.
498 if (r->commit_graph_disabled)
499 return 0;
501 if (r->objects->commit_graph_attempted)
502 return !!r->objects->commit_graph;
503 r->objects->commit_graph_attempted = 1;
505 if (git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD, 0))
506 die("dying as requested by the '%s' variable on commit-graph load!",
507 GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD);
509 prepare_repo_settings(r);
511 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
512 r->settings.core_commit_graph != 1)
514 * This repository is not configured to use commit graphs, so
515 * do not load one. (But report commit_graph_attempted anyway
516 * so that commit graph loading is not attempted again for this
517 * repository.)
519 return 0;
521 if (!commit_graph_compatible(r))
522 return 0;
524 prepare_alt_odb(r);
525 for (odb = r->objects->odb;
526 !r->objects->commit_graph && odb;
527 odb = odb->next)
528 prepare_commit_graph_one(r, odb);
529 return !!r->objects->commit_graph;
532 int generation_numbers_enabled(struct repository *r)
534 uint32_t first_generation;
535 struct commit_graph *g;
536 if (!prepare_commit_graph(r))
537 return 0;
539 g = r->objects->commit_graph;
541 if (!g->num_commits)
542 return 0;
544 first_generation = get_be32(g->chunk_commit_data +
545 g->hash_len + 8) >> 2;
547 return !!first_generation;
550 static void close_commit_graph_one(struct commit_graph *g)
552 if (!g)
553 return;
555 close_commit_graph_one(g->base_graph);
556 free_commit_graph(g);
559 void close_commit_graph(struct raw_object_store *o)
561 close_commit_graph_one(o->commit_graph);
562 o->commit_graph = NULL;
565 static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
567 return bsearch_hash(oid->hash, g->chunk_oid_fanout,
568 g->chunk_oid_lookup, g->hash_len, pos);
571 static void load_oid_from_graph(struct commit_graph *g,
572 uint32_t pos,
573 struct object_id *oid)
575 uint32_t lex_index;
577 while (g && pos < g->num_commits_in_base)
578 g = g->base_graph;
580 if (!g)
581 BUG("NULL commit-graph");
583 if (pos >= g->num_commits + g->num_commits_in_base)
584 die(_("invalid commit position. commit-graph is likely corrupt"));
586 lex_index = pos - g->num_commits_in_base;
588 hashcpy(oid->hash, g->chunk_oid_lookup + g->hash_len * lex_index);
591 static struct commit_list **insert_parent_or_die(struct repository *r,
592 struct commit_graph *g,
593 uint32_t pos,
594 struct commit_list **pptr)
596 struct commit *c;
597 struct object_id oid;
599 if (pos >= g->num_commits + g->num_commits_in_base)
600 die("invalid parent position %"PRIu32, pos);
602 load_oid_from_graph(g, pos, &oid);
603 c = lookup_commit(r, &oid);
604 if (!c)
605 die(_("could not find commit %s"), oid_to_hex(&oid));
606 c->graph_pos = pos;
607 return &commit_list_insert(c, pptr)->next;
610 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
612 const unsigned char *commit_data;
613 uint32_t lex_index;
615 while (pos < g->num_commits_in_base)
616 g = g->base_graph;
618 lex_index = pos - g->num_commits_in_base;
619 commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * lex_index;
620 item->graph_pos = pos;
621 item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
624 static inline void set_commit_tree(struct commit *c, struct tree *t)
626 c->maybe_tree = t;
629 static int fill_commit_in_graph(struct repository *r,
630 struct commit *item,
631 struct commit_graph *g, uint32_t pos)
633 uint32_t edge_value;
634 uint32_t *parent_data_ptr;
635 uint64_t date_low, date_high;
636 struct commit_list **pptr;
637 const unsigned char *commit_data;
638 uint32_t lex_index;
640 while (pos < g->num_commits_in_base)
641 g = g->base_graph;
643 if (pos >= g->num_commits + g->num_commits_in_base)
644 die(_("invalid commit position. commit-graph is likely corrupt"));
647 * Store the "full" position, but then use the
648 * "local" position for the rest of the calculation.
650 item->graph_pos = pos;
651 lex_index = pos - g->num_commits_in_base;
653 commit_data = g->chunk_commit_data + (g->hash_len + 16) * lex_index;
655 item->object.parsed = 1;
657 set_commit_tree(item, NULL);
659 date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
660 date_low = get_be32(commit_data + g->hash_len + 12);
661 item->date = (timestamp_t)((date_high << 32) | date_low);
663 item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
665 pptr = &item->parents;
667 edge_value = get_be32(commit_data + g->hash_len);
668 if (edge_value == GRAPH_PARENT_NONE)
669 return 1;
670 pptr = insert_parent_or_die(r, g, edge_value, pptr);
672 edge_value = get_be32(commit_data + g->hash_len + 4);
673 if (edge_value == GRAPH_PARENT_NONE)
674 return 1;
675 if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
676 pptr = insert_parent_or_die(r, g, edge_value, pptr);
677 return 1;
680 parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
681 4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
682 do {
683 edge_value = get_be32(parent_data_ptr);
684 pptr = insert_parent_or_die(r, g,
685 edge_value & GRAPH_EDGE_LAST_MASK,
686 pptr);
687 parent_data_ptr++;
688 } while (!(edge_value & GRAPH_LAST_EDGE));
690 return 1;
693 static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
695 if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
696 *pos = item->graph_pos;
697 return 1;
698 } else {
699 struct commit_graph *cur_g = g;
700 uint32_t lex_index;
702 while (cur_g && !bsearch_graph(cur_g, &(item->object.oid), &lex_index))
703 cur_g = cur_g->base_graph;
705 if (cur_g) {
706 *pos = lex_index + cur_g->num_commits_in_base;
707 return 1;
710 return 0;
714 static int parse_commit_in_graph_one(struct repository *r,
715 struct commit_graph *g,
716 struct commit *item)
718 uint32_t pos;
720 if (item->object.parsed)
721 return 1;
723 if (find_commit_in_graph(item, g, &pos))
724 return fill_commit_in_graph(r, item, g, pos);
726 return 0;
729 int parse_commit_in_graph(struct repository *r, struct commit *item)
731 if (!prepare_commit_graph(r))
732 return 0;
733 return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
736 void load_commit_graph_info(struct repository *r, struct commit *item)
738 uint32_t pos;
739 if (!prepare_commit_graph(r))
740 return;
741 if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
742 fill_commit_graph_info(item, r->objects->commit_graph, pos);
745 static struct tree *load_tree_for_commit(struct repository *r,
746 struct commit_graph *g,
747 struct commit *c)
749 struct object_id oid;
750 const unsigned char *commit_data;
752 while (c->graph_pos < g->num_commits_in_base)
753 g = g->base_graph;
755 commit_data = g->chunk_commit_data +
756 GRAPH_DATA_WIDTH * (c->graph_pos - g->num_commits_in_base);
758 hashcpy(oid.hash, commit_data);
759 set_commit_tree(c, lookup_tree(r, &oid));
761 return c->maybe_tree;
764 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
765 struct commit_graph *g,
766 const struct commit *c)
768 if (c->maybe_tree)
769 return c->maybe_tree;
770 if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
771 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
773 return load_tree_for_commit(r, g, (struct commit *)c);
776 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
778 return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
781 struct packed_commit_list {
782 struct commit **list;
783 int nr;
784 int alloc;
787 struct packed_oid_list {
788 struct object_id *list;
789 int nr;
790 int alloc;
793 struct write_commit_graph_context {
794 struct repository *r;
795 struct object_directory *odb;
796 char *graph_name;
797 struct packed_oid_list oids;
798 struct packed_commit_list commits;
799 int num_extra_edges;
800 unsigned long approx_nr_objects;
801 struct progress *progress;
802 int progress_done;
803 uint64_t progress_cnt;
805 char *base_graph_name;
806 int num_commit_graphs_before;
807 int num_commit_graphs_after;
808 char **commit_graph_filenames_before;
809 char **commit_graph_filenames_after;
810 char **commit_graph_hash_after;
811 uint32_t new_num_commits_in_base;
812 struct commit_graph *new_base_graph;
814 unsigned append:1,
815 report_progress:1,
816 split:1,
817 check_oids:1,
818 changed_paths:1;
820 const struct split_commit_graph_opts *split_opts;
821 size_t total_bloom_filter_data_size;
824 static void write_graph_chunk_fanout(struct hashfile *f,
825 struct write_commit_graph_context *ctx)
827 int i, count = 0;
828 struct commit **list = ctx->commits.list;
831 * Write the first-level table (the list is sorted,
832 * but we use a 256-entry lookup to be able to avoid
833 * having to do eight extra binary search iterations).
835 for (i = 0; i < 256; i++) {
836 while (count < ctx->commits.nr) {
837 if ((*list)->object.oid.hash[0] != i)
838 break;
839 display_progress(ctx->progress, ++ctx->progress_cnt);
840 count++;
841 list++;
844 hashwrite_be32(f, count);
848 static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
849 struct write_commit_graph_context *ctx)
851 struct commit **list = ctx->commits.list;
852 int count;
853 for (count = 0; count < ctx->commits.nr; count++, list++) {
854 display_progress(ctx->progress, ++ctx->progress_cnt);
855 hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
859 static const unsigned char *commit_to_sha1(size_t index, void *table)
861 struct commit **commits = table;
862 return commits[index]->object.oid.hash;
865 static void write_graph_chunk_data(struct hashfile *f, int hash_len,
866 struct write_commit_graph_context *ctx)
868 struct commit **list = ctx->commits.list;
869 struct commit **last = ctx->commits.list + ctx->commits.nr;
870 uint32_t num_extra_edges = 0;
872 while (list < last) {
873 struct commit_list *parent;
874 struct object_id *tree;
875 int edge_value;
876 uint32_t packedDate[2];
877 display_progress(ctx->progress, ++ctx->progress_cnt);
879 if (parse_commit_no_graph(*list))
880 die(_("unable to parse commit %s"),
881 oid_to_hex(&(*list)->object.oid));
882 tree = get_commit_tree_oid(*list);
883 hashwrite(f, tree->hash, hash_len);
885 parent = (*list)->parents;
887 if (!parent)
888 edge_value = GRAPH_PARENT_NONE;
889 else {
890 edge_value = sha1_pos(parent->item->object.oid.hash,
891 ctx->commits.list,
892 ctx->commits.nr,
893 commit_to_sha1);
895 if (edge_value >= 0)
896 edge_value += ctx->new_num_commits_in_base;
897 else {
898 uint32_t pos;
899 if (find_commit_in_graph(parent->item,
900 ctx->new_base_graph,
901 &pos))
902 edge_value = pos;
905 if (edge_value < 0)
906 BUG("missing parent %s for commit %s",
907 oid_to_hex(&parent->item->object.oid),
908 oid_to_hex(&(*list)->object.oid));
911 hashwrite_be32(f, edge_value);
913 if (parent)
914 parent = parent->next;
916 if (!parent)
917 edge_value = GRAPH_PARENT_NONE;
918 else if (parent->next)
919 edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
920 else {
921 edge_value = sha1_pos(parent->item->object.oid.hash,
922 ctx->commits.list,
923 ctx->commits.nr,
924 commit_to_sha1);
926 if (edge_value >= 0)
927 edge_value += ctx->new_num_commits_in_base;
928 else {
929 uint32_t pos;
930 if (find_commit_in_graph(parent->item,
931 ctx->new_base_graph,
932 &pos))
933 edge_value = pos;
936 if (edge_value < 0)
937 BUG("missing parent %s for commit %s",
938 oid_to_hex(&parent->item->object.oid),
939 oid_to_hex(&(*list)->object.oid));
942 hashwrite_be32(f, edge_value);
944 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
945 do {
946 num_extra_edges++;
947 parent = parent->next;
948 } while (parent);
951 if (sizeof((*list)->date) > 4)
952 packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
953 else
954 packedDate[0] = 0;
956 packedDate[0] |= htonl((*list)->generation << 2);
958 packedDate[1] = htonl((*list)->date);
959 hashwrite(f, packedDate, 8);
961 list++;
965 static void write_graph_chunk_extra_edges(struct hashfile *f,
966 struct write_commit_graph_context *ctx)
968 struct commit **list = ctx->commits.list;
969 struct commit **last = ctx->commits.list + ctx->commits.nr;
970 struct commit_list *parent;
972 while (list < last) {
973 int num_parents = 0;
975 display_progress(ctx->progress, ++ctx->progress_cnt);
977 for (parent = (*list)->parents; num_parents < 3 && parent;
978 parent = parent->next)
979 num_parents++;
981 if (num_parents <= 2) {
982 list++;
983 continue;
986 /* Since num_parents > 2, this initializer is safe. */
987 for (parent = (*list)->parents->next; parent; parent = parent->next) {
988 int edge_value = sha1_pos(parent->item->object.oid.hash,
989 ctx->commits.list,
990 ctx->commits.nr,
991 commit_to_sha1);
993 if (edge_value >= 0)
994 edge_value += ctx->new_num_commits_in_base;
995 else {
996 uint32_t pos;
997 if (find_commit_in_graph(parent->item,
998 ctx->new_base_graph,
999 &pos))
1000 edge_value = pos;
1003 if (edge_value < 0)
1004 BUG("missing parent %s for commit %s",
1005 oid_to_hex(&parent->item->object.oid),
1006 oid_to_hex(&(*list)->object.oid));
1007 else if (!parent->next)
1008 edge_value |= GRAPH_LAST_EDGE;
1010 hashwrite_be32(f, edge_value);
1013 list++;
1017 static int oid_compare(const void *_a, const void *_b)
1019 const struct object_id *a = (const struct object_id *)_a;
1020 const struct object_id *b = (const struct object_id *)_b;
1021 return oidcmp(a, b);
1024 static int add_packed_commits(const struct object_id *oid,
1025 struct packed_git *pack,
1026 uint32_t pos,
1027 void *data)
1029 struct write_commit_graph_context *ctx = (struct write_commit_graph_context*)data;
1030 enum object_type type;
1031 off_t offset = nth_packed_object_offset(pack, pos);
1032 struct object_info oi = OBJECT_INFO_INIT;
1034 if (ctx->progress)
1035 display_progress(ctx->progress, ++ctx->progress_done);
1037 oi.typep = &type;
1038 if (packed_object_info(ctx->r, pack, offset, &oi) < 0)
1039 die(_("unable to get type of object %s"), oid_to_hex(oid));
1041 if (type != OBJ_COMMIT)
1042 return 0;
1044 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1045 oidcpy(&(ctx->oids.list[ctx->oids.nr]), oid);
1046 ctx->oids.nr++;
1048 set_commit_pos(ctx->r, oid);
1050 return 0;
1053 static void add_missing_parents(struct write_commit_graph_context *ctx, struct commit *commit)
1055 struct commit_list *parent;
1056 for (parent = commit->parents; parent; parent = parent->next) {
1057 if (!(parent->item->object.flags & REACHABLE)) {
1058 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1059 oidcpy(&ctx->oids.list[ctx->oids.nr], &(parent->item->object.oid));
1060 ctx->oids.nr++;
1061 parent->item->object.flags |= REACHABLE;
1066 static void close_reachable(struct write_commit_graph_context *ctx)
1068 int i;
1069 struct commit *commit;
1071 if (ctx->report_progress)
1072 ctx->progress = start_delayed_progress(
1073 _("Loading known commits in commit graph"),
1074 ctx->oids.nr);
1075 for (i = 0; i < ctx->oids.nr; i++) {
1076 display_progress(ctx->progress, i + 1);
1077 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1078 if (commit)
1079 commit->object.flags |= REACHABLE;
1081 stop_progress(&ctx->progress);
1084 * As this loop runs, ctx->oids.nr may grow, but not more
1085 * than the number of missing commits in the reachable
1086 * closure.
1088 if (ctx->report_progress)
1089 ctx->progress = start_delayed_progress(
1090 _("Expanding reachable commits in commit graph"),
1092 for (i = 0; i < ctx->oids.nr; i++) {
1093 display_progress(ctx->progress, i + 1);
1094 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1096 if (!commit)
1097 continue;
1098 if (ctx->split) {
1099 if (!parse_commit(commit) &&
1100 commit->graph_pos == COMMIT_NOT_FROM_GRAPH)
1101 add_missing_parents(ctx, commit);
1102 } else if (!parse_commit_no_graph(commit))
1103 add_missing_parents(ctx, commit);
1105 stop_progress(&ctx->progress);
1107 if (ctx->report_progress)
1108 ctx->progress = start_delayed_progress(
1109 _("Clearing commit marks in commit graph"),
1110 ctx->oids.nr);
1111 for (i = 0; i < ctx->oids.nr; i++) {
1112 display_progress(ctx->progress, i + 1);
1113 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1115 if (commit)
1116 commit->object.flags &= ~REACHABLE;
1118 stop_progress(&ctx->progress);
1121 static void compute_generation_numbers(struct write_commit_graph_context *ctx)
1123 int i;
1124 struct commit_list *list = NULL;
1126 if (ctx->report_progress)
1127 ctx->progress = start_delayed_progress(
1128 _("Computing commit graph generation numbers"),
1129 ctx->commits.nr);
1130 for (i = 0; i < ctx->commits.nr; i++) {
1131 display_progress(ctx->progress, i + 1);
1132 if (ctx->commits.list[i]->generation != GENERATION_NUMBER_INFINITY &&
1133 ctx->commits.list[i]->generation != GENERATION_NUMBER_ZERO)
1134 continue;
1136 commit_list_insert(ctx->commits.list[i], &list);
1137 while (list) {
1138 struct commit *current = list->item;
1139 struct commit_list *parent;
1140 int all_parents_computed = 1;
1141 uint32_t max_generation = 0;
1143 for (parent = current->parents; parent; parent = parent->next) {
1144 if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
1145 parent->item->generation == GENERATION_NUMBER_ZERO) {
1146 all_parents_computed = 0;
1147 commit_list_insert(parent->item, &list);
1148 break;
1149 } else if (parent->item->generation > max_generation) {
1150 max_generation = parent->item->generation;
1154 if (all_parents_computed) {
1155 current->generation = max_generation + 1;
1156 pop_commit(&list);
1158 if (current->generation > GENERATION_NUMBER_MAX)
1159 current->generation = GENERATION_NUMBER_MAX;
1163 stop_progress(&ctx->progress);
1166 static void compute_bloom_filters(struct write_commit_graph_context *ctx)
1168 int i;
1169 struct progress *progress = NULL;
1170 struct commit **sorted_commits;
1172 init_bloom_filters();
1174 if (ctx->report_progress)
1175 progress = start_delayed_progress(
1176 _("Computing commit changed paths Bloom filters"),
1177 ctx->commits.nr);
1179 ALLOC_ARRAY(sorted_commits, ctx->commits.nr);
1180 COPY_ARRAY(sorted_commits, ctx->commits.list, ctx->commits.nr);
1181 QSORT(sorted_commits, ctx->commits.nr, commit_pos_cmp);
1183 for (i = 0; i < ctx->commits.nr; i++) {
1184 struct commit *c = sorted_commits[i];
1185 struct bloom_filter *filter = get_bloom_filter(ctx->r, c);
1186 ctx->total_bloom_filter_data_size += sizeof(unsigned char) * filter->len;
1187 display_progress(progress, i + 1);
1190 free(sorted_commits);
1191 stop_progress(&progress);
1194 static int add_ref_to_list(const char *refname,
1195 const struct object_id *oid,
1196 int flags, void *cb_data)
1198 struct string_list *list = (struct string_list *)cb_data;
1200 string_list_append(list, oid_to_hex(oid));
1201 return 0;
1204 int write_commit_graph_reachable(struct object_directory *odb,
1205 enum commit_graph_write_flags flags,
1206 const struct split_commit_graph_opts *split_opts)
1208 struct string_list list = STRING_LIST_INIT_DUP;
1209 int result;
1211 for_each_ref(add_ref_to_list, &list);
1212 result = write_commit_graph(odb, NULL, &list,
1213 flags, split_opts);
1215 string_list_clear(&list, 0);
1216 return result;
1219 static int fill_oids_from_packs(struct write_commit_graph_context *ctx,
1220 struct string_list *pack_indexes)
1222 uint32_t i;
1223 struct strbuf progress_title = STRBUF_INIT;
1224 struct strbuf packname = STRBUF_INIT;
1225 int dirlen;
1227 strbuf_addf(&packname, "%s/pack/", ctx->odb->path);
1228 dirlen = packname.len;
1229 if (ctx->report_progress) {
1230 strbuf_addf(&progress_title,
1231 Q_("Finding commits for commit graph in %d pack",
1232 "Finding commits for commit graph in %d packs",
1233 pack_indexes->nr),
1234 pack_indexes->nr);
1235 ctx->progress = start_delayed_progress(progress_title.buf, 0);
1236 ctx->progress_done = 0;
1238 for (i = 0; i < pack_indexes->nr; i++) {
1239 struct packed_git *p;
1240 strbuf_setlen(&packname, dirlen);
1241 strbuf_addstr(&packname, pack_indexes->items[i].string);
1242 p = add_packed_git(packname.buf, packname.len, 1);
1243 if (!p) {
1244 error(_("error adding pack %s"), packname.buf);
1245 return -1;
1247 if (open_pack_index(p)) {
1248 error(_("error opening index for %s"), packname.buf);
1249 return -1;
1251 for_each_object_in_pack(p, add_packed_commits, ctx,
1252 FOR_EACH_OBJECT_PACK_ORDER);
1253 close_pack(p);
1254 free(p);
1257 stop_progress(&ctx->progress);
1258 strbuf_release(&progress_title);
1259 strbuf_release(&packname);
1261 return 0;
1264 static int fill_oids_from_commit_hex(struct write_commit_graph_context *ctx,
1265 struct string_list *commit_hex)
1267 uint32_t i;
1268 struct strbuf progress_title = STRBUF_INIT;
1270 if (ctx->report_progress) {
1271 strbuf_addf(&progress_title,
1272 Q_("Finding commits for commit graph from %d ref",
1273 "Finding commits for commit graph from %d refs",
1274 commit_hex->nr),
1275 commit_hex->nr);
1276 ctx->progress = start_delayed_progress(
1277 progress_title.buf,
1278 commit_hex->nr);
1280 for (i = 0; i < commit_hex->nr; i++) {
1281 const char *end;
1282 struct object_id oid;
1283 struct commit *result;
1285 display_progress(ctx->progress, i + 1);
1286 if (!parse_oid_hex(commit_hex->items[i].string, &oid, &end) &&
1287 (result = lookup_commit_reference_gently(ctx->r, &oid, 1))) {
1288 ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1289 oidcpy(&ctx->oids.list[ctx->oids.nr], &(result->object.oid));
1290 ctx->oids.nr++;
1291 } else if (ctx->check_oids) {
1292 error(_("invalid commit object id: %s"),
1293 commit_hex->items[i].string);
1294 return -1;
1297 stop_progress(&ctx->progress);
1298 strbuf_release(&progress_title);
1300 return 0;
1303 static void fill_oids_from_all_packs(struct write_commit_graph_context *ctx)
1305 if (ctx->report_progress)
1306 ctx->progress = start_delayed_progress(
1307 _("Finding commits for commit graph among packed objects"),
1308 ctx->approx_nr_objects);
1309 for_each_packed_object(add_packed_commits, ctx,
1310 FOR_EACH_OBJECT_PACK_ORDER);
1311 if (ctx->progress_done < ctx->approx_nr_objects)
1312 display_progress(ctx->progress, ctx->approx_nr_objects);
1313 stop_progress(&ctx->progress);
1316 static uint32_t count_distinct_commits(struct write_commit_graph_context *ctx)
1318 uint32_t i, count_distinct = 1;
1320 if (ctx->report_progress)
1321 ctx->progress = start_delayed_progress(
1322 _("Counting distinct commits in commit graph"),
1323 ctx->oids.nr);
1324 display_progress(ctx->progress, 0); /* TODO: Measure QSORT() progress */
1325 QSORT(ctx->oids.list, ctx->oids.nr, oid_compare);
1327 for (i = 1; i < ctx->oids.nr; i++) {
1328 display_progress(ctx->progress, i + 1);
1329 if (!oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i])) {
1330 if (ctx->split) {
1331 struct commit *c = lookup_commit(ctx->r, &ctx->oids.list[i]);
1333 if (!c || c->graph_pos != COMMIT_NOT_FROM_GRAPH)
1334 continue;
1337 count_distinct++;
1340 stop_progress(&ctx->progress);
1342 return count_distinct;
1345 static void copy_oids_to_commits(struct write_commit_graph_context *ctx)
1347 uint32_t i;
1349 ctx->num_extra_edges = 0;
1350 if (ctx->report_progress)
1351 ctx->progress = start_delayed_progress(
1352 _("Finding extra edges in commit graph"),
1353 ctx->oids.nr);
1354 for (i = 0; i < ctx->oids.nr; i++) {
1355 unsigned int num_parents;
1357 display_progress(ctx->progress, i + 1);
1358 if (i > 0 && oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i]))
1359 continue;
1361 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + 1, ctx->commits.alloc);
1362 ctx->commits.list[ctx->commits.nr] = lookup_commit(ctx->r, &ctx->oids.list[i]);
1364 if (ctx->split &&
1365 ctx->commits.list[ctx->commits.nr]->graph_pos != COMMIT_NOT_FROM_GRAPH)
1366 continue;
1368 parse_commit_no_graph(ctx->commits.list[ctx->commits.nr]);
1370 num_parents = commit_list_count(ctx->commits.list[ctx->commits.nr]->parents);
1371 if (num_parents > 2)
1372 ctx->num_extra_edges += num_parents - 1;
1374 ctx->commits.nr++;
1376 stop_progress(&ctx->progress);
1379 static int write_graph_chunk_base_1(struct hashfile *f,
1380 struct commit_graph *g)
1382 int num = 0;
1384 if (!g)
1385 return 0;
1387 num = write_graph_chunk_base_1(f, g->base_graph);
1388 hashwrite(f, g->oid.hash, the_hash_algo->rawsz);
1389 return num + 1;
1392 static int write_graph_chunk_base(struct hashfile *f,
1393 struct write_commit_graph_context *ctx)
1395 int num = write_graph_chunk_base_1(f, ctx->new_base_graph);
1397 if (num != ctx->num_commit_graphs_after - 1) {
1398 error(_("failed to write correct number of base graph ids"));
1399 return -1;
1402 return 0;
1405 static int write_commit_graph_file(struct write_commit_graph_context *ctx)
1407 uint32_t i;
1408 int fd;
1409 struct hashfile *f;
1410 struct lock_file lk = LOCK_INIT;
1411 uint32_t chunk_ids[MAX_NUM_CHUNKS + 1];
1412 uint64_t chunk_offsets[MAX_NUM_CHUNKS + 1];
1413 const unsigned hashsz = the_hash_algo->rawsz;
1414 struct strbuf progress_title = STRBUF_INIT;
1415 int num_chunks = 3;
1416 struct object_id file_hash;
1418 if (ctx->split) {
1419 struct strbuf tmp_file = STRBUF_INIT;
1421 strbuf_addf(&tmp_file,
1422 "%s/info/commit-graphs/tmp_graph_XXXXXX",
1423 ctx->odb->path);
1424 ctx->graph_name = strbuf_detach(&tmp_file, NULL);
1425 } else {
1426 ctx->graph_name = get_commit_graph_filename(ctx->odb);
1429 if (safe_create_leading_directories(ctx->graph_name)) {
1430 UNLEAK(ctx->graph_name);
1431 error(_("unable to create leading directories of %s"),
1432 ctx->graph_name);
1433 return -1;
1436 if (ctx->split) {
1437 char *lock_name = get_chain_filename(ctx->odb);
1439 hold_lock_file_for_update(&lk, lock_name, LOCK_DIE_ON_ERROR);
1441 fd = git_mkstemp_mode(ctx->graph_name, 0444);
1442 if (fd < 0) {
1443 error(_("unable to create '%s'"), ctx->graph_name);
1444 return -1;
1447 f = hashfd(fd, ctx->graph_name);
1448 } else {
1449 hold_lock_file_for_update(&lk, ctx->graph_name, LOCK_DIE_ON_ERROR);
1450 fd = lk.tempfile->fd;
1451 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
1454 chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
1455 chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
1456 chunk_ids[2] = GRAPH_CHUNKID_DATA;
1457 if (ctx->num_extra_edges) {
1458 chunk_ids[num_chunks] = GRAPH_CHUNKID_EXTRAEDGES;
1459 num_chunks++;
1461 if (ctx->num_commit_graphs_after > 1) {
1462 chunk_ids[num_chunks] = GRAPH_CHUNKID_BASE;
1463 num_chunks++;
1466 chunk_ids[num_chunks] = 0;
1468 chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
1469 chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
1470 chunk_offsets[2] = chunk_offsets[1] + hashsz * ctx->commits.nr;
1471 chunk_offsets[3] = chunk_offsets[2] + (hashsz + 16) * ctx->commits.nr;
1473 num_chunks = 3;
1474 if (ctx->num_extra_edges) {
1475 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1476 4 * ctx->num_extra_edges;
1477 num_chunks++;
1479 if (ctx->num_commit_graphs_after > 1) {
1480 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1481 hashsz * (ctx->num_commit_graphs_after - 1);
1482 num_chunks++;
1485 hashwrite_be32(f, GRAPH_SIGNATURE);
1487 hashwrite_u8(f, GRAPH_VERSION);
1488 hashwrite_u8(f, oid_version());
1489 hashwrite_u8(f, num_chunks);
1490 hashwrite_u8(f, ctx->num_commit_graphs_after - 1);
1492 for (i = 0; i <= num_chunks; i++) {
1493 uint32_t chunk_write[3];
1495 chunk_write[0] = htonl(chunk_ids[i]);
1496 chunk_write[1] = htonl(chunk_offsets[i] >> 32);
1497 chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
1498 hashwrite(f, chunk_write, 12);
1501 if (ctx->report_progress) {
1502 strbuf_addf(&progress_title,
1503 Q_("Writing out commit graph in %d pass",
1504 "Writing out commit graph in %d passes",
1505 num_chunks),
1506 num_chunks);
1507 ctx->progress = start_delayed_progress(
1508 progress_title.buf,
1509 num_chunks * ctx->commits.nr);
1511 write_graph_chunk_fanout(f, ctx);
1512 write_graph_chunk_oids(f, hashsz, ctx);
1513 write_graph_chunk_data(f, hashsz, ctx);
1514 if (ctx->num_extra_edges)
1515 write_graph_chunk_extra_edges(f, ctx);
1516 if (ctx->num_commit_graphs_after > 1 &&
1517 write_graph_chunk_base(f, ctx)) {
1518 return -1;
1520 stop_progress(&ctx->progress);
1521 strbuf_release(&progress_title);
1523 if (ctx->split && ctx->base_graph_name && ctx->num_commit_graphs_after > 1) {
1524 char *new_base_hash = xstrdup(oid_to_hex(&ctx->new_base_graph->oid));
1525 char *new_base_name = get_split_graph_filename(ctx->new_base_graph->odb, new_base_hash);
1527 free(ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2]);
1528 free(ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2]);
1529 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2] = new_base_name;
1530 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2] = new_base_hash;
1533 close_commit_graph(ctx->r->objects);
1534 finalize_hashfile(f, file_hash.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
1536 if (ctx->split) {
1537 FILE *chainf = fdopen_lock_file(&lk, "w");
1538 char *final_graph_name;
1539 int result;
1541 close(fd);
1543 if (!chainf) {
1544 error(_("unable to open commit-graph chain file"));
1545 return -1;
1548 if (ctx->base_graph_name) {
1549 const char *dest = ctx->commit_graph_filenames_after[
1550 ctx->num_commit_graphs_after - 2];
1552 if (strcmp(ctx->base_graph_name, dest)) {
1553 result = rename(ctx->base_graph_name, dest);
1555 if (result) {
1556 error(_("failed to rename base commit-graph file"));
1557 return -1;
1560 } else {
1561 char *graph_name = get_commit_graph_filename(ctx->odb);
1562 unlink(graph_name);
1565 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1] = xstrdup(oid_to_hex(&file_hash));
1566 final_graph_name = get_split_graph_filename(ctx->odb,
1567 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1]);
1568 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 1] = final_graph_name;
1570 result = rename(ctx->graph_name, final_graph_name);
1572 for (i = 0; i < ctx->num_commit_graphs_after; i++)
1573 fprintf(lk.tempfile->fp, "%s\n", ctx->commit_graph_hash_after[i]);
1575 if (result) {
1576 error(_("failed to rename temporary commit-graph file"));
1577 return -1;
1581 commit_lock_file(&lk);
1583 return 0;
1586 static void split_graph_merge_strategy(struct write_commit_graph_context *ctx)
1588 struct commit_graph *g;
1589 uint32_t num_commits;
1590 uint32_t i;
1592 int max_commits = 0;
1593 int size_mult = 2;
1595 if (ctx->split_opts) {
1596 max_commits = ctx->split_opts->max_commits;
1598 if (ctx->split_opts->size_multiple)
1599 size_mult = ctx->split_opts->size_multiple;
1602 g = ctx->r->objects->commit_graph;
1603 num_commits = ctx->commits.nr;
1604 ctx->num_commit_graphs_after = ctx->num_commit_graphs_before + 1;
1606 while (g && (g->num_commits <= size_mult * num_commits ||
1607 (max_commits && num_commits > max_commits))) {
1608 if (g->odb != ctx->odb)
1609 break;
1611 num_commits += g->num_commits;
1612 g = g->base_graph;
1614 ctx->num_commit_graphs_after--;
1617 ctx->new_base_graph = g;
1619 if (ctx->num_commit_graphs_after == 2) {
1620 char *old_graph_name = get_commit_graph_filename(g->odb);
1622 if (!strcmp(g->filename, old_graph_name) &&
1623 g->odb != ctx->odb) {
1624 ctx->num_commit_graphs_after = 1;
1625 ctx->new_base_graph = NULL;
1628 free(old_graph_name);
1631 ALLOC_ARRAY(ctx->commit_graph_filenames_after, ctx->num_commit_graphs_after);
1632 ALLOC_ARRAY(ctx->commit_graph_hash_after, ctx->num_commit_graphs_after);
1634 for (i = 0; i < ctx->num_commit_graphs_after &&
1635 i < ctx->num_commit_graphs_before; i++)
1636 ctx->commit_graph_filenames_after[i] = xstrdup(ctx->commit_graph_filenames_before[i]);
1638 i = ctx->num_commit_graphs_before - 1;
1639 g = ctx->r->objects->commit_graph;
1641 while (g) {
1642 if (i < ctx->num_commit_graphs_after)
1643 ctx->commit_graph_hash_after[i] = xstrdup(oid_to_hex(&g->oid));
1645 i--;
1646 g = g->base_graph;
1650 static void merge_commit_graph(struct write_commit_graph_context *ctx,
1651 struct commit_graph *g)
1653 uint32_t i;
1654 uint32_t offset = g->num_commits_in_base;
1656 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + g->num_commits, ctx->commits.alloc);
1658 for (i = 0; i < g->num_commits; i++) {
1659 struct object_id oid;
1660 struct commit *result;
1662 display_progress(ctx->progress, i + 1);
1664 load_oid_from_graph(g, i + offset, &oid);
1666 /* only add commits if they still exist in the repo */
1667 result = lookup_commit_reference_gently(ctx->r, &oid, 1);
1669 if (result) {
1670 ctx->commits.list[ctx->commits.nr] = result;
1671 ctx->commits.nr++;
1676 static int commit_compare(const void *_a, const void *_b)
1678 const struct commit *a = *(const struct commit **)_a;
1679 const struct commit *b = *(const struct commit **)_b;
1680 return oidcmp(&a->object.oid, &b->object.oid);
1683 static void sort_and_scan_merged_commits(struct write_commit_graph_context *ctx)
1685 uint32_t i;
1687 if (ctx->report_progress)
1688 ctx->progress = start_delayed_progress(
1689 _("Scanning merged commits"),
1690 ctx->commits.nr);
1692 QSORT(ctx->commits.list, ctx->commits.nr, commit_compare);
1694 ctx->num_extra_edges = 0;
1695 for (i = 0; i < ctx->commits.nr; i++) {
1696 display_progress(ctx->progress, i);
1698 if (i && oideq(&ctx->commits.list[i - 1]->object.oid,
1699 &ctx->commits.list[i]->object.oid)) {
1700 die(_("unexpected duplicate commit id %s"),
1701 oid_to_hex(&ctx->commits.list[i]->object.oid));
1702 } else {
1703 unsigned int num_parents;
1705 num_parents = commit_list_count(ctx->commits.list[i]->parents);
1706 if (num_parents > 2)
1707 ctx->num_extra_edges += num_parents - 1;
1711 stop_progress(&ctx->progress);
1714 static void merge_commit_graphs(struct write_commit_graph_context *ctx)
1716 struct commit_graph *g = ctx->r->objects->commit_graph;
1717 uint32_t current_graph_number = ctx->num_commit_graphs_before;
1719 while (g && current_graph_number >= ctx->num_commit_graphs_after) {
1720 current_graph_number--;
1722 if (ctx->report_progress)
1723 ctx->progress = start_delayed_progress(_("Merging commit-graph"), 0);
1725 merge_commit_graph(ctx, g);
1726 stop_progress(&ctx->progress);
1728 g = g->base_graph;
1731 if (g) {
1732 ctx->new_base_graph = g;
1733 ctx->new_num_commits_in_base = g->num_commits + g->num_commits_in_base;
1736 if (ctx->new_base_graph)
1737 ctx->base_graph_name = xstrdup(ctx->new_base_graph->filename);
1739 sort_and_scan_merged_commits(ctx);
1742 static void mark_commit_graphs(struct write_commit_graph_context *ctx)
1744 uint32_t i;
1745 time_t now = time(NULL);
1747 for (i = ctx->num_commit_graphs_after - 1; i < ctx->num_commit_graphs_before; i++) {
1748 struct stat st;
1749 struct utimbuf updated_time;
1751 stat(ctx->commit_graph_filenames_before[i], &st);
1753 updated_time.actime = st.st_atime;
1754 updated_time.modtime = now;
1755 utime(ctx->commit_graph_filenames_before[i], &updated_time);
1759 static void expire_commit_graphs(struct write_commit_graph_context *ctx)
1761 struct strbuf path = STRBUF_INIT;
1762 DIR *dir;
1763 struct dirent *de;
1764 size_t dirnamelen;
1765 timestamp_t expire_time = time(NULL);
1767 if (ctx->split_opts && ctx->split_opts->expire_time)
1768 expire_time -= ctx->split_opts->expire_time;
1769 if (!ctx->split) {
1770 char *chain_file_name = get_chain_filename(ctx->odb);
1771 unlink(chain_file_name);
1772 free(chain_file_name);
1773 ctx->num_commit_graphs_after = 0;
1776 strbuf_addstr(&path, ctx->odb->path);
1777 strbuf_addstr(&path, "/info/commit-graphs");
1778 dir = opendir(path.buf);
1780 if (!dir)
1781 goto out;
1783 strbuf_addch(&path, '/');
1784 dirnamelen = path.len;
1785 while ((de = readdir(dir)) != NULL) {
1786 struct stat st;
1787 uint32_t i, found = 0;
1789 strbuf_setlen(&path, dirnamelen);
1790 strbuf_addstr(&path, de->d_name);
1792 stat(path.buf, &st);
1794 if (st.st_mtime > expire_time)
1795 continue;
1796 if (path.len < 6 || strcmp(path.buf + path.len - 6, ".graph"))
1797 continue;
1799 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
1800 if (!strcmp(ctx->commit_graph_filenames_after[i],
1801 path.buf)) {
1802 found = 1;
1803 break;
1807 if (!found)
1808 unlink(path.buf);
1811 out:
1812 strbuf_release(&path);
1815 int write_commit_graph(struct object_directory *odb,
1816 struct string_list *pack_indexes,
1817 struct string_list *commit_hex,
1818 enum commit_graph_write_flags flags,
1819 const struct split_commit_graph_opts *split_opts)
1821 struct write_commit_graph_context *ctx;
1822 uint32_t i, count_distinct = 0;
1823 int res = 0;
1825 if (!commit_graph_compatible(the_repository))
1826 return 0;
1828 ctx = xcalloc(1, sizeof(struct write_commit_graph_context));
1829 ctx->r = the_repository;
1830 ctx->odb = odb;
1831 ctx->append = flags & COMMIT_GRAPH_WRITE_APPEND ? 1 : 0;
1832 ctx->report_progress = flags & COMMIT_GRAPH_WRITE_PROGRESS ? 1 : 0;
1833 ctx->split = flags & COMMIT_GRAPH_WRITE_SPLIT ? 1 : 0;
1834 ctx->check_oids = flags & COMMIT_GRAPH_WRITE_CHECK_OIDS ? 1 : 0;
1835 ctx->split_opts = split_opts;
1836 ctx->changed_paths = flags & COMMIT_GRAPH_WRITE_BLOOM_FILTERS ? 1 : 0;
1837 ctx->total_bloom_filter_data_size = 0;
1839 if (ctx->split) {
1840 struct commit_graph *g;
1841 prepare_commit_graph(ctx->r);
1843 g = ctx->r->objects->commit_graph;
1845 while (g) {
1846 ctx->num_commit_graphs_before++;
1847 g = g->base_graph;
1850 if (ctx->num_commit_graphs_before) {
1851 ALLOC_ARRAY(ctx->commit_graph_filenames_before, ctx->num_commit_graphs_before);
1852 i = ctx->num_commit_graphs_before;
1853 g = ctx->r->objects->commit_graph;
1855 while (g) {
1856 ctx->commit_graph_filenames_before[--i] = xstrdup(g->filename);
1857 g = g->base_graph;
1862 ctx->approx_nr_objects = approximate_object_count();
1863 ctx->oids.alloc = ctx->approx_nr_objects / 32;
1865 if (ctx->split && split_opts && ctx->oids.alloc > split_opts->max_commits)
1866 ctx->oids.alloc = split_opts->max_commits;
1868 if (ctx->append) {
1869 prepare_commit_graph_one(ctx->r, ctx->odb);
1870 if (ctx->r->objects->commit_graph)
1871 ctx->oids.alloc += ctx->r->objects->commit_graph->num_commits;
1874 if (ctx->oids.alloc < 1024)
1875 ctx->oids.alloc = 1024;
1876 ALLOC_ARRAY(ctx->oids.list, ctx->oids.alloc);
1878 if (ctx->append && ctx->r->objects->commit_graph) {
1879 struct commit_graph *g = ctx->r->objects->commit_graph;
1880 for (i = 0; i < g->num_commits; i++) {
1881 const unsigned char *hash = g->chunk_oid_lookup + g->hash_len * i;
1882 hashcpy(ctx->oids.list[ctx->oids.nr++].hash, hash);
1886 if (pack_indexes) {
1887 if ((res = fill_oids_from_packs(ctx, pack_indexes)))
1888 goto cleanup;
1891 if (commit_hex) {
1892 if ((res = fill_oids_from_commit_hex(ctx, commit_hex)))
1893 goto cleanup;
1896 if (!pack_indexes && !commit_hex)
1897 fill_oids_from_all_packs(ctx);
1899 close_reachable(ctx);
1901 count_distinct = count_distinct_commits(ctx);
1903 if (count_distinct >= GRAPH_EDGE_LAST_MASK) {
1904 error(_("the commit graph format cannot write %d commits"), count_distinct);
1905 res = -1;
1906 goto cleanup;
1909 ctx->commits.alloc = count_distinct;
1910 ALLOC_ARRAY(ctx->commits.list, ctx->commits.alloc);
1912 copy_oids_to_commits(ctx);
1914 if (ctx->commits.nr >= GRAPH_EDGE_LAST_MASK) {
1915 error(_("too many commits to write graph"));
1916 res = -1;
1917 goto cleanup;
1920 if (!ctx->commits.nr)
1921 goto cleanup;
1923 if (ctx->split) {
1924 split_graph_merge_strategy(ctx);
1926 merge_commit_graphs(ctx);
1927 } else
1928 ctx->num_commit_graphs_after = 1;
1930 compute_generation_numbers(ctx);
1932 if (ctx->changed_paths)
1933 compute_bloom_filters(ctx);
1935 res = write_commit_graph_file(ctx);
1937 if (ctx->split)
1938 mark_commit_graphs(ctx);
1940 expire_commit_graphs(ctx);
1942 cleanup:
1943 free(ctx->graph_name);
1944 free(ctx->commits.list);
1945 free(ctx->oids.list);
1947 if (ctx->commit_graph_filenames_after) {
1948 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
1949 free(ctx->commit_graph_filenames_after[i]);
1950 free(ctx->commit_graph_hash_after[i]);
1953 for (i = 0; i < ctx->num_commit_graphs_before; i++)
1954 free(ctx->commit_graph_filenames_before[i]);
1956 free(ctx->commit_graph_filenames_after);
1957 free(ctx->commit_graph_filenames_before);
1958 free(ctx->commit_graph_hash_after);
1961 free(ctx);
1963 return res;
1966 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
1967 static int verify_commit_graph_error;
1969 static void graph_report(const char *fmt, ...)
1971 va_list ap;
1973 verify_commit_graph_error = 1;
1974 va_start(ap, fmt);
1975 vfprintf(stderr, fmt, ap);
1976 fprintf(stderr, "\n");
1977 va_end(ap);
1980 #define GENERATION_ZERO_EXISTS 1
1981 #define GENERATION_NUMBER_EXISTS 2
1983 int verify_commit_graph(struct repository *r, struct commit_graph *g, int flags)
1985 uint32_t i, cur_fanout_pos = 0;
1986 struct object_id prev_oid, cur_oid, checksum;
1987 int generation_zero = 0;
1988 struct hashfile *f;
1989 int devnull;
1990 struct progress *progress = NULL;
1991 int local_error = 0;
1993 if (!g) {
1994 graph_report("no commit-graph file loaded");
1995 return 1;
1998 verify_commit_graph_error = verify_commit_graph_lite(g);
1999 if (verify_commit_graph_error)
2000 return verify_commit_graph_error;
2002 devnull = open("/dev/null", O_WRONLY);
2003 f = hashfd(devnull, NULL);
2004 hashwrite(f, g->data, g->data_len - g->hash_len);
2005 finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
2006 if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
2007 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
2008 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
2011 for (i = 0; i < g->num_commits; i++) {
2012 struct commit *graph_commit;
2014 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
2016 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
2017 graph_report(_("commit-graph has incorrect OID order: %s then %s"),
2018 oid_to_hex(&prev_oid),
2019 oid_to_hex(&cur_oid));
2021 oidcpy(&prev_oid, &cur_oid);
2023 while (cur_oid.hash[0] > cur_fanout_pos) {
2024 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2026 if (i != fanout_value)
2027 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2028 cur_fanout_pos, fanout_value, i);
2029 cur_fanout_pos++;
2032 graph_commit = lookup_commit(r, &cur_oid);
2033 if (!parse_commit_in_graph_one(r, g, graph_commit))
2034 graph_report(_("failed to parse commit %s from commit-graph"),
2035 oid_to_hex(&cur_oid));
2038 while (cur_fanout_pos < 256) {
2039 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2041 if (g->num_commits != fanout_value)
2042 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2043 cur_fanout_pos, fanout_value, i);
2045 cur_fanout_pos++;
2048 if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
2049 return verify_commit_graph_error;
2051 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
2052 progress = start_progress(_("Verifying commits in commit graph"),
2053 g->num_commits);
2055 for (i = 0; i < g->num_commits; i++) {
2056 struct commit *graph_commit, *odb_commit;
2057 struct commit_list *graph_parents, *odb_parents;
2058 uint32_t max_generation = 0;
2060 display_progress(progress, i + 1);
2061 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
2063 graph_commit = lookup_commit(r, &cur_oid);
2064 odb_commit = (struct commit *)create_object(r, &cur_oid, alloc_commit_node(r));
2065 if (parse_commit_internal(odb_commit, 0, 0)) {
2066 graph_report(_("failed to parse commit %s from object database for commit-graph"),
2067 oid_to_hex(&cur_oid));
2068 continue;
2071 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
2072 get_commit_tree_oid(odb_commit)))
2073 graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
2074 oid_to_hex(&cur_oid),
2075 oid_to_hex(get_commit_tree_oid(graph_commit)),
2076 oid_to_hex(get_commit_tree_oid(odb_commit)));
2078 graph_parents = graph_commit->parents;
2079 odb_parents = odb_commit->parents;
2081 while (graph_parents) {
2082 if (odb_parents == NULL) {
2083 graph_report(_("commit-graph parent list for commit %s is too long"),
2084 oid_to_hex(&cur_oid));
2085 break;
2088 /* parse parent in case it is in a base graph */
2089 parse_commit_in_graph_one(r, g, graph_parents->item);
2091 if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
2092 graph_report(_("commit-graph parent for %s is %s != %s"),
2093 oid_to_hex(&cur_oid),
2094 oid_to_hex(&graph_parents->item->object.oid),
2095 oid_to_hex(&odb_parents->item->object.oid));
2097 if (graph_parents->item->generation > max_generation)
2098 max_generation = graph_parents->item->generation;
2100 graph_parents = graph_parents->next;
2101 odb_parents = odb_parents->next;
2104 if (odb_parents != NULL)
2105 graph_report(_("commit-graph parent list for commit %s terminates early"),
2106 oid_to_hex(&cur_oid));
2108 if (!graph_commit->generation) {
2109 if (generation_zero == GENERATION_NUMBER_EXISTS)
2110 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
2111 oid_to_hex(&cur_oid));
2112 generation_zero = GENERATION_ZERO_EXISTS;
2113 } else if (generation_zero == GENERATION_ZERO_EXISTS)
2114 graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
2115 oid_to_hex(&cur_oid));
2117 if (generation_zero == GENERATION_ZERO_EXISTS)
2118 continue;
2121 * If one of our parents has generation GENERATION_NUMBER_MAX, then
2122 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
2123 * extra logic in the following condition.
2125 if (max_generation == GENERATION_NUMBER_MAX)
2126 max_generation--;
2128 if (graph_commit->generation != max_generation + 1)
2129 graph_report(_("commit-graph generation for commit %s is %u != %u"),
2130 oid_to_hex(&cur_oid),
2131 graph_commit->generation,
2132 max_generation + 1);
2134 if (graph_commit->date != odb_commit->date)
2135 graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
2136 oid_to_hex(&cur_oid),
2137 graph_commit->date,
2138 odb_commit->date);
2140 stop_progress(&progress);
2142 local_error = verify_commit_graph_error;
2144 if (!(flags & COMMIT_GRAPH_VERIFY_SHALLOW) && g->base_graph)
2145 local_error |= verify_commit_graph(r, g->base_graph, flags);
2147 return local_error;
2150 void free_commit_graph(struct commit_graph *g)
2152 if (!g)
2153 return;
2154 if (g->graph_fd >= 0) {
2155 munmap((void *)g->data, g->data_len);
2156 g->data = NULL;
2157 close(g->graph_fd);
2159 free(g->filename);
2160 free(g);
2163 void disable_commit_graph(struct repository *r)
2165 r->commit_graph_disabled = 1;