cocci: remove 'unused.cocci'
[git.git] / commit-graph.c
blobb1e737c01b1df701af6f0c0b8b814ea91ab9424b
1 #include "cache.h"
2 #include "config.h"
3 #include "gettext.h"
4 #include "hex.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 "hash-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"
21 #include "shallow.h"
22 #include "json-writer.h"
23 #include "trace2.h"
24 #include "chunk-format.h"
25 #include "wrapper.h"
27 void git_test_write_commit_graph_or_die(void)
29 int flags = 0;
30 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0))
31 return;
33 if (git_env_bool(GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS, 0))
34 flags = COMMIT_GRAPH_WRITE_BLOOM_FILTERS;
36 if (write_commit_graph_reachable(the_repository->objects->odb,
37 flags, NULL))
38 die("failed to write commit-graph under GIT_TEST_COMMIT_GRAPH");
41 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
42 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
43 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
44 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
45 #define GRAPH_CHUNKID_GENERATION_DATA 0x47444132 /* "GDA2" */
46 #define GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW 0x47444f32 /* "GDO2" */
47 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
48 #define GRAPH_CHUNKID_BLOOMINDEXES 0x42494458 /* "BIDX" */
49 #define GRAPH_CHUNKID_BLOOMDATA 0x42444154 /* "BDAT" */
50 #define GRAPH_CHUNKID_BASE 0x42415345 /* "BASE" */
52 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
54 #define GRAPH_VERSION_1 0x1
55 #define GRAPH_VERSION GRAPH_VERSION_1
57 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
58 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
59 #define GRAPH_PARENT_NONE 0x70000000
61 #define GRAPH_LAST_EDGE 0x80000000
63 #define GRAPH_HEADER_SIZE 8
64 #define GRAPH_FANOUT_SIZE (4 * 256)
65 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * CHUNK_TOC_ENTRY_SIZE \
66 + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
68 #define CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW (1ULL << 31)
70 /* Remember to update object flag allocation in object.h */
71 #define REACHABLE (1u<<15)
73 define_commit_slab(topo_level_slab, uint32_t);
75 /* Keep track of the order in which commits are added to our list. */
76 define_commit_slab(commit_pos, int);
77 static struct commit_pos commit_pos = COMMIT_SLAB_INIT(1, commit_pos);
79 static void set_commit_pos(struct repository *r, const struct object_id *oid)
81 static int32_t max_pos;
82 struct commit *commit = lookup_commit(r, oid);
84 if (!commit)
85 return; /* should never happen, but be lenient */
87 *commit_pos_at(&commit_pos, commit) = max_pos++;
90 static int commit_pos_cmp(const void *va, const void *vb)
92 const struct commit *a = *(const struct commit **)va;
93 const struct commit *b = *(const struct commit **)vb;
94 return commit_pos_at(&commit_pos, a) -
95 commit_pos_at(&commit_pos, b);
98 define_commit_slab(commit_graph_data_slab, struct commit_graph_data);
99 static struct commit_graph_data_slab commit_graph_data_slab =
100 COMMIT_SLAB_INIT(1, commit_graph_data_slab);
102 static int get_configured_generation_version(struct repository *r)
104 int version = 2;
105 repo_config_get_int(r, "commitgraph.generationversion", &version);
106 return version;
109 uint32_t commit_graph_position(const struct commit *c)
111 struct commit_graph_data *data =
112 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
114 return data ? data->graph_pos : COMMIT_NOT_FROM_GRAPH;
117 timestamp_t commit_graph_generation(const struct commit *c)
119 struct commit_graph_data *data =
120 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
122 if (data && data->generation)
123 return data->generation;
125 return GENERATION_NUMBER_INFINITY;
128 static struct commit_graph_data *commit_graph_data_at(const struct commit *c)
130 unsigned int i, nth_slab;
131 struct commit_graph_data *data =
132 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
134 if (data)
135 return data;
137 nth_slab = c->index / commit_graph_data_slab.slab_size;
138 data = commit_graph_data_slab_at(&commit_graph_data_slab, c);
141 * commit-slab initializes elements with zero, overwrite this with
142 * COMMIT_NOT_FROM_GRAPH for graph_pos.
144 * We avoid initializing generation with checking if graph position
145 * is not COMMIT_NOT_FROM_GRAPH.
147 for (i = 0; i < commit_graph_data_slab.slab_size; i++) {
148 commit_graph_data_slab.slab[nth_slab][i].graph_pos =
149 COMMIT_NOT_FROM_GRAPH;
152 return data;
156 * Should be used only while writing commit-graph as it compares
157 * generation value of commits by directly accessing commit-slab.
159 static int commit_gen_cmp(const void *va, const void *vb)
161 const struct commit *a = *(const struct commit **)va;
162 const struct commit *b = *(const struct commit **)vb;
164 const timestamp_t generation_a = commit_graph_data_at(a)->generation;
165 const timestamp_t generation_b = commit_graph_data_at(b)->generation;
166 /* lower generation commits first */
167 if (generation_a < generation_b)
168 return -1;
169 else if (generation_a > generation_b)
170 return 1;
172 /* use date as a heuristic when generations are equal */
173 if (a->date < b->date)
174 return -1;
175 else if (a->date > b->date)
176 return 1;
177 return 0;
180 char *get_commit_graph_filename(struct object_directory *obj_dir)
182 return xstrfmt("%s/info/commit-graph", obj_dir->path);
185 static char *get_split_graph_filename(struct object_directory *odb,
186 const char *oid_hex)
188 return xstrfmt("%s/info/commit-graphs/graph-%s.graph", odb->path,
189 oid_hex);
192 char *get_commit_graph_chain_filename(struct object_directory *odb)
194 return xstrfmt("%s/info/commit-graphs/commit-graph-chain", odb->path);
197 static struct commit_graph *alloc_commit_graph(void)
199 struct commit_graph *g = xcalloc(1, sizeof(*g));
201 return g;
204 extern int read_replace_refs;
206 static int commit_graph_compatible(struct repository *r)
208 if (!r->gitdir)
209 return 0;
211 if (read_replace_refs) {
212 prepare_replace_object(r);
213 if (hashmap_get_size(&r->objects->replace_map->map))
214 return 0;
217 prepare_commit_graft(r);
218 if (r->parsed_objects &&
219 (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent))
220 return 0;
221 if (is_repository_shallow(r))
222 return 0;
224 return 1;
227 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
229 *fd = git_open(graph_file);
230 if (*fd < 0)
231 return 0;
232 if (fstat(*fd, st)) {
233 close(*fd);
234 return 0;
236 return 1;
239 struct commit_graph *load_commit_graph_one_fd_st(struct repository *r,
240 int fd, struct stat *st,
241 struct object_directory *odb)
243 void *graph_map;
244 size_t graph_size;
245 struct commit_graph *ret;
247 graph_size = xsize_t(st->st_size);
249 if (graph_size < GRAPH_MIN_SIZE) {
250 close(fd);
251 error(_("commit-graph file is too small"));
252 return NULL;
254 graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
255 close(fd);
256 prepare_repo_settings(r);
257 ret = parse_commit_graph(&r->settings, graph_map, graph_size);
259 if (ret)
260 ret->odb = odb;
261 else
262 munmap(graph_map, graph_size);
264 return ret;
267 static int verify_commit_graph_lite(struct commit_graph *g)
270 * Basic validation shared between parse_commit_graph()
271 * which'll be called every time the graph is used, and the
272 * much more expensive verify_commit_graph() used by
273 * "commit-graph verify".
275 * There should only be very basic checks here to ensure that
276 * we don't e.g. segfault in fill_commit_in_graph(), but
277 * because this is a very hot codepath nothing that e.g. loops
278 * over g->num_commits, or runs a checksum on the commit-graph
279 * itself.
281 if (!g->chunk_oid_fanout) {
282 error("commit-graph is missing the OID Fanout chunk");
283 return 1;
285 if (!g->chunk_oid_lookup) {
286 error("commit-graph is missing the OID Lookup chunk");
287 return 1;
289 if (!g->chunk_commit_data) {
290 error("commit-graph is missing the Commit Data chunk");
291 return 1;
294 return 0;
297 static int graph_read_oid_lookup(const unsigned char *chunk_start,
298 size_t chunk_size, void *data)
300 struct commit_graph *g = data;
301 g->chunk_oid_lookup = chunk_start;
302 g->num_commits = chunk_size / g->hash_len;
303 return 0;
306 static int graph_read_bloom_data(const unsigned char *chunk_start,
307 size_t chunk_size, void *data)
309 struct commit_graph *g = data;
310 uint32_t hash_version;
311 g->chunk_bloom_data = chunk_start;
312 hash_version = get_be32(chunk_start);
314 if (hash_version != 1)
315 return 0;
317 g->bloom_filter_settings = xmalloc(sizeof(struct bloom_filter_settings));
318 g->bloom_filter_settings->hash_version = hash_version;
319 g->bloom_filter_settings->num_hashes = get_be32(chunk_start + 4);
320 g->bloom_filter_settings->bits_per_entry = get_be32(chunk_start + 8);
321 g->bloom_filter_settings->max_changed_paths = DEFAULT_BLOOM_MAX_CHANGES;
323 return 0;
326 struct commit_graph *parse_commit_graph(struct repo_settings *s,
327 void *graph_map, size_t graph_size)
329 const unsigned char *data;
330 struct commit_graph *graph;
331 uint32_t graph_signature;
332 unsigned char graph_version, hash_version;
333 struct chunkfile *cf = NULL;
335 if (!graph_map)
336 return NULL;
338 if (graph_size < GRAPH_MIN_SIZE)
339 return NULL;
341 data = (const unsigned char *)graph_map;
343 graph_signature = get_be32(data);
344 if (graph_signature != GRAPH_SIGNATURE) {
345 error(_("commit-graph signature %X does not match signature %X"),
346 graph_signature, GRAPH_SIGNATURE);
347 return NULL;
350 graph_version = *(unsigned char*)(data + 4);
351 if (graph_version != GRAPH_VERSION) {
352 error(_("commit-graph version %X does not match version %X"),
353 graph_version, GRAPH_VERSION);
354 return NULL;
357 hash_version = *(unsigned char*)(data + 5);
358 if (hash_version != oid_version(the_hash_algo)) {
359 error(_("commit-graph hash version %X does not match version %X"),
360 hash_version, oid_version(the_hash_algo));
361 return NULL;
364 graph = alloc_commit_graph();
366 graph->hash_len = the_hash_algo->rawsz;
367 graph->num_chunks = *(unsigned char*)(data + 6);
368 graph->data = graph_map;
369 graph->data_len = graph_size;
371 if (graph_size < GRAPH_HEADER_SIZE +
372 (graph->num_chunks + 1) * CHUNK_TOC_ENTRY_SIZE +
373 GRAPH_FANOUT_SIZE + the_hash_algo->rawsz) {
374 error(_("commit-graph file is too small to hold %u chunks"),
375 graph->num_chunks);
376 free(graph);
377 return NULL;
380 cf = init_chunkfile(NULL);
382 if (read_table_of_contents(cf, graph->data, graph_size,
383 GRAPH_HEADER_SIZE, graph->num_chunks))
384 goto free_and_return;
386 pair_chunk(cf, GRAPH_CHUNKID_OIDFANOUT,
387 (const unsigned char **)&graph->chunk_oid_fanout);
388 read_chunk(cf, GRAPH_CHUNKID_OIDLOOKUP, graph_read_oid_lookup, graph);
389 pair_chunk(cf, GRAPH_CHUNKID_DATA, &graph->chunk_commit_data);
390 pair_chunk(cf, GRAPH_CHUNKID_EXTRAEDGES, &graph->chunk_extra_edges);
391 pair_chunk(cf, GRAPH_CHUNKID_BASE, &graph->chunk_base_graphs);
393 if (s->commit_graph_generation_version >= 2) {
394 pair_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA,
395 &graph->chunk_generation_data);
396 pair_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW,
397 &graph->chunk_generation_data_overflow);
399 if (graph->chunk_generation_data)
400 graph->read_generation_data = 1;
403 if (s->commit_graph_read_changed_paths) {
404 pair_chunk(cf, GRAPH_CHUNKID_BLOOMINDEXES,
405 &graph->chunk_bloom_indexes);
406 read_chunk(cf, GRAPH_CHUNKID_BLOOMDATA,
407 graph_read_bloom_data, graph);
410 if (graph->chunk_bloom_indexes && graph->chunk_bloom_data) {
411 init_bloom_filters();
412 } else {
413 /* We need both the bloom chunks to exist together. Else ignore the data */
414 graph->chunk_bloom_indexes = NULL;
415 graph->chunk_bloom_data = NULL;
416 FREE_AND_NULL(graph->bloom_filter_settings);
419 oidread(&graph->oid, graph->data + graph->data_len - graph->hash_len);
421 if (verify_commit_graph_lite(graph))
422 goto free_and_return;
424 free_chunkfile(cf);
425 return graph;
427 free_and_return:
428 free_chunkfile(cf);
429 free(graph->bloom_filter_settings);
430 free(graph);
431 return NULL;
434 static struct commit_graph *load_commit_graph_one(struct repository *r,
435 const char *graph_file,
436 struct object_directory *odb)
439 struct stat st;
440 int fd;
441 struct commit_graph *g;
442 int open_ok = open_commit_graph(graph_file, &fd, &st);
444 if (!open_ok)
445 return NULL;
447 g = load_commit_graph_one_fd_st(r, fd, &st, odb);
449 if (g)
450 g->filename = xstrdup(graph_file);
452 return g;
455 static struct commit_graph *load_commit_graph_v1(struct repository *r,
456 struct object_directory *odb)
458 char *graph_name = get_commit_graph_filename(odb);
459 struct commit_graph *g = load_commit_graph_one(r, graph_name, odb);
460 free(graph_name);
462 return g;
465 static int add_graph_to_chain(struct commit_graph *g,
466 struct commit_graph *chain,
467 struct object_id *oids,
468 int n)
470 struct commit_graph *cur_g = chain;
472 if (n && !g->chunk_base_graphs) {
473 warning(_("commit-graph has no base graphs chunk"));
474 return 0;
477 while (n) {
478 n--;
480 if (!cur_g ||
481 !oideq(&oids[n], &cur_g->oid) ||
482 !hasheq(oids[n].hash, g->chunk_base_graphs + g->hash_len * n)) {
483 warning(_("commit-graph chain does not match"));
484 return 0;
487 cur_g = cur_g->base_graph;
490 g->base_graph = chain;
492 if (chain)
493 g->num_commits_in_base = chain->num_commits + chain->num_commits_in_base;
495 return 1;
498 static struct commit_graph *load_commit_graph_chain(struct repository *r,
499 struct object_directory *odb)
501 struct commit_graph *graph_chain = NULL;
502 struct strbuf line = STRBUF_INIT;
503 struct stat st;
504 struct object_id *oids;
505 int i = 0, valid = 1, count;
506 char *chain_name = get_commit_graph_chain_filename(odb);
507 FILE *fp;
508 int stat_res;
510 fp = fopen(chain_name, "r");
511 stat_res = stat(chain_name, &st);
512 free(chain_name);
514 if (!fp)
515 return NULL;
516 if (stat_res ||
517 st.st_size <= the_hash_algo->hexsz) {
518 fclose(fp);
519 return NULL;
522 count = st.st_size / (the_hash_algo->hexsz + 1);
523 CALLOC_ARRAY(oids, count);
525 prepare_alt_odb(r);
527 for (i = 0; i < count; i++) {
528 struct object_directory *odb;
530 if (strbuf_getline_lf(&line, fp) == EOF)
531 break;
533 if (get_oid_hex(line.buf, &oids[i])) {
534 warning(_("invalid commit-graph chain: line '%s' not a hash"),
535 line.buf);
536 valid = 0;
537 break;
540 valid = 0;
541 for (odb = r->objects->odb; odb; odb = odb->next) {
542 char *graph_name = get_split_graph_filename(odb, line.buf);
543 struct commit_graph *g = load_commit_graph_one(r, graph_name, odb);
545 free(graph_name);
547 if (g) {
548 if (add_graph_to_chain(g, graph_chain, oids, i)) {
549 graph_chain = g;
550 valid = 1;
553 break;
557 if (!valid) {
558 warning(_("unable to find all commit-graph files"));
559 break;
563 free(oids);
564 fclose(fp);
565 strbuf_release(&line);
567 return graph_chain;
571 * returns 1 if and only if all graphs in the chain have
572 * corrected commit dates stored in the generation_data chunk.
574 static int validate_mixed_generation_chain(struct commit_graph *g)
576 int read_generation_data = 1;
577 struct commit_graph *p = g;
579 while (read_generation_data && p) {
580 read_generation_data = p->read_generation_data;
581 p = p->base_graph;
584 if (read_generation_data)
585 return 1;
587 while (g) {
588 g->read_generation_data = 0;
589 g = g->base_graph;
592 return 0;
595 struct commit_graph *read_commit_graph_one(struct repository *r,
596 struct object_directory *odb)
598 struct commit_graph *g = load_commit_graph_v1(r, odb);
600 if (!g)
601 g = load_commit_graph_chain(r, odb);
603 validate_mixed_generation_chain(g);
605 return g;
608 static void prepare_commit_graph_one(struct repository *r,
609 struct object_directory *odb)
612 if (r->objects->commit_graph)
613 return;
615 r->objects->commit_graph = read_commit_graph_one(r, odb);
619 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
621 * On the first invocation, this function attempts to load the commit
622 * graph if the_repository is configured to have one.
624 static int prepare_commit_graph(struct repository *r)
626 struct object_directory *odb;
629 * Early return if there is no git dir or if the commit graph is
630 * disabled.
632 * This must come before the "already attempted?" check below, because
633 * we want to disable even an already-loaded graph file.
635 if (!r->gitdir || r->commit_graph_disabled)
636 return 0;
638 if (r->objects->commit_graph_attempted)
639 return !!r->objects->commit_graph;
640 r->objects->commit_graph_attempted = 1;
642 prepare_repo_settings(r);
644 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
645 r->settings.core_commit_graph != 1)
647 * This repository is not configured to use commit graphs, so
648 * do not load one. (But report commit_graph_attempted anyway
649 * so that commit graph loading is not attempted again for this
650 * repository.)
652 return 0;
654 if (!commit_graph_compatible(r))
655 return 0;
657 prepare_alt_odb(r);
658 for (odb = r->objects->odb;
659 !r->objects->commit_graph && odb;
660 odb = odb->next)
661 prepare_commit_graph_one(r, odb);
662 return !!r->objects->commit_graph;
665 int generation_numbers_enabled(struct repository *r)
667 uint32_t first_generation;
668 struct commit_graph *g;
669 if (!prepare_commit_graph(r))
670 return 0;
672 g = r->objects->commit_graph;
674 if (!g->num_commits)
675 return 0;
677 first_generation = get_be32(g->chunk_commit_data +
678 g->hash_len + 8) >> 2;
680 return !!first_generation;
683 int corrected_commit_dates_enabled(struct repository *r)
685 struct commit_graph *g;
686 if (!prepare_commit_graph(r))
687 return 0;
689 g = r->objects->commit_graph;
691 if (!g->num_commits)
692 return 0;
694 return g->read_generation_data;
697 struct bloom_filter_settings *get_bloom_filter_settings(struct repository *r)
699 struct commit_graph *g = r->objects->commit_graph;
700 while (g) {
701 if (g->bloom_filter_settings)
702 return g->bloom_filter_settings;
703 g = g->base_graph;
705 return NULL;
708 static void close_commit_graph_one(struct commit_graph *g)
710 if (!g)
711 return;
713 clear_commit_graph_data_slab(&commit_graph_data_slab);
714 close_commit_graph_one(g->base_graph);
715 free_commit_graph(g);
718 void close_commit_graph(struct raw_object_store *o)
720 close_commit_graph_one(o->commit_graph);
721 o->commit_graph = NULL;
724 static int bsearch_graph(struct commit_graph *g, const struct object_id *oid, uint32_t *pos)
726 return bsearch_hash(oid->hash, g->chunk_oid_fanout,
727 g->chunk_oid_lookup, g->hash_len, pos);
730 static void load_oid_from_graph(struct commit_graph *g,
731 uint32_t pos,
732 struct object_id *oid)
734 uint32_t lex_index;
736 while (g && pos < g->num_commits_in_base)
737 g = g->base_graph;
739 if (!g)
740 BUG("NULL commit-graph");
742 if (pos >= g->num_commits + g->num_commits_in_base)
743 die(_("invalid commit position. commit-graph is likely corrupt"));
745 lex_index = pos - g->num_commits_in_base;
747 oidread(oid, g->chunk_oid_lookup + g->hash_len * lex_index);
750 static struct commit_list **insert_parent_or_die(struct repository *r,
751 struct commit_graph *g,
752 uint32_t pos,
753 struct commit_list **pptr)
755 struct commit *c;
756 struct object_id oid;
758 if (pos >= g->num_commits + g->num_commits_in_base)
759 die("invalid parent position %"PRIu32, pos);
761 load_oid_from_graph(g, pos, &oid);
762 c = lookup_commit(r, &oid);
763 if (!c)
764 die(_("could not find commit %s"), oid_to_hex(&oid));
765 commit_graph_data_at(c)->graph_pos = pos;
766 return &commit_list_insert(c, pptr)->next;
769 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
771 const unsigned char *commit_data;
772 struct commit_graph_data *graph_data;
773 uint32_t lex_index, offset_pos;
774 uint64_t date_high, date_low, offset;
776 while (pos < g->num_commits_in_base)
777 g = g->base_graph;
779 if (pos >= g->num_commits + g->num_commits_in_base)
780 die(_("invalid commit position. commit-graph is likely corrupt"));
782 lex_index = pos - g->num_commits_in_base;
783 commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * lex_index;
785 graph_data = commit_graph_data_at(item);
786 graph_data->graph_pos = pos;
788 date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
789 date_low = get_be32(commit_data + g->hash_len + 12);
790 item->date = (timestamp_t)((date_high << 32) | date_low);
792 if (g->read_generation_data) {
793 offset = (timestamp_t)get_be32(g->chunk_generation_data + sizeof(uint32_t) * lex_index);
795 if (offset & CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW) {
796 if (!g->chunk_generation_data_overflow)
797 die(_("commit-graph requires overflow generation data but has none"));
799 offset_pos = offset ^ CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW;
800 graph_data->generation = item->date + get_be64(g->chunk_generation_data_overflow + 8 * offset_pos);
801 } else
802 graph_data->generation = item->date + offset;
803 } else
804 graph_data->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
806 if (g->topo_levels)
807 *topo_level_slab_at(g->topo_levels, item) = get_be32(commit_data + g->hash_len + 8) >> 2;
810 static inline void set_commit_tree(struct commit *c, struct tree *t)
812 c->maybe_tree = t;
815 static int fill_commit_in_graph(struct repository *r,
816 struct commit *item,
817 struct commit_graph *g, uint32_t pos)
819 uint32_t edge_value;
820 uint32_t *parent_data_ptr;
821 struct commit_list **pptr;
822 const unsigned char *commit_data;
823 uint32_t lex_index;
825 while (pos < g->num_commits_in_base)
826 g = g->base_graph;
828 fill_commit_graph_info(item, g, pos);
830 lex_index = pos - g->num_commits_in_base;
831 commit_data = g->chunk_commit_data + (g->hash_len + 16) * lex_index;
833 item->object.parsed = 1;
835 set_commit_tree(item, NULL);
837 pptr = &item->parents;
839 edge_value = get_be32(commit_data + g->hash_len);
840 if (edge_value == GRAPH_PARENT_NONE)
841 return 1;
842 pptr = insert_parent_or_die(r, g, edge_value, pptr);
844 edge_value = get_be32(commit_data + g->hash_len + 4);
845 if (edge_value == GRAPH_PARENT_NONE)
846 return 1;
847 if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
848 pptr = insert_parent_or_die(r, g, edge_value, pptr);
849 return 1;
852 parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
853 4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
854 do {
855 edge_value = get_be32(parent_data_ptr);
856 pptr = insert_parent_or_die(r, g,
857 edge_value & GRAPH_EDGE_LAST_MASK,
858 pptr);
859 parent_data_ptr++;
860 } while (!(edge_value & GRAPH_LAST_EDGE));
862 return 1;
865 static int search_commit_pos_in_graph(const struct object_id *id, struct commit_graph *g, uint32_t *pos)
867 struct commit_graph *cur_g = g;
868 uint32_t lex_index;
870 while (cur_g && !bsearch_graph(cur_g, id, &lex_index))
871 cur_g = cur_g->base_graph;
873 if (cur_g) {
874 *pos = lex_index + cur_g->num_commits_in_base;
875 return 1;
878 return 0;
881 static int find_commit_pos_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
883 uint32_t graph_pos = commit_graph_position(item);
884 if (graph_pos != COMMIT_NOT_FROM_GRAPH) {
885 *pos = graph_pos;
886 return 1;
887 } else {
888 return search_commit_pos_in_graph(&item->object.oid, g, pos);
892 int repo_find_commit_pos_in_graph(struct repository *r, struct commit *c,
893 uint32_t *pos)
895 if (!prepare_commit_graph(r))
896 return 0;
897 return find_commit_pos_in_graph(c, r->objects->commit_graph, pos);
900 struct commit *lookup_commit_in_graph(struct repository *repo, const struct object_id *id)
902 struct commit *commit;
903 uint32_t pos;
905 if (!prepare_commit_graph(repo))
906 return NULL;
907 if (!search_commit_pos_in_graph(id, repo->objects->commit_graph, &pos))
908 return NULL;
909 if (!has_object(repo, id, 0))
910 return NULL;
912 commit = lookup_commit(repo, id);
913 if (!commit)
914 return NULL;
915 if (commit->object.parsed)
916 return commit;
918 if (!fill_commit_in_graph(repo, commit, repo->objects->commit_graph, pos))
919 return NULL;
921 return commit;
924 static int parse_commit_in_graph_one(struct repository *r,
925 struct commit_graph *g,
926 struct commit *item)
928 uint32_t pos;
930 if (item->object.parsed)
931 return 1;
933 if (find_commit_pos_in_graph(item, g, &pos))
934 return fill_commit_in_graph(r, item, g, pos);
936 return 0;
939 int parse_commit_in_graph(struct repository *r, struct commit *item)
941 static int checked_env = 0;
943 if (!checked_env &&
944 git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_PARSE, 0))
945 die("dying as requested by the '%s' variable on commit-graph parse!",
946 GIT_TEST_COMMIT_GRAPH_DIE_ON_PARSE);
947 checked_env = 1;
949 if (!prepare_commit_graph(r))
950 return 0;
951 return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
954 void load_commit_graph_info(struct repository *r, struct commit *item)
956 uint32_t pos;
957 if (repo_find_commit_pos_in_graph(r, item, &pos))
958 fill_commit_graph_info(item, r->objects->commit_graph, pos);
961 static struct tree *load_tree_for_commit(struct repository *r,
962 struct commit_graph *g,
963 struct commit *c)
965 struct object_id oid;
966 const unsigned char *commit_data;
967 uint32_t graph_pos = commit_graph_position(c);
969 while (graph_pos < g->num_commits_in_base)
970 g = g->base_graph;
972 commit_data = g->chunk_commit_data +
973 GRAPH_DATA_WIDTH * (graph_pos - g->num_commits_in_base);
975 oidread(&oid, commit_data);
976 set_commit_tree(c, lookup_tree(r, &oid));
978 return c->maybe_tree;
981 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
982 struct commit_graph *g,
983 const struct commit *c)
985 if (c->maybe_tree)
986 return c->maybe_tree;
987 if (commit_graph_position(c) == COMMIT_NOT_FROM_GRAPH)
988 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
990 return load_tree_for_commit(r, g, (struct commit *)c);
993 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
995 return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
998 struct packed_commit_list {
999 struct commit **list;
1000 size_t nr;
1001 size_t alloc;
1004 struct write_commit_graph_context {
1005 struct repository *r;
1006 struct object_directory *odb;
1007 char *graph_name;
1008 struct oid_array oids;
1009 struct packed_commit_list commits;
1010 int num_extra_edges;
1011 int num_generation_data_overflows;
1012 unsigned long approx_nr_objects;
1013 struct progress *progress;
1014 int progress_done;
1015 uint64_t progress_cnt;
1017 char *base_graph_name;
1018 int num_commit_graphs_before;
1019 int num_commit_graphs_after;
1020 char **commit_graph_filenames_before;
1021 char **commit_graph_filenames_after;
1022 char **commit_graph_hash_after;
1023 uint32_t new_num_commits_in_base;
1024 struct commit_graph *new_base_graph;
1026 unsigned append:1,
1027 report_progress:1,
1028 split:1,
1029 changed_paths:1,
1030 order_by_pack:1,
1031 write_generation_data:1,
1032 trust_generation_numbers:1;
1034 struct topo_level_slab *topo_levels;
1035 const struct commit_graph_opts *opts;
1036 size_t total_bloom_filter_data_size;
1037 const struct bloom_filter_settings *bloom_settings;
1039 int count_bloom_filter_computed;
1040 int count_bloom_filter_not_computed;
1041 int count_bloom_filter_trunc_empty;
1042 int count_bloom_filter_trunc_large;
1045 static int write_graph_chunk_fanout(struct hashfile *f,
1046 void *data)
1048 struct write_commit_graph_context *ctx = data;
1049 int i, count = 0;
1050 struct commit **list = ctx->commits.list;
1053 * Write the first-level table (the list is sorted,
1054 * but we use a 256-entry lookup to be able to avoid
1055 * having to do eight extra binary search iterations).
1057 for (i = 0; i < 256; i++) {
1058 while (count < ctx->commits.nr) {
1059 if ((*list)->object.oid.hash[0] != i)
1060 break;
1061 display_progress(ctx->progress, ++ctx->progress_cnt);
1062 count++;
1063 list++;
1066 hashwrite_be32(f, count);
1069 return 0;
1072 static int write_graph_chunk_oids(struct hashfile *f,
1073 void *data)
1075 struct write_commit_graph_context *ctx = data;
1076 struct commit **list = ctx->commits.list;
1077 int count;
1078 for (count = 0; count < ctx->commits.nr; count++, list++) {
1079 display_progress(ctx->progress, ++ctx->progress_cnt);
1080 hashwrite(f, (*list)->object.oid.hash, the_hash_algo->rawsz);
1083 return 0;
1086 static const struct object_id *commit_to_oid(size_t index, const void *table)
1088 const struct commit * const *commits = table;
1089 return &commits[index]->object.oid;
1092 static int write_graph_chunk_data(struct hashfile *f,
1093 void *data)
1095 struct write_commit_graph_context *ctx = data;
1096 struct commit **list = ctx->commits.list;
1097 struct commit **last = ctx->commits.list + ctx->commits.nr;
1098 uint32_t num_extra_edges = 0;
1100 while (list < last) {
1101 struct commit_list *parent;
1102 struct object_id *tree;
1103 int edge_value;
1104 uint32_t packedDate[2];
1105 display_progress(ctx->progress, ++ctx->progress_cnt);
1107 if (repo_parse_commit_no_graph(ctx->r, *list))
1108 die(_("unable to parse commit %s"),
1109 oid_to_hex(&(*list)->object.oid));
1110 tree = get_commit_tree_oid(*list);
1111 hashwrite(f, tree->hash, the_hash_algo->rawsz);
1113 parent = (*list)->parents;
1115 if (!parent)
1116 edge_value = GRAPH_PARENT_NONE;
1117 else {
1118 edge_value = oid_pos(&parent->item->object.oid,
1119 ctx->commits.list,
1120 ctx->commits.nr,
1121 commit_to_oid);
1123 if (edge_value >= 0)
1124 edge_value += ctx->new_num_commits_in_base;
1125 else if (ctx->new_base_graph) {
1126 uint32_t pos;
1127 if (find_commit_pos_in_graph(parent->item,
1128 ctx->new_base_graph,
1129 &pos))
1130 edge_value = pos;
1133 if (edge_value < 0)
1134 BUG("missing parent %s for commit %s",
1135 oid_to_hex(&parent->item->object.oid),
1136 oid_to_hex(&(*list)->object.oid));
1139 hashwrite_be32(f, edge_value);
1141 if (parent)
1142 parent = parent->next;
1144 if (!parent)
1145 edge_value = GRAPH_PARENT_NONE;
1146 else if (parent->next)
1147 edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
1148 else {
1149 edge_value = oid_pos(&parent->item->object.oid,
1150 ctx->commits.list,
1151 ctx->commits.nr,
1152 commit_to_oid);
1154 if (edge_value >= 0)
1155 edge_value += ctx->new_num_commits_in_base;
1156 else if (ctx->new_base_graph) {
1157 uint32_t pos;
1158 if (find_commit_pos_in_graph(parent->item,
1159 ctx->new_base_graph,
1160 &pos))
1161 edge_value = pos;
1164 if (edge_value < 0)
1165 BUG("missing parent %s for commit %s",
1166 oid_to_hex(&parent->item->object.oid),
1167 oid_to_hex(&(*list)->object.oid));
1170 hashwrite_be32(f, edge_value);
1172 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
1173 do {
1174 num_extra_edges++;
1175 parent = parent->next;
1176 } while (parent);
1179 if (sizeof((*list)->date) > 4)
1180 packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
1181 else
1182 packedDate[0] = 0;
1184 packedDate[0] |= htonl(*topo_level_slab_at(ctx->topo_levels, *list) << 2);
1186 packedDate[1] = htonl((*list)->date);
1187 hashwrite(f, packedDate, 8);
1189 list++;
1192 return 0;
1195 static int write_graph_chunk_generation_data(struct hashfile *f,
1196 void *data)
1198 struct write_commit_graph_context *ctx = data;
1199 int i, num_generation_data_overflows = 0;
1201 for (i = 0; i < ctx->commits.nr; i++) {
1202 struct commit *c = ctx->commits.list[i];
1203 timestamp_t offset;
1204 repo_parse_commit(ctx->r, c);
1205 offset = commit_graph_data_at(c)->generation - c->date;
1206 display_progress(ctx->progress, ++ctx->progress_cnt);
1208 if (offset > GENERATION_NUMBER_V2_OFFSET_MAX) {
1209 offset = CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW | num_generation_data_overflows;
1210 num_generation_data_overflows++;
1213 hashwrite_be32(f, offset);
1216 return 0;
1219 static int write_graph_chunk_generation_data_overflow(struct hashfile *f,
1220 void *data)
1222 struct write_commit_graph_context *ctx = data;
1223 int i;
1224 for (i = 0; i < ctx->commits.nr; i++) {
1225 struct commit *c = ctx->commits.list[i];
1226 timestamp_t offset = commit_graph_data_at(c)->generation - c->date;
1227 display_progress(ctx->progress, ++ctx->progress_cnt);
1229 if (offset > GENERATION_NUMBER_V2_OFFSET_MAX) {
1230 hashwrite_be32(f, offset >> 32);
1231 hashwrite_be32(f, (uint32_t) offset);
1235 return 0;
1238 static int write_graph_chunk_extra_edges(struct hashfile *f,
1239 void *data)
1241 struct write_commit_graph_context *ctx = data;
1242 struct commit **list = ctx->commits.list;
1243 struct commit **last = ctx->commits.list + ctx->commits.nr;
1244 struct commit_list *parent;
1246 while (list < last) {
1247 int num_parents = 0;
1249 display_progress(ctx->progress, ++ctx->progress_cnt);
1251 for (parent = (*list)->parents; num_parents < 3 && parent;
1252 parent = parent->next)
1253 num_parents++;
1255 if (num_parents <= 2) {
1256 list++;
1257 continue;
1260 /* Since num_parents > 2, this initializer is safe. */
1261 for (parent = (*list)->parents->next; parent; parent = parent->next) {
1262 int edge_value = oid_pos(&parent->item->object.oid,
1263 ctx->commits.list,
1264 ctx->commits.nr,
1265 commit_to_oid);
1267 if (edge_value >= 0)
1268 edge_value += ctx->new_num_commits_in_base;
1269 else if (ctx->new_base_graph) {
1270 uint32_t pos;
1271 if (find_commit_pos_in_graph(parent->item,
1272 ctx->new_base_graph,
1273 &pos))
1274 edge_value = pos;
1277 if (edge_value < 0)
1278 BUG("missing parent %s for commit %s",
1279 oid_to_hex(&parent->item->object.oid),
1280 oid_to_hex(&(*list)->object.oid));
1281 else if (!parent->next)
1282 edge_value |= GRAPH_LAST_EDGE;
1284 hashwrite_be32(f, edge_value);
1287 list++;
1290 return 0;
1293 static int write_graph_chunk_bloom_indexes(struct hashfile *f,
1294 void *data)
1296 struct write_commit_graph_context *ctx = data;
1297 struct commit **list = ctx->commits.list;
1298 struct commit **last = ctx->commits.list + ctx->commits.nr;
1299 uint32_t cur_pos = 0;
1301 while (list < last) {
1302 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list);
1303 size_t len = filter ? filter->len : 0;
1304 cur_pos += len;
1305 display_progress(ctx->progress, ++ctx->progress_cnt);
1306 hashwrite_be32(f, cur_pos);
1307 list++;
1310 return 0;
1313 static void trace2_bloom_filter_settings(struct write_commit_graph_context *ctx)
1315 struct json_writer jw = JSON_WRITER_INIT;
1317 jw_object_begin(&jw, 0);
1318 jw_object_intmax(&jw, "hash_version", ctx->bloom_settings->hash_version);
1319 jw_object_intmax(&jw, "num_hashes", ctx->bloom_settings->num_hashes);
1320 jw_object_intmax(&jw, "bits_per_entry", ctx->bloom_settings->bits_per_entry);
1321 jw_object_intmax(&jw, "max_changed_paths", ctx->bloom_settings->max_changed_paths);
1322 jw_end(&jw);
1324 trace2_data_json("bloom", ctx->r, "settings", &jw);
1326 jw_release(&jw);
1329 static int write_graph_chunk_bloom_data(struct hashfile *f,
1330 void *data)
1332 struct write_commit_graph_context *ctx = data;
1333 struct commit **list = ctx->commits.list;
1334 struct commit **last = ctx->commits.list + ctx->commits.nr;
1336 trace2_bloom_filter_settings(ctx);
1338 hashwrite_be32(f, ctx->bloom_settings->hash_version);
1339 hashwrite_be32(f, ctx->bloom_settings->num_hashes);
1340 hashwrite_be32(f, ctx->bloom_settings->bits_per_entry);
1342 while (list < last) {
1343 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list);
1344 size_t len = filter ? filter->len : 0;
1346 display_progress(ctx->progress, ++ctx->progress_cnt);
1347 if (len)
1348 hashwrite(f, filter->data, len * sizeof(unsigned char));
1349 list++;
1352 return 0;
1355 static int add_packed_commits(const struct object_id *oid,
1356 struct packed_git *pack,
1357 uint32_t pos,
1358 void *data)
1360 struct write_commit_graph_context *ctx = (struct write_commit_graph_context*)data;
1361 enum object_type type;
1362 off_t offset = nth_packed_object_offset(pack, pos);
1363 struct object_info oi = OBJECT_INFO_INIT;
1365 if (ctx->progress)
1366 display_progress(ctx->progress, ++ctx->progress_done);
1368 oi.typep = &type;
1369 if (packed_object_info(ctx->r, pack, offset, &oi) < 0)
1370 die(_("unable to get type of object %s"), oid_to_hex(oid));
1372 if (type != OBJ_COMMIT)
1373 return 0;
1375 oid_array_append(&ctx->oids, oid);
1376 set_commit_pos(ctx->r, oid);
1378 return 0;
1381 static void add_missing_parents(struct write_commit_graph_context *ctx, struct commit *commit)
1383 struct commit_list *parent;
1384 for (parent = commit->parents; parent; parent = parent->next) {
1385 if (!(parent->item->object.flags & REACHABLE)) {
1386 oid_array_append(&ctx->oids, &parent->item->object.oid);
1387 parent->item->object.flags |= REACHABLE;
1392 static void close_reachable(struct write_commit_graph_context *ctx)
1394 int i;
1395 struct commit *commit;
1396 enum commit_graph_split_flags flags = ctx->opts ?
1397 ctx->opts->split_flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1399 if (ctx->report_progress)
1400 ctx->progress = start_delayed_progress(
1401 _("Loading known commits in commit graph"),
1402 ctx->oids.nr);
1403 for (i = 0; i < ctx->oids.nr; i++) {
1404 display_progress(ctx->progress, i + 1);
1405 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1406 if (commit)
1407 commit->object.flags |= REACHABLE;
1409 stop_progress(&ctx->progress);
1412 * As this loop runs, ctx->oids.nr may grow, but not more
1413 * than the number of missing commits in the reachable
1414 * closure.
1416 if (ctx->report_progress)
1417 ctx->progress = start_delayed_progress(
1418 _("Expanding reachable commits in commit graph"),
1420 for (i = 0; i < ctx->oids.nr; i++) {
1421 display_progress(ctx->progress, i + 1);
1422 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1424 if (!commit)
1425 continue;
1426 if (ctx->split) {
1427 if ((!repo_parse_commit(ctx->r, commit) &&
1428 commit_graph_position(commit) == COMMIT_NOT_FROM_GRAPH) ||
1429 flags == COMMIT_GRAPH_SPLIT_REPLACE)
1430 add_missing_parents(ctx, commit);
1431 } else if (!repo_parse_commit_no_graph(ctx->r, commit))
1432 add_missing_parents(ctx, commit);
1434 stop_progress(&ctx->progress);
1436 if (ctx->report_progress)
1437 ctx->progress = start_delayed_progress(
1438 _("Clearing commit marks in commit graph"),
1439 ctx->oids.nr);
1440 for (i = 0; i < ctx->oids.nr; i++) {
1441 display_progress(ctx->progress, i + 1);
1442 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1444 if (commit)
1445 commit->object.flags &= ~REACHABLE;
1447 stop_progress(&ctx->progress);
1450 struct compute_generation_info {
1451 struct repository *r;
1452 struct packed_commit_list *commits;
1453 struct progress *progress;
1454 int progress_cnt;
1456 timestamp_t (*get_generation)(struct commit *c, void *data);
1457 void (*set_generation)(struct commit *c, timestamp_t gen, void *data);
1458 void *data;
1461 static timestamp_t compute_generation_from_max(struct commit *c,
1462 timestamp_t max_gen,
1463 int generation_version)
1465 switch (generation_version) {
1466 case 1: /* topological levels */
1467 if (max_gen > GENERATION_NUMBER_V1_MAX - 1)
1468 max_gen = GENERATION_NUMBER_V1_MAX - 1;
1469 return max_gen + 1;
1471 case 2: /* corrected commit date */
1472 if (c->date && c->date > max_gen)
1473 max_gen = c->date - 1;
1474 return max_gen + 1;
1476 default:
1477 BUG("attempting unimplemented version");
1481 static void compute_reachable_generation_numbers(
1482 struct compute_generation_info *info,
1483 int generation_version)
1485 int i;
1486 struct commit_list *list = NULL;
1488 for (i = 0; i < info->commits->nr; i++) {
1489 struct commit *c = info->commits->list[i];
1490 timestamp_t gen;
1491 repo_parse_commit(info->r, c);
1492 gen = info->get_generation(c, info->data);
1493 display_progress(info->progress, info->progress_cnt + 1);
1495 if (gen != GENERATION_NUMBER_ZERO && gen != GENERATION_NUMBER_INFINITY)
1496 continue;
1498 commit_list_insert(c, &list);
1499 while (list) {
1500 struct commit *current = list->item;
1501 struct commit_list *parent;
1502 int all_parents_computed = 1;
1503 uint32_t max_gen = 0;
1505 for (parent = current->parents; parent; parent = parent->next) {
1506 repo_parse_commit(info->r, parent->item);
1507 gen = info->get_generation(parent->item, info->data);
1509 if (gen == GENERATION_NUMBER_ZERO) {
1510 all_parents_computed = 0;
1511 commit_list_insert(parent->item, &list);
1512 break;
1515 if (gen > max_gen)
1516 max_gen = gen;
1519 if (all_parents_computed) {
1520 pop_commit(&list);
1521 gen = compute_generation_from_max(
1522 current, max_gen,
1523 generation_version);
1524 info->set_generation(current, gen, info->data);
1530 static timestamp_t get_topo_level(struct commit *c, void *data)
1532 struct write_commit_graph_context *ctx = data;
1533 return *topo_level_slab_at(ctx->topo_levels, c);
1536 static void set_topo_level(struct commit *c, timestamp_t t, void *data)
1538 struct write_commit_graph_context *ctx = data;
1539 *topo_level_slab_at(ctx->topo_levels, c) = (uint32_t)t;
1542 static void compute_topological_levels(struct write_commit_graph_context *ctx)
1544 struct compute_generation_info info = {
1545 .r = ctx->r,
1546 .commits = &ctx->commits,
1547 .get_generation = get_topo_level,
1548 .set_generation = set_topo_level,
1549 .data = ctx,
1552 if (ctx->report_progress)
1553 info.progress = ctx->progress
1554 = start_delayed_progress(
1555 _("Computing commit graph topological levels"),
1556 ctx->commits.nr);
1558 compute_reachable_generation_numbers(&info, 1);
1560 stop_progress(&ctx->progress);
1563 static timestamp_t get_generation_from_graph_data(struct commit *c, void *data)
1565 return commit_graph_data_at(c)->generation;
1568 static void set_generation_v2(struct commit *c, timestamp_t t, void *data)
1570 struct commit_graph_data *g = commit_graph_data_at(c);
1571 g->generation = t;
1574 static void compute_generation_numbers(struct write_commit_graph_context *ctx)
1576 int i;
1577 struct compute_generation_info info = {
1578 .r = ctx->r,
1579 .commits = &ctx->commits,
1580 .get_generation = get_generation_from_graph_data,
1581 .set_generation = set_generation_v2,
1582 .data = ctx,
1585 if (ctx->report_progress)
1586 info.progress = ctx->progress
1587 = start_delayed_progress(
1588 _("Computing commit graph generation numbers"),
1589 ctx->commits.nr);
1591 if (!ctx->trust_generation_numbers) {
1592 for (i = 0; i < ctx->commits.nr; i++) {
1593 struct commit *c = ctx->commits.list[i];
1594 repo_parse_commit(ctx->r, c);
1595 commit_graph_data_at(c)->generation = GENERATION_NUMBER_ZERO;
1599 compute_reachable_generation_numbers(&info, 2);
1601 for (i = 0; i < ctx->commits.nr; i++) {
1602 struct commit *c = ctx->commits.list[i];
1603 timestamp_t offset = commit_graph_data_at(c)->generation - c->date;
1604 if (offset > GENERATION_NUMBER_V2_OFFSET_MAX)
1605 ctx->num_generation_data_overflows++;
1607 stop_progress(&ctx->progress);
1610 static void set_generation_in_graph_data(struct commit *c, timestamp_t t,
1611 void *data)
1613 commit_graph_data_at(c)->generation = t;
1617 * After this method, all commits reachable from those in the given
1618 * list will have non-zero, non-infinite generation numbers.
1620 void ensure_generations_valid(struct repository *r,
1621 struct commit **commits, size_t nr)
1623 int generation_version = get_configured_generation_version(r);
1624 struct packed_commit_list list = {
1625 .list = commits,
1626 .alloc = nr,
1627 .nr = nr,
1629 struct compute_generation_info info = {
1630 .r = r,
1631 .commits = &list,
1632 .get_generation = get_generation_from_graph_data,
1633 .set_generation = set_generation_in_graph_data,
1636 compute_reachable_generation_numbers(&info, generation_version);
1639 static void trace2_bloom_filter_write_statistics(struct write_commit_graph_context *ctx)
1641 trace2_data_intmax("commit-graph", ctx->r, "filter-computed",
1642 ctx->count_bloom_filter_computed);
1643 trace2_data_intmax("commit-graph", ctx->r, "filter-not-computed",
1644 ctx->count_bloom_filter_not_computed);
1645 trace2_data_intmax("commit-graph", ctx->r, "filter-trunc-empty",
1646 ctx->count_bloom_filter_trunc_empty);
1647 trace2_data_intmax("commit-graph", ctx->r, "filter-trunc-large",
1648 ctx->count_bloom_filter_trunc_large);
1651 static void compute_bloom_filters(struct write_commit_graph_context *ctx)
1653 int i;
1654 struct progress *progress = NULL;
1655 struct commit **sorted_commits;
1656 int max_new_filters;
1658 init_bloom_filters();
1660 if (ctx->report_progress)
1661 progress = start_delayed_progress(
1662 _("Computing commit changed paths Bloom filters"),
1663 ctx->commits.nr);
1665 DUP_ARRAY(sorted_commits, ctx->commits.list, ctx->commits.nr);
1667 if (ctx->order_by_pack)
1668 QSORT(sorted_commits, ctx->commits.nr, commit_pos_cmp);
1669 else
1670 QSORT(sorted_commits, ctx->commits.nr, commit_gen_cmp);
1672 max_new_filters = ctx->opts && ctx->opts->max_new_filters >= 0 ?
1673 ctx->opts->max_new_filters : ctx->commits.nr;
1675 for (i = 0; i < ctx->commits.nr; i++) {
1676 enum bloom_filter_computed computed = 0;
1677 struct commit *c = sorted_commits[i];
1678 struct bloom_filter *filter = get_or_compute_bloom_filter(
1679 ctx->r,
1681 ctx->count_bloom_filter_computed < max_new_filters,
1682 ctx->bloom_settings,
1683 &computed);
1684 if (computed & BLOOM_COMPUTED) {
1685 ctx->count_bloom_filter_computed++;
1686 if (computed & BLOOM_TRUNC_EMPTY)
1687 ctx->count_bloom_filter_trunc_empty++;
1688 if (computed & BLOOM_TRUNC_LARGE)
1689 ctx->count_bloom_filter_trunc_large++;
1690 } else if (computed & BLOOM_NOT_COMPUTED)
1691 ctx->count_bloom_filter_not_computed++;
1692 ctx->total_bloom_filter_data_size += filter
1693 ? sizeof(unsigned char) * filter->len : 0;
1694 display_progress(progress, i + 1);
1697 if (trace2_is_enabled())
1698 trace2_bloom_filter_write_statistics(ctx);
1700 free(sorted_commits);
1701 stop_progress(&progress);
1704 struct refs_cb_data {
1705 struct oidset *commits;
1706 struct progress *progress;
1709 static int add_ref_to_set(const char *refname UNUSED,
1710 const struct object_id *oid,
1711 int flags UNUSED, void *cb_data)
1713 struct object_id peeled;
1714 struct refs_cb_data *data = (struct refs_cb_data *)cb_data;
1716 if (!peel_iterated_oid(oid, &peeled))
1717 oid = &peeled;
1718 if (oid_object_info(the_repository, oid, NULL) == OBJ_COMMIT)
1719 oidset_insert(data->commits, oid);
1721 display_progress(data->progress, oidset_size(data->commits));
1723 return 0;
1726 int write_commit_graph_reachable(struct object_directory *odb,
1727 enum commit_graph_write_flags flags,
1728 const struct commit_graph_opts *opts)
1730 struct oidset commits = OIDSET_INIT;
1731 struct refs_cb_data data;
1732 int result;
1734 memset(&data, 0, sizeof(data));
1735 data.commits = &commits;
1736 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
1737 data.progress = start_delayed_progress(
1738 _("Collecting referenced commits"), 0);
1740 for_each_ref(add_ref_to_set, &data);
1742 stop_progress(&data.progress);
1744 result = write_commit_graph(odb, NULL, &commits,
1745 flags, opts);
1747 oidset_clear(&commits);
1748 return result;
1751 static int fill_oids_from_packs(struct write_commit_graph_context *ctx,
1752 const struct string_list *pack_indexes)
1754 uint32_t i;
1755 struct strbuf progress_title = STRBUF_INIT;
1756 struct strbuf packname = STRBUF_INIT;
1757 int dirlen;
1758 int ret = 0;
1760 strbuf_addf(&packname, "%s/pack/", ctx->odb->path);
1761 dirlen = packname.len;
1762 if (ctx->report_progress) {
1763 strbuf_addf(&progress_title,
1764 Q_("Finding commits for commit graph in %"PRIuMAX" pack",
1765 "Finding commits for commit graph in %"PRIuMAX" packs",
1766 pack_indexes->nr),
1767 (uintmax_t)pack_indexes->nr);
1768 ctx->progress = start_delayed_progress(progress_title.buf, 0);
1769 ctx->progress_done = 0;
1771 for (i = 0; i < pack_indexes->nr; i++) {
1772 struct packed_git *p;
1773 strbuf_setlen(&packname, dirlen);
1774 strbuf_addstr(&packname, pack_indexes->items[i].string);
1775 p = add_packed_git(packname.buf, packname.len, 1);
1776 if (!p) {
1777 ret = error(_("error adding pack %s"), packname.buf);
1778 goto cleanup;
1780 if (open_pack_index(p)) {
1781 ret = error(_("error opening index for %s"), packname.buf);
1782 goto cleanup;
1784 for_each_object_in_pack(p, add_packed_commits, ctx,
1785 FOR_EACH_OBJECT_PACK_ORDER);
1786 close_pack(p);
1787 free(p);
1790 cleanup:
1791 stop_progress(&ctx->progress);
1792 strbuf_release(&progress_title);
1793 strbuf_release(&packname);
1795 return ret;
1798 static int fill_oids_from_commits(struct write_commit_graph_context *ctx,
1799 struct oidset *commits)
1801 struct oidset_iter iter;
1802 struct object_id *oid;
1804 if (!oidset_size(commits))
1805 return 0;
1807 oidset_iter_init(commits, &iter);
1808 while ((oid = oidset_iter_next(&iter))) {
1809 oid_array_append(&ctx->oids, oid);
1812 return 0;
1815 static void fill_oids_from_all_packs(struct write_commit_graph_context *ctx)
1817 if (ctx->report_progress)
1818 ctx->progress = start_delayed_progress(
1819 _("Finding commits for commit graph among packed objects"),
1820 ctx->approx_nr_objects);
1821 for_each_packed_object(add_packed_commits, ctx,
1822 FOR_EACH_OBJECT_PACK_ORDER);
1823 if (ctx->progress_done < ctx->approx_nr_objects)
1824 display_progress(ctx->progress, ctx->approx_nr_objects);
1825 stop_progress(&ctx->progress);
1828 static void copy_oids_to_commits(struct write_commit_graph_context *ctx)
1830 uint32_t i;
1831 enum commit_graph_split_flags flags = ctx->opts ?
1832 ctx->opts->split_flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1834 ctx->num_extra_edges = 0;
1835 if (ctx->report_progress)
1836 ctx->progress = start_delayed_progress(
1837 _("Finding extra edges in commit graph"),
1838 ctx->oids.nr);
1839 oid_array_sort(&ctx->oids);
1840 for (i = 0; i < ctx->oids.nr; i = oid_array_next_unique(&ctx->oids, i)) {
1841 unsigned int num_parents;
1843 display_progress(ctx->progress, i + 1);
1845 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + 1, ctx->commits.alloc);
1846 ctx->commits.list[ctx->commits.nr] = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1848 if (ctx->split && flags != COMMIT_GRAPH_SPLIT_REPLACE &&
1849 commit_graph_position(ctx->commits.list[ctx->commits.nr]) != COMMIT_NOT_FROM_GRAPH)
1850 continue;
1852 if (ctx->split && flags == COMMIT_GRAPH_SPLIT_REPLACE)
1853 repo_parse_commit(ctx->r, ctx->commits.list[ctx->commits.nr]);
1854 else
1855 repo_parse_commit_no_graph(ctx->r, ctx->commits.list[ctx->commits.nr]);
1857 num_parents = commit_list_count(ctx->commits.list[ctx->commits.nr]->parents);
1858 if (num_parents > 2)
1859 ctx->num_extra_edges += num_parents - 1;
1861 ctx->commits.nr++;
1863 stop_progress(&ctx->progress);
1866 static int write_graph_chunk_base_1(struct hashfile *f,
1867 struct commit_graph *g)
1869 int num = 0;
1871 if (!g)
1872 return 0;
1874 num = write_graph_chunk_base_1(f, g->base_graph);
1875 hashwrite(f, g->oid.hash, the_hash_algo->rawsz);
1876 return num + 1;
1879 static int write_graph_chunk_base(struct hashfile *f,
1880 void *data)
1882 struct write_commit_graph_context *ctx = data;
1883 int num = write_graph_chunk_base_1(f, ctx->new_base_graph);
1885 if (num != ctx->num_commit_graphs_after - 1) {
1886 error(_("failed to write correct number of base graph ids"));
1887 return -1;
1890 return 0;
1893 static int write_commit_graph_file(struct write_commit_graph_context *ctx)
1895 uint32_t i;
1896 int fd;
1897 struct hashfile *f;
1898 struct lock_file lk = LOCK_INIT;
1899 const unsigned hashsz = the_hash_algo->rawsz;
1900 struct strbuf progress_title = STRBUF_INIT;
1901 struct chunkfile *cf;
1902 unsigned char file_hash[GIT_MAX_RAWSZ];
1904 if (ctx->split) {
1905 struct strbuf tmp_file = STRBUF_INIT;
1907 strbuf_addf(&tmp_file,
1908 "%s/info/commit-graphs/tmp_graph_XXXXXX",
1909 ctx->odb->path);
1910 ctx->graph_name = strbuf_detach(&tmp_file, NULL);
1911 } else {
1912 ctx->graph_name = get_commit_graph_filename(ctx->odb);
1915 if (safe_create_leading_directories(ctx->graph_name)) {
1916 UNLEAK(ctx->graph_name);
1917 error(_("unable to create leading directories of %s"),
1918 ctx->graph_name);
1919 return -1;
1922 if (ctx->split) {
1923 char *lock_name = get_commit_graph_chain_filename(ctx->odb);
1925 hold_lock_file_for_update_mode(&lk, lock_name,
1926 LOCK_DIE_ON_ERROR, 0444);
1927 free(lock_name);
1929 fd = git_mkstemp_mode(ctx->graph_name, 0444);
1930 if (fd < 0) {
1931 error(_("unable to create temporary graph layer"));
1932 return -1;
1935 if (adjust_shared_perm(ctx->graph_name)) {
1936 error(_("unable to adjust shared permissions for '%s'"),
1937 ctx->graph_name);
1938 return -1;
1941 f = hashfd(fd, ctx->graph_name);
1942 } else {
1943 hold_lock_file_for_update_mode(&lk, ctx->graph_name,
1944 LOCK_DIE_ON_ERROR, 0444);
1945 fd = get_lock_file_fd(&lk);
1946 f = hashfd(fd, get_lock_file_path(&lk));
1949 cf = init_chunkfile(f);
1951 add_chunk(cf, GRAPH_CHUNKID_OIDFANOUT, GRAPH_FANOUT_SIZE,
1952 write_graph_chunk_fanout);
1953 add_chunk(cf, GRAPH_CHUNKID_OIDLOOKUP, hashsz * ctx->commits.nr,
1954 write_graph_chunk_oids);
1955 add_chunk(cf, GRAPH_CHUNKID_DATA, (hashsz + 16) * ctx->commits.nr,
1956 write_graph_chunk_data);
1958 if (ctx->write_generation_data)
1959 add_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA,
1960 sizeof(uint32_t) * ctx->commits.nr,
1961 write_graph_chunk_generation_data);
1962 if (ctx->num_generation_data_overflows)
1963 add_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW,
1964 sizeof(timestamp_t) * ctx->num_generation_data_overflows,
1965 write_graph_chunk_generation_data_overflow);
1966 if (ctx->num_extra_edges)
1967 add_chunk(cf, GRAPH_CHUNKID_EXTRAEDGES,
1968 4 * ctx->num_extra_edges,
1969 write_graph_chunk_extra_edges);
1970 if (ctx->changed_paths) {
1971 add_chunk(cf, GRAPH_CHUNKID_BLOOMINDEXES,
1972 sizeof(uint32_t) * ctx->commits.nr,
1973 write_graph_chunk_bloom_indexes);
1974 add_chunk(cf, GRAPH_CHUNKID_BLOOMDATA,
1975 sizeof(uint32_t) * 3
1976 + ctx->total_bloom_filter_data_size,
1977 write_graph_chunk_bloom_data);
1979 if (ctx->num_commit_graphs_after > 1)
1980 add_chunk(cf, GRAPH_CHUNKID_BASE,
1981 hashsz * (ctx->num_commit_graphs_after - 1),
1982 write_graph_chunk_base);
1984 hashwrite_be32(f, GRAPH_SIGNATURE);
1986 hashwrite_u8(f, GRAPH_VERSION);
1987 hashwrite_u8(f, oid_version(the_hash_algo));
1988 hashwrite_u8(f, get_num_chunks(cf));
1989 hashwrite_u8(f, ctx->num_commit_graphs_after - 1);
1991 if (ctx->report_progress) {
1992 strbuf_addf(&progress_title,
1993 Q_("Writing out commit graph in %d pass",
1994 "Writing out commit graph in %d passes",
1995 get_num_chunks(cf)),
1996 get_num_chunks(cf));
1997 ctx->progress = start_delayed_progress(
1998 progress_title.buf,
1999 get_num_chunks(cf) * ctx->commits.nr);
2002 write_chunkfile(cf, ctx);
2004 stop_progress(&ctx->progress);
2005 strbuf_release(&progress_title);
2007 if (ctx->split && ctx->base_graph_name && ctx->num_commit_graphs_after > 1) {
2008 char *new_base_hash = xstrdup(oid_to_hex(&ctx->new_base_graph->oid));
2009 char *new_base_name = get_split_graph_filename(ctx->new_base_graph->odb, new_base_hash);
2011 free(ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2]);
2012 free(ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2]);
2013 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2] = new_base_name;
2014 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2] = new_base_hash;
2017 close_commit_graph(ctx->r->objects);
2018 finalize_hashfile(f, file_hash, FSYNC_COMPONENT_COMMIT_GRAPH,
2019 CSUM_HASH_IN_STREAM | CSUM_FSYNC);
2020 free_chunkfile(cf);
2022 if (ctx->split) {
2023 FILE *chainf = fdopen_lock_file(&lk, "w");
2024 char *final_graph_name;
2025 int result;
2027 close(fd);
2029 if (!chainf) {
2030 error(_("unable to open commit-graph chain file"));
2031 return -1;
2034 if (ctx->base_graph_name) {
2035 const char *dest;
2036 int idx = ctx->num_commit_graphs_after - 1;
2037 if (ctx->num_commit_graphs_after > 1)
2038 idx--;
2040 dest = ctx->commit_graph_filenames_after[idx];
2042 if (strcmp(ctx->base_graph_name, dest)) {
2043 result = rename(ctx->base_graph_name, dest);
2045 if (result) {
2046 error(_("failed to rename base commit-graph file"));
2047 return -1;
2050 } else {
2051 char *graph_name = get_commit_graph_filename(ctx->odb);
2052 unlink(graph_name);
2053 free(graph_name);
2056 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1] = xstrdup(hash_to_hex(file_hash));
2057 final_graph_name = get_split_graph_filename(ctx->odb,
2058 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1]);
2059 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 1] = final_graph_name;
2061 result = rename(ctx->graph_name, final_graph_name);
2063 for (i = 0; i < ctx->num_commit_graphs_after; i++)
2064 fprintf(get_lock_file_fp(&lk), "%s\n", ctx->commit_graph_hash_after[i]);
2066 if (result) {
2067 error(_("failed to rename temporary commit-graph file"));
2068 return -1;
2072 commit_lock_file(&lk);
2074 return 0;
2077 static void split_graph_merge_strategy(struct write_commit_graph_context *ctx)
2079 struct commit_graph *g;
2080 uint32_t num_commits;
2081 enum commit_graph_split_flags flags = COMMIT_GRAPH_SPLIT_UNSPECIFIED;
2082 uint32_t i;
2084 int max_commits = 0;
2085 int size_mult = 2;
2087 if (ctx->opts) {
2088 max_commits = ctx->opts->max_commits;
2090 if (ctx->opts->size_multiple)
2091 size_mult = ctx->opts->size_multiple;
2093 flags = ctx->opts->split_flags;
2096 g = ctx->r->objects->commit_graph;
2097 num_commits = ctx->commits.nr;
2098 if (flags == COMMIT_GRAPH_SPLIT_REPLACE)
2099 ctx->num_commit_graphs_after = 1;
2100 else
2101 ctx->num_commit_graphs_after = ctx->num_commit_graphs_before + 1;
2103 if (flags != COMMIT_GRAPH_SPLIT_MERGE_PROHIBITED &&
2104 flags != COMMIT_GRAPH_SPLIT_REPLACE) {
2105 while (g && (g->num_commits <= size_mult * num_commits ||
2106 (max_commits && num_commits > max_commits))) {
2107 if (g->odb != ctx->odb)
2108 break;
2110 num_commits += g->num_commits;
2111 g = g->base_graph;
2113 ctx->num_commit_graphs_after--;
2117 if (flags != COMMIT_GRAPH_SPLIT_REPLACE)
2118 ctx->new_base_graph = g;
2119 else if (ctx->num_commit_graphs_after != 1)
2120 BUG("split_graph_merge_strategy: num_commit_graphs_after "
2121 "should be 1 with --split=replace");
2123 if (ctx->num_commit_graphs_after == 2) {
2124 char *old_graph_name = get_commit_graph_filename(g->odb);
2126 if (!strcmp(g->filename, old_graph_name) &&
2127 g->odb != ctx->odb) {
2128 ctx->num_commit_graphs_after = 1;
2129 ctx->new_base_graph = NULL;
2132 free(old_graph_name);
2135 CALLOC_ARRAY(ctx->commit_graph_filenames_after, ctx->num_commit_graphs_after);
2136 CALLOC_ARRAY(ctx->commit_graph_hash_after, ctx->num_commit_graphs_after);
2138 for (i = 0; i < ctx->num_commit_graphs_after &&
2139 i < ctx->num_commit_graphs_before; i++)
2140 ctx->commit_graph_filenames_after[i] = xstrdup(ctx->commit_graph_filenames_before[i]);
2142 i = ctx->num_commit_graphs_before - 1;
2143 g = ctx->r->objects->commit_graph;
2145 while (g) {
2146 if (i < ctx->num_commit_graphs_after)
2147 ctx->commit_graph_hash_after[i] = xstrdup(oid_to_hex(&g->oid));
2150 * If the topmost remaining layer has generation data chunk, the
2151 * resultant layer also has generation data chunk.
2153 if (i == ctx->num_commit_graphs_after - 2)
2154 ctx->write_generation_data = !!g->chunk_generation_data;
2156 i--;
2157 g = g->base_graph;
2161 static void merge_commit_graph(struct write_commit_graph_context *ctx,
2162 struct commit_graph *g)
2164 uint32_t i;
2165 uint32_t offset = g->num_commits_in_base;
2167 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + g->num_commits, ctx->commits.alloc);
2169 for (i = 0; i < g->num_commits; i++) {
2170 struct object_id oid;
2171 struct commit *result;
2173 display_progress(ctx->progress, i + 1);
2175 load_oid_from_graph(g, i + offset, &oid);
2177 /* only add commits if they still exist in the repo */
2178 result = lookup_commit_reference_gently(ctx->r, &oid, 1);
2180 if (result) {
2181 ctx->commits.list[ctx->commits.nr] = result;
2182 ctx->commits.nr++;
2187 static int commit_compare(const void *_a, const void *_b)
2189 const struct commit *a = *(const struct commit **)_a;
2190 const struct commit *b = *(const struct commit **)_b;
2191 return oidcmp(&a->object.oid, &b->object.oid);
2194 static void sort_and_scan_merged_commits(struct write_commit_graph_context *ctx)
2196 uint32_t i, dedup_i = 0;
2198 if (ctx->report_progress)
2199 ctx->progress = start_delayed_progress(
2200 _("Scanning merged commits"),
2201 ctx->commits.nr);
2203 QSORT(ctx->commits.list, ctx->commits.nr, commit_compare);
2205 ctx->num_extra_edges = 0;
2206 for (i = 0; i < ctx->commits.nr; i++) {
2207 display_progress(ctx->progress, i + 1);
2209 if (i && oideq(&ctx->commits.list[i - 1]->object.oid,
2210 &ctx->commits.list[i]->object.oid)) {
2212 * Silently ignore duplicates. These were likely
2213 * created due to a commit appearing in multiple
2214 * layers of the chain, which is unexpected but
2215 * not invalid. We should make sure there is a
2216 * unique copy in the new layer.
2218 } else {
2219 unsigned int num_parents;
2221 ctx->commits.list[dedup_i] = ctx->commits.list[i];
2222 dedup_i++;
2224 num_parents = commit_list_count(ctx->commits.list[i]->parents);
2225 if (num_parents > 2)
2226 ctx->num_extra_edges += num_parents - 1;
2230 ctx->commits.nr = dedup_i;
2232 stop_progress(&ctx->progress);
2235 static void merge_commit_graphs(struct write_commit_graph_context *ctx)
2237 struct commit_graph *g = ctx->r->objects->commit_graph;
2238 uint32_t current_graph_number = ctx->num_commit_graphs_before;
2240 while (g && current_graph_number >= ctx->num_commit_graphs_after) {
2241 current_graph_number--;
2243 if (ctx->report_progress)
2244 ctx->progress = start_delayed_progress(_("Merging commit-graph"), 0);
2246 merge_commit_graph(ctx, g);
2247 stop_progress(&ctx->progress);
2249 g = g->base_graph;
2252 if (g) {
2253 ctx->new_base_graph = g;
2254 ctx->new_num_commits_in_base = g->num_commits + g->num_commits_in_base;
2257 if (ctx->new_base_graph)
2258 ctx->base_graph_name = xstrdup(ctx->new_base_graph->filename);
2260 sort_and_scan_merged_commits(ctx);
2263 static void mark_commit_graphs(struct write_commit_graph_context *ctx)
2265 uint32_t i;
2266 time_t now = time(NULL);
2268 for (i = ctx->num_commit_graphs_after - 1; i < ctx->num_commit_graphs_before; i++) {
2269 struct stat st;
2270 struct utimbuf updated_time;
2272 if (stat(ctx->commit_graph_filenames_before[i], &st) < 0)
2273 continue;
2275 updated_time.actime = st.st_atime;
2276 updated_time.modtime = now;
2277 utime(ctx->commit_graph_filenames_before[i], &updated_time);
2281 static void expire_commit_graphs(struct write_commit_graph_context *ctx)
2283 struct strbuf path = STRBUF_INIT;
2284 DIR *dir;
2285 struct dirent *de;
2286 size_t dirnamelen;
2287 timestamp_t expire_time = time(NULL);
2289 if (ctx->opts && ctx->opts->expire_time)
2290 expire_time = ctx->opts->expire_time;
2291 if (!ctx->split) {
2292 char *chain_file_name = get_commit_graph_chain_filename(ctx->odb);
2293 unlink(chain_file_name);
2294 free(chain_file_name);
2295 ctx->num_commit_graphs_after = 0;
2298 strbuf_addstr(&path, ctx->odb->path);
2299 strbuf_addstr(&path, "/info/commit-graphs");
2300 dir = opendir(path.buf);
2302 if (!dir)
2303 goto out;
2305 strbuf_addch(&path, '/');
2306 dirnamelen = path.len;
2307 while ((de = readdir(dir)) != NULL) {
2308 struct stat st;
2309 uint32_t i, found = 0;
2311 strbuf_setlen(&path, dirnamelen);
2312 strbuf_addstr(&path, de->d_name);
2314 if (stat(path.buf, &st) < 0)
2315 continue;
2317 if (st.st_mtime > expire_time)
2318 continue;
2319 if (path.len < 6 || strcmp(path.buf + path.len - 6, ".graph"))
2320 continue;
2322 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
2323 if (!strcmp(ctx->commit_graph_filenames_after[i],
2324 path.buf)) {
2325 found = 1;
2326 break;
2330 if (!found)
2331 unlink(path.buf);
2334 out:
2335 if(dir)
2336 closedir(dir);
2337 strbuf_release(&path);
2340 int write_commit_graph(struct object_directory *odb,
2341 const struct string_list *const pack_indexes,
2342 struct oidset *commits,
2343 enum commit_graph_write_flags flags,
2344 const struct commit_graph_opts *opts)
2346 struct repository *r = the_repository;
2347 struct write_commit_graph_context *ctx;
2348 uint32_t i;
2349 int res = 0;
2350 int replace = 0;
2351 struct bloom_filter_settings bloom_settings = DEFAULT_BLOOM_FILTER_SETTINGS;
2352 struct topo_level_slab topo_levels;
2354 prepare_repo_settings(r);
2355 if (!r->settings.core_commit_graph) {
2356 warning(_("attempting to write a commit-graph, but 'core.commitGraph' is disabled"));
2357 return 0;
2359 if (!commit_graph_compatible(r))
2360 return 0;
2362 CALLOC_ARRAY(ctx, 1);
2363 ctx->r = r;
2364 ctx->odb = odb;
2365 ctx->append = flags & COMMIT_GRAPH_WRITE_APPEND ? 1 : 0;
2366 ctx->report_progress = flags & COMMIT_GRAPH_WRITE_PROGRESS ? 1 : 0;
2367 ctx->split = flags & COMMIT_GRAPH_WRITE_SPLIT ? 1 : 0;
2368 ctx->opts = opts;
2369 ctx->total_bloom_filter_data_size = 0;
2370 ctx->write_generation_data = (get_configured_generation_version(r) == 2);
2371 ctx->num_generation_data_overflows = 0;
2373 bloom_settings.bits_per_entry = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_BITS_PER_ENTRY",
2374 bloom_settings.bits_per_entry);
2375 bloom_settings.num_hashes = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_NUM_HASHES",
2376 bloom_settings.num_hashes);
2377 bloom_settings.max_changed_paths = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS",
2378 bloom_settings.max_changed_paths);
2379 ctx->bloom_settings = &bloom_settings;
2381 init_topo_level_slab(&topo_levels);
2382 ctx->topo_levels = &topo_levels;
2384 prepare_commit_graph(ctx->r);
2385 if (ctx->r->objects->commit_graph) {
2386 struct commit_graph *g = ctx->r->objects->commit_graph;
2388 while (g) {
2389 g->topo_levels = &topo_levels;
2390 g = g->base_graph;
2394 if (flags & COMMIT_GRAPH_WRITE_BLOOM_FILTERS)
2395 ctx->changed_paths = 1;
2396 if (!(flags & COMMIT_GRAPH_NO_WRITE_BLOOM_FILTERS)) {
2397 struct commit_graph *g;
2399 g = ctx->r->objects->commit_graph;
2401 /* We have changed-paths already. Keep them in the next graph */
2402 if (g && g->chunk_bloom_data) {
2403 ctx->changed_paths = 1;
2404 ctx->bloom_settings = g->bloom_filter_settings;
2408 if (ctx->split) {
2409 struct commit_graph *g = ctx->r->objects->commit_graph;
2411 while (g) {
2412 ctx->num_commit_graphs_before++;
2413 g = g->base_graph;
2416 if (ctx->num_commit_graphs_before) {
2417 ALLOC_ARRAY(ctx->commit_graph_filenames_before, ctx->num_commit_graphs_before);
2418 i = ctx->num_commit_graphs_before;
2419 g = ctx->r->objects->commit_graph;
2421 while (g) {
2422 ctx->commit_graph_filenames_before[--i] = xstrdup(g->filename);
2423 g = g->base_graph;
2427 if (ctx->opts)
2428 replace = ctx->opts->split_flags & COMMIT_GRAPH_SPLIT_REPLACE;
2431 ctx->approx_nr_objects = repo_approximate_object_count(the_repository);
2433 if (ctx->append && ctx->r->objects->commit_graph) {
2434 struct commit_graph *g = ctx->r->objects->commit_graph;
2435 for (i = 0; i < g->num_commits; i++) {
2436 struct object_id oid;
2437 oidread(&oid, g->chunk_oid_lookup + g->hash_len * i);
2438 oid_array_append(&ctx->oids, &oid);
2442 if (pack_indexes) {
2443 ctx->order_by_pack = 1;
2444 if ((res = fill_oids_from_packs(ctx, pack_indexes)))
2445 goto cleanup;
2448 if (commits) {
2449 if ((res = fill_oids_from_commits(ctx, commits)))
2450 goto cleanup;
2453 if (!pack_indexes && !commits) {
2454 ctx->order_by_pack = 1;
2455 fill_oids_from_all_packs(ctx);
2458 close_reachable(ctx);
2460 copy_oids_to_commits(ctx);
2462 if (ctx->commits.nr >= GRAPH_EDGE_LAST_MASK) {
2463 error(_("too many commits to write graph"));
2464 res = -1;
2465 goto cleanup;
2468 if (!ctx->commits.nr && !replace)
2469 goto cleanup;
2471 if (ctx->split) {
2472 split_graph_merge_strategy(ctx);
2474 if (!replace)
2475 merge_commit_graphs(ctx);
2476 } else
2477 ctx->num_commit_graphs_after = 1;
2479 ctx->trust_generation_numbers = validate_mixed_generation_chain(ctx->r->objects->commit_graph);
2481 compute_topological_levels(ctx);
2482 if (ctx->write_generation_data)
2483 compute_generation_numbers(ctx);
2485 if (ctx->changed_paths)
2486 compute_bloom_filters(ctx);
2488 res = write_commit_graph_file(ctx);
2490 if (ctx->split)
2491 mark_commit_graphs(ctx);
2493 expire_commit_graphs(ctx);
2495 cleanup:
2496 free(ctx->graph_name);
2497 free(ctx->commits.list);
2498 oid_array_clear(&ctx->oids);
2499 clear_topo_level_slab(&topo_levels);
2501 if (ctx->commit_graph_filenames_after) {
2502 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
2503 free(ctx->commit_graph_filenames_after[i]);
2504 free(ctx->commit_graph_hash_after[i]);
2507 for (i = 0; i < ctx->num_commit_graphs_before; i++)
2508 free(ctx->commit_graph_filenames_before[i]);
2510 free(ctx->commit_graph_filenames_after);
2511 free(ctx->commit_graph_filenames_before);
2512 free(ctx->commit_graph_hash_after);
2515 free(ctx);
2517 return res;
2520 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
2521 static int verify_commit_graph_error;
2523 __attribute__((format (printf, 1, 2)))
2524 static void graph_report(const char *fmt, ...)
2526 va_list ap;
2528 verify_commit_graph_error = 1;
2529 va_start(ap, fmt);
2530 vfprintf(stderr, fmt, ap);
2531 fprintf(stderr, "\n");
2532 va_end(ap);
2535 #define GENERATION_ZERO_EXISTS 1
2536 #define GENERATION_NUMBER_EXISTS 2
2538 static int commit_graph_checksum_valid(struct commit_graph *g)
2540 return hashfile_checksum_valid(g->data, g->data_len);
2543 int verify_commit_graph(struct repository *r, struct commit_graph *g, int flags)
2545 uint32_t i, cur_fanout_pos = 0;
2546 struct object_id prev_oid, cur_oid;
2547 int generation_zero = 0;
2548 struct progress *progress = NULL;
2549 int local_error = 0;
2551 if (!g) {
2552 graph_report("no commit-graph file loaded");
2553 return 1;
2556 verify_commit_graph_error = verify_commit_graph_lite(g);
2557 if (verify_commit_graph_error)
2558 return verify_commit_graph_error;
2560 if (!commit_graph_checksum_valid(g)) {
2561 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
2562 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
2565 for (i = 0; i < g->num_commits; i++) {
2566 struct commit *graph_commit;
2568 oidread(&cur_oid, g->chunk_oid_lookup + g->hash_len * i);
2570 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
2571 graph_report(_("commit-graph has incorrect OID order: %s then %s"),
2572 oid_to_hex(&prev_oid),
2573 oid_to_hex(&cur_oid));
2575 oidcpy(&prev_oid, &cur_oid);
2577 while (cur_oid.hash[0] > cur_fanout_pos) {
2578 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2580 if (i != fanout_value)
2581 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2582 cur_fanout_pos, fanout_value, i);
2583 cur_fanout_pos++;
2586 graph_commit = lookup_commit(r, &cur_oid);
2587 if (!parse_commit_in_graph_one(r, g, graph_commit))
2588 graph_report(_("failed to parse commit %s from commit-graph"),
2589 oid_to_hex(&cur_oid));
2592 while (cur_fanout_pos < 256) {
2593 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2595 if (g->num_commits != fanout_value)
2596 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2597 cur_fanout_pos, fanout_value, i);
2599 cur_fanout_pos++;
2602 if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
2603 return verify_commit_graph_error;
2605 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
2606 progress = start_progress(_("Verifying commits in commit graph"),
2607 g->num_commits);
2609 for (i = 0; i < g->num_commits; i++) {
2610 struct commit *graph_commit, *odb_commit;
2611 struct commit_list *graph_parents, *odb_parents;
2612 timestamp_t max_generation = 0;
2613 timestamp_t generation;
2615 display_progress(progress, i + 1);
2616 oidread(&cur_oid, g->chunk_oid_lookup + g->hash_len * i);
2618 graph_commit = lookup_commit(r, &cur_oid);
2619 odb_commit = (struct commit *)create_object(r, &cur_oid, alloc_commit_node(r));
2620 if (repo_parse_commit_internal(r, odb_commit, 0, 0)) {
2621 graph_report(_("failed to parse commit %s from object database for commit-graph"),
2622 oid_to_hex(&cur_oid));
2623 continue;
2626 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
2627 get_commit_tree_oid(odb_commit)))
2628 graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
2629 oid_to_hex(&cur_oid),
2630 oid_to_hex(get_commit_tree_oid(graph_commit)),
2631 oid_to_hex(get_commit_tree_oid(odb_commit)));
2633 graph_parents = graph_commit->parents;
2634 odb_parents = odb_commit->parents;
2636 while (graph_parents) {
2637 if (!odb_parents) {
2638 graph_report(_("commit-graph parent list for commit %s is too long"),
2639 oid_to_hex(&cur_oid));
2640 break;
2643 /* parse parent in case it is in a base graph */
2644 parse_commit_in_graph_one(r, g, graph_parents->item);
2646 if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
2647 graph_report(_("commit-graph parent for %s is %s != %s"),
2648 oid_to_hex(&cur_oid),
2649 oid_to_hex(&graph_parents->item->object.oid),
2650 oid_to_hex(&odb_parents->item->object.oid));
2652 generation = commit_graph_generation(graph_parents->item);
2653 if (generation > max_generation)
2654 max_generation = generation;
2656 graph_parents = graph_parents->next;
2657 odb_parents = odb_parents->next;
2660 if (odb_parents)
2661 graph_report(_("commit-graph parent list for commit %s terminates early"),
2662 oid_to_hex(&cur_oid));
2664 if (!commit_graph_generation(graph_commit)) {
2665 if (generation_zero == GENERATION_NUMBER_EXISTS)
2666 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
2667 oid_to_hex(&cur_oid));
2668 generation_zero = GENERATION_ZERO_EXISTS;
2669 } else if (generation_zero == GENERATION_ZERO_EXISTS)
2670 graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
2671 oid_to_hex(&cur_oid));
2673 if (generation_zero == GENERATION_ZERO_EXISTS)
2674 continue;
2677 * If we are using topological level and one of our parents has
2678 * generation GENERATION_NUMBER_V1_MAX, then our generation is
2679 * also GENERATION_NUMBER_V1_MAX. Decrement to avoid extra logic
2680 * in the following condition.
2682 if (!g->read_generation_data && max_generation == GENERATION_NUMBER_V1_MAX)
2683 max_generation--;
2685 generation = commit_graph_generation(graph_commit);
2686 if (generation < max_generation + 1)
2687 graph_report(_("commit-graph generation for commit %s is %"PRItime" < %"PRItime),
2688 oid_to_hex(&cur_oid),
2689 generation,
2690 max_generation + 1);
2692 if (graph_commit->date != odb_commit->date)
2693 graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
2694 oid_to_hex(&cur_oid),
2695 graph_commit->date,
2696 odb_commit->date);
2698 stop_progress(&progress);
2700 local_error = verify_commit_graph_error;
2702 if (!(flags & COMMIT_GRAPH_VERIFY_SHALLOW) && g->base_graph)
2703 local_error |= verify_commit_graph(r, g->base_graph, flags);
2705 return local_error;
2708 void free_commit_graph(struct commit_graph *g)
2710 if (!g)
2711 return;
2712 if (g->data) {
2713 munmap((void *)g->data, g->data_len);
2714 g->data = NULL;
2716 free(g->filename);
2717 free(g->bloom_filter_settings);
2718 free(g);
2721 void disable_commit_graph(struct repository *r)
2723 r->commit_graph_disabled = 1;