Merge branch 'ab/send-email-config-fix'
[git/debian.git] / commit-graph.c
blob00614acd65d8417792e4f3060b87113f2699347e
1 #include "git-compat-util.h"
2 #include "config.h"
3 #include "lockfile.h"
4 #include "pack.h"
5 #include "packfile.h"
6 #include "commit.h"
7 #include "object.h"
8 #include "refs.h"
9 #include "revision.h"
10 #include "hash-lookup.h"
11 #include "commit-graph.h"
12 #include "object-store.h"
13 #include "alloc.h"
14 #include "hashmap.h"
15 #include "replace-object.h"
16 #include "progress.h"
17 #include "bloom.h"
18 #include "commit-slab.h"
19 #include "shallow.h"
20 #include "json-writer.h"
21 #include "trace2.h"
22 #include "chunk-format.h"
24 void git_test_write_commit_graph_or_die(void)
26 int flags = 0;
27 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0))
28 return;
30 if (git_env_bool(GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS, 0))
31 flags = COMMIT_GRAPH_WRITE_BLOOM_FILTERS;
33 if (write_commit_graph_reachable(the_repository->objects->odb,
34 flags, NULL))
35 die("failed to write commit-graph under GIT_TEST_COMMIT_GRAPH");
38 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
39 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
40 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
41 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
42 #define GRAPH_CHUNKID_GENERATION_DATA 0x47444154 /* "GDAT" */
43 #define GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW 0x47444f56 /* "GDOV" */
44 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
45 #define GRAPH_CHUNKID_BLOOMINDEXES 0x42494458 /* "BIDX" */
46 #define GRAPH_CHUNKID_BLOOMDATA 0x42444154 /* "BDAT" */
47 #define GRAPH_CHUNKID_BASE 0x42415345 /* "BASE" */
49 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
51 #define GRAPH_VERSION_1 0x1
52 #define GRAPH_VERSION GRAPH_VERSION_1
54 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
55 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
56 #define GRAPH_PARENT_NONE 0x70000000
58 #define GRAPH_LAST_EDGE 0x80000000
60 #define GRAPH_HEADER_SIZE 8
61 #define GRAPH_FANOUT_SIZE (4 * 256)
62 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * CHUNK_TOC_ENTRY_SIZE \
63 + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
65 #define CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW (1ULL << 31)
67 /* Remember to update object flag allocation in object.h */
68 #define REACHABLE (1u<<15)
70 define_commit_slab(topo_level_slab, uint32_t);
72 /* Keep track of the order in which commits are added to our list. */
73 define_commit_slab(commit_pos, int);
74 static struct commit_pos commit_pos = COMMIT_SLAB_INIT(1, commit_pos);
76 static void set_commit_pos(struct repository *r, const struct object_id *oid)
78 static int32_t max_pos;
79 struct commit *commit = lookup_commit(r, oid);
81 if (!commit)
82 return; /* should never happen, but be lenient */
84 *commit_pos_at(&commit_pos, commit) = max_pos++;
87 static int commit_pos_cmp(const void *va, const void *vb)
89 const struct commit *a = *(const struct commit **)va;
90 const struct commit *b = *(const struct commit **)vb;
91 return commit_pos_at(&commit_pos, a) -
92 commit_pos_at(&commit_pos, b);
95 define_commit_slab(commit_graph_data_slab, struct commit_graph_data);
96 static struct commit_graph_data_slab commit_graph_data_slab =
97 COMMIT_SLAB_INIT(1, commit_graph_data_slab);
99 static int get_configured_generation_version(struct repository *r)
101 int version = 2;
102 repo_config_get_int(r, "commitgraph.generationversion", &version);
103 return version;
106 uint32_t commit_graph_position(const struct commit *c)
108 struct commit_graph_data *data =
109 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
111 return data ? data->graph_pos : COMMIT_NOT_FROM_GRAPH;
114 timestamp_t commit_graph_generation(const struct commit *c)
116 struct commit_graph_data *data =
117 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
119 if (!data)
120 return GENERATION_NUMBER_INFINITY;
121 else if (data->graph_pos == COMMIT_NOT_FROM_GRAPH)
122 return GENERATION_NUMBER_INFINITY;
124 return data->generation;
127 static struct commit_graph_data *commit_graph_data_at(const struct commit *c)
129 unsigned int i, nth_slab;
130 struct commit_graph_data *data =
131 commit_graph_data_slab_peek(&commit_graph_data_slab, c);
133 if (data)
134 return data;
136 nth_slab = c->index / commit_graph_data_slab.slab_size;
137 data = commit_graph_data_slab_at(&commit_graph_data_slab, c);
140 * commit-slab initializes elements with zero, overwrite this with
141 * COMMIT_NOT_FROM_GRAPH for graph_pos.
143 * We avoid initializing generation with checking if graph position
144 * is not COMMIT_NOT_FROM_GRAPH.
146 for (i = 0; i < commit_graph_data_slab.slab_size; i++) {
147 commit_graph_data_slab.slab[nth_slab][i].graph_pos =
148 COMMIT_NOT_FROM_GRAPH;
151 return data;
155 * Should be used only while writing commit-graph as it compares
156 * generation value of commits by directly accessing commit-slab.
158 static int commit_gen_cmp(const void *va, const void *vb)
160 const struct commit *a = *(const struct commit **)va;
161 const struct commit *b = *(const struct commit **)vb;
163 const timestamp_t generation_a = commit_graph_data_at(a)->generation;
164 const timestamp_t generation_b = commit_graph_data_at(b)->generation;
165 /* lower generation commits first */
166 if (generation_a < generation_b)
167 return -1;
168 else if (generation_a > generation_b)
169 return 1;
171 /* use date as a heuristic when generations are equal */
172 if (a->date < b->date)
173 return -1;
174 else if (a->date > b->date)
175 return 1;
176 return 0;
179 char *get_commit_graph_filename(struct object_directory *obj_dir)
181 return xstrfmt("%s/info/commit-graph", obj_dir->path);
184 static char *get_split_graph_filename(struct object_directory *odb,
185 const char *oid_hex)
187 return xstrfmt("%s/info/commit-graphs/graph-%s.graph", odb->path,
188 oid_hex);
191 char *get_commit_graph_chain_filename(struct object_directory *odb)
193 return xstrfmt("%s/info/commit-graphs/commit-graph-chain", odb->path);
196 static uint8_t oid_version(void)
198 switch (hash_algo_by_ptr(the_hash_algo)) {
199 case GIT_HASH_SHA1:
200 return 1;
201 case GIT_HASH_SHA256:
202 return 2;
203 default:
204 die(_("invalid hash version"));
208 static struct commit_graph *alloc_commit_graph(void)
210 struct commit_graph *g = xcalloc(1, sizeof(*g));
212 return g;
215 extern int read_replace_refs;
217 static int commit_graph_compatible(struct repository *r)
219 if (!r->gitdir)
220 return 0;
222 if (read_replace_refs) {
223 prepare_replace_object(r);
224 if (hashmap_get_size(&r->objects->replace_map->map))
225 return 0;
228 prepare_commit_graft(r);
229 if (r->parsed_objects &&
230 (r->parsed_objects->grafts_nr || r->parsed_objects->substituted_parent))
231 return 0;
232 if (is_repository_shallow(r))
233 return 0;
235 return 1;
238 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
240 *fd = git_open(graph_file);
241 if (*fd < 0)
242 return 0;
243 if (fstat(*fd, st)) {
244 close(*fd);
245 return 0;
247 return 1;
250 struct commit_graph *load_commit_graph_one_fd_st(struct repository *r,
251 int fd, struct stat *st,
252 struct object_directory *odb)
254 void *graph_map;
255 size_t graph_size;
256 struct commit_graph *ret;
258 graph_size = xsize_t(st->st_size);
260 if (graph_size < GRAPH_MIN_SIZE) {
261 close(fd);
262 error(_("commit-graph file is too small"));
263 return NULL;
265 graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
266 close(fd);
267 ret = parse_commit_graph(r, graph_map, graph_size);
269 if (ret)
270 ret->odb = odb;
271 else
272 munmap(graph_map, graph_size);
274 return ret;
277 static int verify_commit_graph_lite(struct commit_graph *g)
280 * Basic validation shared between parse_commit_graph()
281 * which'll be called every time the graph is used, and the
282 * much more expensive verify_commit_graph() used by
283 * "commit-graph verify".
285 * There should only be very basic checks here to ensure that
286 * we don't e.g. segfault in fill_commit_in_graph(), but
287 * because this is a very hot codepath nothing that e.g. loops
288 * over g->num_commits, or runs a checksum on the commit-graph
289 * itself.
291 if (!g->chunk_oid_fanout) {
292 error("commit-graph is missing the OID Fanout chunk");
293 return 1;
295 if (!g->chunk_oid_lookup) {
296 error("commit-graph is missing the OID Lookup chunk");
297 return 1;
299 if (!g->chunk_commit_data) {
300 error("commit-graph is missing the Commit Data chunk");
301 return 1;
304 return 0;
307 static int graph_read_oid_lookup(const unsigned char *chunk_start,
308 size_t chunk_size, void *data)
310 struct commit_graph *g = data;
311 g->chunk_oid_lookup = chunk_start;
312 g->num_commits = chunk_size / g->hash_len;
313 return 0;
316 static int graph_read_bloom_data(const unsigned char *chunk_start,
317 size_t chunk_size, void *data)
319 struct commit_graph *g = data;
320 uint32_t hash_version;
321 g->chunk_bloom_data = chunk_start;
322 hash_version = get_be32(chunk_start);
324 if (hash_version != 1)
325 return 0;
327 g->bloom_filter_settings = xmalloc(sizeof(struct bloom_filter_settings));
328 g->bloom_filter_settings->hash_version = hash_version;
329 g->bloom_filter_settings->num_hashes = get_be32(chunk_start + 4);
330 g->bloom_filter_settings->bits_per_entry = get_be32(chunk_start + 8);
331 g->bloom_filter_settings->max_changed_paths = DEFAULT_BLOOM_MAX_CHANGES;
333 return 0;
336 struct commit_graph *parse_commit_graph(struct repository *r,
337 void *graph_map, size_t graph_size)
339 const unsigned char *data;
340 struct commit_graph *graph;
341 uint32_t graph_signature;
342 unsigned char graph_version, hash_version;
343 struct chunkfile *cf = NULL;
345 if (!graph_map)
346 return NULL;
348 if (graph_size < GRAPH_MIN_SIZE)
349 return NULL;
351 data = (const unsigned char *)graph_map;
353 graph_signature = get_be32(data);
354 if (graph_signature != GRAPH_SIGNATURE) {
355 error(_("commit-graph signature %X does not match signature %X"),
356 graph_signature, GRAPH_SIGNATURE);
357 return NULL;
360 graph_version = *(unsigned char*)(data + 4);
361 if (graph_version != GRAPH_VERSION) {
362 error(_("commit-graph version %X does not match version %X"),
363 graph_version, GRAPH_VERSION);
364 return NULL;
367 hash_version = *(unsigned char*)(data + 5);
368 if (hash_version != oid_version()) {
369 error(_("commit-graph hash version %X does not match version %X"),
370 hash_version, oid_version());
371 return NULL;
374 prepare_repo_settings(r);
376 graph = alloc_commit_graph();
378 graph->hash_len = the_hash_algo->rawsz;
379 graph->num_chunks = *(unsigned char*)(data + 6);
380 graph->data = graph_map;
381 graph->data_len = graph_size;
383 if (graph_size < GRAPH_HEADER_SIZE +
384 (graph->num_chunks + 1) * CHUNK_TOC_ENTRY_SIZE +
385 GRAPH_FANOUT_SIZE + the_hash_algo->rawsz) {
386 error(_("commit-graph file is too small to hold %u chunks"),
387 graph->num_chunks);
388 free(graph);
389 return NULL;
392 cf = init_chunkfile(NULL);
394 if (read_table_of_contents(cf, graph->data, graph_size,
395 GRAPH_HEADER_SIZE, graph->num_chunks))
396 goto free_and_return;
398 pair_chunk(cf, GRAPH_CHUNKID_OIDFANOUT,
399 (const unsigned char **)&graph->chunk_oid_fanout);
400 read_chunk(cf, GRAPH_CHUNKID_OIDLOOKUP, graph_read_oid_lookup, graph);
401 pair_chunk(cf, GRAPH_CHUNKID_DATA, &graph->chunk_commit_data);
402 pair_chunk(cf, GRAPH_CHUNKID_EXTRAEDGES, &graph->chunk_extra_edges);
403 pair_chunk(cf, GRAPH_CHUNKID_BASE, &graph->chunk_base_graphs);
405 if (get_configured_generation_version(r) >= 2) {
406 pair_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA,
407 &graph->chunk_generation_data);
408 pair_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW,
409 &graph->chunk_generation_data_overflow);
412 if (r->settings.commit_graph_read_changed_paths) {
413 pair_chunk(cf, GRAPH_CHUNKID_BLOOMINDEXES,
414 &graph->chunk_bloom_indexes);
415 read_chunk(cf, GRAPH_CHUNKID_BLOOMDATA,
416 graph_read_bloom_data, graph);
419 if (graph->chunk_bloom_indexes && graph->chunk_bloom_data) {
420 init_bloom_filters();
421 } else {
422 /* We need both the bloom chunks to exist together. Else ignore the data */
423 graph->chunk_bloom_indexes = NULL;
424 graph->chunk_bloom_data = NULL;
425 FREE_AND_NULL(graph->bloom_filter_settings);
428 oidread(&graph->oid, graph->data + graph->data_len - graph->hash_len);
430 if (verify_commit_graph_lite(graph))
431 goto free_and_return;
433 free_chunkfile(cf);
434 return graph;
436 free_and_return:
437 free_chunkfile(cf);
438 free(graph->bloom_filter_settings);
439 free(graph);
440 return NULL;
443 static struct commit_graph *load_commit_graph_one(struct repository *r,
444 const char *graph_file,
445 struct object_directory *odb)
448 struct stat st;
449 int fd;
450 struct commit_graph *g;
451 int open_ok = open_commit_graph(graph_file, &fd, &st);
453 if (!open_ok)
454 return NULL;
456 g = load_commit_graph_one_fd_st(r, fd, &st, odb);
458 if (g)
459 g->filename = xstrdup(graph_file);
461 return g;
464 static struct commit_graph *load_commit_graph_v1(struct repository *r,
465 struct object_directory *odb)
467 char *graph_name = get_commit_graph_filename(odb);
468 struct commit_graph *g = load_commit_graph_one(r, graph_name, odb);
469 free(graph_name);
471 return g;
474 static int add_graph_to_chain(struct commit_graph *g,
475 struct commit_graph *chain,
476 struct object_id *oids,
477 int n)
479 struct commit_graph *cur_g = chain;
481 if (n && !g->chunk_base_graphs) {
482 warning(_("commit-graph has no base graphs chunk"));
483 return 0;
486 while (n) {
487 n--;
489 if (!cur_g ||
490 !oideq(&oids[n], &cur_g->oid) ||
491 !hasheq(oids[n].hash, g->chunk_base_graphs + g->hash_len * n)) {
492 warning(_("commit-graph chain does not match"));
493 return 0;
496 cur_g = cur_g->base_graph;
499 g->base_graph = chain;
501 if (chain)
502 g->num_commits_in_base = chain->num_commits + chain->num_commits_in_base;
504 return 1;
507 static struct commit_graph *load_commit_graph_chain(struct repository *r,
508 struct object_directory *odb)
510 struct commit_graph *graph_chain = NULL;
511 struct strbuf line = STRBUF_INIT;
512 struct stat st;
513 struct object_id *oids;
514 int i = 0, valid = 1, count;
515 char *chain_name = get_commit_graph_chain_filename(odb);
516 FILE *fp;
517 int stat_res;
519 fp = fopen(chain_name, "r");
520 stat_res = stat(chain_name, &st);
521 free(chain_name);
523 if (!fp ||
524 stat_res ||
525 st.st_size <= the_hash_algo->hexsz)
526 return NULL;
528 count = st.st_size / (the_hash_algo->hexsz + 1);
529 CALLOC_ARRAY(oids, count);
531 prepare_alt_odb(r);
533 for (i = 0; i < count; i++) {
534 struct object_directory *odb;
536 if (strbuf_getline_lf(&line, fp) == EOF)
537 break;
539 if (get_oid_hex(line.buf, &oids[i])) {
540 warning(_("invalid commit-graph chain: line '%s' not a hash"),
541 line.buf);
542 valid = 0;
543 break;
546 valid = 0;
547 for (odb = r->objects->odb; odb; odb = odb->next) {
548 char *graph_name = get_split_graph_filename(odb, line.buf);
549 struct commit_graph *g = load_commit_graph_one(r, graph_name, odb);
551 free(graph_name);
553 if (g) {
554 if (add_graph_to_chain(g, graph_chain, oids, i)) {
555 graph_chain = g;
556 valid = 1;
559 break;
563 if (!valid) {
564 warning(_("unable to find all commit-graph files"));
565 break;
569 free(oids);
570 fclose(fp);
571 strbuf_release(&line);
573 return graph_chain;
577 * returns 1 if and only if all graphs in the chain have
578 * corrected commit dates stored in the generation_data chunk.
580 static int validate_mixed_generation_chain(struct commit_graph *g)
582 int read_generation_data = 1;
583 struct commit_graph *p = g;
585 while (read_generation_data && p) {
586 read_generation_data = p->read_generation_data;
587 p = p->base_graph;
590 if (read_generation_data)
591 return 1;
593 while (g) {
594 g->read_generation_data = 0;
595 g = g->base_graph;
598 return 0;
601 struct commit_graph *read_commit_graph_one(struct repository *r,
602 struct object_directory *odb)
604 struct commit_graph *g = load_commit_graph_v1(r, odb);
606 if (!g)
607 g = load_commit_graph_chain(r, odb);
609 validate_mixed_generation_chain(g);
611 return g;
614 static void prepare_commit_graph_one(struct repository *r,
615 struct object_directory *odb)
618 if (r->objects->commit_graph)
619 return;
621 r->objects->commit_graph = read_commit_graph_one(r, odb);
625 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
627 * On the first invocation, this function attempts to load the commit
628 * graph if the_repository is configured to have one.
630 static int prepare_commit_graph(struct repository *r)
632 struct object_directory *odb;
635 * This must come before the "already attempted?" check below, because
636 * we want to disable even an already-loaded graph file.
638 if (r->commit_graph_disabled)
639 return 0;
641 if (r->objects->commit_graph_attempted)
642 return !!r->objects->commit_graph;
643 r->objects->commit_graph_attempted = 1;
645 prepare_repo_settings(r);
647 if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
648 r->settings.core_commit_graph != 1)
650 * This repository is not configured to use commit graphs, so
651 * do not load one. (But report commit_graph_attempted anyway
652 * so that commit graph loading is not attempted again for this
653 * repository.)
655 return 0;
657 if (!commit_graph_compatible(r))
658 return 0;
660 prepare_alt_odb(r);
661 for (odb = r->objects->odb;
662 !r->objects->commit_graph && odb;
663 odb = odb->next)
664 prepare_commit_graph_one(r, odb);
665 return !!r->objects->commit_graph;
668 int generation_numbers_enabled(struct repository *r)
670 uint32_t first_generation;
671 struct commit_graph *g;
672 if (!prepare_commit_graph(r))
673 return 0;
675 g = r->objects->commit_graph;
677 if (!g->num_commits)
678 return 0;
680 first_generation = get_be32(g->chunk_commit_data +
681 g->hash_len + 8) >> 2;
683 return !!first_generation;
686 int corrected_commit_dates_enabled(struct repository *r)
688 struct commit_graph *g;
689 if (!prepare_commit_graph(r))
690 return 0;
692 g = r->objects->commit_graph;
694 if (!g->num_commits)
695 return 0;
697 return g->read_generation_data;
700 struct bloom_filter_settings *get_bloom_filter_settings(struct repository *r)
702 struct commit_graph *g = r->objects->commit_graph;
703 while (g) {
704 if (g->bloom_filter_settings)
705 return g->bloom_filter_settings;
706 g = g->base_graph;
708 return NULL;
711 static void close_commit_graph_one(struct commit_graph *g)
713 if (!g)
714 return;
716 close_commit_graph_one(g->base_graph);
717 free_commit_graph(g);
720 void close_commit_graph(struct raw_object_store *o)
722 close_commit_graph_one(o->commit_graph);
723 o->commit_graph = NULL;
726 static int bsearch_graph(struct commit_graph *g, const struct object_id *oid, uint32_t *pos)
728 return bsearch_hash(oid->hash, g->chunk_oid_fanout,
729 g->chunk_oid_lookup, g->hash_len, pos);
732 static void load_oid_from_graph(struct commit_graph *g,
733 uint32_t pos,
734 struct object_id *oid)
736 uint32_t lex_index;
738 while (g && pos < g->num_commits_in_base)
739 g = g->base_graph;
741 if (!g)
742 BUG("NULL commit-graph");
744 if (pos >= g->num_commits + g->num_commits_in_base)
745 die(_("invalid commit position. commit-graph is likely corrupt"));
747 lex_index = pos - g->num_commits_in_base;
749 oidread(oid, g->chunk_oid_lookup + g->hash_len * lex_index);
752 static struct commit_list **insert_parent_or_die(struct repository *r,
753 struct commit_graph *g,
754 uint32_t pos,
755 struct commit_list **pptr)
757 struct commit *c;
758 struct object_id oid;
760 if (pos >= g->num_commits + g->num_commits_in_base)
761 die("invalid parent position %"PRIu32, pos);
763 load_oid_from_graph(g, pos, &oid);
764 c = lookup_commit(r, &oid);
765 if (!c)
766 die(_("could not find commit %s"), oid_to_hex(&oid));
767 commit_graph_data_at(c)->graph_pos = pos;
768 return &commit_list_insert(c, pptr)->next;
771 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
773 const unsigned char *commit_data;
774 struct commit_graph_data *graph_data;
775 uint32_t lex_index, offset_pos;
776 uint64_t date_high, date_low, offset;
778 while (pos < g->num_commits_in_base)
779 g = g->base_graph;
781 if (pos >= g->num_commits + g->num_commits_in_base)
782 die(_("invalid commit position. commit-graph is likely corrupt"));
784 lex_index = pos - g->num_commits_in_base;
785 commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * lex_index;
787 graph_data = commit_graph_data_at(item);
788 graph_data->graph_pos = pos;
790 date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
791 date_low = get_be32(commit_data + g->hash_len + 12);
792 item->date = (timestamp_t)((date_high << 32) | date_low);
794 if (g->read_generation_data) {
795 offset = (timestamp_t)get_be32(g->chunk_generation_data + sizeof(uint32_t) * lex_index);
797 if (offset & CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW) {
798 if (!g->chunk_generation_data_overflow)
799 die(_("commit-graph requires overflow generation data but has none"));
801 offset_pos = offset ^ CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW;
802 graph_data->generation = get_be64(g->chunk_generation_data_overflow + 8 * offset_pos);
803 } else
804 graph_data->generation = item->date + offset;
805 } else
806 graph_data->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
808 if (g->topo_levels)
809 *topo_level_slab_at(g->topo_levels, item) = get_be32(commit_data + g->hash_len + 8) >> 2;
812 static inline void set_commit_tree(struct commit *c, struct tree *t)
814 c->maybe_tree = t;
817 static int fill_commit_in_graph(struct repository *r,
818 struct commit *item,
819 struct commit_graph *g, uint32_t pos)
821 uint32_t edge_value;
822 uint32_t *parent_data_ptr;
823 struct commit_list **pptr;
824 const unsigned char *commit_data;
825 uint32_t lex_index;
827 while (pos < g->num_commits_in_base)
828 g = g->base_graph;
830 fill_commit_graph_info(item, g, pos);
832 lex_index = pos - g->num_commits_in_base;
833 commit_data = g->chunk_commit_data + (g->hash_len + 16) * lex_index;
835 item->object.parsed = 1;
837 set_commit_tree(item, NULL);
839 pptr = &item->parents;
841 edge_value = get_be32(commit_data + g->hash_len);
842 if (edge_value == GRAPH_PARENT_NONE)
843 return 1;
844 pptr = insert_parent_or_die(r, g, edge_value, pptr);
846 edge_value = get_be32(commit_data + g->hash_len + 4);
847 if (edge_value == GRAPH_PARENT_NONE)
848 return 1;
849 if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
850 pptr = insert_parent_or_die(r, g, edge_value, pptr);
851 return 1;
854 parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
855 4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
856 do {
857 edge_value = get_be32(parent_data_ptr);
858 pptr = insert_parent_or_die(r, g,
859 edge_value & GRAPH_EDGE_LAST_MASK,
860 pptr);
861 parent_data_ptr++;
862 } while (!(edge_value & GRAPH_LAST_EDGE));
864 return 1;
867 static int search_commit_pos_in_graph(const struct object_id *id, struct commit_graph *g, uint32_t *pos)
869 struct commit_graph *cur_g = g;
870 uint32_t lex_index;
872 while (cur_g && !bsearch_graph(cur_g, id, &lex_index))
873 cur_g = cur_g->base_graph;
875 if (cur_g) {
876 *pos = lex_index + cur_g->num_commits_in_base;
877 return 1;
880 return 0;
883 static int find_commit_pos_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
885 uint32_t graph_pos = commit_graph_position(item);
886 if (graph_pos != COMMIT_NOT_FROM_GRAPH) {
887 *pos = graph_pos;
888 return 1;
889 } else {
890 return search_commit_pos_in_graph(&item->object.oid, g, pos);
894 struct commit *lookup_commit_in_graph(struct repository *repo, const struct object_id *id)
896 struct commit *commit;
897 uint32_t pos;
899 if (!repo->objects->commit_graph)
900 return NULL;
901 if (!search_commit_pos_in_graph(id, repo->objects->commit_graph, &pos))
902 return NULL;
903 if (!repo_has_object_file(repo, id))
904 return NULL;
906 commit = lookup_commit(repo, id);
907 if (!commit)
908 return NULL;
909 if (commit->object.parsed)
910 return commit;
912 if (!fill_commit_in_graph(repo, commit, repo->objects->commit_graph, pos))
913 return NULL;
915 return commit;
918 static int parse_commit_in_graph_one(struct repository *r,
919 struct commit_graph *g,
920 struct commit *item)
922 uint32_t pos;
924 if (item->object.parsed)
925 return 1;
927 if (find_commit_pos_in_graph(item, g, &pos))
928 return fill_commit_in_graph(r, item, g, pos);
930 return 0;
933 int parse_commit_in_graph(struct repository *r, struct commit *item)
935 static int checked_env = 0;
937 if (!checked_env &&
938 git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_PARSE, 0))
939 die("dying as requested by the '%s' variable on commit-graph parse!",
940 GIT_TEST_COMMIT_GRAPH_DIE_ON_PARSE);
941 checked_env = 1;
943 if (!prepare_commit_graph(r))
944 return 0;
945 return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
948 void load_commit_graph_info(struct repository *r, struct commit *item)
950 uint32_t pos;
951 if (!prepare_commit_graph(r))
952 return;
953 if (find_commit_pos_in_graph(item, r->objects->commit_graph, &pos))
954 fill_commit_graph_info(item, r->objects->commit_graph, pos);
957 static struct tree *load_tree_for_commit(struct repository *r,
958 struct commit_graph *g,
959 struct commit *c)
961 struct object_id oid;
962 const unsigned char *commit_data;
963 uint32_t graph_pos = commit_graph_position(c);
965 while (graph_pos < g->num_commits_in_base)
966 g = g->base_graph;
968 commit_data = g->chunk_commit_data +
969 GRAPH_DATA_WIDTH * (graph_pos - g->num_commits_in_base);
971 oidread(&oid, commit_data);
972 set_commit_tree(c, lookup_tree(r, &oid));
974 return c->maybe_tree;
977 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
978 struct commit_graph *g,
979 const struct commit *c)
981 if (c->maybe_tree)
982 return c->maybe_tree;
983 if (commit_graph_position(c) == COMMIT_NOT_FROM_GRAPH)
984 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
986 return load_tree_for_commit(r, g, (struct commit *)c);
989 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
991 return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
994 struct packed_commit_list {
995 struct commit **list;
996 size_t nr;
997 size_t alloc;
1000 struct write_commit_graph_context {
1001 struct repository *r;
1002 struct object_directory *odb;
1003 char *graph_name;
1004 struct oid_array oids;
1005 struct packed_commit_list commits;
1006 int num_extra_edges;
1007 int num_generation_data_overflows;
1008 unsigned long approx_nr_objects;
1009 struct progress *progress;
1010 int progress_done;
1011 uint64_t progress_cnt;
1013 char *base_graph_name;
1014 int num_commit_graphs_before;
1015 int num_commit_graphs_after;
1016 char **commit_graph_filenames_before;
1017 char **commit_graph_filenames_after;
1018 char **commit_graph_hash_after;
1019 uint32_t new_num_commits_in_base;
1020 struct commit_graph *new_base_graph;
1022 unsigned append:1,
1023 report_progress:1,
1024 split:1,
1025 changed_paths:1,
1026 order_by_pack:1,
1027 write_generation_data:1,
1028 trust_generation_numbers:1;
1030 struct topo_level_slab *topo_levels;
1031 const struct commit_graph_opts *opts;
1032 size_t total_bloom_filter_data_size;
1033 const struct bloom_filter_settings *bloom_settings;
1035 int count_bloom_filter_computed;
1036 int count_bloom_filter_not_computed;
1037 int count_bloom_filter_trunc_empty;
1038 int count_bloom_filter_trunc_large;
1041 static int write_graph_chunk_fanout(struct hashfile *f,
1042 void *data)
1044 struct write_commit_graph_context *ctx = data;
1045 int i, count = 0;
1046 struct commit **list = ctx->commits.list;
1049 * Write the first-level table (the list is sorted,
1050 * but we use a 256-entry lookup to be able to avoid
1051 * having to do eight extra binary search iterations).
1053 for (i = 0; i < 256; i++) {
1054 while (count < ctx->commits.nr) {
1055 if ((*list)->object.oid.hash[0] != i)
1056 break;
1057 display_progress(ctx->progress, ++ctx->progress_cnt);
1058 count++;
1059 list++;
1062 hashwrite_be32(f, count);
1065 return 0;
1068 static int write_graph_chunk_oids(struct hashfile *f,
1069 void *data)
1071 struct write_commit_graph_context *ctx = data;
1072 struct commit **list = ctx->commits.list;
1073 int count;
1074 for (count = 0; count < ctx->commits.nr; count++, list++) {
1075 display_progress(ctx->progress, ++ctx->progress_cnt);
1076 hashwrite(f, (*list)->object.oid.hash, the_hash_algo->rawsz);
1079 return 0;
1082 static const struct object_id *commit_to_oid(size_t index, const void *table)
1084 const struct commit * const *commits = table;
1085 return &commits[index]->object.oid;
1088 static int write_graph_chunk_data(struct hashfile *f,
1089 void *data)
1091 struct write_commit_graph_context *ctx = data;
1092 struct commit **list = ctx->commits.list;
1093 struct commit **last = ctx->commits.list + ctx->commits.nr;
1094 uint32_t num_extra_edges = 0;
1096 while (list < last) {
1097 struct commit_list *parent;
1098 struct object_id *tree;
1099 int edge_value;
1100 uint32_t packedDate[2];
1101 display_progress(ctx->progress, ++ctx->progress_cnt);
1103 if (repo_parse_commit_no_graph(ctx->r, *list))
1104 die(_("unable to parse commit %s"),
1105 oid_to_hex(&(*list)->object.oid));
1106 tree = get_commit_tree_oid(*list);
1107 hashwrite(f, tree->hash, the_hash_algo->rawsz);
1109 parent = (*list)->parents;
1111 if (!parent)
1112 edge_value = GRAPH_PARENT_NONE;
1113 else {
1114 edge_value = oid_pos(&parent->item->object.oid,
1115 ctx->commits.list,
1116 ctx->commits.nr,
1117 commit_to_oid);
1119 if (edge_value >= 0)
1120 edge_value += ctx->new_num_commits_in_base;
1121 else if (ctx->new_base_graph) {
1122 uint32_t pos;
1123 if (find_commit_pos_in_graph(parent->item,
1124 ctx->new_base_graph,
1125 &pos))
1126 edge_value = pos;
1129 if (edge_value < 0)
1130 BUG("missing parent %s for commit %s",
1131 oid_to_hex(&parent->item->object.oid),
1132 oid_to_hex(&(*list)->object.oid));
1135 hashwrite_be32(f, edge_value);
1137 if (parent)
1138 parent = parent->next;
1140 if (!parent)
1141 edge_value = GRAPH_PARENT_NONE;
1142 else if (parent->next)
1143 edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
1144 else {
1145 edge_value = oid_pos(&parent->item->object.oid,
1146 ctx->commits.list,
1147 ctx->commits.nr,
1148 commit_to_oid);
1150 if (edge_value >= 0)
1151 edge_value += ctx->new_num_commits_in_base;
1152 else if (ctx->new_base_graph) {
1153 uint32_t pos;
1154 if (find_commit_pos_in_graph(parent->item,
1155 ctx->new_base_graph,
1156 &pos))
1157 edge_value = pos;
1160 if (edge_value < 0)
1161 BUG("missing parent %s for commit %s",
1162 oid_to_hex(&parent->item->object.oid),
1163 oid_to_hex(&(*list)->object.oid));
1166 hashwrite_be32(f, edge_value);
1168 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
1169 do {
1170 num_extra_edges++;
1171 parent = parent->next;
1172 } while (parent);
1175 if (sizeof((*list)->date) > 4)
1176 packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
1177 else
1178 packedDate[0] = 0;
1180 packedDate[0] |= htonl(*topo_level_slab_at(ctx->topo_levels, *list) << 2);
1182 packedDate[1] = htonl((*list)->date);
1183 hashwrite(f, packedDate, 8);
1185 list++;
1188 return 0;
1191 static int write_graph_chunk_generation_data(struct hashfile *f,
1192 void *data)
1194 struct write_commit_graph_context *ctx = data;
1195 int i, num_generation_data_overflows = 0;
1197 for (i = 0; i < ctx->commits.nr; i++) {
1198 struct commit *c = ctx->commits.list[i];
1199 timestamp_t offset;
1200 repo_parse_commit(ctx->r, c);
1201 offset = commit_graph_data_at(c)->generation - c->date;
1202 display_progress(ctx->progress, ++ctx->progress_cnt);
1204 if (offset > GENERATION_NUMBER_V2_OFFSET_MAX) {
1205 offset = CORRECTED_COMMIT_DATE_OFFSET_OVERFLOW | num_generation_data_overflows;
1206 num_generation_data_overflows++;
1209 hashwrite_be32(f, offset);
1212 return 0;
1215 static int write_graph_chunk_generation_data_overflow(struct hashfile *f,
1216 void *data)
1218 struct write_commit_graph_context *ctx = data;
1219 int i;
1220 for (i = 0; i < ctx->commits.nr; i++) {
1221 struct commit *c = ctx->commits.list[i];
1222 timestamp_t offset = commit_graph_data_at(c)->generation - c->date;
1223 display_progress(ctx->progress, ++ctx->progress_cnt);
1225 if (offset > GENERATION_NUMBER_V2_OFFSET_MAX) {
1226 hashwrite_be32(f, offset >> 32);
1227 hashwrite_be32(f, (uint32_t) offset);
1231 return 0;
1234 static int write_graph_chunk_extra_edges(struct hashfile *f,
1235 void *data)
1237 struct write_commit_graph_context *ctx = data;
1238 struct commit **list = ctx->commits.list;
1239 struct commit **last = ctx->commits.list + ctx->commits.nr;
1240 struct commit_list *parent;
1242 while (list < last) {
1243 int num_parents = 0;
1245 display_progress(ctx->progress, ++ctx->progress_cnt);
1247 for (parent = (*list)->parents; num_parents < 3 && parent;
1248 parent = parent->next)
1249 num_parents++;
1251 if (num_parents <= 2) {
1252 list++;
1253 continue;
1256 /* Since num_parents > 2, this initializer is safe. */
1257 for (parent = (*list)->parents->next; parent; parent = parent->next) {
1258 int edge_value = oid_pos(&parent->item->object.oid,
1259 ctx->commits.list,
1260 ctx->commits.nr,
1261 commit_to_oid);
1263 if (edge_value >= 0)
1264 edge_value += ctx->new_num_commits_in_base;
1265 else if (ctx->new_base_graph) {
1266 uint32_t pos;
1267 if (find_commit_pos_in_graph(parent->item,
1268 ctx->new_base_graph,
1269 &pos))
1270 edge_value = pos;
1273 if (edge_value < 0)
1274 BUG("missing parent %s for commit %s",
1275 oid_to_hex(&parent->item->object.oid),
1276 oid_to_hex(&(*list)->object.oid));
1277 else if (!parent->next)
1278 edge_value |= GRAPH_LAST_EDGE;
1280 hashwrite_be32(f, edge_value);
1283 list++;
1286 return 0;
1289 static int write_graph_chunk_bloom_indexes(struct hashfile *f,
1290 void *data)
1292 struct write_commit_graph_context *ctx = data;
1293 struct commit **list = ctx->commits.list;
1294 struct commit **last = ctx->commits.list + ctx->commits.nr;
1295 uint32_t cur_pos = 0;
1297 while (list < last) {
1298 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list);
1299 size_t len = filter ? filter->len : 0;
1300 cur_pos += len;
1301 display_progress(ctx->progress, ++ctx->progress_cnt);
1302 hashwrite_be32(f, cur_pos);
1303 list++;
1306 return 0;
1309 static void trace2_bloom_filter_settings(struct write_commit_graph_context *ctx)
1311 struct json_writer jw = JSON_WRITER_INIT;
1313 jw_object_begin(&jw, 0);
1314 jw_object_intmax(&jw, "hash_version", ctx->bloom_settings->hash_version);
1315 jw_object_intmax(&jw, "num_hashes", ctx->bloom_settings->num_hashes);
1316 jw_object_intmax(&jw, "bits_per_entry", ctx->bloom_settings->bits_per_entry);
1317 jw_object_intmax(&jw, "max_changed_paths", ctx->bloom_settings->max_changed_paths);
1318 jw_end(&jw);
1320 trace2_data_json("bloom", ctx->r, "settings", &jw);
1322 jw_release(&jw);
1325 static int write_graph_chunk_bloom_data(struct hashfile *f,
1326 void *data)
1328 struct write_commit_graph_context *ctx = data;
1329 struct commit **list = ctx->commits.list;
1330 struct commit **last = ctx->commits.list + ctx->commits.nr;
1332 trace2_bloom_filter_settings(ctx);
1334 hashwrite_be32(f, ctx->bloom_settings->hash_version);
1335 hashwrite_be32(f, ctx->bloom_settings->num_hashes);
1336 hashwrite_be32(f, ctx->bloom_settings->bits_per_entry);
1338 while (list < last) {
1339 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list);
1340 size_t len = filter ? filter->len : 0;
1342 display_progress(ctx->progress, ++ctx->progress_cnt);
1343 if (len)
1344 hashwrite(f, filter->data, len * sizeof(unsigned char));
1345 list++;
1348 return 0;
1351 static int add_packed_commits(const struct object_id *oid,
1352 struct packed_git *pack,
1353 uint32_t pos,
1354 void *data)
1356 struct write_commit_graph_context *ctx = (struct write_commit_graph_context*)data;
1357 enum object_type type;
1358 off_t offset = nth_packed_object_offset(pack, pos);
1359 struct object_info oi = OBJECT_INFO_INIT;
1361 if (ctx->progress)
1362 display_progress(ctx->progress, ++ctx->progress_done);
1364 oi.typep = &type;
1365 if (packed_object_info(ctx->r, pack, offset, &oi) < 0)
1366 die(_("unable to get type of object %s"), oid_to_hex(oid));
1368 if (type != OBJ_COMMIT)
1369 return 0;
1371 oid_array_append(&ctx->oids, oid);
1372 set_commit_pos(ctx->r, oid);
1374 return 0;
1377 static void add_missing_parents(struct write_commit_graph_context *ctx, struct commit *commit)
1379 struct commit_list *parent;
1380 for (parent = commit->parents; parent; parent = parent->next) {
1381 if (!(parent->item->object.flags & REACHABLE)) {
1382 oid_array_append(&ctx->oids, &parent->item->object.oid);
1383 parent->item->object.flags |= REACHABLE;
1388 static void close_reachable(struct write_commit_graph_context *ctx)
1390 int i;
1391 struct commit *commit;
1392 enum commit_graph_split_flags flags = ctx->opts ?
1393 ctx->opts->split_flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1395 if (ctx->report_progress)
1396 ctx->progress = start_delayed_progress(
1397 _("Loading known commits in commit graph"),
1398 ctx->oids.nr);
1399 for (i = 0; i < ctx->oids.nr; i++) {
1400 display_progress(ctx->progress, i + 1);
1401 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1402 if (commit)
1403 commit->object.flags |= REACHABLE;
1405 stop_progress(&ctx->progress);
1408 * As this loop runs, ctx->oids.nr may grow, but not more
1409 * than the number of missing commits in the reachable
1410 * closure.
1412 if (ctx->report_progress)
1413 ctx->progress = start_delayed_progress(
1414 _("Expanding reachable commits in commit graph"),
1416 for (i = 0; i < ctx->oids.nr; i++) {
1417 display_progress(ctx->progress, i + 1);
1418 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1420 if (!commit)
1421 continue;
1422 if (ctx->split) {
1423 if ((!repo_parse_commit(ctx->r, commit) &&
1424 commit_graph_position(commit) == COMMIT_NOT_FROM_GRAPH) ||
1425 flags == COMMIT_GRAPH_SPLIT_REPLACE)
1426 add_missing_parents(ctx, commit);
1427 } else if (!repo_parse_commit_no_graph(ctx->r, commit))
1428 add_missing_parents(ctx, commit);
1430 stop_progress(&ctx->progress);
1432 if (ctx->report_progress)
1433 ctx->progress = start_delayed_progress(
1434 _("Clearing commit marks in commit graph"),
1435 ctx->oids.nr);
1436 for (i = 0; i < ctx->oids.nr; i++) {
1437 display_progress(ctx->progress, i + 1);
1438 commit = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1440 if (commit)
1441 commit->object.flags &= ~REACHABLE;
1443 stop_progress(&ctx->progress);
1446 static void compute_topological_levels(struct write_commit_graph_context *ctx)
1448 int i;
1449 struct commit_list *list = NULL;
1451 if (ctx->report_progress)
1452 ctx->progress = start_delayed_progress(
1453 _("Computing commit graph topological levels"),
1454 ctx->commits.nr);
1455 for (i = 0; i < ctx->commits.nr; i++) {
1456 struct commit *c = ctx->commits.list[i];
1457 uint32_t level;
1459 repo_parse_commit(ctx->r, c);
1460 level = *topo_level_slab_at(ctx->topo_levels, c);
1462 display_progress(ctx->progress, i + 1);
1463 if (level != GENERATION_NUMBER_ZERO)
1464 continue;
1466 commit_list_insert(c, &list);
1467 while (list) {
1468 struct commit *current = list->item;
1469 struct commit_list *parent;
1470 int all_parents_computed = 1;
1471 uint32_t max_level = 0;
1473 for (parent = current->parents; parent; parent = parent->next) {
1474 repo_parse_commit(ctx->r, parent->item);
1475 level = *topo_level_slab_at(ctx->topo_levels, parent->item);
1477 if (level == GENERATION_NUMBER_ZERO) {
1478 all_parents_computed = 0;
1479 commit_list_insert(parent->item, &list);
1480 break;
1483 if (level > max_level)
1484 max_level = level;
1487 if (all_parents_computed) {
1488 pop_commit(&list);
1490 if (max_level > GENERATION_NUMBER_V1_MAX - 1)
1491 max_level = GENERATION_NUMBER_V1_MAX - 1;
1492 *topo_level_slab_at(ctx->topo_levels, current) = max_level + 1;
1496 stop_progress(&ctx->progress);
1499 static void compute_generation_numbers(struct write_commit_graph_context *ctx)
1501 int i;
1502 struct commit_list *list = NULL;
1504 if (ctx->report_progress)
1505 ctx->progress = start_delayed_progress(
1506 _("Computing commit graph generation numbers"),
1507 ctx->commits.nr);
1509 if (!ctx->trust_generation_numbers) {
1510 for (i = 0; i < ctx->commits.nr; i++) {
1511 struct commit *c = ctx->commits.list[i];
1512 repo_parse_commit(ctx->r, c);
1513 commit_graph_data_at(c)->generation = GENERATION_NUMBER_ZERO;
1517 for (i = 0; i < ctx->commits.nr; i++) {
1518 struct commit *c = ctx->commits.list[i];
1519 timestamp_t corrected_commit_date;
1521 repo_parse_commit(ctx->r, c);
1522 corrected_commit_date = commit_graph_data_at(c)->generation;
1524 display_progress(ctx->progress, i + 1);
1525 if (corrected_commit_date != GENERATION_NUMBER_ZERO)
1526 continue;
1528 commit_list_insert(c, &list);
1529 while (list) {
1530 struct commit *current = list->item;
1531 struct commit_list *parent;
1532 int all_parents_computed = 1;
1533 timestamp_t max_corrected_commit_date = 0;
1535 for (parent = current->parents; parent; parent = parent->next) {
1536 repo_parse_commit(ctx->r, parent->item);
1537 corrected_commit_date = commit_graph_data_at(parent->item)->generation;
1539 if (corrected_commit_date == GENERATION_NUMBER_ZERO) {
1540 all_parents_computed = 0;
1541 commit_list_insert(parent->item, &list);
1542 break;
1545 if (corrected_commit_date > max_corrected_commit_date)
1546 max_corrected_commit_date = corrected_commit_date;
1549 if (all_parents_computed) {
1550 pop_commit(&list);
1552 if (current->date && current->date > max_corrected_commit_date)
1553 max_corrected_commit_date = current->date - 1;
1554 commit_graph_data_at(current)->generation = max_corrected_commit_date + 1;
1556 if (commit_graph_data_at(current)->generation - current->date > GENERATION_NUMBER_V2_OFFSET_MAX)
1557 ctx->num_generation_data_overflows++;
1561 stop_progress(&ctx->progress);
1564 static void trace2_bloom_filter_write_statistics(struct write_commit_graph_context *ctx)
1566 trace2_data_intmax("commit-graph", ctx->r, "filter-computed",
1567 ctx->count_bloom_filter_computed);
1568 trace2_data_intmax("commit-graph", ctx->r, "filter-not-computed",
1569 ctx->count_bloom_filter_not_computed);
1570 trace2_data_intmax("commit-graph", ctx->r, "filter-trunc-empty",
1571 ctx->count_bloom_filter_trunc_empty);
1572 trace2_data_intmax("commit-graph", ctx->r, "filter-trunc-large",
1573 ctx->count_bloom_filter_trunc_large);
1576 static void compute_bloom_filters(struct write_commit_graph_context *ctx)
1578 int i;
1579 struct progress *progress = NULL;
1580 struct commit **sorted_commits;
1581 int max_new_filters;
1583 init_bloom_filters();
1585 if (ctx->report_progress)
1586 progress = start_delayed_progress(
1587 _("Computing commit changed paths Bloom filters"),
1588 ctx->commits.nr);
1590 ALLOC_ARRAY(sorted_commits, ctx->commits.nr);
1591 COPY_ARRAY(sorted_commits, ctx->commits.list, ctx->commits.nr);
1593 if (ctx->order_by_pack)
1594 QSORT(sorted_commits, ctx->commits.nr, commit_pos_cmp);
1595 else
1596 QSORT(sorted_commits, ctx->commits.nr, commit_gen_cmp);
1598 max_new_filters = ctx->opts && ctx->opts->max_new_filters >= 0 ?
1599 ctx->opts->max_new_filters : ctx->commits.nr;
1601 for (i = 0; i < ctx->commits.nr; i++) {
1602 enum bloom_filter_computed computed = 0;
1603 struct commit *c = sorted_commits[i];
1604 struct bloom_filter *filter = get_or_compute_bloom_filter(
1605 ctx->r,
1607 ctx->count_bloom_filter_computed < max_new_filters,
1608 ctx->bloom_settings,
1609 &computed);
1610 if (computed & BLOOM_COMPUTED) {
1611 ctx->count_bloom_filter_computed++;
1612 if (computed & BLOOM_TRUNC_EMPTY)
1613 ctx->count_bloom_filter_trunc_empty++;
1614 if (computed & BLOOM_TRUNC_LARGE)
1615 ctx->count_bloom_filter_trunc_large++;
1616 } else if (computed & BLOOM_NOT_COMPUTED)
1617 ctx->count_bloom_filter_not_computed++;
1618 ctx->total_bloom_filter_data_size += filter
1619 ? sizeof(unsigned char) * filter->len : 0;
1620 display_progress(progress, i + 1);
1623 if (trace2_is_enabled())
1624 trace2_bloom_filter_write_statistics(ctx);
1626 free(sorted_commits);
1627 stop_progress(&progress);
1630 struct refs_cb_data {
1631 struct oidset *commits;
1632 struct progress *progress;
1635 static int add_ref_to_set(const char *refname,
1636 const struct object_id *oid,
1637 int flags, void *cb_data)
1639 struct object_id peeled;
1640 struct refs_cb_data *data = (struct refs_cb_data *)cb_data;
1642 if (!peel_iterated_oid(oid, &peeled))
1643 oid = &peeled;
1644 if (oid_object_info(the_repository, oid, NULL) == OBJ_COMMIT)
1645 oidset_insert(data->commits, oid);
1647 display_progress(data->progress, oidset_size(data->commits));
1649 return 0;
1652 int write_commit_graph_reachable(struct object_directory *odb,
1653 enum commit_graph_write_flags flags,
1654 const struct commit_graph_opts *opts)
1656 struct oidset commits = OIDSET_INIT;
1657 struct refs_cb_data data;
1658 int result;
1660 memset(&data, 0, sizeof(data));
1661 data.commits = &commits;
1662 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
1663 data.progress = start_delayed_progress(
1664 _("Collecting referenced commits"), 0);
1666 for_each_ref(add_ref_to_set, &data);
1668 stop_progress(&data.progress);
1670 result = write_commit_graph(odb, NULL, &commits,
1671 flags, opts);
1673 oidset_clear(&commits);
1674 return result;
1677 static int fill_oids_from_packs(struct write_commit_graph_context *ctx,
1678 struct string_list *pack_indexes)
1680 uint32_t i;
1681 struct strbuf progress_title = STRBUF_INIT;
1682 struct strbuf packname = STRBUF_INIT;
1683 int dirlen;
1685 strbuf_addf(&packname, "%s/pack/", ctx->odb->path);
1686 dirlen = packname.len;
1687 if (ctx->report_progress) {
1688 strbuf_addf(&progress_title,
1689 Q_("Finding commits for commit graph in %d pack",
1690 "Finding commits for commit graph in %d packs",
1691 pack_indexes->nr),
1692 pack_indexes->nr);
1693 ctx->progress = start_delayed_progress(progress_title.buf, 0);
1694 ctx->progress_done = 0;
1696 for (i = 0; i < pack_indexes->nr; i++) {
1697 struct packed_git *p;
1698 strbuf_setlen(&packname, dirlen);
1699 strbuf_addstr(&packname, pack_indexes->items[i].string);
1700 p = add_packed_git(packname.buf, packname.len, 1);
1701 if (!p) {
1702 error(_("error adding pack %s"), packname.buf);
1703 return -1;
1705 if (open_pack_index(p)) {
1706 error(_("error opening index for %s"), packname.buf);
1707 return -1;
1709 for_each_object_in_pack(p, add_packed_commits, ctx,
1710 FOR_EACH_OBJECT_PACK_ORDER);
1711 close_pack(p);
1712 free(p);
1715 stop_progress(&ctx->progress);
1716 strbuf_release(&progress_title);
1717 strbuf_release(&packname);
1719 return 0;
1722 static int fill_oids_from_commits(struct write_commit_graph_context *ctx,
1723 struct oidset *commits)
1725 struct oidset_iter iter;
1726 struct object_id *oid;
1728 if (!oidset_size(commits))
1729 return 0;
1731 oidset_iter_init(commits, &iter);
1732 while ((oid = oidset_iter_next(&iter))) {
1733 oid_array_append(&ctx->oids, oid);
1736 return 0;
1739 static void fill_oids_from_all_packs(struct write_commit_graph_context *ctx)
1741 if (ctx->report_progress)
1742 ctx->progress = start_delayed_progress(
1743 _("Finding commits for commit graph among packed objects"),
1744 ctx->approx_nr_objects);
1745 for_each_packed_object(add_packed_commits, ctx,
1746 FOR_EACH_OBJECT_PACK_ORDER);
1747 if (ctx->progress_done < ctx->approx_nr_objects)
1748 display_progress(ctx->progress, ctx->approx_nr_objects);
1749 stop_progress(&ctx->progress);
1752 static void copy_oids_to_commits(struct write_commit_graph_context *ctx)
1754 uint32_t i;
1755 enum commit_graph_split_flags flags = ctx->opts ?
1756 ctx->opts->split_flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1758 ctx->num_extra_edges = 0;
1759 if (ctx->report_progress)
1760 ctx->progress = start_delayed_progress(
1761 _("Finding extra edges in commit graph"),
1762 ctx->oids.nr);
1763 oid_array_sort(&ctx->oids);
1764 for (i = 0; i < ctx->oids.nr; i = oid_array_next_unique(&ctx->oids, i)) {
1765 unsigned int num_parents;
1767 display_progress(ctx->progress, i + 1);
1769 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + 1, ctx->commits.alloc);
1770 ctx->commits.list[ctx->commits.nr] = lookup_commit(ctx->r, &ctx->oids.oid[i]);
1772 if (ctx->split && flags != COMMIT_GRAPH_SPLIT_REPLACE &&
1773 commit_graph_position(ctx->commits.list[ctx->commits.nr]) != COMMIT_NOT_FROM_GRAPH)
1774 continue;
1776 if (ctx->split && flags == COMMIT_GRAPH_SPLIT_REPLACE)
1777 repo_parse_commit(ctx->r, ctx->commits.list[ctx->commits.nr]);
1778 else
1779 repo_parse_commit_no_graph(ctx->r, ctx->commits.list[ctx->commits.nr]);
1781 num_parents = commit_list_count(ctx->commits.list[ctx->commits.nr]->parents);
1782 if (num_parents > 2)
1783 ctx->num_extra_edges += num_parents - 1;
1785 ctx->commits.nr++;
1787 stop_progress(&ctx->progress);
1790 static int write_graph_chunk_base_1(struct hashfile *f,
1791 struct commit_graph *g)
1793 int num = 0;
1795 if (!g)
1796 return 0;
1798 num = write_graph_chunk_base_1(f, g->base_graph);
1799 hashwrite(f, g->oid.hash, the_hash_algo->rawsz);
1800 return num + 1;
1803 static int write_graph_chunk_base(struct hashfile *f,
1804 void *data)
1806 struct write_commit_graph_context *ctx = data;
1807 int num = write_graph_chunk_base_1(f, ctx->new_base_graph);
1809 if (num != ctx->num_commit_graphs_after - 1) {
1810 error(_("failed to write correct number of base graph ids"));
1811 return -1;
1814 return 0;
1817 static int write_commit_graph_file(struct write_commit_graph_context *ctx)
1819 uint32_t i;
1820 int fd;
1821 struct hashfile *f;
1822 struct lock_file lk = LOCK_INIT;
1823 const unsigned hashsz = the_hash_algo->rawsz;
1824 struct strbuf progress_title = STRBUF_INIT;
1825 struct chunkfile *cf;
1826 unsigned char file_hash[GIT_MAX_RAWSZ];
1828 if (ctx->split) {
1829 struct strbuf tmp_file = STRBUF_INIT;
1831 strbuf_addf(&tmp_file,
1832 "%s/info/commit-graphs/tmp_graph_XXXXXX",
1833 ctx->odb->path);
1834 ctx->graph_name = strbuf_detach(&tmp_file, NULL);
1835 } else {
1836 ctx->graph_name = get_commit_graph_filename(ctx->odb);
1839 if (safe_create_leading_directories(ctx->graph_name)) {
1840 UNLEAK(ctx->graph_name);
1841 error(_("unable to create leading directories of %s"),
1842 ctx->graph_name);
1843 return -1;
1846 if (ctx->split) {
1847 char *lock_name = get_commit_graph_chain_filename(ctx->odb);
1849 hold_lock_file_for_update_mode(&lk, lock_name,
1850 LOCK_DIE_ON_ERROR, 0444);
1852 fd = git_mkstemp_mode(ctx->graph_name, 0444);
1853 if (fd < 0) {
1854 error(_("unable to create temporary graph layer"));
1855 return -1;
1858 if (adjust_shared_perm(ctx->graph_name)) {
1859 error(_("unable to adjust shared permissions for '%s'"),
1860 ctx->graph_name);
1861 return -1;
1864 f = hashfd(fd, ctx->graph_name);
1865 } else {
1866 hold_lock_file_for_update_mode(&lk, ctx->graph_name,
1867 LOCK_DIE_ON_ERROR, 0444);
1868 fd = get_lock_file_fd(&lk);
1869 f = hashfd(fd, get_lock_file_path(&lk));
1872 cf = init_chunkfile(f);
1874 add_chunk(cf, GRAPH_CHUNKID_OIDFANOUT, GRAPH_FANOUT_SIZE,
1875 write_graph_chunk_fanout);
1876 add_chunk(cf, GRAPH_CHUNKID_OIDLOOKUP, hashsz * ctx->commits.nr,
1877 write_graph_chunk_oids);
1878 add_chunk(cf, GRAPH_CHUNKID_DATA, (hashsz + 16) * ctx->commits.nr,
1879 write_graph_chunk_data);
1881 if (ctx->write_generation_data)
1882 add_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA,
1883 sizeof(uint32_t) * ctx->commits.nr,
1884 write_graph_chunk_generation_data);
1885 if (ctx->num_generation_data_overflows)
1886 add_chunk(cf, GRAPH_CHUNKID_GENERATION_DATA_OVERFLOW,
1887 sizeof(timestamp_t) * ctx->num_generation_data_overflows,
1888 write_graph_chunk_generation_data_overflow);
1889 if (ctx->num_extra_edges)
1890 add_chunk(cf, GRAPH_CHUNKID_EXTRAEDGES,
1891 4 * ctx->num_extra_edges,
1892 write_graph_chunk_extra_edges);
1893 if (ctx->changed_paths) {
1894 add_chunk(cf, GRAPH_CHUNKID_BLOOMINDEXES,
1895 sizeof(uint32_t) * ctx->commits.nr,
1896 write_graph_chunk_bloom_indexes);
1897 add_chunk(cf, GRAPH_CHUNKID_BLOOMDATA,
1898 sizeof(uint32_t) * 3
1899 + ctx->total_bloom_filter_data_size,
1900 write_graph_chunk_bloom_data);
1902 if (ctx->num_commit_graphs_after > 1)
1903 add_chunk(cf, GRAPH_CHUNKID_BASE,
1904 hashsz * (ctx->num_commit_graphs_after - 1),
1905 write_graph_chunk_base);
1907 hashwrite_be32(f, GRAPH_SIGNATURE);
1909 hashwrite_u8(f, GRAPH_VERSION);
1910 hashwrite_u8(f, oid_version());
1911 hashwrite_u8(f, get_num_chunks(cf));
1912 hashwrite_u8(f, ctx->num_commit_graphs_after - 1);
1914 if (ctx->report_progress) {
1915 strbuf_addf(&progress_title,
1916 Q_("Writing out commit graph in %d pass",
1917 "Writing out commit graph in %d passes",
1918 get_num_chunks(cf)),
1919 get_num_chunks(cf));
1920 ctx->progress = start_delayed_progress(
1921 progress_title.buf,
1922 get_num_chunks(cf) * ctx->commits.nr);
1925 write_chunkfile(cf, ctx);
1927 stop_progress(&ctx->progress);
1928 strbuf_release(&progress_title);
1930 if (ctx->split && ctx->base_graph_name && ctx->num_commit_graphs_after > 1) {
1931 char *new_base_hash = xstrdup(oid_to_hex(&ctx->new_base_graph->oid));
1932 char *new_base_name = get_split_graph_filename(ctx->new_base_graph->odb, new_base_hash);
1934 free(ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2]);
1935 free(ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2]);
1936 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2] = new_base_name;
1937 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2] = new_base_hash;
1940 close_commit_graph(ctx->r->objects);
1941 finalize_hashfile(f, file_hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
1942 free_chunkfile(cf);
1944 if (ctx->split) {
1945 FILE *chainf = fdopen_lock_file(&lk, "w");
1946 char *final_graph_name;
1947 int result;
1949 close(fd);
1951 if (!chainf) {
1952 error(_("unable to open commit-graph chain file"));
1953 return -1;
1956 if (ctx->base_graph_name) {
1957 const char *dest;
1958 int idx = ctx->num_commit_graphs_after - 1;
1959 if (ctx->num_commit_graphs_after > 1)
1960 idx--;
1962 dest = ctx->commit_graph_filenames_after[idx];
1964 if (strcmp(ctx->base_graph_name, dest)) {
1965 result = rename(ctx->base_graph_name, dest);
1967 if (result) {
1968 error(_("failed to rename base commit-graph file"));
1969 return -1;
1972 } else {
1973 char *graph_name = get_commit_graph_filename(ctx->odb);
1974 unlink(graph_name);
1977 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1] = xstrdup(hash_to_hex(file_hash));
1978 final_graph_name = get_split_graph_filename(ctx->odb,
1979 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1]);
1980 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 1] = final_graph_name;
1982 result = rename(ctx->graph_name, final_graph_name);
1984 for (i = 0; i < ctx->num_commit_graphs_after; i++)
1985 fprintf(get_lock_file_fp(&lk), "%s\n", ctx->commit_graph_hash_after[i]);
1987 if (result) {
1988 error(_("failed to rename temporary commit-graph file"));
1989 return -1;
1993 commit_lock_file(&lk);
1995 return 0;
1998 static void split_graph_merge_strategy(struct write_commit_graph_context *ctx)
2000 struct commit_graph *g;
2001 uint32_t num_commits;
2002 enum commit_graph_split_flags flags = COMMIT_GRAPH_SPLIT_UNSPECIFIED;
2003 uint32_t i;
2005 int max_commits = 0;
2006 int size_mult = 2;
2008 if (ctx->opts) {
2009 max_commits = ctx->opts->max_commits;
2011 if (ctx->opts->size_multiple)
2012 size_mult = ctx->opts->size_multiple;
2014 flags = ctx->opts->split_flags;
2017 g = ctx->r->objects->commit_graph;
2018 num_commits = ctx->commits.nr;
2019 if (flags == COMMIT_GRAPH_SPLIT_REPLACE)
2020 ctx->num_commit_graphs_after = 1;
2021 else
2022 ctx->num_commit_graphs_after = ctx->num_commit_graphs_before + 1;
2024 if (flags != COMMIT_GRAPH_SPLIT_MERGE_PROHIBITED &&
2025 flags != COMMIT_GRAPH_SPLIT_REPLACE) {
2026 while (g && (g->num_commits <= size_mult * num_commits ||
2027 (max_commits && num_commits > max_commits))) {
2028 if (g->odb != ctx->odb)
2029 break;
2031 num_commits += g->num_commits;
2032 g = g->base_graph;
2034 ctx->num_commit_graphs_after--;
2038 if (flags != COMMIT_GRAPH_SPLIT_REPLACE)
2039 ctx->new_base_graph = g;
2040 else if (ctx->num_commit_graphs_after != 1)
2041 BUG("split_graph_merge_strategy: num_commit_graphs_after "
2042 "should be 1 with --split=replace");
2044 if (ctx->num_commit_graphs_after == 2) {
2045 char *old_graph_name = get_commit_graph_filename(g->odb);
2047 if (!strcmp(g->filename, old_graph_name) &&
2048 g->odb != ctx->odb) {
2049 ctx->num_commit_graphs_after = 1;
2050 ctx->new_base_graph = NULL;
2053 free(old_graph_name);
2056 CALLOC_ARRAY(ctx->commit_graph_filenames_after, ctx->num_commit_graphs_after);
2057 CALLOC_ARRAY(ctx->commit_graph_hash_after, ctx->num_commit_graphs_after);
2059 for (i = 0; i < ctx->num_commit_graphs_after &&
2060 i < ctx->num_commit_graphs_before; i++)
2061 ctx->commit_graph_filenames_after[i] = xstrdup(ctx->commit_graph_filenames_before[i]);
2063 i = ctx->num_commit_graphs_before - 1;
2064 g = ctx->r->objects->commit_graph;
2066 while (g) {
2067 if (i < ctx->num_commit_graphs_after)
2068 ctx->commit_graph_hash_after[i] = xstrdup(oid_to_hex(&g->oid));
2071 * If the topmost remaining layer has generation data chunk, the
2072 * resultant layer also has generation data chunk.
2074 if (i == ctx->num_commit_graphs_after - 2)
2075 ctx->write_generation_data = !!g->chunk_generation_data;
2077 i--;
2078 g = g->base_graph;
2082 static void merge_commit_graph(struct write_commit_graph_context *ctx,
2083 struct commit_graph *g)
2085 uint32_t i;
2086 uint32_t offset = g->num_commits_in_base;
2088 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + g->num_commits, ctx->commits.alloc);
2090 for (i = 0; i < g->num_commits; i++) {
2091 struct object_id oid;
2092 struct commit *result;
2094 display_progress(ctx->progress, i + 1);
2096 load_oid_from_graph(g, i + offset, &oid);
2098 /* only add commits if they still exist in the repo */
2099 result = lookup_commit_reference_gently(ctx->r, &oid, 1);
2101 if (result) {
2102 ctx->commits.list[ctx->commits.nr] = result;
2103 ctx->commits.nr++;
2108 static int commit_compare(const void *_a, const void *_b)
2110 const struct commit *a = *(const struct commit **)_a;
2111 const struct commit *b = *(const struct commit **)_b;
2112 return oidcmp(&a->object.oid, &b->object.oid);
2115 static void sort_and_scan_merged_commits(struct write_commit_graph_context *ctx)
2117 uint32_t i, dedup_i = 0;
2119 if (ctx->report_progress)
2120 ctx->progress = start_delayed_progress(
2121 _("Scanning merged commits"),
2122 ctx->commits.nr);
2124 QSORT(ctx->commits.list, ctx->commits.nr, commit_compare);
2126 ctx->num_extra_edges = 0;
2127 for (i = 0; i < ctx->commits.nr; i++) {
2128 display_progress(ctx->progress, i);
2130 if (i && oideq(&ctx->commits.list[i - 1]->object.oid,
2131 &ctx->commits.list[i]->object.oid)) {
2133 * Silently ignore duplicates. These were likely
2134 * created due to a commit appearing in multiple
2135 * layers of the chain, which is unexpected but
2136 * not invalid. We should make sure there is a
2137 * unique copy in the new layer.
2139 } else {
2140 unsigned int num_parents;
2142 ctx->commits.list[dedup_i] = ctx->commits.list[i];
2143 dedup_i++;
2145 num_parents = commit_list_count(ctx->commits.list[i]->parents);
2146 if (num_parents > 2)
2147 ctx->num_extra_edges += num_parents - 1;
2151 ctx->commits.nr = dedup_i;
2153 stop_progress(&ctx->progress);
2156 static void merge_commit_graphs(struct write_commit_graph_context *ctx)
2158 struct commit_graph *g = ctx->r->objects->commit_graph;
2159 uint32_t current_graph_number = ctx->num_commit_graphs_before;
2161 while (g && current_graph_number >= ctx->num_commit_graphs_after) {
2162 current_graph_number--;
2164 if (ctx->report_progress)
2165 ctx->progress = start_delayed_progress(_("Merging commit-graph"), 0);
2167 merge_commit_graph(ctx, g);
2168 stop_progress(&ctx->progress);
2170 g = g->base_graph;
2173 if (g) {
2174 ctx->new_base_graph = g;
2175 ctx->new_num_commits_in_base = g->num_commits + g->num_commits_in_base;
2178 if (ctx->new_base_graph)
2179 ctx->base_graph_name = xstrdup(ctx->new_base_graph->filename);
2181 sort_and_scan_merged_commits(ctx);
2184 static void mark_commit_graphs(struct write_commit_graph_context *ctx)
2186 uint32_t i;
2187 time_t now = time(NULL);
2189 for (i = ctx->num_commit_graphs_after - 1; i < ctx->num_commit_graphs_before; i++) {
2190 struct stat st;
2191 struct utimbuf updated_time;
2193 stat(ctx->commit_graph_filenames_before[i], &st);
2195 updated_time.actime = st.st_atime;
2196 updated_time.modtime = now;
2197 utime(ctx->commit_graph_filenames_before[i], &updated_time);
2201 static void expire_commit_graphs(struct write_commit_graph_context *ctx)
2203 struct strbuf path = STRBUF_INIT;
2204 DIR *dir;
2205 struct dirent *de;
2206 size_t dirnamelen;
2207 timestamp_t expire_time = time(NULL);
2209 if (ctx->opts && ctx->opts->expire_time)
2210 expire_time = ctx->opts->expire_time;
2211 if (!ctx->split) {
2212 char *chain_file_name = get_commit_graph_chain_filename(ctx->odb);
2213 unlink(chain_file_name);
2214 free(chain_file_name);
2215 ctx->num_commit_graphs_after = 0;
2218 strbuf_addstr(&path, ctx->odb->path);
2219 strbuf_addstr(&path, "/info/commit-graphs");
2220 dir = opendir(path.buf);
2222 if (!dir)
2223 goto out;
2225 strbuf_addch(&path, '/');
2226 dirnamelen = path.len;
2227 while ((de = readdir(dir)) != NULL) {
2228 struct stat st;
2229 uint32_t i, found = 0;
2231 strbuf_setlen(&path, dirnamelen);
2232 strbuf_addstr(&path, de->d_name);
2234 stat(path.buf, &st);
2236 if (st.st_mtime > expire_time)
2237 continue;
2238 if (path.len < 6 || strcmp(path.buf + path.len - 6, ".graph"))
2239 continue;
2241 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
2242 if (!strcmp(ctx->commit_graph_filenames_after[i],
2243 path.buf)) {
2244 found = 1;
2245 break;
2249 if (!found)
2250 unlink(path.buf);
2253 out:
2254 strbuf_release(&path);
2257 int write_commit_graph(struct object_directory *odb,
2258 struct string_list *pack_indexes,
2259 struct oidset *commits,
2260 enum commit_graph_write_flags flags,
2261 const struct commit_graph_opts *opts)
2263 struct repository *r = the_repository;
2264 struct write_commit_graph_context *ctx;
2265 uint32_t i;
2266 int res = 0;
2267 int replace = 0;
2268 struct bloom_filter_settings bloom_settings = DEFAULT_BLOOM_FILTER_SETTINGS;
2269 struct topo_level_slab topo_levels;
2271 prepare_repo_settings(r);
2272 if (!r->settings.core_commit_graph) {
2273 warning(_("attempting to write a commit-graph, but 'core.commitGraph' is disabled"));
2274 return 0;
2276 if (!commit_graph_compatible(r))
2277 return 0;
2279 CALLOC_ARRAY(ctx, 1);
2280 ctx->r = r;
2281 ctx->odb = odb;
2282 ctx->append = flags & COMMIT_GRAPH_WRITE_APPEND ? 1 : 0;
2283 ctx->report_progress = flags & COMMIT_GRAPH_WRITE_PROGRESS ? 1 : 0;
2284 ctx->split = flags & COMMIT_GRAPH_WRITE_SPLIT ? 1 : 0;
2285 ctx->opts = opts;
2286 ctx->total_bloom_filter_data_size = 0;
2287 ctx->write_generation_data = (get_configured_generation_version(r) == 2);
2288 ctx->num_generation_data_overflows = 0;
2290 bloom_settings.bits_per_entry = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_BITS_PER_ENTRY",
2291 bloom_settings.bits_per_entry);
2292 bloom_settings.num_hashes = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_NUM_HASHES",
2293 bloom_settings.num_hashes);
2294 bloom_settings.max_changed_paths = git_env_ulong("GIT_TEST_BLOOM_SETTINGS_MAX_CHANGED_PATHS",
2295 bloom_settings.max_changed_paths);
2296 ctx->bloom_settings = &bloom_settings;
2298 init_topo_level_slab(&topo_levels);
2299 ctx->topo_levels = &topo_levels;
2301 prepare_commit_graph(ctx->r);
2302 if (ctx->r->objects->commit_graph) {
2303 struct commit_graph *g = ctx->r->objects->commit_graph;
2305 while (g) {
2306 g->topo_levels = &topo_levels;
2307 g = g->base_graph;
2311 if (flags & COMMIT_GRAPH_WRITE_BLOOM_FILTERS)
2312 ctx->changed_paths = 1;
2313 if (!(flags & COMMIT_GRAPH_NO_WRITE_BLOOM_FILTERS)) {
2314 struct commit_graph *g;
2316 g = ctx->r->objects->commit_graph;
2318 /* We have changed-paths already. Keep them in the next graph */
2319 if (g && g->chunk_bloom_data) {
2320 ctx->changed_paths = 1;
2321 ctx->bloom_settings = g->bloom_filter_settings;
2325 if (ctx->split) {
2326 struct commit_graph *g = ctx->r->objects->commit_graph;
2328 while (g) {
2329 ctx->num_commit_graphs_before++;
2330 g = g->base_graph;
2333 if (ctx->num_commit_graphs_before) {
2334 ALLOC_ARRAY(ctx->commit_graph_filenames_before, ctx->num_commit_graphs_before);
2335 i = ctx->num_commit_graphs_before;
2336 g = ctx->r->objects->commit_graph;
2338 while (g) {
2339 ctx->commit_graph_filenames_before[--i] = xstrdup(g->filename);
2340 g = g->base_graph;
2344 if (ctx->opts)
2345 replace = ctx->opts->split_flags & COMMIT_GRAPH_SPLIT_REPLACE;
2348 ctx->approx_nr_objects = approximate_object_count();
2350 if (ctx->append && ctx->r->objects->commit_graph) {
2351 struct commit_graph *g = ctx->r->objects->commit_graph;
2352 for (i = 0; i < g->num_commits; i++) {
2353 struct object_id oid;
2354 oidread(&oid, g->chunk_oid_lookup + g->hash_len * i);
2355 oid_array_append(&ctx->oids, &oid);
2359 if (pack_indexes) {
2360 ctx->order_by_pack = 1;
2361 if ((res = fill_oids_from_packs(ctx, pack_indexes)))
2362 goto cleanup;
2365 if (commits) {
2366 if ((res = fill_oids_from_commits(ctx, commits)))
2367 goto cleanup;
2370 if (!pack_indexes && !commits) {
2371 ctx->order_by_pack = 1;
2372 fill_oids_from_all_packs(ctx);
2375 close_reachable(ctx);
2377 copy_oids_to_commits(ctx);
2379 if (ctx->commits.nr >= GRAPH_EDGE_LAST_MASK) {
2380 error(_("too many commits to write graph"));
2381 res = -1;
2382 goto cleanup;
2385 if (!ctx->commits.nr && !replace)
2386 goto cleanup;
2388 if (ctx->split) {
2389 split_graph_merge_strategy(ctx);
2391 if (!replace)
2392 merge_commit_graphs(ctx);
2393 } else
2394 ctx->num_commit_graphs_after = 1;
2396 ctx->trust_generation_numbers = validate_mixed_generation_chain(ctx->r->objects->commit_graph);
2398 compute_topological_levels(ctx);
2399 if (ctx->write_generation_data)
2400 compute_generation_numbers(ctx);
2402 if (ctx->changed_paths)
2403 compute_bloom_filters(ctx);
2405 res = write_commit_graph_file(ctx);
2407 if (ctx->split)
2408 mark_commit_graphs(ctx);
2410 expire_commit_graphs(ctx);
2412 cleanup:
2413 free(ctx->graph_name);
2414 free(ctx->commits.list);
2415 oid_array_clear(&ctx->oids);
2416 clear_topo_level_slab(&topo_levels);
2418 if (ctx->commit_graph_filenames_after) {
2419 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
2420 free(ctx->commit_graph_filenames_after[i]);
2421 free(ctx->commit_graph_hash_after[i]);
2424 for (i = 0; i < ctx->num_commit_graphs_before; i++)
2425 free(ctx->commit_graph_filenames_before[i]);
2427 free(ctx->commit_graph_filenames_after);
2428 free(ctx->commit_graph_filenames_before);
2429 free(ctx->commit_graph_hash_after);
2432 free(ctx);
2434 return res;
2437 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
2438 static int verify_commit_graph_error;
2440 __attribute__((format (printf, 1, 2)))
2441 static void graph_report(const char *fmt, ...)
2443 va_list ap;
2445 verify_commit_graph_error = 1;
2446 va_start(ap, fmt);
2447 vfprintf(stderr, fmt, ap);
2448 fprintf(stderr, "\n");
2449 va_end(ap);
2452 #define GENERATION_ZERO_EXISTS 1
2453 #define GENERATION_NUMBER_EXISTS 2
2455 static int commit_graph_checksum_valid(struct commit_graph *g)
2457 return hashfile_checksum_valid(g->data, g->data_len);
2460 int verify_commit_graph(struct repository *r, struct commit_graph *g, int flags)
2462 uint32_t i, cur_fanout_pos = 0;
2463 struct object_id prev_oid, cur_oid;
2464 int generation_zero = 0;
2465 struct progress *progress = NULL;
2466 int local_error = 0;
2468 if (!g) {
2469 graph_report("no commit-graph file loaded");
2470 return 1;
2473 verify_commit_graph_error = verify_commit_graph_lite(g);
2474 if (verify_commit_graph_error)
2475 return verify_commit_graph_error;
2477 if (!commit_graph_checksum_valid(g)) {
2478 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
2479 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
2482 for (i = 0; i < g->num_commits; i++) {
2483 struct commit *graph_commit;
2485 oidread(&cur_oid, g->chunk_oid_lookup + g->hash_len * i);
2487 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
2488 graph_report(_("commit-graph has incorrect OID order: %s then %s"),
2489 oid_to_hex(&prev_oid),
2490 oid_to_hex(&cur_oid));
2492 oidcpy(&prev_oid, &cur_oid);
2494 while (cur_oid.hash[0] > cur_fanout_pos) {
2495 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2497 if (i != fanout_value)
2498 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2499 cur_fanout_pos, fanout_value, i);
2500 cur_fanout_pos++;
2503 graph_commit = lookup_commit(r, &cur_oid);
2504 if (!parse_commit_in_graph_one(r, g, graph_commit))
2505 graph_report(_("failed to parse commit %s from commit-graph"),
2506 oid_to_hex(&cur_oid));
2509 while (cur_fanout_pos < 256) {
2510 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2512 if (g->num_commits != fanout_value)
2513 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2514 cur_fanout_pos, fanout_value, i);
2516 cur_fanout_pos++;
2519 if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
2520 return verify_commit_graph_error;
2522 if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
2523 progress = start_progress(_("Verifying commits in commit graph"),
2524 g->num_commits);
2526 for (i = 0; i < g->num_commits; i++) {
2527 struct commit *graph_commit, *odb_commit;
2528 struct commit_list *graph_parents, *odb_parents;
2529 timestamp_t max_generation = 0;
2530 timestamp_t generation;
2532 display_progress(progress, i + 1);
2533 oidread(&cur_oid, g->chunk_oid_lookup + g->hash_len * i);
2535 graph_commit = lookup_commit(r, &cur_oid);
2536 odb_commit = (struct commit *)create_object(r, &cur_oid, alloc_commit_node(r));
2537 if (parse_commit_internal(odb_commit, 0, 0)) {
2538 graph_report(_("failed to parse commit %s from object database for commit-graph"),
2539 oid_to_hex(&cur_oid));
2540 continue;
2543 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
2544 get_commit_tree_oid(odb_commit)))
2545 graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
2546 oid_to_hex(&cur_oid),
2547 oid_to_hex(get_commit_tree_oid(graph_commit)),
2548 oid_to_hex(get_commit_tree_oid(odb_commit)));
2550 graph_parents = graph_commit->parents;
2551 odb_parents = odb_commit->parents;
2553 while (graph_parents) {
2554 if (odb_parents == NULL) {
2555 graph_report(_("commit-graph parent list for commit %s is too long"),
2556 oid_to_hex(&cur_oid));
2557 break;
2560 /* parse parent in case it is in a base graph */
2561 parse_commit_in_graph_one(r, g, graph_parents->item);
2563 if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
2564 graph_report(_("commit-graph parent for %s is %s != %s"),
2565 oid_to_hex(&cur_oid),
2566 oid_to_hex(&graph_parents->item->object.oid),
2567 oid_to_hex(&odb_parents->item->object.oid));
2569 generation = commit_graph_generation(graph_parents->item);
2570 if (generation > max_generation)
2571 max_generation = generation;
2573 graph_parents = graph_parents->next;
2574 odb_parents = odb_parents->next;
2577 if (odb_parents != NULL)
2578 graph_report(_("commit-graph parent list for commit %s terminates early"),
2579 oid_to_hex(&cur_oid));
2581 if (!commit_graph_generation(graph_commit)) {
2582 if (generation_zero == GENERATION_NUMBER_EXISTS)
2583 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
2584 oid_to_hex(&cur_oid));
2585 generation_zero = GENERATION_ZERO_EXISTS;
2586 } else if (generation_zero == GENERATION_ZERO_EXISTS)
2587 graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
2588 oid_to_hex(&cur_oid));
2590 if (generation_zero == GENERATION_ZERO_EXISTS)
2591 continue;
2594 * If we are using topological level and one of our parents has
2595 * generation GENERATION_NUMBER_V1_MAX, then our generation is
2596 * also GENERATION_NUMBER_V1_MAX. Decrement to avoid extra logic
2597 * in the following condition.
2599 if (!g->read_generation_data && max_generation == GENERATION_NUMBER_V1_MAX)
2600 max_generation--;
2602 generation = commit_graph_generation(graph_commit);
2603 if (generation < max_generation + 1)
2604 graph_report(_("commit-graph generation for commit %s is %"PRItime" < %"PRItime),
2605 oid_to_hex(&cur_oid),
2606 generation,
2607 max_generation + 1);
2609 if (graph_commit->date != odb_commit->date)
2610 graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
2611 oid_to_hex(&cur_oid),
2612 graph_commit->date,
2613 odb_commit->date);
2615 stop_progress(&progress);
2617 local_error = verify_commit_graph_error;
2619 if (!(flags & COMMIT_GRAPH_VERIFY_SHALLOW) && g->base_graph)
2620 local_error |= verify_commit_graph(r, g->base_graph, flags);
2622 return local_error;
2625 void free_commit_graph(struct commit_graph *g)
2627 if (!g)
2628 return;
2629 if (g->data) {
2630 munmap((void *)g->data, g->data_len);
2631 g->data = NULL;
2633 free(g->filename);
2634 free(g->bloom_filter_settings);
2635 free(g);
2638 void disable_commit_graph(struct repository *r)
2640 r->commit_graph_disabled = 1;