commit-graph.c: show progress of finding reachable commits
[git] / commit-graph.c
1 #include "cache.h"
2 #include "config.h"
3 #include "dir.h"
4 #include "git-compat-util.h"
5 #include "lockfile.h"
6 #include "pack.h"
7 #include "packfile.h"
8 #include "commit.h"
9 #include "object.h"
10 #include "refs.h"
11 #include "revision.h"
12 #include "sha1-lookup.h"
13 #include "commit-graph.h"
14 #include "object-store.h"
15 #include "alloc.h"
16 #include "hashmap.h"
17 #include "replace-object.h"
18 #include "progress.h"
19 #include "bloom.h"
20 #include "commit-slab.h"
21
22 void git_test_write_commit_graph_or_die(void)
23 {
24         int flags = 0;
25         if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0))
26                 return;
27
28         if (git_env_bool(GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS, 0))
29                 flags = COMMIT_GRAPH_WRITE_BLOOM_FILTERS;
30
31         if (write_commit_graph_reachable(the_repository->objects->odb,
32                                          flags, NULL))
33                 die("failed to write commit-graph under GIT_TEST_COMMIT_GRAPH");
34 }
35
36 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
37 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
38 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
39 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
40 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
41 #define GRAPH_CHUNKID_BLOOMINDEXES 0x42494458 /* "BIDX" */
42 #define GRAPH_CHUNKID_BLOOMDATA 0x42444154 /* "BDAT" */
43 #define GRAPH_CHUNKID_BASE 0x42415345 /* "BASE" */
44 #define MAX_NUM_CHUNKS 7
45
46 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
47
48 #define GRAPH_VERSION_1 0x1
49 #define GRAPH_VERSION GRAPH_VERSION_1
50
51 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
52 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
53 #define GRAPH_PARENT_NONE 0x70000000
54
55 #define GRAPH_LAST_EDGE 0x80000000
56
57 #define GRAPH_HEADER_SIZE 8
58 #define GRAPH_FANOUT_SIZE (4 * 256)
59 #define GRAPH_CHUNKLOOKUP_WIDTH 12
60 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
61                         + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
62
63 /* Remember to update object flag allocation in object.h */
64 #define REACHABLE       (1u<<15)
65
66 /* Keep track of the order in which commits are added to our list. */
67 define_commit_slab(commit_pos, int);
68 static struct commit_pos commit_pos = COMMIT_SLAB_INIT(1, commit_pos);
69
70 static void set_commit_pos(struct repository *r, const struct object_id *oid)
71 {
72         static int32_t max_pos;
73         struct commit *commit = lookup_commit(r, oid);
74
75         if (!commit)
76                 return; /* should never happen, but be lenient */
77
78         *commit_pos_at(&commit_pos, commit) = max_pos++;
79 }
80
81 static int commit_pos_cmp(const void *va, const void *vb)
82 {
83         const struct commit *a = *(const struct commit **)va;
84         const struct commit *b = *(const struct commit **)vb;
85         return commit_pos_at(&commit_pos, a) -
86                commit_pos_at(&commit_pos, b);
87 }
88
89 static int commit_gen_cmp(const void *va, const void *vb)
90 {
91         const struct commit *a = *(const struct commit **)va;
92         const struct commit *b = *(const struct commit **)vb;
93
94         /* lower generation commits first */
95         if (a->generation < b->generation)
96                 return -1;
97         else if (a->generation > b->generation)
98                 return 1;
99
100         /* use date as a heuristic when generations are equal */
101         if (a->date < b->date)
102                 return -1;
103         else if (a->date > b->date)
104                 return 1;
105         return 0;
106 }
107
108 char *get_commit_graph_filename(struct object_directory *obj_dir)
109 {
110         return xstrfmt("%s/info/commit-graph", obj_dir->path);
111 }
112
113 static char *get_split_graph_filename(struct object_directory *odb,
114                                       const char *oid_hex)
115 {
116         return xstrfmt("%s/info/commit-graphs/graph-%s.graph", odb->path,
117                        oid_hex);
118 }
119
120 static char *get_chain_filename(struct object_directory *odb)
121 {
122         return xstrfmt("%s/info/commit-graphs/commit-graph-chain", odb->path);
123 }
124
125 static uint8_t oid_version(void)
126 {
127         return 1;
128 }
129
130 static struct commit_graph *alloc_commit_graph(void)
131 {
132         struct commit_graph *g = xcalloc(1, sizeof(*g));
133
134         return g;
135 }
136
137 extern int read_replace_refs;
138
139 static int commit_graph_compatible(struct repository *r)
140 {
141         if (!r->gitdir)
142                 return 0;
143
144         if (read_replace_refs) {
145                 prepare_replace_object(r);
146                 if (hashmap_get_size(&r->objects->replace_map->map))
147                         return 0;
148         }
149
150         prepare_commit_graft(r);
151         if (r->parsed_objects && r->parsed_objects->grafts_nr)
152                 return 0;
153         if (is_repository_shallow(r))
154                 return 0;
155
156         return 1;
157 }
158
159 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
160 {
161         *fd = git_open(graph_file);
162         if (*fd < 0)
163                 return 0;
164         if (fstat(*fd, st)) {
165                 close(*fd);
166                 return 0;
167         }
168         return 1;
169 }
170
171 struct commit_graph *load_commit_graph_one_fd_st(int fd, struct stat *st,
172                                                  struct object_directory *odb)
173 {
174         void *graph_map;
175         size_t graph_size;
176         struct commit_graph *ret;
177
178         graph_size = xsize_t(st->st_size);
179
180         if (graph_size < GRAPH_MIN_SIZE) {
181                 close(fd);
182                 error(_("commit-graph file is too small"));
183                 return NULL;
184         }
185         graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
186         close(fd);
187         ret = parse_commit_graph(graph_map, graph_size);
188
189         if (ret)
190                 ret->odb = odb;
191         else
192                 munmap(graph_map, graph_size);
193
194         return ret;
195 }
196
197 static int verify_commit_graph_lite(struct commit_graph *g)
198 {
199         /*
200          * Basic validation shared between parse_commit_graph()
201          * which'll be called every time the graph is used, and the
202          * much more expensive verify_commit_graph() used by
203          * "commit-graph verify".
204          *
205          * There should only be very basic checks here to ensure that
206          * we don't e.g. segfault in fill_commit_in_graph(), but
207          * because this is a very hot codepath nothing that e.g. loops
208          * over g->num_commits, or runs a checksum on the commit-graph
209          * itself.
210          */
211         if (!g->chunk_oid_fanout) {
212                 error("commit-graph is missing the OID Fanout chunk");
213                 return 1;
214         }
215         if (!g->chunk_oid_lookup) {
216                 error("commit-graph is missing the OID Lookup chunk");
217                 return 1;
218         }
219         if (!g->chunk_commit_data) {
220                 error("commit-graph is missing the Commit Data chunk");
221                 return 1;
222         }
223
224         return 0;
225 }
226
227 struct commit_graph *parse_commit_graph(void *graph_map, size_t graph_size)
228 {
229         const unsigned char *data, *chunk_lookup;
230         uint32_t i;
231         struct commit_graph *graph;
232         uint64_t last_chunk_offset;
233         uint32_t last_chunk_id;
234         uint32_t graph_signature;
235         unsigned char graph_version, hash_version;
236
237         if (!graph_map)
238                 return NULL;
239
240         if (graph_size < GRAPH_MIN_SIZE)
241                 return NULL;
242
243         data = (const unsigned char *)graph_map;
244
245         graph_signature = get_be32(data);
246         if (graph_signature != GRAPH_SIGNATURE) {
247                 error(_("commit-graph signature %X does not match signature %X"),
248                       graph_signature, GRAPH_SIGNATURE);
249                 return NULL;
250         }
251
252         graph_version = *(unsigned char*)(data + 4);
253         if (graph_version != GRAPH_VERSION) {
254                 error(_("commit-graph version %X does not match version %X"),
255                       graph_version, GRAPH_VERSION);
256                 return NULL;
257         }
258
259         hash_version = *(unsigned char*)(data + 5);
260         if (hash_version != oid_version()) {
261                 error(_("commit-graph hash version %X does not match version %X"),
262                       hash_version, oid_version());
263                 return NULL;
264         }
265
266         graph = alloc_commit_graph();
267
268         graph->hash_len = the_hash_algo->rawsz;
269         graph->num_chunks = *(unsigned char*)(data + 6);
270         graph->data = graph_map;
271         graph->data_len = graph_size;
272
273         last_chunk_id = 0;
274         last_chunk_offset = 8;
275         chunk_lookup = data + 8;
276         for (i = 0; i < graph->num_chunks; i++) {
277                 uint32_t chunk_id;
278                 uint64_t chunk_offset;
279                 int chunk_repeated = 0;
280
281                 if (data + graph_size - chunk_lookup <
282                     GRAPH_CHUNKLOOKUP_WIDTH) {
283                         error(_("commit-graph chunk lookup table entry missing; file may be incomplete"));
284                         free(graph);
285                         return NULL;
286                 }
287
288                 chunk_id = get_be32(chunk_lookup + 0);
289                 chunk_offset = get_be64(chunk_lookup + 4);
290
291                 chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
292
293                 if (chunk_offset > graph_size - the_hash_algo->rawsz) {
294                         error(_("commit-graph improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
295                               (uint32_t)chunk_offset);
296                         free(graph);
297                         return NULL;
298                 }
299
300                 switch (chunk_id) {
301                 case GRAPH_CHUNKID_OIDFANOUT:
302                         if (graph->chunk_oid_fanout)
303                                 chunk_repeated = 1;
304                         else
305                                 graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
306                         break;
307
308                 case GRAPH_CHUNKID_OIDLOOKUP:
309                         if (graph->chunk_oid_lookup)
310                                 chunk_repeated = 1;
311                         else
312                                 graph->chunk_oid_lookup = data + chunk_offset;
313                         break;
314
315                 case GRAPH_CHUNKID_DATA:
316                         if (graph->chunk_commit_data)
317                                 chunk_repeated = 1;
318                         else
319                                 graph->chunk_commit_data = data + chunk_offset;
320                         break;
321
322                 case GRAPH_CHUNKID_EXTRAEDGES:
323                         if (graph->chunk_extra_edges)
324                                 chunk_repeated = 1;
325                         else
326                                 graph->chunk_extra_edges = data + chunk_offset;
327                         break;
328
329                 case GRAPH_CHUNKID_BASE:
330                         if (graph->chunk_base_graphs)
331                                 chunk_repeated = 1;
332                         else
333                                 graph->chunk_base_graphs = data + chunk_offset;
334                         break;
335
336                 case GRAPH_CHUNKID_BLOOMINDEXES:
337                         if (graph->chunk_bloom_indexes)
338                                 chunk_repeated = 1;
339                         else
340                                 graph->chunk_bloom_indexes = data + chunk_offset;
341                         break;
342
343                 case GRAPH_CHUNKID_BLOOMDATA:
344                         if (graph->chunk_bloom_data)
345                                 chunk_repeated = 1;
346                         else {
347                                 uint32_t hash_version;
348                                 graph->chunk_bloom_data = data + chunk_offset;
349                                 hash_version = get_be32(data + chunk_offset);
350
351                                 if (hash_version != 1)
352                                         break;
353
354                                 graph->bloom_filter_settings = xmalloc(sizeof(struct bloom_filter_settings));
355                                 graph->bloom_filter_settings->hash_version = hash_version;
356                                 graph->bloom_filter_settings->num_hashes = get_be32(data + chunk_offset + 4);
357                                 graph->bloom_filter_settings->bits_per_entry = get_be32(data + chunk_offset + 8);
358                         }
359                         break;
360                 }
361
362                 if (chunk_repeated) {
363                         error(_("commit-graph chunk id %08x appears multiple times"), chunk_id);
364                         free(graph);
365                         return NULL;
366                 }
367
368                 if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
369                 {
370                         graph->num_commits = (chunk_offset - last_chunk_offset)
371                                              / graph->hash_len;
372                 }
373
374                 last_chunk_id = chunk_id;
375                 last_chunk_offset = chunk_offset;
376         }
377
378         if (graph->chunk_bloom_indexes && graph->chunk_bloom_data) {
379                 init_bloom_filters();
380         } else {
381                 /* We need both the bloom chunks to exist together. Else ignore the data */
382                 graph->chunk_bloom_indexes = NULL;
383                 graph->chunk_bloom_data = NULL;
384                 graph->bloom_filter_settings = NULL;
385         }
386
387         hashcpy(graph->oid.hash, graph->data + graph->data_len - graph->hash_len);
388
389         if (verify_commit_graph_lite(graph)) {
390                 free(graph);
391                 return NULL;
392         }
393
394         return graph;
395 }
396
397 static struct commit_graph *load_commit_graph_one(const char *graph_file,
398                                                   struct object_directory *odb)
399 {
400
401         struct stat st;
402         int fd;
403         struct commit_graph *g;
404         int open_ok = open_commit_graph(graph_file, &fd, &st);
405
406         if (!open_ok)
407                 return NULL;
408
409         g = load_commit_graph_one_fd_st(fd, &st, odb);
410
411         if (g)
412                 g->filename = xstrdup(graph_file);
413
414         return g;
415 }
416
417 static struct commit_graph *load_commit_graph_v1(struct repository *r,
418                                                  struct object_directory *odb)
419 {
420         char *graph_name = get_commit_graph_filename(odb);
421         struct commit_graph *g = load_commit_graph_one(graph_name, odb);
422         free(graph_name);
423
424         return g;
425 }
426
427 static int add_graph_to_chain(struct commit_graph *g,
428                               struct commit_graph *chain,
429                               struct object_id *oids,
430                               int n)
431 {
432         struct commit_graph *cur_g = chain;
433
434         if (n && !g->chunk_base_graphs) {
435                 warning(_("commit-graph has no base graphs chunk"));
436                 return 0;
437         }
438
439         while (n) {
440                 n--;
441
442                 if (!cur_g ||
443                     !oideq(&oids[n], &cur_g->oid) ||
444                     !hasheq(oids[n].hash, g->chunk_base_graphs + g->hash_len * n)) {
445                         warning(_("commit-graph chain does not match"));
446                         return 0;
447                 }
448
449                 cur_g = cur_g->base_graph;
450         }
451
452         g->base_graph = chain;
453
454         if (chain)
455                 g->num_commits_in_base = chain->num_commits + chain->num_commits_in_base;
456
457         return 1;
458 }
459
460 static struct commit_graph *load_commit_graph_chain(struct repository *r,
461                                                     struct object_directory *odb)
462 {
463         struct commit_graph *graph_chain = NULL;
464         struct strbuf line = STRBUF_INIT;
465         struct stat st;
466         struct object_id *oids;
467         int i = 0, valid = 1, count;
468         char *chain_name = get_chain_filename(odb);
469         FILE *fp;
470         int stat_res;
471
472         fp = fopen(chain_name, "r");
473         stat_res = stat(chain_name, &st);
474         free(chain_name);
475
476         if (!fp ||
477             stat_res ||
478             st.st_size <= the_hash_algo->hexsz)
479                 return NULL;
480
481         count = st.st_size / (the_hash_algo->hexsz + 1);
482         oids = xcalloc(count, sizeof(struct object_id));
483
484         prepare_alt_odb(r);
485
486         for (i = 0; i < count; i++) {
487                 struct object_directory *odb;
488
489                 if (strbuf_getline_lf(&line, fp) == EOF)
490                         break;
491
492                 if (get_oid_hex(line.buf, &oids[i])) {
493                         warning(_("invalid commit-graph chain: line '%s' not a hash"),
494                                 line.buf);
495                         valid = 0;
496                         break;
497                 }
498
499                 valid = 0;
500                 for (odb = r->objects->odb; odb; odb = odb->next) {
501                         char *graph_name = get_split_graph_filename(odb, line.buf);
502                         struct commit_graph *g = load_commit_graph_one(graph_name, odb);
503
504                         free(graph_name);
505
506                         if (g) {
507                                 if (add_graph_to_chain(g, graph_chain, oids, i)) {
508                                         graph_chain = g;
509                                         valid = 1;
510                                 }
511
512                                 break;
513                         }
514                 }
515
516                 if (!valid) {
517                         warning(_("unable to find all commit-graph files"));
518                         break;
519                 }
520         }
521
522         free(oids);
523         fclose(fp);
524         strbuf_release(&line);
525
526         return graph_chain;
527 }
528
529 struct commit_graph *read_commit_graph_one(struct repository *r,
530                                            struct object_directory *odb)
531 {
532         struct commit_graph *g = load_commit_graph_v1(r, odb);
533
534         if (!g)
535                 g = load_commit_graph_chain(r, odb);
536
537         return g;
538 }
539
540 static void prepare_commit_graph_one(struct repository *r,
541                                      struct object_directory *odb)
542 {
543
544         if (r->objects->commit_graph)
545                 return;
546
547         r->objects->commit_graph = read_commit_graph_one(r, odb);
548 }
549
550 /*
551  * Return 1 if commit_graph is non-NULL, and 0 otherwise.
552  *
553  * On the first invocation, this function attempts to load the commit
554  * graph if the_repository is configured to have one.
555  */
556 static int prepare_commit_graph(struct repository *r)
557 {
558         struct object_directory *odb;
559
560         /*
561          * This must come before the "already attempted?" check below, because
562          * we want to disable even an already-loaded graph file.
563          */
564         if (r->commit_graph_disabled)
565                 return 0;
566
567         if (r->objects->commit_graph_attempted)
568                 return !!r->objects->commit_graph;
569         r->objects->commit_graph_attempted = 1;
570
571         if (git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD, 0))
572                 die("dying as requested by the '%s' variable on commit-graph load!",
573                     GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD);
574
575         prepare_repo_settings(r);
576
577         if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
578             r->settings.core_commit_graph != 1)
579                 /*
580                  * This repository is not configured to use commit graphs, so
581                  * do not load one. (But report commit_graph_attempted anyway
582                  * so that commit graph loading is not attempted again for this
583                  * repository.)
584                  */
585                 return 0;
586
587         if (!commit_graph_compatible(r))
588                 return 0;
589
590         prepare_alt_odb(r);
591         for (odb = r->objects->odb;
592              !r->objects->commit_graph && odb;
593              odb = odb->next)
594                 prepare_commit_graph_one(r, odb);
595         return !!r->objects->commit_graph;
596 }
597
598 int generation_numbers_enabled(struct repository *r)
599 {
600         uint32_t first_generation;
601         struct commit_graph *g;
602         if (!prepare_commit_graph(r))
603                return 0;
604
605         g = r->objects->commit_graph;
606
607         if (!g->num_commits)
608                 return 0;
609
610         first_generation = get_be32(g->chunk_commit_data +
611                                     g->hash_len + 8) >> 2;
612
613         return !!first_generation;
614 }
615
616 static void close_commit_graph_one(struct commit_graph *g)
617 {
618         if (!g)
619                 return;
620
621         close_commit_graph_one(g->base_graph);
622         free_commit_graph(g);
623 }
624
625 void close_commit_graph(struct raw_object_store *o)
626 {
627         close_commit_graph_one(o->commit_graph);
628         o->commit_graph = NULL;
629 }
630
631 static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
632 {
633         return bsearch_hash(oid->hash, g->chunk_oid_fanout,
634                             g->chunk_oid_lookup, g->hash_len, pos);
635 }
636
637 static void load_oid_from_graph(struct commit_graph *g,
638                                 uint32_t pos,
639                                 struct object_id *oid)
640 {
641         uint32_t lex_index;
642
643         while (g && pos < g->num_commits_in_base)
644                 g = g->base_graph;
645
646         if (!g)
647                 BUG("NULL commit-graph");
648
649         if (pos >= g->num_commits + g->num_commits_in_base)
650                 die(_("invalid commit position. commit-graph is likely corrupt"));
651
652         lex_index = pos - g->num_commits_in_base;
653
654         hashcpy(oid->hash, g->chunk_oid_lookup + g->hash_len * lex_index);
655 }
656
657 static struct commit_list **insert_parent_or_die(struct repository *r,
658                                                  struct commit_graph *g,
659                                                  uint32_t pos,
660                                                  struct commit_list **pptr)
661 {
662         struct commit *c;
663         struct object_id oid;
664
665         if (pos >= g->num_commits + g->num_commits_in_base)
666                 die("invalid parent position %"PRIu32, pos);
667
668         load_oid_from_graph(g, pos, &oid);
669         c = lookup_commit(r, &oid);
670         if (!c)
671                 die(_("could not find commit %s"), oid_to_hex(&oid));
672         c->graph_pos = pos;
673         return &commit_list_insert(c, pptr)->next;
674 }
675
676 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
677 {
678         const unsigned char *commit_data;
679         uint32_t lex_index;
680
681         while (pos < g->num_commits_in_base)
682                 g = g->base_graph;
683
684         lex_index = pos - g->num_commits_in_base;
685         commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * lex_index;
686         item->graph_pos = pos;
687         item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
688 }
689
690 static inline void set_commit_tree(struct commit *c, struct tree *t)
691 {
692         c->maybe_tree = t;
693 }
694
695 static int fill_commit_in_graph(struct repository *r,
696                                 struct commit *item,
697                                 struct commit_graph *g, uint32_t pos)
698 {
699         uint32_t edge_value;
700         uint32_t *parent_data_ptr;
701         uint64_t date_low, date_high;
702         struct commit_list **pptr;
703         const unsigned char *commit_data;
704         uint32_t lex_index;
705
706         while (pos < g->num_commits_in_base)
707                 g = g->base_graph;
708
709         if (pos >= g->num_commits + g->num_commits_in_base)
710                 die(_("invalid commit position. commit-graph is likely corrupt"));
711
712         /*
713          * Store the "full" position, but then use the
714          * "local" position for the rest of the calculation.
715          */
716         item->graph_pos = pos;
717         lex_index = pos - g->num_commits_in_base;
718
719         commit_data = g->chunk_commit_data + (g->hash_len + 16) * lex_index;
720
721         item->object.parsed = 1;
722
723         set_commit_tree(item, NULL);
724
725         date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
726         date_low = get_be32(commit_data + g->hash_len + 12);
727         item->date = (timestamp_t)((date_high << 32) | date_low);
728
729         item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
730
731         pptr = &item->parents;
732
733         edge_value = get_be32(commit_data + g->hash_len);
734         if (edge_value == GRAPH_PARENT_NONE)
735                 return 1;
736         pptr = insert_parent_or_die(r, g, edge_value, pptr);
737
738         edge_value = get_be32(commit_data + g->hash_len + 4);
739         if (edge_value == GRAPH_PARENT_NONE)
740                 return 1;
741         if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
742                 pptr = insert_parent_or_die(r, g, edge_value, pptr);
743                 return 1;
744         }
745
746         parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
747                           4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
748         do {
749                 edge_value = get_be32(parent_data_ptr);
750                 pptr = insert_parent_or_die(r, g,
751                                             edge_value & GRAPH_EDGE_LAST_MASK,
752                                             pptr);
753                 parent_data_ptr++;
754         } while (!(edge_value & GRAPH_LAST_EDGE));
755
756         return 1;
757 }
758
759 static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
760 {
761         if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
762                 *pos = item->graph_pos;
763                 return 1;
764         } else {
765                 struct commit_graph *cur_g = g;
766                 uint32_t lex_index;
767
768                 while (cur_g && !bsearch_graph(cur_g, &(item->object.oid), &lex_index))
769                         cur_g = cur_g->base_graph;
770
771                 if (cur_g) {
772                         *pos = lex_index + cur_g->num_commits_in_base;
773                         return 1;
774                 }
775
776                 return 0;
777         }
778 }
779
780 static int parse_commit_in_graph_one(struct repository *r,
781                                      struct commit_graph *g,
782                                      struct commit *item)
783 {
784         uint32_t pos;
785
786         if (item->object.parsed)
787                 return 1;
788
789         if (find_commit_in_graph(item, g, &pos))
790                 return fill_commit_in_graph(r, item, g, pos);
791
792         return 0;
793 }
794
795 int parse_commit_in_graph(struct repository *r, struct commit *item)
796 {
797         if (!prepare_commit_graph(r))
798                 return 0;
799         return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
800 }
801
802 void load_commit_graph_info(struct repository *r, struct commit *item)
803 {
804         uint32_t pos;
805         if (!prepare_commit_graph(r))
806                 return;
807         if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
808                 fill_commit_graph_info(item, r->objects->commit_graph, pos);
809 }
810
811 static struct tree *load_tree_for_commit(struct repository *r,
812                                          struct commit_graph *g,
813                                          struct commit *c)
814 {
815         struct object_id oid;
816         const unsigned char *commit_data;
817
818         while (c->graph_pos < g->num_commits_in_base)
819                 g = g->base_graph;
820
821         commit_data = g->chunk_commit_data +
822                         GRAPH_DATA_WIDTH * (c->graph_pos - g->num_commits_in_base);
823
824         hashcpy(oid.hash, commit_data);
825         set_commit_tree(c, lookup_tree(r, &oid));
826
827         return c->maybe_tree;
828 }
829
830 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
831                                                  struct commit_graph *g,
832                                                  const struct commit *c)
833 {
834         if (c->maybe_tree)
835                 return c->maybe_tree;
836         if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
837                 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
838
839         return load_tree_for_commit(r, g, (struct commit *)c);
840 }
841
842 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
843 {
844         return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
845 }
846
847 struct packed_commit_list {
848         struct commit **list;
849         int nr;
850         int alloc;
851 };
852
853 struct packed_oid_list {
854         struct object_id *list;
855         int nr;
856         int alloc;
857 };
858
859 struct write_commit_graph_context {
860         struct repository *r;
861         struct object_directory *odb;
862         char *graph_name;
863         struct packed_oid_list oids;
864         struct packed_commit_list commits;
865         int num_extra_edges;
866         unsigned long approx_nr_objects;
867         struct progress *progress;
868         int progress_done;
869         uint64_t progress_cnt;
870
871         char *base_graph_name;
872         int num_commit_graphs_before;
873         int num_commit_graphs_after;
874         char **commit_graph_filenames_before;
875         char **commit_graph_filenames_after;
876         char **commit_graph_hash_after;
877         uint32_t new_num_commits_in_base;
878         struct commit_graph *new_base_graph;
879
880         unsigned append:1,
881                  report_progress:1,
882                  split:1,
883                  check_oids:1,
884                  changed_paths:1,
885                  order_by_pack:1;
886
887         const struct split_commit_graph_opts *split_opts;
888         size_t total_bloom_filter_data_size;
889 };
890
891 static void write_graph_chunk_fanout(struct hashfile *f,
892                                      struct write_commit_graph_context *ctx)
893 {
894         int i, count = 0;
895         struct commit **list = ctx->commits.list;
896
897         /*
898          * Write the first-level table (the list is sorted,
899          * but we use a 256-entry lookup to be able to avoid
900          * having to do eight extra binary search iterations).
901          */
902         for (i = 0; i < 256; i++) {
903                 while (count < ctx->commits.nr) {
904                         if ((*list)->object.oid.hash[0] != i)
905                                 break;
906                         display_progress(ctx->progress, ++ctx->progress_cnt);
907                         count++;
908                         list++;
909                 }
910
911                 hashwrite_be32(f, count);
912         }
913 }
914
915 static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
916                                    struct write_commit_graph_context *ctx)
917 {
918         struct commit **list = ctx->commits.list;
919         int count;
920         for (count = 0; count < ctx->commits.nr; count++, list++) {
921                 display_progress(ctx->progress, ++ctx->progress_cnt);
922                 hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
923         }
924 }
925
926 static const unsigned char *commit_to_sha1(size_t index, void *table)
927 {
928         struct commit **commits = table;
929         return commits[index]->object.oid.hash;
930 }
931
932 static void write_graph_chunk_data(struct hashfile *f, int hash_len,
933                                    struct write_commit_graph_context *ctx)
934 {
935         struct commit **list = ctx->commits.list;
936         struct commit **last = ctx->commits.list + ctx->commits.nr;
937         uint32_t num_extra_edges = 0;
938
939         while (list < last) {
940                 struct commit_list *parent;
941                 struct object_id *tree;
942                 int edge_value;
943                 uint32_t packedDate[2];
944                 display_progress(ctx->progress, ++ctx->progress_cnt);
945
946                 if (parse_commit_no_graph(*list))
947                         die(_("unable to parse commit %s"),
948                                 oid_to_hex(&(*list)->object.oid));
949                 tree = get_commit_tree_oid(*list);
950                 hashwrite(f, tree->hash, hash_len);
951
952                 parent = (*list)->parents;
953
954                 if (!parent)
955                         edge_value = GRAPH_PARENT_NONE;
956                 else {
957                         edge_value = sha1_pos(parent->item->object.oid.hash,
958                                               ctx->commits.list,
959                                               ctx->commits.nr,
960                                               commit_to_sha1);
961
962                         if (edge_value >= 0)
963                                 edge_value += ctx->new_num_commits_in_base;
964                         else if (ctx->new_base_graph) {
965                                 uint32_t pos;
966                                 if (find_commit_in_graph(parent->item,
967                                                          ctx->new_base_graph,
968                                                          &pos))
969                                         edge_value = pos;
970                         }
971
972                         if (edge_value < 0)
973                                 BUG("missing parent %s for commit %s",
974                                     oid_to_hex(&parent->item->object.oid),
975                                     oid_to_hex(&(*list)->object.oid));
976                 }
977
978                 hashwrite_be32(f, edge_value);
979
980                 if (parent)
981                         parent = parent->next;
982
983                 if (!parent)
984                         edge_value = GRAPH_PARENT_NONE;
985                 else if (parent->next)
986                         edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
987                 else {
988                         edge_value = sha1_pos(parent->item->object.oid.hash,
989                                               ctx->commits.list,
990                                               ctx->commits.nr,
991                                               commit_to_sha1);
992
993                         if (edge_value >= 0)
994                                 edge_value += ctx->new_num_commits_in_base;
995                         else if (ctx->new_base_graph) {
996                                 uint32_t pos;
997                                 if (find_commit_in_graph(parent->item,
998                                                          ctx->new_base_graph,
999                                                          &pos))
1000                                         edge_value = pos;
1001                         }
1002
1003                         if (edge_value < 0)
1004                                 BUG("missing parent %s for commit %s",
1005                                     oid_to_hex(&parent->item->object.oid),
1006                                     oid_to_hex(&(*list)->object.oid));
1007                 }
1008
1009                 hashwrite_be32(f, edge_value);
1010
1011                 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
1012                         do {
1013                                 num_extra_edges++;
1014                                 parent = parent->next;
1015                         } while (parent);
1016                 }
1017
1018                 if (sizeof((*list)->date) > 4)
1019                         packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
1020                 else
1021                         packedDate[0] = 0;
1022
1023                 packedDate[0] |= htonl((*list)->generation << 2);
1024
1025                 packedDate[1] = htonl((*list)->date);
1026                 hashwrite(f, packedDate, 8);
1027
1028                 list++;
1029         }
1030 }
1031
1032 static void write_graph_chunk_extra_edges(struct hashfile *f,
1033                                           struct write_commit_graph_context *ctx)
1034 {
1035         struct commit **list = ctx->commits.list;
1036         struct commit **last = ctx->commits.list + ctx->commits.nr;
1037         struct commit_list *parent;
1038
1039         while (list < last) {
1040                 int num_parents = 0;
1041
1042                 display_progress(ctx->progress, ++ctx->progress_cnt);
1043
1044                 for (parent = (*list)->parents; num_parents < 3 && parent;
1045                      parent = parent->next)
1046                         num_parents++;
1047
1048                 if (num_parents <= 2) {
1049                         list++;
1050                         continue;
1051                 }
1052
1053                 /* Since num_parents > 2, this initializer is safe. */
1054                 for (parent = (*list)->parents->next; parent; parent = parent->next) {
1055                         int edge_value = sha1_pos(parent->item->object.oid.hash,
1056                                                   ctx->commits.list,
1057                                                   ctx->commits.nr,
1058                                                   commit_to_sha1);
1059
1060                         if (edge_value >= 0)
1061                                 edge_value += ctx->new_num_commits_in_base;
1062                         else if (ctx->new_base_graph) {
1063                                 uint32_t pos;
1064                                 if (find_commit_in_graph(parent->item,
1065                                                          ctx->new_base_graph,
1066                                                          &pos))
1067                                         edge_value = pos;
1068                         }
1069
1070                         if (edge_value < 0)
1071                                 BUG("missing parent %s for commit %s",
1072                                     oid_to_hex(&parent->item->object.oid),
1073                                     oid_to_hex(&(*list)->object.oid));
1074                         else if (!parent->next)
1075                                 edge_value |= GRAPH_LAST_EDGE;
1076
1077                         hashwrite_be32(f, edge_value);
1078                 }
1079
1080                 list++;
1081         }
1082 }
1083
1084 static void write_graph_chunk_bloom_indexes(struct hashfile *f,
1085                                             struct write_commit_graph_context *ctx)
1086 {
1087         struct commit **list = ctx->commits.list;
1088         struct commit **last = ctx->commits.list + ctx->commits.nr;
1089         uint32_t cur_pos = 0;
1090         struct progress *progress = NULL;
1091         int i = 0;
1092
1093         if (ctx->report_progress)
1094                 progress = start_delayed_progress(
1095                         _("Writing changed paths Bloom filters index"),
1096                         ctx->commits.nr);
1097
1098         while (list < last) {
1099                 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list, 0);
1100                 cur_pos += filter->len;
1101                 display_progress(progress, ++i);
1102                 hashwrite_be32(f, cur_pos);
1103                 list++;
1104         }
1105
1106         stop_progress(&progress);
1107 }
1108
1109 static void write_graph_chunk_bloom_data(struct hashfile *f,
1110                                          struct write_commit_graph_context *ctx,
1111                                          const struct bloom_filter_settings *settings)
1112 {
1113         struct commit **list = ctx->commits.list;
1114         struct commit **last = ctx->commits.list + ctx->commits.nr;
1115         struct progress *progress = NULL;
1116         int i = 0;
1117
1118         if (ctx->report_progress)
1119                 progress = start_delayed_progress(
1120                         _("Writing changed paths Bloom filters data"),
1121                         ctx->commits.nr);
1122
1123         hashwrite_be32(f, settings->hash_version);
1124         hashwrite_be32(f, settings->num_hashes);
1125         hashwrite_be32(f, settings->bits_per_entry);
1126
1127         while (list < last) {
1128                 struct bloom_filter *filter = get_bloom_filter(ctx->r, *list, 0);
1129                 display_progress(progress, ++i);
1130                 hashwrite(f, filter->data, filter->len * sizeof(unsigned char));
1131                 list++;
1132         }
1133
1134         stop_progress(&progress);
1135 }
1136
1137 static int oid_compare(const void *_a, const void *_b)
1138 {
1139         const struct object_id *a = (const struct object_id *)_a;
1140         const struct object_id *b = (const struct object_id *)_b;
1141         return oidcmp(a, b);
1142 }
1143
1144 static int add_packed_commits(const struct object_id *oid,
1145                               struct packed_git *pack,
1146                               uint32_t pos,
1147                               void *data)
1148 {
1149         struct write_commit_graph_context *ctx = (struct write_commit_graph_context*)data;
1150         enum object_type type;
1151         off_t offset = nth_packed_object_offset(pack, pos);
1152         struct object_info oi = OBJECT_INFO_INIT;
1153
1154         if (ctx->progress)
1155                 display_progress(ctx->progress, ++ctx->progress_done);
1156
1157         oi.typep = &type;
1158         if (packed_object_info(ctx->r, pack, offset, &oi) < 0)
1159                 die(_("unable to get type of object %s"), oid_to_hex(oid));
1160
1161         if (type != OBJ_COMMIT)
1162                 return 0;
1163
1164         ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1165         oidcpy(&(ctx->oids.list[ctx->oids.nr]), oid);
1166         ctx->oids.nr++;
1167
1168         set_commit_pos(ctx->r, oid);
1169
1170         return 0;
1171 }
1172
1173 static void add_missing_parents(struct write_commit_graph_context *ctx, struct commit *commit)
1174 {
1175         struct commit_list *parent;
1176         for (parent = commit->parents; parent; parent = parent->next) {
1177                 if (!(parent->item->object.flags & REACHABLE)) {
1178                         ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1179                         oidcpy(&ctx->oids.list[ctx->oids.nr], &(parent->item->object.oid));
1180                         ctx->oids.nr++;
1181                         parent->item->object.flags |= REACHABLE;
1182                 }
1183         }
1184 }
1185
1186 static void close_reachable(struct write_commit_graph_context *ctx)
1187 {
1188         int i;
1189         struct commit *commit;
1190         enum commit_graph_split_flags flags = ctx->split_opts ?
1191                 ctx->split_opts->flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1192
1193         if (ctx->report_progress)
1194                 ctx->progress = start_delayed_progress(
1195                                         _("Loading known commits in commit graph"),
1196                                         ctx->oids.nr);
1197         for (i = 0; i < ctx->oids.nr; i++) {
1198                 display_progress(ctx->progress, i + 1);
1199                 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1200                 if (commit)
1201                         commit->object.flags |= REACHABLE;
1202         }
1203         stop_progress(&ctx->progress);
1204
1205         /*
1206          * As this loop runs, ctx->oids.nr may grow, but not more
1207          * than the number of missing commits in the reachable
1208          * closure.
1209          */
1210         if (ctx->report_progress)
1211                 ctx->progress = start_delayed_progress(
1212                                         _("Expanding reachable commits in commit graph"),
1213                                         0);
1214         for (i = 0; i < ctx->oids.nr; i++) {
1215                 display_progress(ctx->progress, i + 1);
1216                 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1217
1218                 if (!commit)
1219                         continue;
1220                 if (ctx->split) {
1221                         if ((!parse_commit(commit) &&
1222                              commit->graph_pos == COMMIT_NOT_FROM_GRAPH) ||
1223                             flags == COMMIT_GRAPH_SPLIT_REPLACE)
1224                                 add_missing_parents(ctx, commit);
1225                 } else if (!parse_commit_no_graph(commit))
1226                         add_missing_parents(ctx, commit);
1227         }
1228         stop_progress(&ctx->progress);
1229
1230         if (ctx->report_progress)
1231                 ctx->progress = start_delayed_progress(
1232                                         _("Clearing commit marks in commit graph"),
1233                                         ctx->oids.nr);
1234         for (i = 0; i < ctx->oids.nr; i++) {
1235                 display_progress(ctx->progress, i + 1);
1236                 commit = lookup_commit(ctx->r, &ctx->oids.list[i]);
1237
1238                 if (commit)
1239                         commit->object.flags &= ~REACHABLE;
1240         }
1241         stop_progress(&ctx->progress);
1242 }
1243
1244 static void compute_generation_numbers(struct write_commit_graph_context *ctx)
1245 {
1246         int i;
1247         struct commit_list *list = NULL;
1248
1249         if (ctx->report_progress)
1250                 ctx->progress = start_delayed_progress(
1251                                         _("Computing commit graph generation numbers"),
1252                                         ctx->commits.nr);
1253         for (i = 0; i < ctx->commits.nr; i++) {
1254                 display_progress(ctx->progress, i + 1);
1255                 if (ctx->commits.list[i]->generation != GENERATION_NUMBER_INFINITY &&
1256                     ctx->commits.list[i]->generation != GENERATION_NUMBER_ZERO)
1257                         continue;
1258
1259                 commit_list_insert(ctx->commits.list[i], &list);
1260                 while (list) {
1261                         struct commit *current = list->item;
1262                         struct commit_list *parent;
1263                         int all_parents_computed = 1;
1264                         uint32_t max_generation = 0;
1265
1266                         for (parent = current->parents; parent; parent = parent->next) {
1267                                 if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
1268                                     parent->item->generation == GENERATION_NUMBER_ZERO) {
1269                                         all_parents_computed = 0;
1270                                         commit_list_insert(parent->item, &list);
1271                                         break;
1272                                 } else if (parent->item->generation > max_generation) {
1273                                         max_generation = parent->item->generation;
1274                                 }
1275                         }
1276
1277                         if (all_parents_computed) {
1278                                 current->generation = max_generation + 1;
1279                                 pop_commit(&list);
1280
1281                                 if (current->generation > GENERATION_NUMBER_MAX)
1282                                         current->generation = GENERATION_NUMBER_MAX;
1283                         }
1284                 }
1285         }
1286         stop_progress(&ctx->progress);
1287 }
1288
1289 static void compute_bloom_filters(struct write_commit_graph_context *ctx)
1290 {
1291         int i;
1292         struct progress *progress = NULL;
1293         struct commit **sorted_commits;
1294
1295         init_bloom_filters();
1296
1297         if (ctx->report_progress)
1298                 progress = start_delayed_progress(
1299                         _("Computing commit changed paths Bloom filters"),
1300                         ctx->commits.nr);
1301
1302         ALLOC_ARRAY(sorted_commits, ctx->commits.nr);
1303         COPY_ARRAY(sorted_commits, ctx->commits.list, ctx->commits.nr);
1304
1305         if (ctx->order_by_pack)
1306                 QSORT(sorted_commits, ctx->commits.nr, commit_pos_cmp);
1307         else
1308                 QSORT(sorted_commits, ctx->commits.nr, commit_gen_cmp);
1309
1310         for (i = 0; i < ctx->commits.nr; i++) {
1311                 struct commit *c = sorted_commits[i];
1312                 struct bloom_filter *filter = get_bloom_filter(ctx->r, c, 1);
1313                 ctx->total_bloom_filter_data_size += sizeof(unsigned char) * filter->len;
1314                 display_progress(progress, i + 1);
1315         }
1316
1317         free(sorted_commits);
1318         stop_progress(&progress);
1319 }
1320
1321 struct refs_cb_data {
1322         struct oidset *commits;
1323         struct progress *progress;
1324 };
1325
1326 static int add_ref_to_set(const char *refname,
1327                           const struct object_id *oid,
1328                           int flags, void *cb_data)
1329 {
1330         struct refs_cb_data *data = (struct refs_cb_data *)cb_data;
1331
1332         oidset_insert(data->commits, oid);
1333
1334         display_progress(data->progress, oidset_size(data->commits));
1335
1336         return 0;
1337 }
1338
1339 int write_commit_graph_reachable(struct object_directory *odb,
1340                                  enum commit_graph_write_flags flags,
1341                                  const struct split_commit_graph_opts *split_opts)
1342 {
1343         struct oidset commits = OIDSET_INIT;
1344         struct refs_cb_data data;
1345         int result;
1346
1347         memset(&data, 0, sizeof(data));
1348         data.commits = &commits;
1349         if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
1350                 data.progress = start_delayed_progress(
1351                         _("Collecting referenced commits"), 0);
1352
1353         for_each_ref(add_ref_to_set, &data);
1354         result = write_commit_graph(odb, NULL, &commits,
1355                                     flags, split_opts);
1356
1357         oidset_clear(&commits);
1358         if (data.progress)
1359                 stop_progress(&data.progress);
1360         return result;
1361 }
1362
1363 static int fill_oids_from_packs(struct write_commit_graph_context *ctx,
1364                                 struct string_list *pack_indexes)
1365 {
1366         uint32_t i;
1367         struct strbuf progress_title = STRBUF_INIT;
1368         struct strbuf packname = STRBUF_INIT;
1369         int dirlen;
1370
1371         strbuf_addf(&packname, "%s/pack/", ctx->odb->path);
1372         dirlen = packname.len;
1373         if (ctx->report_progress) {
1374                 strbuf_addf(&progress_title,
1375                             Q_("Finding commits for commit graph in %d pack",
1376                                "Finding commits for commit graph in %d packs",
1377                                pack_indexes->nr),
1378                             pack_indexes->nr);
1379                 ctx->progress = start_delayed_progress(progress_title.buf, 0);
1380                 ctx->progress_done = 0;
1381         }
1382         for (i = 0; i < pack_indexes->nr; i++) {
1383                 struct packed_git *p;
1384                 strbuf_setlen(&packname, dirlen);
1385                 strbuf_addstr(&packname, pack_indexes->items[i].string);
1386                 p = add_packed_git(packname.buf, packname.len, 1);
1387                 if (!p) {
1388                         error(_("error adding pack %s"), packname.buf);
1389                         return -1;
1390                 }
1391                 if (open_pack_index(p)) {
1392                         error(_("error opening index for %s"), packname.buf);
1393                         return -1;
1394                 }
1395                 for_each_object_in_pack(p, add_packed_commits, ctx,
1396                                         FOR_EACH_OBJECT_PACK_ORDER);
1397                 close_pack(p);
1398                 free(p);
1399         }
1400
1401         stop_progress(&ctx->progress);
1402         strbuf_release(&progress_title);
1403         strbuf_release(&packname);
1404
1405         return 0;
1406 }
1407
1408 static int fill_oids_from_commits(struct write_commit_graph_context *ctx,
1409                                   struct oidset *commits)
1410 {
1411         uint32_t i = 0;
1412         struct strbuf progress_title = STRBUF_INIT;
1413         struct oidset_iter iter;
1414         struct object_id *oid;
1415
1416         if (!oidset_size(commits))
1417                 return 0;
1418
1419         if (ctx->report_progress) {
1420                 strbuf_addf(&progress_title,
1421                             Q_("Finding commits for commit graph from %d ref",
1422                                "Finding commits for commit graph from %d refs",
1423                                oidset_size(commits)),
1424                             oidset_size(commits));
1425                 ctx->progress = start_delayed_progress(
1426                                         progress_title.buf,
1427                                         oidset_size(commits));
1428         }
1429
1430         oidset_iter_init(commits, &iter);
1431         while ((oid = oidset_iter_next(&iter))) {
1432                 struct commit *result;
1433
1434                 display_progress(ctx->progress, ++i);
1435
1436                 result = lookup_commit_reference_gently(ctx->r, oid, 1);
1437                 if (result) {
1438                         ALLOC_GROW(ctx->oids.list, ctx->oids.nr + 1, ctx->oids.alloc);
1439                         oidcpy(&ctx->oids.list[ctx->oids.nr], &(result->object.oid));
1440                         ctx->oids.nr++;
1441                 } else if (ctx->check_oids) {
1442                         error(_("invalid commit object id: %s"),
1443                               oid_to_hex(oid));
1444                         return -1;
1445                 }
1446         }
1447
1448         stop_progress(&ctx->progress);
1449         strbuf_release(&progress_title);
1450
1451         return 0;
1452 }
1453
1454 static void fill_oids_from_all_packs(struct write_commit_graph_context *ctx)
1455 {
1456         if (ctx->report_progress)
1457                 ctx->progress = start_delayed_progress(
1458                         _("Finding commits for commit graph among packed objects"),
1459                         ctx->approx_nr_objects);
1460         for_each_packed_object(add_packed_commits, ctx,
1461                                FOR_EACH_OBJECT_PACK_ORDER);
1462         if (ctx->progress_done < ctx->approx_nr_objects)
1463                 display_progress(ctx->progress, ctx->approx_nr_objects);
1464         stop_progress(&ctx->progress);
1465 }
1466
1467 static uint32_t count_distinct_commits(struct write_commit_graph_context *ctx)
1468 {
1469         uint32_t i, count_distinct = 1;
1470
1471         if (ctx->report_progress)
1472                 ctx->progress = start_delayed_progress(
1473                         _("Counting distinct commits in commit graph"),
1474                         ctx->oids.nr);
1475         display_progress(ctx->progress, 0); /* TODO: Measure QSORT() progress */
1476         QSORT(ctx->oids.list, ctx->oids.nr, oid_compare);
1477
1478         for (i = 1; i < ctx->oids.nr; i++) {
1479                 display_progress(ctx->progress, i + 1);
1480                 if (!oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i])) {
1481                         if (ctx->split) {
1482                                 struct commit *c = lookup_commit(ctx->r, &ctx->oids.list[i]);
1483
1484                                 if (!c || c->graph_pos != COMMIT_NOT_FROM_GRAPH)
1485                                         continue;
1486                         }
1487
1488                         count_distinct++;
1489                 }
1490         }
1491         stop_progress(&ctx->progress);
1492
1493         return count_distinct;
1494 }
1495
1496 static void copy_oids_to_commits(struct write_commit_graph_context *ctx)
1497 {
1498         uint32_t i;
1499         enum commit_graph_split_flags flags = ctx->split_opts ?
1500                 ctx->split_opts->flags : COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1501
1502         ctx->num_extra_edges = 0;
1503         if (ctx->report_progress)
1504                 ctx->progress = start_delayed_progress(
1505                         _("Finding extra edges in commit graph"),
1506                         ctx->oids.nr);
1507         for (i = 0; i < ctx->oids.nr; i++) {
1508                 unsigned int num_parents;
1509
1510                 display_progress(ctx->progress, i + 1);
1511                 if (i > 0 && oideq(&ctx->oids.list[i - 1], &ctx->oids.list[i]))
1512                         continue;
1513
1514                 ALLOC_GROW(ctx->commits.list, ctx->commits.nr + 1, ctx->commits.alloc);
1515                 ctx->commits.list[ctx->commits.nr] = lookup_commit(ctx->r, &ctx->oids.list[i]);
1516
1517                 if (ctx->split && flags != COMMIT_GRAPH_SPLIT_REPLACE &&
1518                     ctx->commits.list[ctx->commits.nr]->graph_pos != COMMIT_NOT_FROM_GRAPH)
1519                         continue;
1520
1521                 if (ctx->split && flags == COMMIT_GRAPH_SPLIT_REPLACE)
1522                         parse_commit(ctx->commits.list[ctx->commits.nr]);
1523                 else
1524                         parse_commit_no_graph(ctx->commits.list[ctx->commits.nr]);
1525
1526                 num_parents = commit_list_count(ctx->commits.list[ctx->commits.nr]->parents);
1527                 if (num_parents > 2)
1528                         ctx->num_extra_edges += num_parents - 1;
1529
1530                 ctx->commits.nr++;
1531         }
1532         stop_progress(&ctx->progress);
1533 }
1534
1535 static int write_graph_chunk_base_1(struct hashfile *f,
1536                                     struct commit_graph *g)
1537 {
1538         int num = 0;
1539
1540         if (!g)
1541                 return 0;
1542
1543         num = write_graph_chunk_base_1(f, g->base_graph);
1544         hashwrite(f, g->oid.hash, the_hash_algo->rawsz);
1545         return num + 1;
1546 }
1547
1548 static int write_graph_chunk_base(struct hashfile *f,
1549                                   struct write_commit_graph_context *ctx)
1550 {
1551         int num = write_graph_chunk_base_1(f, ctx->new_base_graph);
1552
1553         if (num != ctx->num_commit_graphs_after - 1) {
1554                 error(_("failed to write correct number of base graph ids"));
1555                 return -1;
1556         }
1557
1558         return 0;
1559 }
1560
1561 static int write_commit_graph_file(struct write_commit_graph_context *ctx)
1562 {
1563         uint32_t i;
1564         int fd;
1565         struct hashfile *f;
1566         struct lock_file lk = LOCK_INIT;
1567         uint32_t chunk_ids[MAX_NUM_CHUNKS + 1];
1568         uint64_t chunk_offsets[MAX_NUM_CHUNKS + 1];
1569         const unsigned hashsz = the_hash_algo->rawsz;
1570         struct strbuf progress_title = STRBUF_INIT;
1571         int num_chunks = 3;
1572         struct object_id file_hash;
1573         const struct bloom_filter_settings bloom_settings = DEFAULT_BLOOM_FILTER_SETTINGS;
1574
1575         if (ctx->split) {
1576                 struct strbuf tmp_file = STRBUF_INIT;
1577
1578                 strbuf_addf(&tmp_file,
1579                             "%s/info/commit-graphs/tmp_graph_XXXXXX",
1580                             ctx->odb->path);
1581                 ctx->graph_name = strbuf_detach(&tmp_file, NULL);
1582         } else {
1583                 ctx->graph_name = get_commit_graph_filename(ctx->odb);
1584         }
1585
1586         if (safe_create_leading_directories(ctx->graph_name)) {
1587                 UNLEAK(ctx->graph_name);
1588                 error(_("unable to create leading directories of %s"),
1589                         ctx->graph_name);
1590                 return -1;
1591         }
1592
1593         if (ctx->split) {
1594                 char *lock_name = get_chain_filename(ctx->odb);
1595
1596                 hold_lock_file_for_update(&lk, lock_name, LOCK_DIE_ON_ERROR);
1597
1598                 fd = git_mkstemp_mode(ctx->graph_name, 0444);
1599                 if (fd < 0) {
1600                         error(_("unable to create temporary graph layer"));
1601                         return -1;
1602                 }
1603
1604                 f = hashfd(fd, ctx->graph_name);
1605         } else {
1606                 hold_lock_file_for_update(&lk, ctx->graph_name, LOCK_DIE_ON_ERROR);
1607                 fd = lk.tempfile->fd;
1608                 f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
1609         }
1610
1611         chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
1612         chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
1613         chunk_ids[2] = GRAPH_CHUNKID_DATA;
1614         if (ctx->num_extra_edges) {
1615                 chunk_ids[num_chunks] = GRAPH_CHUNKID_EXTRAEDGES;
1616                 num_chunks++;
1617         }
1618         if (ctx->changed_paths) {
1619                 chunk_ids[num_chunks] = GRAPH_CHUNKID_BLOOMINDEXES;
1620                 num_chunks++;
1621                 chunk_ids[num_chunks] = GRAPH_CHUNKID_BLOOMDATA;
1622                 num_chunks++;
1623         }
1624         if (ctx->num_commit_graphs_after > 1) {
1625                 chunk_ids[num_chunks] = GRAPH_CHUNKID_BASE;
1626                 num_chunks++;
1627         }
1628
1629         chunk_ids[num_chunks] = 0;
1630
1631         chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
1632         chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
1633         chunk_offsets[2] = chunk_offsets[1] + hashsz * ctx->commits.nr;
1634         chunk_offsets[3] = chunk_offsets[2] + (hashsz + 16) * ctx->commits.nr;
1635
1636         num_chunks = 3;
1637         if (ctx->num_extra_edges) {
1638                 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1639                                                 4 * ctx->num_extra_edges;
1640                 num_chunks++;
1641         }
1642         if (ctx->changed_paths) {
1643                 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1644                                                 sizeof(uint32_t) * ctx->commits.nr;
1645                 num_chunks++;
1646
1647                 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1648                                                 sizeof(uint32_t) * 3 + ctx->total_bloom_filter_data_size;
1649                 num_chunks++;
1650         }
1651         if (ctx->num_commit_graphs_after > 1) {
1652                 chunk_offsets[num_chunks + 1] = chunk_offsets[num_chunks] +
1653                                                 hashsz * (ctx->num_commit_graphs_after - 1);
1654                 num_chunks++;
1655         }
1656
1657         hashwrite_be32(f, GRAPH_SIGNATURE);
1658
1659         hashwrite_u8(f, GRAPH_VERSION);
1660         hashwrite_u8(f, oid_version());
1661         hashwrite_u8(f, num_chunks);
1662         hashwrite_u8(f, ctx->num_commit_graphs_after - 1);
1663
1664         for (i = 0; i <= num_chunks; i++) {
1665                 uint32_t chunk_write[3];
1666
1667                 chunk_write[0] = htonl(chunk_ids[i]);
1668                 chunk_write[1] = htonl(chunk_offsets[i] >> 32);
1669                 chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
1670                 hashwrite(f, chunk_write, 12);
1671         }
1672
1673         if (ctx->report_progress) {
1674                 strbuf_addf(&progress_title,
1675                             Q_("Writing out commit graph in %d pass",
1676                                "Writing out commit graph in %d passes",
1677                                num_chunks),
1678                             num_chunks);
1679                 ctx->progress = start_delayed_progress(
1680                         progress_title.buf,
1681                         num_chunks * ctx->commits.nr);
1682         }
1683         write_graph_chunk_fanout(f, ctx);
1684         write_graph_chunk_oids(f, hashsz, ctx);
1685         write_graph_chunk_data(f, hashsz, ctx);
1686         if (ctx->num_extra_edges)
1687                 write_graph_chunk_extra_edges(f, ctx);
1688         if (ctx->changed_paths) {
1689                 write_graph_chunk_bloom_indexes(f, ctx);
1690                 write_graph_chunk_bloom_data(f, ctx, &bloom_settings);
1691         }
1692         if (ctx->num_commit_graphs_after > 1 &&
1693             write_graph_chunk_base(f, ctx)) {
1694                 return -1;
1695         }
1696         stop_progress(&ctx->progress);
1697         strbuf_release(&progress_title);
1698
1699         if (ctx->split && ctx->base_graph_name && ctx->num_commit_graphs_after > 1) {
1700                 char *new_base_hash = xstrdup(oid_to_hex(&ctx->new_base_graph->oid));
1701                 char *new_base_name = get_split_graph_filename(ctx->new_base_graph->odb, new_base_hash);
1702
1703                 free(ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2]);
1704                 free(ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2]);
1705                 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 2] = new_base_name;
1706                 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 2] = new_base_hash;
1707         }
1708
1709         close_commit_graph(ctx->r->objects);
1710         finalize_hashfile(f, file_hash.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
1711
1712         if (ctx->split) {
1713                 FILE *chainf = fdopen_lock_file(&lk, "w");
1714                 char *final_graph_name;
1715                 int result;
1716
1717                 close(fd);
1718
1719                 if (!chainf) {
1720                         error(_("unable to open commit-graph chain file"));
1721                         return -1;
1722                 }
1723
1724                 if (ctx->base_graph_name) {
1725                         const char *dest;
1726                         int idx = ctx->num_commit_graphs_after - 1;
1727                         if (ctx->num_commit_graphs_after > 1)
1728                                 idx--;
1729
1730                         dest = ctx->commit_graph_filenames_after[idx];
1731
1732                         if (strcmp(ctx->base_graph_name, dest)) {
1733                                 result = rename(ctx->base_graph_name, dest);
1734
1735                                 if (result) {
1736                                         error(_("failed to rename base commit-graph file"));
1737                                         return -1;
1738                                 }
1739                         }
1740                 } else {
1741                         char *graph_name = get_commit_graph_filename(ctx->odb);
1742                         unlink(graph_name);
1743                 }
1744
1745                 ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1] = xstrdup(oid_to_hex(&file_hash));
1746                 final_graph_name = get_split_graph_filename(ctx->odb,
1747                                         ctx->commit_graph_hash_after[ctx->num_commit_graphs_after - 1]);
1748                 ctx->commit_graph_filenames_after[ctx->num_commit_graphs_after - 1] = final_graph_name;
1749
1750                 result = rename(ctx->graph_name, final_graph_name);
1751
1752                 for (i = 0; i < ctx->num_commit_graphs_after; i++)
1753                         fprintf(lk.tempfile->fp, "%s\n", ctx->commit_graph_hash_after[i]);
1754
1755                 if (result) {
1756                         error(_("failed to rename temporary commit-graph file"));
1757                         return -1;
1758                 }
1759         }
1760
1761         commit_lock_file(&lk);
1762
1763         return 0;
1764 }
1765
1766 static void split_graph_merge_strategy(struct write_commit_graph_context *ctx)
1767 {
1768         struct commit_graph *g;
1769         uint32_t num_commits;
1770         enum commit_graph_split_flags flags = COMMIT_GRAPH_SPLIT_UNSPECIFIED;
1771         uint32_t i;
1772
1773         int max_commits = 0;
1774         int size_mult = 2;
1775
1776         if (ctx->split_opts) {
1777                 max_commits = ctx->split_opts->max_commits;
1778
1779                 if (ctx->split_opts->size_multiple)
1780                         size_mult = ctx->split_opts->size_multiple;
1781
1782                 flags = ctx->split_opts->flags;
1783         }
1784
1785         g = ctx->r->objects->commit_graph;
1786         num_commits = ctx->commits.nr;
1787         if (flags == COMMIT_GRAPH_SPLIT_REPLACE)
1788                 ctx->num_commit_graphs_after = 1;
1789         else
1790                 ctx->num_commit_graphs_after = ctx->num_commit_graphs_before + 1;
1791
1792         if (flags != COMMIT_GRAPH_SPLIT_MERGE_PROHIBITED &&
1793             flags != COMMIT_GRAPH_SPLIT_REPLACE) {
1794                 while (g && (g->num_commits <= size_mult * num_commits ||
1795                             (max_commits && num_commits > max_commits))) {
1796                         if (g->odb != ctx->odb)
1797                                 break;
1798
1799                         num_commits += g->num_commits;
1800                         g = g->base_graph;
1801
1802                         ctx->num_commit_graphs_after--;
1803                 }
1804         }
1805
1806         if (flags != COMMIT_GRAPH_SPLIT_REPLACE)
1807                 ctx->new_base_graph = g;
1808         else if (ctx->num_commit_graphs_after != 1)
1809                 BUG("split_graph_merge_strategy: num_commit_graphs_after "
1810                     "should be 1 with --split=replace");
1811
1812         if (ctx->num_commit_graphs_after == 2) {
1813                 char *old_graph_name = get_commit_graph_filename(g->odb);
1814
1815                 if (!strcmp(g->filename, old_graph_name) &&
1816                     g->odb != ctx->odb) {
1817                         ctx->num_commit_graphs_after = 1;
1818                         ctx->new_base_graph = NULL;
1819                 }
1820
1821                 free(old_graph_name);
1822         }
1823
1824         CALLOC_ARRAY(ctx->commit_graph_filenames_after, ctx->num_commit_graphs_after);
1825         CALLOC_ARRAY(ctx->commit_graph_hash_after, ctx->num_commit_graphs_after);
1826
1827         for (i = 0; i < ctx->num_commit_graphs_after &&
1828                     i < ctx->num_commit_graphs_before; i++)
1829                 ctx->commit_graph_filenames_after[i] = xstrdup(ctx->commit_graph_filenames_before[i]);
1830
1831         i = ctx->num_commit_graphs_before - 1;
1832         g = ctx->r->objects->commit_graph;
1833
1834         while (g) {
1835                 if (i < ctx->num_commit_graphs_after)
1836                         ctx->commit_graph_hash_after[i] = xstrdup(oid_to_hex(&g->oid));
1837
1838                 i--;
1839                 g = g->base_graph;
1840         }
1841 }
1842
1843 static void merge_commit_graph(struct write_commit_graph_context *ctx,
1844                                struct commit_graph *g)
1845 {
1846         uint32_t i;
1847         uint32_t offset = g->num_commits_in_base;
1848
1849         ALLOC_GROW(ctx->commits.list, ctx->commits.nr + g->num_commits, ctx->commits.alloc);
1850
1851         for (i = 0; i < g->num_commits; i++) {
1852                 struct object_id oid;
1853                 struct commit *result;
1854
1855                 display_progress(ctx->progress, i + 1);
1856
1857                 load_oid_from_graph(g, i + offset, &oid);
1858
1859                 /* only add commits if they still exist in the repo */
1860                 result = lookup_commit_reference_gently(ctx->r, &oid, 1);
1861
1862                 if (result) {
1863                         ctx->commits.list[ctx->commits.nr] = result;
1864                         ctx->commits.nr++;
1865                 }
1866         }
1867 }
1868
1869 static int commit_compare(const void *_a, const void *_b)
1870 {
1871         const struct commit *a = *(const struct commit **)_a;
1872         const struct commit *b = *(const struct commit **)_b;
1873         return oidcmp(&a->object.oid, &b->object.oid);
1874 }
1875
1876 static void sort_and_scan_merged_commits(struct write_commit_graph_context *ctx)
1877 {
1878         uint32_t i;
1879
1880         if (ctx->report_progress)
1881                 ctx->progress = start_delayed_progress(
1882                                         _("Scanning merged commits"),
1883                                         ctx->commits.nr);
1884
1885         QSORT(ctx->commits.list, ctx->commits.nr, commit_compare);
1886
1887         ctx->num_extra_edges = 0;
1888         for (i = 0; i < ctx->commits.nr; i++) {
1889                 display_progress(ctx->progress, i);
1890
1891                 if (i && oideq(&ctx->commits.list[i - 1]->object.oid,
1892                           &ctx->commits.list[i]->object.oid)) {
1893                         die(_("unexpected duplicate commit id %s"),
1894                             oid_to_hex(&ctx->commits.list[i]->object.oid));
1895                 } else {
1896                         unsigned int num_parents;
1897
1898                         num_parents = commit_list_count(ctx->commits.list[i]->parents);
1899                         if (num_parents > 2)
1900                                 ctx->num_extra_edges += num_parents - 1;
1901                 }
1902         }
1903
1904         stop_progress(&ctx->progress);
1905 }
1906
1907 static void merge_commit_graphs(struct write_commit_graph_context *ctx)
1908 {
1909         struct commit_graph *g = ctx->r->objects->commit_graph;
1910         uint32_t current_graph_number = ctx->num_commit_graphs_before;
1911
1912         while (g && current_graph_number >= ctx->num_commit_graphs_after) {
1913                 current_graph_number--;
1914
1915                 if (ctx->report_progress)
1916                         ctx->progress = start_delayed_progress(_("Merging commit-graph"), 0);
1917
1918                 merge_commit_graph(ctx, g);
1919                 stop_progress(&ctx->progress);
1920
1921                 g = g->base_graph;
1922         }
1923
1924         if (g) {
1925                 ctx->new_base_graph = g;
1926                 ctx->new_num_commits_in_base = g->num_commits + g->num_commits_in_base;
1927         }
1928
1929         if (ctx->new_base_graph)
1930                 ctx->base_graph_name = xstrdup(ctx->new_base_graph->filename);
1931
1932         sort_and_scan_merged_commits(ctx);
1933 }
1934
1935 static void mark_commit_graphs(struct write_commit_graph_context *ctx)
1936 {
1937         uint32_t i;
1938         time_t now = time(NULL);
1939
1940         for (i = ctx->num_commit_graphs_after - 1; i < ctx->num_commit_graphs_before; i++) {
1941                 struct stat st;
1942                 struct utimbuf updated_time;
1943
1944                 stat(ctx->commit_graph_filenames_before[i], &st);
1945
1946                 updated_time.actime = st.st_atime;
1947                 updated_time.modtime = now;
1948                 utime(ctx->commit_graph_filenames_before[i], &updated_time);
1949         }
1950 }
1951
1952 static void expire_commit_graphs(struct write_commit_graph_context *ctx)
1953 {
1954         struct strbuf path = STRBUF_INIT;
1955         DIR *dir;
1956         struct dirent *de;
1957         size_t dirnamelen;
1958         timestamp_t expire_time = time(NULL);
1959
1960         if (ctx->split_opts && ctx->split_opts->expire_time)
1961                 expire_time = ctx->split_opts->expire_time;
1962         if (!ctx->split) {
1963                 char *chain_file_name = get_chain_filename(ctx->odb);
1964                 unlink(chain_file_name);
1965                 free(chain_file_name);
1966                 ctx->num_commit_graphs_after = 0;
1967         }
1968
1969         strbuf_addstr(&path, ctx->odb->path);
1970         strbuf_addstr(&path, "/info/commit-graphs");
1971         dir = opendir(path.buf);
1972
1973         if (!dir)
1974                 goto out;
1975
1976         strbuf_addch(&path, '/');
1977         dirnamelen = path.len;
1978         while ((de = readdir(dir)) != NULL) {
1979                 struct stat st;
1980                 uint32_t i, found = 0;
1981
1982                 strbuf_setlen(&path, dirnamelen);
1983                 strbuf_addstr(&path, de->d_name);
1984
1985                 stat(path.buf, &st);
1986
1987                 if (st.st_mtime > expire_time)
1988                         continue;
1989                 if (path.len < 6 || strcmp(path.buf + path.len - 6, ".graph"))
1990                         continue;
1991
1992                 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
1993                         if (!strcmp(ctx->commit_graph_filenames_after[i],
1994                                     path.buf)) {
1995                                 found = 1;
1996                                 break;
1997                         }
1998                 }
1999
2000                 if (!found)
2001                         unlink(path.buf);
2002         }
2003
2004 out:
2005         strbuf_release(&path);
2006 }
2007
2008 int write_commit_graph(struct object_directory *odb,
2009                        struct string_list *pack_indexes,
2010                        struct oidset *commits,
2011                        enum commit_graph_write_flags flags,
2012                        const struct split_commit_graph_opts *split_opts)
2013 {
2014         struct write_commit_graph_context *ctx;
2015         uint32_t i, count_distinct = 0;
2016         int res = 0;
2017         int replace = 0;
2018
2019         if (!commit_graph_compatible(the_repository))
2020                 return 0;
2021
2022         ctx = xcalloc(1, sizeof(struct write_commit_graph_context));
2023         ctx->r = the_repository;
2024         ctx->odb = odb;
2025         ctx->append = flags & COMMIT_GRAPH_WRITE_APPEND ? 1 : 0;
2026         ctx->report_progress = flags & COMMIT_GRAPH_WRITE_PROGRESS ? 1 : 0;
2027         ctx->split = flags & COMMIT_GRAPH_WRITE_SPLIT ? 1 : 0;
2028         ctx->check_oids = flags & COMMIT_GRAPH_WRITE_CHECK_OIDS ? 1 : 0;
2029         ctx->split_opts = split_opts;
2030         ctx->changed_paths = flags & COMMIT_GRAPH_WRITE_BLOOM_FILTERS ? 1 : 0;
2031         ctx->total_bloom_filter_data_size = 0;
2032
2033         if (ctx->split) {
2034                 struct commit_graph *g;
2035                 prepare_commit_graph(ctx->r);
2036
2037                 g = ctx->r->objects->commit_graph;
2038
2039                 while (g) {
2040                         ctx->num_commit_graphs_before++;
2041                         g = g->base_graph;
2042                 }
2043
2044                 if (ctx->num_commit_graphs_before) {
2045                         ALLOC_ARRAY(ctx->commit_graph_filenames_before, ctx->num_commit_graphs_before);
2046                         i = ctx->num_commit_graphs_before;
2047                         g = ctx->r->objects->commit_graph;
2048
2049                         while (g) {
2050                                 ctx->commit_graph_filenames_before[--i] = xstrdup(g->filename);
2051                                 g = g->base_graph;
2052                         }
2053                 }
2054
2055                 if (ctx->split_opts)
2056                         replace = ctx->split_opts->flags & COMMIT_GRAPH_SPLIT_REPLACE;
2057         }
2058
2059         ctx->approx_nr_objects = approximate_object_count();
2060         ctx->oids.alloc = ctx->approx_nr_objects / 32;
2061
2062         if (ctx->split && split_opts && ctx->oids.alloc > split_opts->max_commits)
2063                 ctx->oids.alloc = split_opts->max_commits;
2064
2065         if (ctx->append) {
2066                 prepare_commit_graph_one(ctx->r, ctx->odb);
2067                 if (ctx->r->objects->commit_graph)
2068                         ctx->oids.alloc += ctx->r->objects->commit_graph->num_commits;
2069         }
2070
2071         if (ctx->oids.alloc < 1024)
2072                 ctx->oids.alloc = 1024;
2073         ALLOC_ARRAY(ctx->oids.list, ctx->oids.alloc);
2074
2075         if (ctx->append && ctx->r->objects->commit_graph) {
2076                 struct commit_graph *g = ctx->r->objects->commit_graph;
2077                 for (i = 0; i < g->num_commits; i++) {
2078                         const unsigned char *hash = g->chunk_oid_lookup + g->hash_len * i;
2079                         hashcpy(ctx->oids.list[ctx->oids.nr++].hash, hash);
2080                 }
2081         }
2082
2083         if (pack_indexes) {
2084                 ctx->order_by_pack = 1;
2085                 if ((res = fill_oids_from_packs(ctx, pack_indexes)))
2086                         goto cleanup;
2087         }
2088
2089         if (commits) {
2090                 if ((res = fill_oids_from_commits(ctx, commits)))
2091                         goto cleanup;
2092         }
2093
2094         if (!pack_indexes && !commits) {
2095                 ctx->order_by_pack = 1;
2096                 fill_oids_from_all_packs(ctx);
2097         }
2098
2099         close_reachable(ctx);
2100
2101         count_distinct = count_distinct_commits(ctx);
2102
2103         if (count_distinct >= GRAPH_EDGE_LAST_MASK) {
2104                 error(_("the commit graph format cannot write %d commits"), count_distinct);
2105                 res = -1;
2106                 goto cleanup;
2107         }
2108
2109         ctx->commits.alloc = count_distinct;
2110         ALLOC_ARRAY(ctx->commits.list, ctx->commits.alloc);
2111
2112         copy_oids_to_commits(ctx);
2113
2114         if (ctx->commits.nr >= GRAPH_EDGE_LAST_MASK) {
2115                 error(_("too many commits to write graph"));
2116                 res = -1;
2117                 goto cleanup;
2118         }
2119
2120         if (!ctx->commits.nr && !replace)
2121                 goto cleanup;
2122
2123         if (ctx->split) {
2124                 split_graph_merge_strategy(ctx);
2125
2126                 if (!replace)
2127                         merge_commit_graphs(ctx);
2128         } else
2129                 ctx->num_commit_graphs_after = 1;
2130
2131         compute_generation_numbers(ctx);
2132
2133         if (ctx->changed_paths)
2134                 compute_bloom_filters(ctx);
2135
2136         res = write_commit_graph_file(ctx);
2137
2138         if (ctx->split)
2139                 mark_commit_graphs(ctx);
2140
2141         expire_commit_graphs(ctx);
2142
2143 cleanup:
2144         free(ctx->graph_name);
2145         free(ctx->commits.list);
2146         free(ctx->oids.list);
2147
2148         if (ctx->commit_graph_filenames_after) {
2149                 for (i = 0; i < ctx->num_commit_graphs_after; i++) {
2150                         free(ctx->commit_graph_filenames_after[i]);
2151                         free(ctx->commit_graph_hash_after[i]);
2152                 }
2153
2154                 for (i = 0; i < ctx->num_commit_graphs_before; i++)
2155                         free(ctx->commit_graph_filenames_before[i]);
2156
2157                 free(ctx->commit_graph_filenames_after);
2158                 free(ctx->commit_graph_filenames_before);
2159                 free(ctx->commit_graph_hash_after);
2160         }
2161
2162         free(ctx);
2163
2164         return res;
2165 }
2166
2167 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
2168 static int verify_commit_graph_error;
2169
2170 static void graph_report(const char *fmt, ...)
2171 {
2172         va_list ap;
2173
2174         verify_commit_graph_error = 1;
2175         va_start(ap, fmt);
2176         vfprintf(stderr, fmt, ap);
2177         fprintf(stderr, "\n");
2178         va_end(ap);
2179 }
2180
2181 #define GENERATION_ZERO_EXISTS 1
2182 #define GENERATION_NUMBER_EXISTS 2
2183
2184 int verify_commit_graph(struct repository *r, struct commit_graph *g, int flags)
2185 {
2186         uint32_t i, cur_fanout_pos = 0;
2187         struct object_id prev_oid, cur_oid, checksum;
2188         int generation_zero = 0;
2189         struct hashfile *f;
2190         int devnull;
2191         struct progress *progress = NULL;
2192         int local_error = 0;
2193
2194         if (!g) {
2195                 graph_report("no commit-graph file loaded");
2196                 return 1;
2197         }
2198
2199         verify_commit_graph_error = verify_commit_graph_lite(g);
2200         if (verify_commit_graph_error)
2201                 return verify_commit_graph_error;
2202
2203         devnull = open("/dev/null", O_WRONLY);
2204         f = hashfd(devnull, NULL);
2205         hashwrite(f, g->data, g->data_len - g->hash_len);
2206         finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
2207         if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
2208                 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
2209                 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
2210         }
2211
2212         for (i = 0; i < g->num_commits; i++) {
2213                 struct commit *graph_commit;
2214
2215                 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
2216
2217                 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
2218                         graph_report(_("commit-graph has incorrect OID order: %s then %s"),
2219                                      oid_to_hex(&prev_oid),
2220                                      oid_to_hex(&cur_oid));
2221
2222                 oidcpy(&prev_oid, &cur_oid);
2223
2224                 while (cur_oid.hash[0] > cur_fanout_pos) {
2225                         uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2226
2227                         if (i != fanout_value)
2228                                 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2229                                              cur_fanout_pos, fanout_value, i);
2230                         cur_fanout_pos++;
2231                 }
2232
2233                 graph_commit = lookup_commit(r, &cur_oid);
2234                 if (!parse_commit_in_graph_one(r, g, graph_commit))
2235                         graph_report(_("failed to parse commit %s from commit-graph"),
2236                                      oid_to_hex(&cur_oid));
2237         }
2238
2239         while (cur_fanout_pos < 256) {
2240                 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
2241
2242                 if (g->num_commits != fanout_value)
2243                         graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
2244                                      cur_fanout_pos, fanout_value, i);
2245
2246                 cur_fanout_pos++;
2247         }
2248
2249         if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
2250                 return verify_commit_graph_error;
2251
2252         if (flags & COMMIT_GRAPH_WRITE_PROGRESS)
2253                 progress = start_progress(_("Verifying commits in commit graph"),
2254                                         g->num_commits);
2255
2256         for (i = 0; i < g->num_commits; i++) {
2257                 struct commit *graph_commit, *odb_commit;
2258                 struct commit_list *graph_parents, *odb_parents;
2259                 uint32_t max_generation = 0;
2260
2261                 display_progress(progress, i + 1);
2262                 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
2263
2264                 graph_commit = lookup_commit(r, &cur_oid);
2265                 odb_commit = (struct commit *)create_object(r, &cur_oid, alloc_commit_node(r));
2266                 if (parse_commit_internal(odb_commit, 0, 0)) {
2267                         graph_report(_("failed to parse commit %s from object database for commit-graph"),
2268                                      oid_to_hex(&cur_oid));
2269                         continue;
2270                 }
2271
2272                 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
2273                            get_commit_tree_oid(odb_commit)))
2274                         graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
2275                                      oid_to_hex(&cur_oid),
2276                                      oid_to_hex(get_commit_tree_oid(graph_commit)),
2277                                      oid_to_hex(get_commit_tree_oid(odb_commit)));
2278
2279                 graph_parents = graph_commit->parents;
2280                 odb_parents = odb_commit->parents;
2281
2282                 while (graph_parents) {
2283                         if (odb_parents == NULL) {
2284                                 graph_report(_("commit-graph parent list for commit %s is too long"),
2285                                              oid_to_hex(&cur_oid));
2286                                 break;
2287                         }
2288
2289                         /* parse parent in case it is in a base graph */
2290                         parse_commit_in_graph_one(r, g, graph_parents->item);
2291
2292                         if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
2293                                 graph_report(_("commit-graph parent for %s is %s != %s"),
2294                                              oid_to_hex(&cur_oid),
2295                                              oid_to_hex(&graph_parents->item->object.oid),
2296                                              oid_to_hex(&odb_parents->item->object.oid));
2297
2298                         if (graph_parents->item->generation > max_generation)
2299                                 max_generation = graph_parents->item->generation;
2300
2301                         graph_parents = graph_parents->next;
2302                         odb_parents = odb_parents->next;
2303                 }
2304
2305                 if (odb_parents != NULL)
2306                         graph_report(_("commit-graph parent list for commit %s terminates early"),
2307                                      oid_to_hex(&cur_oid));
2308
2309                 if (!graph_commit->generation) {
2310                         if (generation_zero == GENERATION_NUMBER_EXISTS)
2311                                 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
2312                                              oid_to_hex(&cur_oid));
2313                         generation_zero = GENERATION_ZERO_EXISTS;
2314                 } else if (generation_zero == GENERATION_ZERO_EXISTS)
2315                         graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
2316                                      oid_to_hex(&cur_oid));
2317
2318                 if (generation_zero == GENERATION_ZERO_EXISTS)
2319                         continue;
2320
2321                 /*
2322                  * If one of our parents has generation GENERATION_NUMBER_MAX, then
2323                  * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
2324                  * extra logic in the following condition.
2325                  */
2326                 if (max_generation == GENERATION_NUMBER_MAX)
2327                         max_generation--;
2328
2329                 if (graph_commit->generation != max_generation + 1)
2330                         graph_report(_("commit-graph generation for commit %s is %u != %u"),
2331                                      oid_to_hex(&cur_oid),
2332                                      graph_commit->generation,
2333                                      max_generation + 1);
2334
2335                 if (graph_commit->date != odb_commit->date)
2336                         graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
2337                                      oid_to_hex(&cur_oid),
2338                                      graph_commit->date,
2339                                      odb_commit->date);
2340         }
2341         stop_progress(&progress);
2342
2343         local_error = verify_commit_graph_error;
2344
2345         if (!(flags & COMMIT_GRAPH_VERIFY_SHALLOW) && g->base_graph)
2346                 local_error |= verify_commit_graph(r, g->base_graph, flags);
2347
2348         return local_error;
2349 }
2350
2351 void free_commit_graph(struct commit_graph *g)
2352 {
2353         if (!g)
2354                 return;
2355         if (g->data) {
2356                 munmap((void *)g->data, g->data_len);
2357                 g->data = NULL;
2358         }
2359         free(g->filename);
2360         free(g->bloom_filter_settings);
2361         free(g);
2362 }
2363
2364 void disable_commit_graph(struct repository *r)
2365 {
2366         r->commit_graph_disabled = 1;
2367 }