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