Merge branch 'jk/perf-aggregate-wo-libjson'
[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
20 #define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
21 #define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
22 #define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
23 #define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
24 #define GRAPH_CHUNKID_EXTRAEDGES 0x45444745 /* "EDGE" */
25
26 #define GRAPH_DATA_WIDTH (the_hash_algo->rawsz + 16)
27
28 #define GRAPH_VERSION_1 0x1
29 #define GRAPH_VERSION GRAPH_VERSION_1
30
31 #define GRAPH_EXTRA_EDGES_NEEDED 0x80000000
32 #define GRAPH_EDGE_LAST_MASK 0x7fffffff
33 #define GRAPH_PARENT_NONE 0x70000000
34
35 #define GRAPH_LAST_EDGE 0x80000000
36
37 #define GRAPH_HEADER_SIZE 8
38 #define GRAPH_FANOUT_SIZE (4 * 256)
39 #define GRAPH_CHUNKLOOKUP_WIDTH 12
40 #define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
41                         + GRAPH_FANOUT_SIZE + the_hash_algo->rawsz)
42
43 char *get_commit_graph_filename(const char *obj_dir)
44 {
45         return xstrfmt("%s/info/commit-graph", obj_dir);
46 }
47
48 static uint8_t oid_version(void)
49 {
50         return 1;
51 }
52
53 static struct commit_graph *alloc_commit_graph(void)
54 {
55         struct commit_graph *g = xcalloc(1, sizeof(*g));
56         g->graph_fd = -1;
57
58         return g;
59 }
60
61 extern int read_replace_refs;
62
63 static int commit_graph_compatible(struct repository *r)
64 {
65         if (!r->gitdir)
66                 return 0;
67
68         if (read_replace_refs) {
69                 prepare_replace_object(r);
70                 if (hashmap_get_size(&r->objects->replace_map->map))
71                         return 0;
72         }
73
74         prepare_commit_graft(r);
75         if (r->parsed_objects && r->parsed_objects->grafts_nr)
76                 return 0;
77         if (is_repository_shallow(r))
78                 return 0;
79
80         return 1;
81 }
82
83 int open_commit_graph(const char *graph_file, int *fd, struct stat *st)
84 {
85         *fd = git_open(graph_file);
86         if (*fd < 0)
87                 return 0;
88         if (fstat(*fd, st)) {
89                 close(*fd);
90                 return 0;
91         }
92         return 1;
93 }
94
95 struct commit_graph *load_commit_graph_one_fd_st(int fd, struct stat *st)
96 {
97         void *graph_map;
98         size_t graph_size;
99         struct commit_graph *ret;
100
101         graph_size = xsize_t(st->st_size);
102
103         if (graph_size < GRAPH_MIN_SIZE) {
104                 close(fd);
105                 error(_("commit-graph file is too small"));
106                 return NULL;
107         }
108         graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
109         ret = parse_commit_graph(graph_map, fd, graph_size);
110
111         if (!ret) {
112                 munmap(graph_map, graph_size);
113                 close(fd);
114         }
115
116         return ret;
117 }
118
119 static int verify_commit_graph_lite(struct commit_graph *g)
120 {
121         /*
122          * Basic validation shared between parse_commit_graph()
123          * which'll be called every time the graph is used, and the
124          * much more expensive verify_commit_graph() used by
125          * "commit-graph verify".
126          *
127          * There should only be very basic checks here to ensure that
128          * we don't e.g. segfault in fill_commit_in_graph(), but
129          * because this is a very hot codepath nothing that e.g. loops
130          * over g->num_commits, or runs a checksum on the commit-graph
131          * itself.
132          */
133         if (!g->chunk_oid_fanout) {
134                 error("commit-graph is missing the OID Fanout chunk");
135                 return 1;
136         }
137         if (!g->chunk_oid_lookup) {
138                 error("commit-graph is missing the OID Lookup chunk");
139                 return 1;
140         }
141         if (!g->chunk_commit_data) {
142                 error("commit-graph is missing the Commit Data chunk");
143                 return 1;
144         }
145
146         return 0;
147 }
148
149 struct commit_graph *parse_commit_graph(void *graph_map, int fd,
150                                         size_t graph_size)
151 {
152         const unsigned char *data, *chunk_lookup;
153         uint32_t i;
154         struct commit_graph *graph;
155         uint64_t last_chunk_offset;
156         uint32_t last_chunk_id;
157         uint32_t graph_signature;
158         unsigned char graph_version, hash_version;
159
160         if (!graph_map)
161                 return NULL;
162
163         if (graph_size < GRAPH_MIN_SIZE)
164                 return NULL;
165
166         data = (const unsigned char *)graph_map;
167
168         graph_signature = get_be32(data);
169         if (graph_signature != GRAPH_SIGNATURE) {
170                 error(_("commit-graph signature %X does not match signature %X"),
171                       graph_signature, GRAPH_SIGNATURE);
172                 return NULL;
173         }
174
175         graph_version = *(unsigned char*)(data + 4);
176         if (graph_version != GRAPH_VERSION) {
177                 error(_("commit-graph version %X does not match version %X"),
178                       graph_version, GRAPH_VERSION);
179                 return NULL;
180         }
181
182         hash_version = *(unsigned char*)(data + 5);
183         if (hash_version != oid_version()) {
184                 error(_("commit-graph hash version %X does not match version %X"),
185                       hash_version, oid_version());
186                 return NULL;
187         }
188
189         graph = alloc_commit_graph();
190
191         graph->hash_len = the_hash_algo->rawsz;
192         graph->num_chunks = *(unsigned char*)(data + 6);
193         graph->graph_fd = fd;
194         graph->data = graph_map;
195         graph->data_len = graph_size;
196
197         last_chunk_id = 0;
198         last_chunk_offset = 8;
199         chunk_lookup = data + 8;
200         for (i = 0; i < graph->num_chunks; i++) {
201                 uint32_t chunk_id;
202                 uint64_t chunk_offset;
203                 int chunk_repeated = 0;
204
205                 if (data + graph_size - chunk_lookup <
206                     GRAPH_CHUNKLOOKUP_WIDTH) {
207                         error(_("commit-graph chunk lookup table entry missing; file may be incomplete"));
208                         free(graph);
209                         return NULL;
210                 }
211
212                 chunk_id = get_be32(chunk_lookup + 0);
213                 chunk_offset = get_be64(chunk_lookup + 4);
214
215                 chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
216
217                 if (chunk_offset > graph_size - the_hash_algo->rawsz) {
218                         error(_("commit-graph improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
219                               (uint32_t)chunk_offset);
220                         free(graph);
221                         return NULL;
222                 }
223
224                 switch (chunk_id) {
225                 case GRAPH_CHUNKID_OIDFANOUT:
226                         if (graph->chunk_oid_fanout)
227                                 chunk_repeated = 1;
228                         else
229                                 graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
230                         break;
231
232                 case GRAPH_CHUNKID_OIDLOOKUP:
233                         if (graph->chunk_oid_lookup)
234                                 chunk_repeated = 1;
235                         else
236                                 graph->chunk_oid_lookup = data + chunk_offset;
237                         break;
238
239                 case GRAPH_CHUNKID_DATA:
240                         if (graph->chunk_commit_data)
241                                 chunk_repeated = 1;
242                         else
243                                 graph->chunk_commit_data = data + chunk_offset;
244                         break;
245
246                 case GRAPH_CHUNKID_EXTRAEDGES:
247                         if (graph->chunk_extra_edges)
248                                 chunk_repeated = 1;
249                         else
250                                 graph->chunk_extra_edges = data + chunk_offset;
251                         break;
252                 }
253
254                 if (chunk_repeated) {
255                         error(_("commit-graph chunk id %08x appears multiple times"), chunk_id);
256                         free(graph);
257                         return NULL;
258                 }
259
260                 if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
261                 {
262                         graph->num_commits = (chunk_offset - last_chunk_offset)
263                                              / graph->hash_len;
264                 }
265
266                 last_chunk_id = chunk_id;
267                 last_chunk_offset = chunk_offset;
268         }
269
270         if (verify_commit_graph_lite(graph))
271                 return NULL;
272
273         return graph;
274 }
275
276 static struct commit_graph *load_commit_graph_one(const char *graph_file)
277 {
278
279         struct stat st;
280         int fd;
281         int open_ok = open_commit_graph(graph_file, &fd, &st);
282
283         if (!open_ok)
284                 return NULL;
285
286         return load_commit_graph_one_fd_st(fd, &st);
287 }
288
289 static void prepare_commit_graph_one(struct repository *r, const char *obj_dir)
290 {
291         char *graph_name;
292
293         if (r->objects->commit_graph)
294                 return;
295
296         graph_name = get_commit_graph_filename(obj_dir);
297         r->objects->commit_graph =
298                 load_commit_graph_one(graph_name);
299
300         FREE_AND_NULL(graph_name);
301 }
302
303 /*
304  * Return 1 if commit_graph is non-NULL, and 0 otherwise.
305  *
306  * On the first invocation, this function attemps to load the commit
307  * graph if the_repository is configured to have one.
308  */
309 static int prepare_commit_graph(struct repository *r)
310 {
311         struct object_directory *odb;
312         int config_value;
313
314         if (git_env_bool(GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD, 0))
315                 die("dying as requested by the '%s' variable on commit-graph load!",
316                     GIT_TEST_COMMIT_GRAPH_DIE_ON_LOAD);
317
318         if (r->objects->commit_graph_attempted)
319                 return !!r->objects->commit_graph;
320         r->objects->commit_graph_attempted = 1;
321
322         if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
323             (repo_config_get_bool(r, "core.commitgraph", &config_value) ||
324             !config_value))
325                 /*
326                  * This repository is not configured to use commit graphs, so
327                  * do not load one. (But report commit_graph_attempted anyway
328                  * so that commit graph loading is not attempted again for this
329                  * repository.)
330                  */
331                 return 0;
332
333         if (!commit_graph_compatible(r))
334                 return 0;
335
336         prepare_alt_odb(r);
337         for (odb = r->objects->odb;
338              !r->objects->commit_graph && odb;
339              odb = odb->next)
340                 prepare_commit_graph_one(r, odb->path);
341         return !!r->objects->commit_graph;
342 }
343
344 int generation_numbers_enabled(struct repository *r)
345 {
346         uint32_t first_generation;
347         struct commit_graph *g;
348         if (!prepare_commit_graph(r))
349                return 0;
350
351         g = r->objects->commit_graph;
352
353         if (!g->num_commits)
354                 return 0;
355
356         first_generation = get_be32(g->chunk_commit_data +
357                                     g->hash_len + 8) >> 2;
358
359         return !!first_generation;
360 }
361
362 void close_commit_graph(struct repository *r)
363 {
364         free_commit_graph(r->objects->commit_graph);
365         r->objects->commit_graph = NULL;
366 }
367
368 static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
369 {
370         return bsearch_hash(oid->hash, g->chunk_oid_fanout,
371                             g->chunk_oid_lookup, g->hash_len, pos);
372 }
373
374 static struct commit_list **insert_parent_or_die(struct repository *r,
375                                                  struct commit_graph *g,
376                                                  uint64_t pos,
377                                                  struct commit_list **pptr)
378 {
379         struct commit *c;
380         struct object_id oid;
381
382         if (pos >= g->num_commits)
383                 die("invalid parent position %"PRIu64, pos);
384
385         hashcpy(oid.hash, g->chunk_oid_lookup + g->hash_len * pos);
386         c = lookup_commit(r, &oid);
387         if (!c)
388                 die(_("could not find commit %s"), oid_to_hex(&oid));
389         c->graph_pos = pos;
390         return &commit_list_insert(c, pptr)->next;
391 }
392
393 static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
394 {
395         const unsigned char *commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * pos;
396         item->graph_pos = pos;
397         item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
398 }
399
400 static inline void set_commit_tree(struct commit *c, struct tree *t)
401 {
402         c->maybe_tree = t;
403 }
404
405 static int fill_commit_in_graph(struct repository *r,
406                                 struct commit *item,
407                                 struct commit_graph *g, uint32_t pos)
408 {
409         uint32_t edge_value;
410         uint32_t *parent_data_ptr;
411         uint64_t date_low, date_high;
412         struct commit_list **pptr;
413         const unsigned char *commit_data = g->chunk_commit_data + (g->hash_len + 16) * pos;
414
415         item->object.parsed = 1;
416         item->graph_pos = pos;
417
418         set_commit_tree(item, NULL);
419
420         date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
421         date_low = get_be32(commit_data + g->hash_len + 12);
422         item->date = (timestamp_t)((date_high << 32) | date_low);
423
424         item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
425
426         pptr = &item->parents;
427
428         edge_value = get_be32(commit_data + g->hash_len);
429         if (edge_value == GRAPH_PARENT_NONE)
430                 return 1;
431         pptr = insert_parent_or_die(r, g, edge_value, pptr);
432
433         edge_value = get_be32(commit_data + g->hash_len + 4);
434         if (edge_value == GRAPH_PARENT_NONE)
435                 return 1;
436         if (!(edge_value & GRAPH_EXTRA_EDGES_NEEDED)) {
437                 pptr = insert_parent_or_die(r, g, edge_value, pptr);
438                 return 1;
439         }
440
441         parent_data_ptr = (uint32_t*)(g->chunk_extra_edges +
442                           4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
443         do {
444                 edge_value = get_be32(parent_data_ptr);
445                 pptr = insert_parent_or_die(r, g,
446                                             edge_value & GRAPH_EDGE_LAST_MASK,
447                                             pptr);
448                 parent_data_ptr++;
449         } while (!(edge_value & GRAPH_LAST_EDGE));
450
451         return 1;
452 }
453
454 static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
455 {
456         if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
457                 *pos = item->graph_pos;
458                 return 1;
459         } else {
460                 return bsearch_graph(g, &(item->object.oid), pos);
461         }
462 }
463
464 static int parse_commit_in_graph_one(struct repository *r,
465                                      struct commit_graph *g,
466                                      struct commit *item)
467 {
468         uint32_t pos;
469
470         if (item->object.parsed)
471                 return 1;
472
473         if (find_commit_in_graph(item, g, &pos))
474                 return fill_commit_in_graph(r, item, g, pos);
475
476         return 0;
477 }
478
479 int parse_commit_in_graph(struct repository *r, struct commit *item)
480 {
481         if (!prepare_commit_graph(r))
482                 return 0;
483         return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
484 }
485
486 void load_commit_graph_info(struct repository *r, struct commit *item)
487 {
488         uint32_t pos;
489         if (!prepare_commit_graph(r))
490                 return;
491         if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
492                 fill_commit_graph_info(item, r->objects->commit_graph, pos);
493 }
494
495 static struct tree *load_tree_for_commit(struct repository *r,
496                                          struct commit_graph *g,
497                                          struct commit *c)
498 {
499         struct object_id oid;
500         const unsigned char *commit_data = g->chunk_commit_data +
501                                            GRAPH_DATA_WIDTH * (c->graph_pos);
502
503         hashcpy(oid.hash, commit_data);
504         set_commit_tree(c, lookup_tree(r, &oid));
505
506         return c->maybe_tree;
507 }
508
509 static struct tree *get_commit_tree_in_graph_one(struct repository *r,
510                                                  struct commit_graph *g,
511                                                  const struct commit *c)
512 {
513         if (c->maybe_tree)
514                 return c->maybe_tree;
515         if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
516                 BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
517
518         return load_tree_for_commit(r, g, (struct commit *)c);
519 }
520
521 struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
522 {
523         return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
524 }
525
526 static void write_graph_chunk_fanout(struct hashfile *f,
527                                      struct commit **commits,
528                                      int nr_commits,
529                                      struct progress *progress,
530                                      uint64_t *progress_cnt)
531 {
532         int i, count = 0;
533         struct commit **list = commits;
534
535         /*
536          * Write the first-level table (the list is sorted,
537          * but we use a 256-entry lookup to be able to avoid
538          * having to do eight extra binary search iterations).
539          */
540         for (i = 0; i < 256; i++) {
541                 while (count < nr_commits) {
542                         if ((*list)->object.oid.hash[0] != i)
543                                 break;
544                         display_progress(progress, ++*progress_cnt);
545                         count++;
546                         list++;
547                 }
548
549                 hashwrite_be32(f, count);
550         }
551 }
552
553 static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
554                                    struct commit **commits, int nr_commits,
555                                    struct progress *progress,
556                                    uint64_t *progress_cnt)
557 {
558         struct commit **list = commits;
559         int count;
560         for (count = 0; count < nr_commits; count++, list++) {
561                 display_progress(progress, ++*progress_cnt);
562                 hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
563         }
564 }
565
566 static const unsigned char *commit_to_sha1(size_t index, void *table)
567 {
568         struct commit **commits = table;
569         return commits[index]->object.oid.hash;
570 }
571
572 static void write_graph_chunk_data(struct hashfile *f, int hash_len,
573                                    struct commit **commits, int nr_commits,
574                                    struct progress *progress,
575                                    uint64_t *progress_cnt)
576 {
577         struct commit **list = commits;
578         struct commit **last = commits + nr_commits;
579         uint32_t num_extra_edges = 0;
580
581         while (list < last) {
582                 struct commit_list *parent;
583                 int edge_value;
584                 uint32_t packedDate[2];
585                 display_progress(progress, ++*progress_cnt);
586
587                 parse_commit_no_graph(*list);
588                 hashwrite(f, get_commit_tree_oid(*list)->hash, hash_len);
589
590                 parent = (*list)->parents;
591
592                 if (!parent)
593                         edge_value = GRAPH_PARENT_NONE;
594                 else {
595                         edge_value = sha1_pos(parent->item->object.oid.hash,
596                                               commits,
597                                               nr_commits,
598                                               commit_to_sha1);
599
600                         if (edge_value < 0)
601                                 BUG("missing parent %s for commit %s",
602                                     oid_to_hex(&parent->item->object.oid),
603                                     oid_to_hex(&(*list)->object.oid));
604                 }
605
606                 hashwrite_be32(f, edge_value);
607
608                 if (parent)
609                         parent = parent->next;
610
611                 if (!parent)
612                         edge_value = GRAPH_PARENT_NONE;
613                 else if (parent->next)
614                         edge_value = GRAPH_EXTRA_EDGES_NEEDED | num_extra_edges;
615                 else {
616                         edge_value = sha1_pos(parent->item->object.oid.hash,
617                                               commits,
618                                               nr_commits,
619                                               commit_to_sha1);
620                         if (edge_value < 0)
621                                 BUG("missing parent %s for commit %s",
622                                     oid_to_hex(&parent->item->object.oid),
623                                     oid_to_hex(&(*list)->object.oid));
624                 }
625
626                 hashwrite_be32(f, edge_value);
627
628                 if (edge_value & GRAPH_EXTRA_EDGES_NEEDED) {
629                         do {
630                                 num_extra_edges++;
631                                 parent = parent->next;
632                         } while (parent);
633                 }
634
635                 if (sizeof((*list)->date) > 4)
636                         packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
637                 else
638                         packedDate[0] = 0;
639
640                 packedDate[0] |= htonl((*list)->generation << 2);
641
642                 packedDate[1] = htonl((*list)->date);
643                 hashwrite(f, packedDate, 8);
644
645                 list++;
646         }
647 }
648
649 static void write_graph_chunk_extra_edges(struct hashfile *f,
650                                           struct commit **commits,
651                                           int nr_commits,
652                                           struct progress *progress,
653                                           uint64_t *progress_cnt)
654 {
655         struct commit **list = commits;
656         struct commit **last = commits + nr_commits;
657         struct commit_list *parent;
658
659         while (list < last) {
660                 int num_parents = 0;
661
662                 display_progress(progress, ++*progress_cnt);
663
664                 for (parent = (*list)->parents; num_parents < 3 && parent;
665                      parent = parent->next)
666                         num_parents++;
667
668                 if (num_parents <= 2) {
669                         list++;
670                         continue;
671                 }
672
673                 /* Since num_parents > 2, this initializer is safe. */
674                 for (parent = (*list)->parents->next; parent; parent = parent->next) {
675                         int edge_value = sha1_pos(parent->item->object.oid.hash,
676                                                   commits,
677                                                   nr_commits,
678                                                   commit_to_sha1);
679
680                         if (edge_value < 0)
681                                 BUG("missing parent %s for commit %s",
682                                     oid_to_hex(&parent->item->object.oid),
683                                     oid_to_hex(&(*list)->object.oid));
684                         else if (!parent->next)
685                                 edge_value |= GRAPH_LAST_EDGE;
686
687                         hashwrite_be32(f, edge_value);
688                 }
689
690                 list++;
691         }
692 }
693
694 static int commit_compare(const void *_a, const void *_b)
695 {
696         const struct object_id *a = (const struct object_id *)_a;
697         const struct object_id *b = (const struct object_id *)_b;
698         return oidcmp(a, b);
699 }
700
701 struct packed_commit_list {
702         struct commit **list;
703         int nr;
704         int alloc;
705 };
706
707 struct packed_oid_list {
708         struct object_id *list;
709         int nr;
710         int alloc;
711         struct progress *progress;
712         int progress_done;
713 };
714
715 static int add_packed_commits(const struct object_id *oid,
716                               struct packed_git *pack,
717                               uint32_t pos,
718                               void *data)
719 {
720         struct packed_oid_list *list = (struct packed_oid_list*)data;
721         enum object_type type;
722         off_t offset = nth_packed_object_offset(pack, pos);
723         struct object_info oi = OBJECT_INFO_INIT;
724
725         if (list->progress)
726                 display_progress(list->progress, ++list->progress_done);
727
728         oi.typep = &type;
729         if (packed_object_info(the_repository, pack, offset, &oi) < 0)
730                 die(_("unable to get type of object %s"), oid_to_hex(oid));
731
732         if (type != OBJ_COMMIT)
733                 return 0;
734
735         ALLOC_GROW(list->list, list->nr + 1, list->alloc);
736         oidcpy(&(list->list[list->nr]), oid);
737         list->nr++;
738
739         return 0;
740 }
741
742 static void add_missing_parents(struct packed_oid_list *oids, struct commit *commit)
743 {
744         struct commit_list *parent;
745         for (parent = commit->parents; parent; parent = parent->next) {
746                 if (!(parent->item->object.flags & UNINTERESTING)) {
747                         ALLOC_GROW(oids->list, oids->nr + 1, oids->alloc);
748                         oidcpy(&oids->list[oids->nr], &(parent->item->object.oid));
749                         oids->nr++;
750                         parent->item->object.flags |= UNINTERESTING;
751                 }
752         }
753 }
754
755 static void close_reachable(struct packed_oid_list *oids, int report_progress)
756 {
757         int i;
758         struct commit *commit;
759         struct progress *progress = NULL;
760
761         if (report_progress)
762                 progress = start_delayed_progress(
763                         _("Loading known commits in commit graph"), oids->nr);
764         for (i = 0; i < oids->nr; i++) {
765                 display_progress(progress, i + 1);
766                 commit = lookup_commit(the_repository, &oids->list[i]);
767                 if (commit)
768                         commit->object.flags |= UNINTERESTING;
769         }
770         stop_progress(&progress);
771
772         /*
773          * As this loop runs, oids->nr may grow, but not more
774          * than the number of missing commits in the reachable
775          * closure.
776          */
777         if (report_progress)
778                 progress = start_delayed_progress(
779                         _("Expanding reachable commits in commit graph"), oids->nr);
780         for (i = 0; i < oids->nr; i++) {
781                 display_progress(progress, i + 1);
782                 commit = lookup_commit(the_repository, &oids->list[i]);
783
784                 if (commit && !parse_commit_no_graph(commit))
785                         add_missing_parents(oids, commit);
786         }
787         stop_progress(&progress);
788
789         if (report_progress)
790                 progress = start_delayed_progress(
791                         _("Clearing commit marks in commit graph"), oids->nr);
792         for (i = 0; i < oids->nr; i++) {
793                 display_progress(progress, i + 1);
794                 commit = lookup_commit(the_repository, &oids->list[i]);
795
796                 if (commit)
797                         commit->object.flags &= ~UNINTERESTING;
798         }
799         stop_progress(&progress);
800 }
801
802 static void compute_generation_numbers(struct packed_commit_list* commits,
803                                        int report_progress)
804 {
805         int i;
806         struct commit_list *list = NULL;
807         struct progress *progress = NULL;
808
809         if (report_progress)
810                 progress = start_progress(
811                         _("Computing commit graph generation numbers"),
812                         commits->nr);
813         for (i = 0; i < commits->nr; i++) {
814                 display_progress(progress, i + 1);
815                 if (commits->list[i]->generation != GENERATION_NUMBER_INFINITY &&
816                     commits->list[i]->generation != GENERATION_NUMBER_ZERO)
817                         continue;
818
819                 commit_list_insert(commits->list[i], &list);
820                 while (list) {
821                         struct commit *current = list->item;
822                         struct commit_list *parent;
823                         int all_parents_computed = 1;
824                         uint32_t max_generation = 0;
825
826                         for (parent = current->parents; parent; parent = parent->next) {
827                                 if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
828                                     parent->item->generation == GENERATION_NUMBER_ZERO) {
829                                         all_parents_computed = 0;
830                                         commit_list_insert(parent->item, &list);
831                                         break;
832                                 } else if (parent->item->generation > max_generation) {
833                                         max_generation = parent->item->generation;
834                                 }
835                         }
836
837                         if (all_parents_computed) {
838                                 current->generation = max_generation + 1;
839                                 pop_commit(&list);
840
841                                 if (current->generation > GENERATION_NUMBER_MAX)
842                                         current->generation = GENERATION_NUMBER_MAX;
843                         }
844                 }
845         }
846         stop_progress(&progress);
847 }
848
849 static int add_ref_to_list(const char *refname,
850                            const struct object_id *oid,
851                            int flags, void *cb_data)
852 {
853         struct string_list *list = (struct string_list *)cb_data;
854
855         string_list_append(list, oid_to_hex(oid));
856         return 0;
857 }
858
859 void write_commit_graph_reachable(const char *obj_dir, int append,
860                                   int report_progress)
861 {
862         struct string_list list = STRING_LIST_INIT_DUP;
863
864         for_each_ref(add_ref_to_list, &list);
865         write_commit_graph(obj_dir, NULL, &list, append, report_progress);
866
867         string_list_clear(&list, 0);
868 }
869
870 void write_commit_graph(const char *obj_dir,
871                         struct string_list *pack_indexes,
872                         struct string_list *commit_hex,
873                         int append, int report_progress)
874 {
875         struct packed_oid_list oids;
876         struct packed_commit_list commits;
877         struct hashfile *f;
878         uint32_t i, count_distinct = 0;
879         char *graph_name;
880         struct lock_file lk = LOCK_INIT;
881         uint32_t chunk_ids[5];
882         uint64_t chunk_offsets[5];
883         int num_chunks;
884         int num_extra_edges;
885         struct commit_list *parent;
886         struct progress *progress = NULL;
887         const unsigned hashsz = the_hash_algo->rawsz;
888         uint64_t progress_cnt = 0;
889         struct strbuf progress_title = STRBUF_INIT;
890         unsigned long approx_nr_objects;
891
892         if (!commit_graph_compatible(the_repository))
893                 return;
894
895         oids.nr = 0;
896         approx_nr_objects = approximate_object_count();
897         oids.alloc = approx_nr_objects / 32;
898         oids.progress = NULL;
899         oids.progress_done = 0;
900
901         if (append) {
902                 prepare_commit_graph_one(the_repository, obj_dir);
903                 if (the_repository->objects->commit_graph)
904                         oids.alloc += the_repository->objects->commit_graph->num_commits;
905         }
906
907         if (oids.alloc < 1024)
908                 oids.alloc = 1024;
909         ALLOC_ARRAY(oids.list, oids.alloc);
910
911         if (append && the_repository->objects->commit_graph) {
912                 struct commit_graph *commit_graph =
913                         the_repository->objects->commit_graph;
914                 for (i = 0; i < commit_graph->num_commits; i++) {
915                         const unsigned char *hash = commit_graph->chunk_oid_lookup +
916                                 commit_graph->hash_len * i;
917                         hashcpy(oids.list[oids.nr++].hash, hash);
918                 }
919         }
920
921         if (pack_indexes) {
922                 struct strbuf packname = STRBUF_INIT;
923                 int dirlen;
924                 strbuf_addf(&packname, "%s/pack/", obj_dir);
925                 dirlen = packname.len;
926                 if (report_progress) {
927                         strbuf_addf(&progress_title,
928                                     Q_("Finding commits for commit graph in %d pack",
929                                        "Finding commits for commit graph in %d packs",
930                                        pack_indexes->nr),
931                                     pack_indexes->nr);
932                         oids.progress = start_delayed_progress(progress_title.buf, 0);
933                         oids.progress_done = 0;
934                 }
935                 for (i = 0; i < pack_indexes->nr; i++) {
936                         struct packed_git *p;
937                         strbuf_setlen(&packname, dirlen);
938                         strbuf_addstr(&packname, pack_indexes->items[i].string);
939                         p = add_packed_git(packname.buf, packname.len, 1);
940                         if (!p)
941                                 die(_("error adding pack %s"), packname.buf);
942                         if (open_pack_index(p))
943                                 die(_("error opening index for %s"), packname.buf);
944                         for_each_object_in_pack(p, add_packed_commits, &oids,
945                                                 FOR_EACH_OBJECT_PACK_ORDER);
946                         close_pack(p);
947                         free(p);
948                 }
949                 stop_progress(&oids.progress);
950                 strbuf_reset(&progress_title);
951                 strbuf_release(&packname);
952         }
953
954         if (commit_hex) {
955                 if (report_progress) {
956                         strbuf_addf(&progress_title,
957                                     Q_("Finding commits for commit graph from %d ref",
958                                        "Finding commits for commit graph from %d refs",
959                                        commit_hex->nr),
960                                     commit_hex->nr);
961                         progress = start_delayed_progress(progress_title.buf,
962                                                           commit_hex->nr);
963                 }
964                 for (i = 0; i < commit_hex->nr; i++) {
965                         const char *end;
966                         struct object_id oid;
967                         struct commit *result;
968
969                         display_progress(progress, i + 1);
970                         if (commit_hex->items[i].string &&
971                             parse_oid_hex(commit_hex->items[i].string, &oid, &end))
972                                 continue;
973
974                         result = lookup_commit_reference_gently(the_repository, &oid, 1);
975
976                         if (result) {
977                                 ALLOC_GROW(oids.list, oids.nr + 1, oids.alloc);
978                                 oidcpy(&oids.list[oids.nr], &(result->object.oid));
979                                 oids.nr++;
980                         }
981                 }
982                 stop_progress(&progress);
983                 strbuf_reset(&progress_title);
984         }
985
986         if (!pack_indexes && !commit_hex) {
987                 if (report_progress)
988                         oids.progress = start_delayed_progress(
989                                 _("Finding commits for commit graph among packed objects"),
990                                 approx_nr_objects);
991                 for_each_packed_object(add_packed_commits, &oids,
992                                        FOR_EACH_OBJECT_PACK_ORDER);
993                 if (oids.progress_done < approx_nr_objects)
994                         display_progress(oids.progress, approx_nr_objects);
995                 stop_progress(&oids.progress);
996         }
997
998         close_reachable(&oids, report_progress);
999
1000         if (report_progress)
1001                 progress = start_delayed_progress(
1002                         _("Counting distinct commits in commit graph"),
1003                         oids.nr);
1004         display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1005         QSORT(oids.list, oids.nr, commit_compare);
1006         count_distinct = 1;
1007         for (i = 1; i < oids.nr; i++) {
1008                 display_progress(progress, i + 1);
1009                 if (!oideq(&oids.list[i - 1], &oids.list[i]))
1010                         count_distinct++;
1011         }
1012         stop_progress(&progress);
1013
1014         if (count_distinct >= GRAPH_EDGE_LAST_MASK)
1015                 die(_("the commit graph format cannot write %d commits"), count_distinct);
1016
1017         commits.nr = 0;
1018         commits.alloc = count_distinct;
1019         ALLOC_ARRAY(commits.list, commits.alloc);
1020
1021         num_extra_edges = 0;
1022         if (report_progress)
1023                 progress = start_delayed_progress(
1024                         _("Finding extra edges in commit graph"),
1025                         oids.nr);
1026         for (i = 0; i < oids.nr; i++) {
1027                 int num_parents = 0;
1028                 display_progress(progress, i + 1);
1029                 if (i > 0 && oideq(&oids.list[i - 1], &oids.list[i]))
1030                         continue;
1031
1032                 commits.list[commits.nr] = lookup_commit(the_repository, &oids.list[i]);
1033                 parse_commit_no_graph(commits.list[commits.nr]);
1034
1035                 for (parent = commits.list[commits.nr]->parents;
1036                      parent; parent = parent->next)
1037                         num_parents++;
1038
1039                 if (num_parents > 2)
1040                         num_extra_edges += num_parents - 1;
1041
1042                 commits.nr++;
1043         }
1044         num_chunks = num_extra_edges ? 4 : 3;
1045         stop_progress(&progress);
1046
1047         if (commits.nr >= GRAPH_EDGE_LAST_MASK)
1048                 die(_("too many commits to write graph"));
1049
1050         compute_generation_numbers(&commits, report_progress);
1051
1052         graph_name = get_commit_graph_filename(obj_dir);
1053         if (safe_create_leading_directories(graph_name)) {
1054                 UNLEAK(graph_name);
1055                 die_errno(_("unable to create leading directories of %s"),
1056                           graph_name);
1057         }
1058
1059         hold_lock_file_for_update(&lk, graph_name, LOCK_DIE_ON_ERROR);
1060         f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
1061
1062         hashwrite_be32(f, GRAPH_SIGNATURE);
1063
1064         hashwrite_u8(f, GRAPH_VERSION);
1065         hashwrite_u8(f, oid_version());
1066         hashwrite_u8(f, num_chunks);
1067         hashwrite_u8(f, 0); /* unused padding byte */
1068
1069         chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
1070         chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
1071         chunk_ids[2] = GRAPH_CHUNKID_DATA;
1072         if (num_extra_edges)
1073                 chunk_ids[3] = GRAPH_CHUNKID_EXTRAEDGES;
1074         else
1075                 chunk_ids[3] = 0;
1076         chunk_ids[4] = 0;
1077
1078         chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
1079         chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
1080         chunk_offsets[2] = chunk_offsets[1] + hashsz * commits.nr;
1081         chunk_offsets[3] = chunk_offsets[2] + (hashsz + 16) * commits.nr;
1082         chunk_offsets[4] = chunk_offsets[3] + 4 * num_extra_edges;
1083
1084         for (i = 0; i <= num_chunks; i++) {
1085                 uint32_t chunk_write[3];
1086
1087                 chunk_write[0] = htonl(chunk_ids[i]);
1088                 chunk_write[1] = htonl(chunk_offsets[i] >> 32);
1089                 chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
1090                 hashwrite(f, chunk_write, 12);
1091         }
1092
1093         if (report_progress) {
1094                 strbuf_addf(&progress_title,
1095                             Q_("Writing out commit graph in %d pass",
1096                                "Writing out commit graph in %d passes",
1097                                num_chunks),
1098                             num_chunks);
1099                 progress = start_delayed_progress(
1100                         progress_title.buf,
1101                         num_chunks * commits.nr);
1102         }
1103         write_graph_chunk_fanout(f, commits.list, commits.nr, progress, &progress_cnt);
1104         write_graph_chunk_oids(f, hashsz, commits.list, commits.nr, progress, &progress_cnt);
1105         write_graph_chunk_data(f, hashsz, commits.list, commits.nr, progress, &progress_cnt);
1106         if (num_extra_edges)
1107                 write_graph_chunk_extra_edges(f, commits.list, commits.nr, progress, &progress_cnt);
1108         stop_progress(&progress);
1109         strbuf_release(&progress_title);
1110
1111         close_commit_graph(the_repository);
1112         finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
1113         commit_lock_file(&lk);
1114
1115         free(graph_name);
1116         free(commits.list);
1117         free(oids.list);
1118 }
1119
1120 #define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
1121 static int verify_commit_graph_error;
1122
1123 static void graph_report(const char *fmt, ...)
1124 {
1125         va_list ap;
1126
1127         verify_commit_graph_error = 1;
1128         va_start(ap, fmt);
1129         vfprintf(stderr, fmt, ap);
1130         fprintf(stderr, "\n");
1131         va_end(ap);
1132 }
1133
1134 #define GENERATION_ZERO_EXISTS 1
1135 #define GENERATION_NUMBER_EXISTS 2
1136
1137 int verify_commit_graph(struct repository *r, struct commit_graph *g)
1138 {
1139         uint32_t i, cur_fanout_pos = 0;
1140         struct object_id prev_oid, cur_oid, checksum;
1141         int generation_zero = 0;
1142         struct hashfile *f;
1143         int devnull;
1144         struct progress *progress = NULL;
1145
1146         if (!g) {
1147                 graph_report("no commit-graph file loaded");
1148                 return 1;
1149         }
1150
1151         verify_commit_graph_error = verify_commit_graph_lite(g);
1152         if (verify_commit_graph_error)
1153                 return verify_commit_graph_error;
1154
1155         devnull = open("/dev/null", O_WRONLY);
1156         f = hashfd(devnull, NULL);
1157         hashwrite(f, g->data, g->data_len - g->hash_len);
1158         finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
1159         if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
1160                 graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
1161                 verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
1162         }
1163
1164         for (i = 0; i < g->num_commits; i++) {
1165                 struct commit *graph_commit;
1166
1167                 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1168
1169                 if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
1170                         graph_report(_("commit-graph has incorrect OID order: %s then %s"),
1171                                      oid_to_hex(&prev_oid),
1172                                      oid_to_hex(&cur_oid));
1173
1174                 oidcpy(&prev_oid, &cur_oid);
1175
1176                 while (cur_oid.hash[0] > cur_fanout_pos) {
1177                         uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1178
1179                         if (i != fanout_value)
1180                                 graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
1181                                              cur_fanout_pos, fanout_value, i);
1182                         cur_fanout_pos++;
1183                 }
1184
1185                 graph_commit = lookup_commit(r, &cur_oid);
1186                 if (!parse_commit_in_graph_one(r, g, graph_commit))
1187                         graph_report(_("failed to parse commit %s from commit-graph"),
1188                                      oid_to_hex(&cur_oid));
1189         }
1190
1191         while (cur_fanout_pos < 256) {
1192                 uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1193
1194                 if (g->num_commits != fanout_value)
1195                         graph_report(_("commit-graph has incorrect fanout value: fanout[%d] = %u != %u"),
1196                                      cur_fanout_pos, fanout_value, i);
1197
1198                 cur_fanout_pos++;
1199         }
1200
1201         if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
1202                 return verify_commit_graph_error;
1203
1204         progress = start_progress(_("Verifying commits in commit graph"),
1205                                   g->num_commits);
1206         for (i = 0; i < g->num_commits; i++) {
1207                 struct commit *graph_commit, *odb_commit;
1208                 struct commit_list *graph_parents, *odb_parents;
1209                 uint32_t max_generation = 0;
1210
1211                 display_progress(progress, i + 1);
1212                 hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1213
1214                 graph_commit = lookup_commit(r, &cur_oid);
1215                 odb_commit = (struct commit *)create_object(r, cur_oid.hash, alloc_commit_node(r));
1216                 if (parse_commit_internal(odb_commit, 0, 0)) {
1217                         graph_report(_("failed to parse commit %s from object database for commit-graph"),
1218                                      oid_to_hex(&cur_oid));
1219                         continue;
1220                 }
1221
1222                 if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
1223                            get_commit_tree_oid(odb_commit)))
1224                         graph_report(_("root tree OID for commit %s in commit-graph is %s != %s"),
1225                                      oid_to_hex(&cur_oid),
1226                                      oid_to_hex(get_commit_tree_oid(graph_commit)),
1227                                      oid_to_hex(get_commit_tree_oid(odb_commit)));
1228
1229                 graph_parents = graph_commit->parents;
1230                 odb_parents = odb_commit->parents;
1231
1232                 while (graph_parents) {
1233                         if (odb_parents == NULL) {
1234                                 graph_report(_("commit-graph parent list for commit %s is too long"),
1235                                              oid_to_hex(&cur_oid));
1236                                 break;
1237                         }
1238
1239                         if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
1240                                 graph_report(_("commit-graph parent for %s is %s != %s"),
1241                                              oid_to_hex(&cur_oid),
1242                                              oid_to_hex(&graph_parents->item->object.oid),
1243                                              oid_to_hex(&odb_parents->item->object.oid));
1244
1245                         if (graph_parents->item->generation > max_generation)
1246                                 max_generation = graph_parents->item->generation;
1247
1248                         graph_parents = graph_parents->next;
1249                         odb_parents = odb_parents->next;
1250                 }
1251
1252                 if (odb_parents != NULL)
1253                         graph_report(_("commit-graph parent list for commit %s terminates early"),
1254                                      oid_to_hex(&cur_oid));
1255
1256                 if (!graph_commit->generation) {
1257                         if (generation_zero == GENERATION_NUMBER_EXISTS)
1258                                 graph_report(_("commit-graph has generation number zero for commit %s, but non-zero elsewhere"),
1259                                              oid_to_hex(&cur_oid));
1260                         generation_zero = GENERATION_ZERO_EXISTS;
1261                 } else if (generation_zero == GENERATION_ZERO_EXISTS)
1262                         graph_report(_("commit-graph has non-zero generation number for commit %s, but zero elsewhere"),
1263                                      oid_to_hex(&cur_oid));
1264
1265                 if (generation_zero == GENERATION_ZERO_EXISTS)
1266                         continue;
1267
1268                 /*
1269                  * If one of our parents has generation GENERATION_NUMBER_MAX, then
1270                  * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
1271                  * extra logic in the following condition.
1272                  */
1273                 if (max_generation == GENERATION_NUMBER_MAX)
1274                         max_generation--;
1275
1276                 if (graph_commit->generation != max_generation + 1)
1277                         graph_report(_("commit-graph generation for commit %s is %u != %u"),
1278                                      oid_to_hex(&cur_oid),
1279                                      graph_commit->generation,
1280                                      max_generation + 1);
1281
1282                 if (graph_commit->date != odb_commit->date)
1283                         graph_report(_("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime),
1284                                      oid_to_hex(&cur_oid),
1285                                      graph_commit->date,
1286                                      odb_commit->date);
1287         }
1288         stop_progress(&progress);
1289
1290         return verify_commit_graph_error;
1291 }
1292
1293 void free_commit_graph(struct commit_graph *g)
1294 {
1295         if (!g)
1296                 return;
1297         if (g->graph_fd >= 0) {
1298                 munmap((void *)g->data, g->data_len);
1299                 g->data = NULL;
1300                 close(g->graph_fd);
1301         }
1302         free(g);
1303 }