Merge branch 'cd/commit-graph-doc'
[git] / pack-bitmap.c
1 #include "cache.h"
2 #include "commit.h"
3 #include "tag.h"
4 #include "diff.h"
5 #include "revision.h"
6 #include "progress.h"
7 #include "list-objects.h"
8 #include "pack.h"
9 #include "pack-bitmap.h"
10 #include "pack-revindex.h"
11 #include "pack-objects.h"
12 #include "packfile.h"
13 #include "repository.h"
14 #include "object-store.h"
15 #include "list-objects-filter-options.h"
16
17 /*
18  * An entry on the bitmap index, representing the bitmap for a given
19  * commit.
20  */
21 struct stored_bitmap {
22         struct object_id oid;
23         struct ewah_bitmap *root;
24         struct stored_bitmap *xor;
25         int flags;
26 };
27
28 /*
29  * The active bitmap index for a repository. By design, repositories only have
30  * a single bitmap index available (the index for the biggest packfile in
31  * the repository), since bitmap indexes need full closure.
32  *
33  * If there is more than one bitmap index available (e.g. because of alternates),
34  * the active bitmap index is the largest one.
35  */
36 struct bitmap_index {
37         /* Packfile to which this bitmap index belongs to */
38         struct packed_git *pack;
39
40         /*
41          * Mark the first `reuse_objects` in the packfile as reused:
42          * they will be sent as-is without using them for repacking
43          * calculations
44          */
45         uint32_t reuse_objects;
46
47         /* mmapped buffer of the whole bitmap index */
48         unsigned char *map;
49         size_t map_size; /* size of the mmaped buffer */
50         size_t map_pos; /* current position when loading the index */
51
52         /*
53          * Type indexes.
54          *
55          * Each bitmap marks which objects in the packfile  are of the given
56          * type. This provides type information when yielding the objects from
57          * the packfile during a walk, which allows for better delta bases.
58          */
59         struct ewah_bitmap *commits;
60         struct ewah_bitmap *trees;
61         struct ewah_bitmap *blobs;
62         struct ewah_bitmap *tags;
63
64         /* Map from object ID -> `stored_bitmap` for all the bitmapped commits */
65         kh_oid_map_t *bitmaps;
66
67         /* Number of bitmapped commits */
68         uint32_t entry_count;
69
70         /* If not NULL, this is a name-hash cache pointing into map. */
71         uint32_t *hashes;
72
73         /*
74          * Extended index.
75          *
76          * When trying to perform bitmap operations with objects that are not
77          * packed in `pack`, these objects are added to this "fake index" and
78          * are assumed to appear at the end of the packfile for all operations
79          */
80         struct eindex {
81                 struct object **objects;
82                 uint32_t *hashes;
83                 uint32_t count, alloc;
84                 kh_oid_pos_t *positions;
85         } ext_index;
86
87         /* Bitmap result of the last performed walk */
88         struct bitmap *result;
89
90         /* "have" bitmap from the last performed walk */
91         struct bitmap *haves;
92
93         /* Version of the bitmap index */
94         unsigned int version;
95 };
96
97 static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st)
98 {
99         struct ewah_bitmap *parent;
100         struct ewah_bitmap *composed;
101
102         if (st->xor == NULL)
103                 return st->root;
104
105         composed = ewah_pool_new();
106         parent = lookup_stored_bitmap(st->xor);
107         ewah_xor(st->root, parent, composed);
108
109         ewah_pool_free(st->root);
110         st->root = composed;
111         st->xor = NULL;
112
113         return composed;
114 }
115
116 /*
117  * Read a bitmap from the current read position on the mmaped
118  * index, and increase the read position accordingly
119  */
120 static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index)
121 {
122         struct ewah_bitmap *b = ewah_pool_new();
123
124         ssize_t bitmap_size = ewah_read_mmap(b,
125                 index->map + index->map_pos,
126                 index->map_size - index->map_pos);
127
128         if (bitmap_size < 0) {
129                 error("Failed to load bitmap index (corrupted?)");
130                 ewah_pool_free(b);
131                 return NULL;
132         }
133
134         index->map_pos += bitmap_size;
135         return b;
136 }
137
138 static int load_bitmap_header(struct bitmap_index *index)
139 {
140         struct bitmap_disk_header *header = (void *)index->map;
141
142         if (index->map_size < sizeof(*header) + the_hash_algo->rawsz)
143                 return error("Corrupted bitmap index (missing header data)");
144
145         if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0)
146                 return error("Corrupted bitmap index file (wrong header)");
147
148         index->version = ntohs(header->version);
149         if (index->version != 1)
150                 return error("Unsupported version for bitmap index file (%d)", index->version);
151
152         /* Parse known bitmap format options */
153         {
154                 uint32_t flags = ntohs(header->options);
155
156                 if ((flags & BITMAP_OPT_FULL_DAG) == 0)
157                         return error("Unsupported options for bitmap index file "
158                                 "(Git requires BITMAP_OPT_FULL_DAG)");
159
160                 if (flags & BITMAP_OPT_HASH_CACHE) {
161                         unsigned char *end = index->map + index->map_size - the_hash_algo->rawsz;
162                         index->hashes = ((uint32_t *)end) - index->pack->num_objects;
163                 }
164         }
165
166         index->entry_count = ntohl(header->entry_count);
167         index->map_pos += sizeof(*header) - GIT_MAX_RAWSZ + the_hash_algo->rawsz;
168         return 0;
169 }
170
171 static struct stored_bitmap *store_bitmap(struct bitmap_index *index,
172                                           struct ewah_bitmap *root,
173                                           const struct object_id *oid,
174                                           struct stored_bitmap *xor_with,
175                                           int flags)
176 {
177         struct stored_bitmap *stored;
178         khiter_t hash_pos;
179         int ret;
180
181         stored = xmalloc(sizeof(struct stored_bitmap));
182         stored->root = root;
183         stored->xor = xor_with;
184         stored->flags = flags;
185         oidcpy(&stored->oid, oid);
186
187         hash_pos = kh_put_oid_map(index->bitmaps, stored->oid, &ret);
188
189         /* a 0 return code means the insertion succeeded with no changes,
190          * because the SHA1 already existed on the map. this is bad, there
191          * shouldn't be duplicated commits in the index */
192         if (ret == 0) {
193                 error("Duplicate entry in bitmap index: %s", oid_to_hex(oid));
194                 return NULL;
195         }
196
197         kh_value(index->bitmaps, hash_pos) = stored;
198         return stored;
199 }
200
201 static inline uint32_t read_be32(const unsigned char *buffer, size_t *pos)
202 {
203         uint32_t result = get_be32(buffer + *pos);
204         (*pos) += sizeof(result);
205         return result;
206 }
207
208 static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
209 {
210         return buffer[(*pos)++];
211 }
212
213 #define MAX_XOR_OFFSET 160
214
215 static int load_bitmap_entries_v1(struct bitmap_index *index)
216 {
217         uint32_t i;
218         struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
219
220         for (i = 0; i < index->entry_count; ++i) {
221                 int xor_offset, flags;
222                 struct ewah_bitmap *bitmap = NULL;
223                 struct stored_bitmap *xor_bitmap = NULL;
224                 uint32_t commit_idx_pos;
225                 struct object_id oid;
226
227                 commit_idx_pos = read_be32(index->map, &index->map_pos);
228                 xor_offset = read_u8(index->map, &index->map_pos);
229                 flags = read_u8(index->map, &index->map_pos);
230
231                 nth_packed_object_id(&oid, index->pack, commit_idx_pos);
232
233                 bitmap = read_bitmap_1(index);
234                 if (!bitmap)
235                         return -1;
236
237                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
238                         return error("Corrupted bitmap pack index");
239
240                 if (xor_offset > 0) {
241                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
242
243                         if (xor_bitmap == NULL)
244                                 return error("Invalid XOR offset in bitmap pack index");
245                 }
246
247                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
248                         index, bitmap, &oid, xor_bitmap, flags);
249         }
250
251         return 0;
252 }
253
254 static char *pack_bitmap_filename(struct packed_git *p)
255 {
256         size_t len;
257
258         if (!strip_suffix(p->pack_name, ".pack", &len))
259                 BUG("pack_name does not end in .pack");
260         return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
261 }
262
263 static int open_pack_bitmap_1(struct bitmap_index *bitmap_git, struct packed_git *packfile)
264 {
265         int fd;
266         struct stat st;
267         char *idx_name;
268
269         if (open_pack_index(packfile))
270                 return -1;
271
272         idx_name = pack_bitmap_filename(packfile);
273         fd = git_open(idx_name);
274         free(idx_name);
275
276         if (fd < 0)
277                 return -1;
278
279         if (fstat(fd, &st)) {
280                 close(fd);
281                 return -1;
282         }
283
284         if (bitmap_git->pack) {
285                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
286                 close(fd);
287                 return -1;
288         }
289
290         bitmap_git->pack = packfile;
291         bitmap_git->map_size = xsize_t(st.st_size);
292         bitmap_git->map = xmmap(NULL, bitmap_git->map_size, PROT_READ, MAP_PRIVATE, fd, 0);
293         bitmap_git->map_pos = 0;
294         close(fd);
295
296         if (load_bitmap_header(bitmap_git) < 0) {
297                 munmap(bitmap_git->map, bitmap_git->map_size);
298                 bitmap_git->map = NULL;
299                 bitmap_git->map_size = 0;
300                 return -1;
301         }
302
303         return 0;
304 }
305
306 static int load_pack_bitmap(struct bitmap_index *bitmap_git)
307 {
308         assert(bitmap_git->map);
309
310         bitmap_git->bitmaps = kh_init_oid_map();
311         bitmap_git->ext_index.positions = kh_init_oid_pos();
312         if (load_pack_revindex(bitmap_git->pack))
313                 goto failed;
314
315         if (!(bitmap_git->commits = read_bitmap_1(bitmap_git)) ||
316                 !(bitmap_git->trees = read_bitmap_1(bitmap_git)) ||
317                 !(bitmap_git->blobs = read_bitmap_1(bitmap_git)) ||
318                 !(bitmap_git->tags = read_bitmap_1(bitmap_git)))
319                 goto failed;
320
321         if (load_bitmap_entries_v1(bitmap_git) < 0)
322                 goto failed;
323
324         return 0;
325
326 failed:
327         munmap(bitmap_git->map, bitmap_git->map_size);
328         bitmap_git->map = NULL;
329         bitmap_git->map_size = 0;
330
331         kh_destroy_oid_map(bitmap_git->bitmaps);
332         bitmap_git->bitmaps = NULL;
333
334         kh_destroy_oid_pos(bitmap_git->ext_index.positions);
335         bitmap_git->ext_index.positions = NULL;
336
337         return -1;
338 }
339
340 static int open_pack_bitmap(struct repository *r,
341                             struct bitmap_index *bitmap_git)
342 {
343         struct packed_git *p;
344         int ret = -1;
345
346         assert(!bitmap_git->map);
347
348         for (p = get_all_packs(r); p; p = p->next) {
349                 if (open_pack_bitmap_1(bitmap_git, p) == 0)
350                         ret = 0;
351         }
352
353         return ret;
354 }
355
356 struct bitmap_index *prepare_bitmap_git(struct repository *r)
357 {
358         struct bitmap_index *bitmap_git = xcalloc(1, sizeof(*bitmap_git));
359
360         if (!open_pack_bitmap(r, bitmap_git) && !load_pack_bitmap(bitmap_git))
361                 return bitmap_git;
362
363         free_bitmap_index(bitmap_git);
364         return NULL;
365 }
366
367 struct include_data {
368         struct bitmap_index *bitmap_git;
369         struct bitmap *base;
370         struct bitmap *seen;
371 };
372
373 static inline int bitmap_position_extended(struct bitmap_index *bitmap_git,
374                                            const struct object_id *oid)
375 {
376         kh_oid_pos_t *positions = bitmap_git->ext_index.positions;
377         khiter_t pos = kh_get_oid_pos(positions, *oid);
378
379         if (pos < kh_end(positions)) {
380                 int bitmap_pos = kh_value(positions, pos);
381                 return bitmap_pos + bitmap_git->pack->num_objects;
382         }
383
384         return -1;
385 }
386
387 static inline int bitmap_position_packfile(struct bitmap_index *bitmap_git,
388                                            const struct object_id *oid)
389 {
390         off_t offset = find_pack_entry_one(oid->hash, bitmap_git->pack);
391         if (!offset)
392                 return -1;
393
394         return find_revindex_position(bitmap_git->pack, offset);
395 }
396
397 static int bitmap_position(struct bitmap_index *bitmap_git,
398                            const struct object_id *oid)
399 {
400         int pos = bitmap_position_packfile(bitmap_git, oid);
401         return (pos >= 0) ? pos : bitmap_position_extended(bitmap_git, oid);
402 }
403
404 static int ext_index_add_object(struct bitmap_index *bitmap_git,
405                                 struct object *object, const char *name)
406 {
407         struct eindex *eindex = &bitmap_git->ext_index;
408
409         khiter_t hash_pos;
410         int hash_ret;
411         int bitmap_pos;
412
413         hash_pos = kh_put_oid_pos(eindex->positions, object->oid, &hash_ret);
414         if (hash_ret > 0) {
415                 if (eindex->count >= eindex->alloc) {
416                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
417                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
418                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
419                 }
420
421                 bitmap_pos = eindex->count;
422                 eindex->objects[eindex->count] = object;
423                 eindex->hashes[eindex->count] = pack_name_hash(name);
424                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
425                 eindex->count++;
426         } else {
427                 bitmap_pos = kh_value(eindex->positions, hash_pos);
428         }
429
430         return bitmap_pos + bitmap_git->pack->num_objects;
431 }
432
433 struct bitmap_show_data {
434         struct bitmap_index *bitmap_git;
435         struct bitmap *base;
436 };
437
438 static void show_object(struct object *object, const char *name, void *data_)
439 {
440         struct bitmap_show_data *data = data_;
441         int bitmap_pos;
442
443         bitmap_pos = bitmap_position(data->bitmap_git, &object->oid);
444
445         if (bitmap_pos < 0)
446                 bitmap_pos = ext_index_add_object(data->bitmap_git, object,
447                                                   name);
448
449         bitmap_set(data->base, bitmap_pos);
450 }
451
452 static void show_commit(struct commit *commit, void *data)
453 {
454 }
455
456 static int add_to_include_set(struct bitmap_index *bitmap_git,
457                               struct include_data *data,
458                               const struct object_id *oid,
459                               int bitmap_pos)
460 {
461         khiter_t hash_pos;
462
463         if (data->seen && bitmap_get(data->seen, bitmap_pos))
464                 return 0;
465
466         if (bitmap_get(data->base, bitmap_pos))
467                 return 0;
468
469         hash_pos = kh_get_oid_map(bitmap_git->bitmaps, *oid);
470         if (hash_pos < kh_end(bitmap_git->bitmaps)) {
471                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, hash_pos);
472                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
473                 return 0;
474         }
475
476         bitmap_set(data->base, bitmap_pos);
477         return 1;
478 }
479
480 static int should_include(struct commit *commit, void *_data)
481 {
482         struct include_data *data = _data;
483         int bitmap_pos;
484
485         bitmap_pos = bitmap_position(data->bitmap_git, &commit->object.oid);
486         if (bitmap_pos < 0)
487                 bitmap_pos = ext_index_add_object(data->bitmap_git,
488                                                   (struct object *)commit,
489                                                   NULL);
490
491         if (!add_to_include_set(data->bitmap_git, data, &commit->object.oid,
492                                 bitmap_pos)) {
493                 struct commit_list *parent = commit->parents;
494
495                 while (parent) {
496                         parent->item->object.flags |= SEEN;
497                         parent = parent->next;
498                 }
499
500                 return 0;
501         }
502
503         return 1;
504 }
505
506 static struct bitmap *find_objects(struct bitmap_index *bitmap_git,
507                                    struct rev_info *revs,
508                                    struct object_list *roots,
509                                    struct bitmap *seen,
510                                    struct list_objects_filter_options *filter)
511 {
512         struct bitmap *base = NULL;
513         int needs_walk = 0;
514
515         struct object_list *not_mapped = NULL;
516
517         /*
518          * Go through all the roots for the walk. The ones that have bitmaps
519          * on the bitmap index will be `or`ed together to form an initial
520          * global reachability analysis.
521          *
522          * The ones without bitmaps in the index will be stored in the
523          * `not_mapped_list` for further processing.
524          */
525         while (roots) {
526                 struct object *object = roots->item;
527                 roots = roots->next;
528
529                 if (object->type == OBJ_COMMIT) {
530                         khiter_t pos = kh_get_oid_map(bitmap_git->bitmaps, object->oid);
531
532                         if (pos < kh_end(bitmap_git->bitmaps)) {
533                                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
534                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
535
536                                 if (base == NULL)
537                                         base = ewah_to_bitmap(or_with);
538                                 else
539                                         bitmap_or_ewah(base, or_with);
540
541                                 object->flags |= SEEN;
542                                 continue;
543                         }
544                 }
545
546                 object_list_insert(object, &not_mapped);
547         }
548
549         /*
550          * Best case scenario: We found bitmaps for all the roots,
551          * so the resulting `or` bitmap has the full reachability analysis
552          */
553         if (not_mapped == NULL)
554                 return base;
555
556         roots = not_mapped;
557
558         /*
559          * Let's iterate through all the roots that don't have bitmaps to
560          * check if we can determine them to be reachable from the existing
561          * global bitmap.
562          *
563          * If we cannot find them in the existing global bitmap, we'll need
564          * to push them to an actual walk and run it until we can confirm
565          * they are reachable
566          */
567         while (roots) {
568                 struct object *object = roots->item;
569                 int pos;
570
571                 roots = roots->next;
572                 pos = bitmap_position(bitmap_git, &object->oid);
573
574                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
575                         object->flags &= ~UNINTERESTING;
576                         add_pending_object(revs, object, "");
577                         needs_walk = 1;
578                 } else {
579                         object->flags |= SEEN;
580                 }
581         }
582
583         if (needs_walk) {
584                 struct include_data incdata;
585                 struct bitmap_show_data show_data;
586
587                 if (base == NULL)
588                         base = bitmap_new();
589
590                 incdata.bitmap_git = bitmap_git;
591                 incdata.base = base;
592                 incdata.seen = seen;
593
594                 revs->include_check = should_include;
595                 revs->include_check_data = &incdata;
596
597                 if (prepare_revision_walk(revs))
598                         die("revision walk setup failed");
599
600                 show_data.bitmap_git = bitmap_git;
601                 show_data.base = base;
602
603                 traverse_commit_list_filtered(filter, revs,
604                                               show_commit, show_object,
605                                               &show_data, NULL);
606         }
607
608         return base;
609 }
610
611 static void show_extended_objects(struct bitmap_index *bitmap_git,
612                                   struct rev_info *revs,
613                                   show_reachable_fn show_reach)
614 {
615         struct bitmap *objects = bitmap_git->result;
616         struct eindex *eindex = &bitmap_git->ext_index;
617         uint32_t i;
618
619         for (i = 0; i < eindex->count; ++i) {
620                 struct object *obj;
621
622                 if (!bitmap_get(objects, bitmap_git->pack->num_objects + i))
623                         continue;
624
625                 obj = eindex->objects[i];
626                 if ((obj->type == OBJ_BLOB && !revs->blob_objects) ||
627                     (obj->type == OBJ_TREE && !revs->tree_objects) ||
628                     (obj->type == OBJ_TAG && !revs->tag_objects))
629                         continue;
630
631                 show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
632         }
633 }
634
635 static void init_type_iterator(struct ewah_iterator *it,
636                                struct bitmap_index *bitmap_git,
637                                enum object_type type)
638 {
639         switch (type) {
640         case OBJ_COMMIT:
641                 ewah_iterator_init(it, bitmap_git->commits);
642                 break;
643
644         case OBJ_TREE:
645                 ewah_iterator_init(it, bitmap_git->trees);
646                 break;
647
648         case OBJ_BLOB:
649                 ewah_iterator_init(it, bitmap_git->blobs);
650                 break;
651
652         case OBJ_TAG:
653                 ewah_iterator_init(it, bitmap_git->tags);
654                 break;
655
656         default:
657                 BUG("object type %d not stored by bitmap type index", type);
658                 break;
659         }
660 }
661
662 static void show_objects_for_type(
663         struct bitmap_index *bitmap_git,
664         enum object_type object_type,
665         show_reachable_fn show_reach)
666 {
667         size_t i = 0;
668         uint32_t offset;
669
670         struct ewah_iterator it;
671         eword_t filter;
672
673         struct bitmap *objects = bitmap_git->result;
674
675         init_type_iterator(&it, bitmap_git, object_type);
676
677         for (i = 0; i < objects->word_alloc &&
678                         ewah_iterator_next(&filter, &it); i++) {
679                 eword_t word = objects->words[i] & filter;
680                 size_t pos = (i * BITS_IN_EWORD);
681
682                 if (!word)
683                         continue;
684
685                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
686                         struct object_id oid;
687                         struct revindex_entry *entry;
688                         uint32_t hash = 0;
689
690                         if ((word >> offset) == 0)
691                                 break;
692
693                         offset += ewah_bit_ctz64(word >> offset);
694
695                         entry = &bitmap_git->pack->revindex[pos + offset];
696                         nth_packed_object_id(&oid, bitmap_git->pack, entry->nr);
697
698                         if (bitmap_git->hashes)
699                                 hash = get_be32(bitmap_git->hashes + entry->nr);
700
701                         show_reach(&oid, object_type, 0, hash, bitmap_git->pack, entry->offset);
702                 }
703         }
704 }
705
706 static int in_bitmapped_pack(struct bitmap_index *bitmap_git,
707                              struct object_list *roots)
708 {
709         while (roots) {
710                 struct object *object = roots->item;
711                 roots = roots->next;
712
713                 if (find_pack_entry_one(object->oid.hash, bitmap_git->pack) > 0)
714                         return 1;
715         }
716
717         return 0;
718 }
719
720 static struct bitmap *find_tip_objects(struct bitmap_index *bitmap_git,
721                                        struct object_list *tip_objects,
722                                        enum object_type type)
723 {
724         struct bitmap *result = bitmap_new();
725         struct object_list *p;
726
727         for (p = tip_objects; p; p = p->next) {
728                 int pos;
729
730                 if (p->item->type != type)
731                         continue;
732
733                 pos = bitmap_position(bitmap_git, &p->item->oid);
734                 if (pos < 0)
735                         continue;
736
737                 bitmap_set(result, pos);
738         }
739
740         return result;
741 }
742
743 static void filter_bitmap_exclude_type(struct bitmap_index *bitmap_git,
744                                        struct object_list *tip_objects,
745                                        struct bitmap *to_filter,
746                                        enum object_type type)
747 {
748         struct eindex *eindex = &bitmap_git->ext_index;
749         struct bitmap *tips;
750         struct ewah_iterator it;
751         eword_t mask;
752         uint32_t i;
753
754         if (type != OBJ_BLOB && type != OBJ_TREE)
755                 BUG("filter_bitmap_exclude_type: unsupported type '%d'", type);
756
757         /*
758          * The non-bitmap version of this filter never removes
759          * objects which the other side specifically asked for,
760          * so we must match that behavior.
761          */
762         tips = find_tip_objects(bitmap_git, tip_objects, type);
763
764         /*
765          * We can use the blob type-bitmap to work in whole words
766          * for the objects that are actually in the bitmapped packfile.
767          */
768         for (i = 0, init_type_iterator(&it, bitmap_git, type);
769              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
770              i++) {
771                 if (i < tips->word_alloc)
772                         mask &= ~tips->words[i];
773                 to_filter->words[i] &= ~mask;
774         }
775
776         /*
777          * Clear any blobs that weren't in the packfile (and so would not have
778          * been caught by the loop above. We'll have to check them
779          * individually.
780          */
781         for (i = 0; i < eindex->count; i++) {
782                 uint32_t pos = i + bitmap_git->pack->num_objects;
783                 if (eindex->objects[i]->type == type &&
784                     bitmap_get(to_filter, pos) &&
785                     !bitmap_get(tips, pos))
786                         bitmap_unset(to_filter, pos);
787         }
788
789         bitmap_free(tips);
790 }
791
792 static void filter_bitmap_blob_none(struct bitmap_index *bitmap_git,
793                                     struct object_list *tip_objects,
794                                     struct bitmap *to_filter)
795 {
796         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
797                                    OBJ_BLOB);
798 }
799
800 static unsigned long get_size_by_pos(struct bitmap_index *bitmap_git,
801                                      uint32_t pos)
802 {
803         struct packed_git *pack = bitmap_git->pack;
804         unsigned long size;
805         struct object_info oi = OBJECT_INFO_INIT;
806
807         oi.sizep = &size;
808
809         if (pos < pack->num_objects) {
810                 struct revindex_entry *entry = &pack->revindex[pos];
811                 if (packed_object_info(the_repository, pack,
812                                        entry->offset, &oi) < 0) {
813                         struct object_id oid;
814                         nth_packed_object_id(&oid, pack, entry->nr);
815                         die(_("unable to get size of %s"), oid_to_hex(&oid));
816                 }
817         } else {
818                 struct eindex *eindex = &bitmap_git->ext_index;
819                 struct object *obj = eindex->objects[pos - pack->num_objects];
820                 if (oid_object_info_extended(the_repository, &obj->oid, &oi, 0) < 0)
821                         die(_("unable to get size of %s"), oid_to_hex(&obj->oid));
822         }
823
824         return size;
825 }
826
827 static void filter_bitmap_blob_limit(struct bitmap_index *bitmap_git,
828                                      struct object_list *tip_objects,
829                                      struct bitmap *to_filter,
830                                      unsigned long limit)
831 {
832         struct eindex *eindex = &bitmap_git->ext_index;
833         struct bitmap *tips;
834         struct ewah_iterator it;
835         eword_t mask;
836         uint32_t i;
837
838         tips = find_tip_objects(bitmap_git, tip_objects, OBJ_BLOB);
839
840         for (i = 0, init_type_iterator(&it, bitmap_git, OBJ_BLOB);
841              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
842              i++) {
843                 eword_t word = to_filter->words[i] & mask;
844                 unsigned offset;
845
846                 for (offset = 0; offset < BITS_IN_EWORD; offset++) {
847                         uint32_t pos;
848
849                         if ((word >> offset) == 0)
850                                 break;
851                         offset += ewah_bit_ctz64(word >> offset);
852                         pos = i * BITS_IN_EWORD + offset;
853
854                         if (!bitmap_get(tips, pos) &&
855                             get_size_by_pos(bitmap_git, pos) >= limit)
856                                 bitmap_unset(to_filter, pos);
857                 }
858         }
859
860         for (i = 0; i < eindex->count; i++) {
861                 uint32_t pos = i + bitmap_git->pack->num_objects;
862                 if (eindex->objects[i]->type == OBJ_BLOB &&
863                     bitmap_get(to_filter, pos) &&
864                     !bitmap_get(tips, pos) &&
865                     get_size_by_pos(bitmap_git, pos) >= limit)
866                         bitmap_unset(to_filter, pos);
867         }
868
869         bitmap_free(tips);
870 }
871
872 static void filter_bitmap_tree_depth(struct bitmap_index *bitmap_git,
873                                      struct object_list *tip_objects,
874                                      struct bitmap *to_filter,
875                                      unsigned long limit)
876 {
877         if (limit)
878                 BUG("filter_bitmap_tree_depth given non-zero limit");
879
880         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
881                                    OBJ_TREE);
882         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
883                                    OBJ_BLOB);
884 }
885
886 static int filter_bitmap(struct bitmap_index *bitmap_git,
887                          struct object_list *tip_objects,
888                          struct bitmap *to_filter,
889                          struct list_objects_filter_options *filter)
890 {
891         if (!filter || filter->choice == LOFC_DISABLED)
892                 return 0;
893
894         if (filter->choice == LOFC_BLOB_NONE) {
895                 if (bitmap_git)
896                         filter_bitmap_blob_none(bitmap_git, tip_objects,
897                                                 to_filter);
898                 return 0;
899         }
900
901         if (filter->choice == LOFC_BLOB_LIMIT) {
902                 if (bitmap_git)
903                         filter_bitmap_blob_limit(bitmap_git, tip_objects,
904                                                  to_filter,
905                                                  filter->blob_limit_value);
906                 return 0;
907         }
908
909         if (filter->choice == LOFC_TREE_DEPTH &&
910             filter->tree_exclude_depth == 0) {
911                 if (bitmap_git)
912                         filter_bitmap_tree_depth(bitmap_git, tip_objects,
913                                                  to_filter,
914                                                  filter->tree_exclude_depth);
915                 return 0;
916         }
917
918         /* filter choice not handled */
919         return -1;
920 }
921
922 static int can_filter_bitmap(struct list_objects_filter_options *filter)
923 {
924         return !filter_bitmap(NULL, NULL, NULL, filter);
925 }
926
927 struct bitmap_index *prepare_bitmap_walk(struct rev_info *revs,
928                                          struct list_objects_filter_options *filter)
929 {
930         unsigned int i;
931
932         struct object_list *wants = NULL;
933         struct object_list *haves = NULL;
934
935         struct bitmap *wants_bitmap = NULL;
936         struct bitmap *haves_bitmap = NULL;
937
938         struct bitmap_index *bitmap_git;
939
940         /*
941          * We can't do pathspec limiting with bitmaps, because we don't know
942          * which commits are associated with which object changes (let alone
943          * even which objects are associated with which paths).
944          */
945         if (revs->prune)
946                 return NULL;
947
948         if (!can_filter_bitmap(filter))
949                 return NULL;
950
951         /* try to open a bitmapped pack, but don't parse it yet
952          * because we may not need to use it */
953         bitmap_git = xcalloc(1, sizeof(*bitmap_git));
954         if (open_pack_bitmap(revs->repo, bitmap_git) < 0)
955                 goto cleanup;
956
957         for (i = 0; i < revs->pending.nr; ++i) {
958                 struct object *object = revs->pending.objects[i].item;
959
960                 if (object->type == OBJ_NONE)
961                         parse_object_or_die(&object->oid, NULL);
962
963                 while (object->type == OBJ_TAG) {
964                         struct tag *tag = (struct tag *) object;
965
966                         if (object->flags & UNINTERESTING)
967                                 object_list_insert(object, &haves);
968                         else
969                                 object_list_insert(object, &wants);
970
971                         object = parse_object_or_die(get_tagged_oid(tag), NULL);
972                 }
973
974                 if (object->flags & UNINTERESTING)
975                         object_list_insert(object, &haves);
976                 else
977                         object_list_insert(object, &wants);
978         }
979
980         /*
981          * if we have a HAVES list, but none of those haves is contained
982          * in the packfile that has a bitmap, we don't have anything to
983          * optimize here
984          */
985         if (haves && !in_bitmapped_pack(bitmap_git, haves))
986                 goto cleanup;
987
988         /* if we don't want anything, we're done here */
989         if (!wants)
990                 goto cleanup;
991
992         /*
993          * now we're going to use bitmaps, so load the actual bitmap entries
994          * from disk. this is the point of no return; after this the rev_list
995          * becomes invalidated and we must perform the revwalk through bitmaps
996          */
997         if (load_pack_bitmap(bitmap_git) < 0)
998                 goto cleanup;
999
1000         object_array_clear(&revs->pending);
1001
1002         if (haves) {
1003                 revs->ignore_missing_links = 1;
1004                 haves_bitmap = find_objects(bitmap_git, revs, haves, NULL,
1005                                             filter);
1006                 reset_revision_walk();
1007                 revs->ignore_missing_links = 0;
1008
1009                 if (haves_bitmap == NULL)
1010                         BUG("failed to perform bitmap walk");
1011         }
1012
1013         wants_bitmap = find_objects(bitmap_git, revs, wants, haves_bitmap,
1014                                     filter);
1015
1016         if (!wants_bitmap)
1017                 BUG("failed to perform bitmap walk");
1018
1019         if (haves_bitmap)
1020                 bitmap_and_not(wants_bitmap, haves_bitmap);
1021
1022         filter_bitmap(bitmap_git, wants, wants_bitmap, filter);
1023
1024         bitmap_git->result = wants_bitmap;
1025         bitmap_git->haves = haves_bitmap;
1026
1027         object_list_free(&wants);
1028         object_list_free(&haves);
1029
1030         return bitmap_git;
1031
1032 cleanup:
1033         free_bitmap_index(bitmap_git);
1034         object_list_free(&wants);
1035         object_list_free(&haves);
1036         return NULL;
1037 }
1038
1039 static void try_partial_reuse(struct bitmap_index *bitmap_git,
1040                               size_t pos,
1041                               struct bitmap *reuse,
1042                               struct pack_window **w_curs)
1043 {
1044         struct revindex_entry *revidx;
1045         off_t offset;
1046         enum object_type type;
1047         unsigned long size;
1048
1049         if (pos >= bitmap_git->pack->num_objects)
1050                 return; /* not actually in the pack */
1051
1052         revidx = &bitmap_git->pack->revindex[pos];
1053         offset = revidx->offset;
1054         type = unpack_object_header(bitmap_git->pack, w_curs, &offset, &size);
1055         if (type < 0)
1056                 return; /* broken packfile, punt */
1057
1058         if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA) {
1059                 off_t base_offset;
1060                 int base_pos;
1061
1062                 /*
1063                  * Find the position of the base object so we can look it up
1064                  * in our bitmaps. If we can't come up with an offset, or if
1065                  * that offset is not in the revidx, the pack is corrupt.
1066                  * There's nothing we can do, so just punt on this object,
1067                  * and the normal slow path will complain about it in
1068                  * more detail.
1069                  */
1070                 base_offset = get_delta_base(bitmap_git->pack, w_curs,
1071                                              &offset, type, revidx->offset);
1072                 if (!base_offset)
1073                         return;
1074                 base_pos = find_revindex_position(bitmap_git->pack, base_offset);
1075                 if (base_pos < 0)
1076                         return;
1077
1078                 /*
1079                  * We assume delta dependencies always point backwards. This
1080                  * lets us do a single pass, and is basically always true
1081                  * due to the way OFS_DELTAs work. You would not typically
1082                  * find REF_DELTA in a bitmapped pack, since we only bitmap
1083                  * packs we write fresh, and OFS_DELTA is the default). But
1084                  * let's double check to make sure the pack wasn't written with
1085                  * odd parameters.
1086                  */
1087                 if (base_pos >= pos)
1088                         return;
1089
1090                 /*
1091                  * And finally, if we're not sending the base as part of our
1092                  * reuse chunk, then don't send this object either. The base
1093                  * would come after us, along with other objects not
1094                  * necessarily in the pack, which means we'd need to convert
1095                  * to REF_DELTA on the fly. Better to just let the normal
1096                  * object_entry code path handle it.
1097                  */
1098                 if (!bitmap_get(reuse, base_pos))
1099                         return;
1100         }
1101
1102         /*
1103          * If we got here, then the object is OK to reuse. Mark it.
1104          */
1105         bitmap_set(reuse, pos);
1106 }
1107
1108 int reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
1109                                        struct packed_git **packfile_out,
1110                                        uint32_t *entries,
1111                                        struct bitmap **reuse_out)
1112 {
1113         struct bitmap *result = bitmap_git->result;
1114         struct bitmap *reuse;
1115         struct pack_window *w_curs = NULL;
1116         size_t i = 0;
1117         uint32_t offset;
1118
1119         assert(result);
1120
1121         while (i < result->word_alloc && result->words[i] == (eword_t)~0)
1122                 i++;
1123
1124         /* Don't mark objects not in the packfile */
1125         if (i > bitmap_git->pack->num_objects / BITS_IN_EWORD)
1126                 i = bitmap_git->pack->num_objects / BITS_IN_EWORD;
1127
1128         reuse = bitmap_word_alloc(i);
1129         memset(reuse->words, 0xFF, i * sizeof(eword_t));
1130
1131         for (; i < result->word_alloc; ++i) {
1132                 eword_t word = result->words[i];
1133                 size_t pos = (i * BITS_IN_EWORD);
1134
1135                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1136                         if ((word >> offset) == 0)
1137                                 break;
1138
1139                         offset += ewah_bit_ctz64(word >> offset);
1140                         try_partial_reuse(bitmap_git, pos + offset, reuse, &w_curs);
1141                 }
1142         }
1143
1144         unuse_pack(&w_curs);
1145
1146         *entries = bitmap_popcount(reuse);
1147         if (!*entries) {
1148                 bitmap_free(reuse);
1149                 return -1;
1150         }
1151
1152         /*
1153          * Drop any reused objects from the result, since they will not
1154          * need to be handled separately.
1155          */
1156         bitmap_and_not(result, reuse);
1157         *packfile_out = bitmap_git->pack;
1158         *reuse_out = reuse;
1159         return 0;
1160 }
1161
1162 int bitmap_walk_contains(struct bitmap_index *bitmap_git,
1163                          struct bitmap *bitmap, const struct object_id *oid)
1164 {
1165         int idx;
1166
1167         if (!bitmap)
1168                 return 0;
1169
1170         idx = bitmap_position(bitmap_git, oid);
1171         return idx >= 0 && bitmap_get(bitmap, idx);
1172 }
1173
1174 void traverse_bitmap_commit_list(struct bitmap_index *bitmap_git,
1175                                  struct rev_info *revs,
1176                                  show_reachable_fn show_reachable)
1177 {
1178         assert(bitmap_git->result);
1179
1180         show_objects_for_type(bitmap_git, OBJ_COMMIT, show_reachable);
1181         if (revs->tree_objects)
1182                 show_objects_for_type(bitmap_git, OBJ_TREE, show_reachable);
1183         if (revs->blob_objects)
1184                 show_objects_for_type(bitmap_git, OBJ_BLOB, show_reachable);
1185         if (revs->tag_objects)
1186                 show_objects_for_type(bitmap_git, OBJ_TAG, show_reachable);
1187
1188         show_extended_objects(bitmap_git, revs, show_reachable);
1189 }
1190
1191 static uint32_t count_object_type(struct bitmap_index *bitmap_git,
1192                                   enum object_type type)
1193 {
1194         struct bitmap *objects = bitmap_git->result;
1195         struct eindex *eindex = &bitmap_git->ext_index;
1196
1197         uint32_t i = 0, count = 0;
1198         struct ewah_iterator it;
1199         eword_t filter;
1200
1201         init_type_iterator(&it, bitmap_git, type);
1202
1203         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
1204                 eword_t word = objects->words[i++] & filter;
1205                 count += ewah_bit_popcount64(word);
1206         }
1207
1208         for (i = 0; i < eindex->count; ++i) {
1209                 if (eindex->objects[i]->type == type &&
1210                         bitmap_get(objects, bitmap_git->pack->num_objects + i))
1211                         count++;
1212         }
1213
1214         return count;
1215 }
1216
1217 void count_bitmap_commit_list(struct bitmap_index *bitmap_git,
1218                               uint32_t *commits, uint32_t *trees,
1219                               uint32_t *blobs, uint32_t *tags)
1220 {
1221         assert(bitmap_git->result);
1222
1223         if (commits)
1224                 *commits = count_object_type(bitmap_git, OBJ_COMMIT);
1225
1226         if (trees)
1227                 *trees = count_object_type(bitmap_git, OBJ_TREE);
1228
1229         if (blobs)
1230                 *blobs = count_object_type(bitmap_git, OBJ_BLOB);
1231
1232         if (tags)
1233                 *tags = count_object_type(bitmap_git, OBJ_TAG);
1234 }
1235
1236 struct bitmap_test_data {
1237         struct bitmap_index *bitmap_git;
1238         struct bitmap *base;
1239         struct progress *prg;
1240         size_t seen;
1241 };
1242
1243 static void test_show_object(struct object *object, const char *name,
1244                              void *data)
1245 {
1246         struct bitmap_test_data *tdata = data;
1247         int bitmap_pos;
1248
1249         bitmap_pos = bitmap_position(tdata->bitmap_git, &object->oid);
1250         if (bitmap_pos < 0)
1251                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
1252
1253         bitmap_set(tdata->base, bitmap_pos);
1254         display_progress(tdata->prg, ++tdata->seen);
1255 }
1256
1257 static void test_show_commit(struct commit *commit, void *data)
1258 {
1259         struct bitmap_test_data *tdata = data;
1260         int bitmap_pos;
1261
1262         bitmap_pos = bitmap_position(tdata->bitmap_git,
1263                                      &commit->object.oid);
1264         if (bitmap_pos < 0)
1265                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
1266
1267         bitmap_set(tdata->base, bitmap_pos);
1268         display_progress(tdata->prg, ++tdata->seen);
1269 }
1270
1271 void test_bitmap_walk(struct rev_info *revs)
1272 {
1273         struct object *root;
1274         struct bitmap *result = NULL;
1275         khiter_t pos;
1276         size_t result_popcnt;
1277         struct bitmap_test_data tdata;
1278         struct bitmap_index *bitmap_git;
1279
1280         if (!(bitmap_git = prepare_bitmap_git(revs->repo)))
1281                 die("failed to load bitmap indexes");
1282
1283         if (revs->pending.nr != 1)
1284                 die("you must specify exactly one commit to test");
1285
1286         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
1287                 bitmap_git->version, bitmap_git->entry_count);
1288
1289         root = revs->pending.objects[0].item;
1290         pos = kh_get_oid_map(bitmap_git->bitmaps, root->oid);
1291
1292         if (pos < kh_end(bitmap_git->bitmaps)) {
1293                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
1294                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
1295
1296                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
1297                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
1298
1299                 result = ewah_to_bitmap(bm);
1300         }
1301
1302         if (result == NULL)
1303                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
1304
1305         revs->tag_objects = 1;
1306         revs->tree_objects = 1;
1307         revs->blob_objects = 1;
1308
1309         result_popcnt = bitmap_popcount(result);
1310
1311         if (prepare_revision_walk(revs))
1312                 die("revision walk setup failed");
1313
1314         tdata.bitmap_git = bitmap_git;
1315         tdata.base = bitmap_new();
1316         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
1317         tdata.seen = 0;
1318
1319         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
1320
1321         stop_progress(&tdata.prg);
1322
1323         if (bitmap_equals(result, tdata.base))
1324                 fprintf(stderr, "OK!\n");
1325         else
1326                 fprintf(stderr, "Mismatch!\n");
1327
1328         free_bitmap_index(bitmap_git);
1329 }
1330
1331 static int rebuild_bitmap(uint32_t *reposition,
1332                           struct ewah_bitmap *source,
1333                           struct bitmap *dest)
1334 {
1335         uint32_t pos = 0;
1336         struct ewah_iterator it;
1337         eword_t word;
1338
1339         ewah_iterator_init(&it, source);
1340
1341         while (ewah_iterator_next(&word, &it)) {
1342                 uint32_t offset, bit_pos;
1343
1344                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1345                         if ((word >> offset) == 0)
1346                                 break;
1347
1348                         offset += ewah_bit_ctz64(word >> offset);
1349
1350                         bit_pos = reposition[pos + offset];
1351                         if (bit_pos > 0)
1352                                 bitmap_set(dest, bit_pos - 1);
1353                         else /* can't reuse, we don't have the object */
1354                                 return -1;
1355                 }
1356
1357                 pos += BITS_IN_EWORD;
1358         }
1359         return 0;
1360 }
1361
1362 int rebuild_existing_bitmaps(struct bitmap_index *bitmap_git,
1363                              struct packing_data *mapping,
1364                              kh_oid_map_t *reused_bitmaps,
1365                              int show_progress)
1366 {
1367         uint32_t i, num_objects;
1368         uint32_t *reposition;
1369         struct bitmap *rebuild;
1370         struct stored_bitmap *stored;
1371         struct progress *progress = NULL;
1372
1373         khiter_t hash_pos;
1374         int hash_ret;
1375
1376         num_objects = bitmap_git->pack->num_objects;
1377         reposition = xcalloc(num_objects, sizeof(uint32_t));
1378
1379         for (i = 0; i < num_objects; ++i) {
1380                 struct object_id oid;
1381                 struct revindex_entry *entry;
1382                 struct object_entry *oe;
1383
1384                 entry = &bitmap_git->pack->revindex[i];
1385                 nth_packed_object_id(&oid, bitmap_git->pack, entry->nr);
1386                 oe = packlist_find(mapping, &oid);
1387
1388                 if (oe)
1389                         reposition[i] = oe_in_pack_pos(mapping, oe) + 1;
1390         }
1391
1392         rebuild = bitmap_new();
1393         i = 0;
1394
1395         if (show_progress)
1396                 progress = start_progress("Reusing bitmaps", 0);
1397
1398         kh_foreach_value(bitmap_git->bitmaps, stored, {
1399                 if (stored->flags & BITMAP_FLAG_REUSE) {
1400                         if (!rebuild_bitmap(reposition,
1401                                             lookup_stored_bitmap(stored),
1402                                             rebuild)) {
1403                                 hash_pos = kh_put_oid_map(reused_bitmaps,
1404                                                           stored->oid,
1405                                                           &hash_ret);
1406                                 kh_value(reused_bitmaps, hash_pos) =
1407                                         bitmap_to_ewah(rebuild);
1408                         }
1409                         bitmap_reset(rebuild);
1410                         display_progress(progress, ++i);
1411                 }
1412         });
1413
1414         stop_progress(&progress);
1415
1416         free(reposition);
1417         bitmap_free(rebuild);
1418         return 0;
1419 }
1420
1421 void free_bitmap_index(struct bitmap_index *b)
1422 {
1423         if (!b)
1424                 return;
1425
1426         if (b->map)
1427                 munmap(b->map, b->map_size);
1428         ewah_pool_free(b->commits);
1429         ewah_pool_free(b->trees);
1430         ewah_pool_free(b->blobs);
1431         ewah_pool_free(b->tags);
1432         kh_destroy_oid_map(b->bitmaps);
1433         free(b->ext_index.objects);
1434         free(b->ext_index.hashes);
1435         bitmap_free(b->result);
1436         bitmap_free(b->haves);
1437         free(b);
1438 }
1439
1440 int bitmap_has_oid_in_uninteresting(struct bitmap_index *bitmap_git,
1441                                     const struct object_id *oid)
1442 {
1443         return bitmap_git &&
1444                 bitmap_walk_contains(bitmap_git, bitmap_git->haves, oid);
1445 }