t5504: make hash algorithm independent
[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
16 /*
17  * An entry on the bitmap index, representing the bitmap for a given
18  * commit.
19  */
20 struct stored_bitmap {
21         struct object_id oid;
22         struct ewah_bitmap *root;
23         struct stored_bitmap *xor;
24         int flags;
25 };
26
27 /*
28  * The active bitmap index for a repository. By design, repositories only have
29  * a single bitmap index available (the index for the biggest packfile in
30  * the repository), since bitmap indexes need full closure.
31  *
32  * If there is more than one bitmap index available (e.g. because of alternates),
33  * the active bitmap index is the largest one.
34  */
35 struct bitmap_index {
36         /* Packfile to which this bitmap index belongs to */
37         struct packed_git *pack;
38
39         /*
40          * Mark the first `reuse_objects` in the packfile as reused:
41          * they will be sent as-is without using them for repacking
42          * calculations
43          */
44         uint32_t reuse_objects;
45
46         /* mmapped buffer of the whole bitmap index */
47         unsigned char *map;
48         size_t map_size; /* size of the mmaped buffer */
49         size_t map_pos; /* current position when loading the index */
50
51         /*
52          * Type indexes.
53          *
54          * Each bitmap marks which objects in the packfile  are of the given
55          * type. This provides type information when yielding the objects from
56          * the packfile during a walk, which allows for better delta bases.
57          */
58         struct ewah_bitmap *commits;
59         struct ewah_bitmap *trees;
60         struct ewah_bitmap *blobs;
61         struct ewah_bitmap *tags;
62
63         /* Map from object ID -> `stored_bitmap` for all the bitmapped commits */
64         kh_oid_map_t *bitmaps;
65
66         /* Number of bitmapped commits */
67         uint32_t entry_count;
68
69         /* If not NULL, this is a name-hash cache pointing into map. */
70         uint32_t *hashes;
71
72         /*
73          * Extended index.
74          *
75          * When trying to perform bitmap operations with objects that are not
76          * packed in `pack`, these objects are added to this "fake index" and
77          * are assumed to appear at the end of the packfile for all operations
78          */
79         struct eindex {
80                 struct object **objects;
81                 uint32_t *hashes;
82                 uint32_t count, alloc;
83                 kh_oid_pos_t *positions;
84         } ext_index;
85
86         /* Bitmap result of the last performed walk */
87         struct bitmap *result;
88
89         /* "have" bitmap from the last performed walk */
90         struct bitmap *haves;
91
92         /* Version of the bitmap index */
93         unsigned int version;
94 };
95
96 static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st)
97 {
98         struct ewah_bitmap *parent;
99         struct ewah_bitmap *composed;
100
101         if (st->xor == NULL)
102                 return st->root;
103
104         composed = ewah_pool_new();
105         parent = lookup_stored_bitmap(st->xor);
106         ewah_xor(st->root, parent, composed);
107
108         ewah_pool_free(st->root);
109         st->root = composed;
110         st->xor = NULL;
111
112         return composed;
113 }
114
115 /*
116  * Read a bitmap from the current read position on the mmaped
117  * index, and increase the read position accordingly
118  */
119 static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index)
120 {
121         struct ewah_bitmap *b = ewah_pool_new();
122
123         ssize_t bitmap_size = ewah_read_mmap(b,
124                 index->map + index->map_pos,
125                 index->map_size - index->map_pos);
126
127         if (bitmap_size < 0) {
128                 error("Failed to load bitmap index (corrupted?)");
129                 ewah_pool_free(b);
130                 return NULL;
131         }
132
133         index->map_pos += bitmap_size;
134         return b;
135 }
136
137 static int load_bitmap_header(struct bitmap_index *index)
138 {
139         struct bitmap_disk_header *header = (void *)index->map;
140
141         if (index->map_size < sizeof(*header) + the_hash_algo->rawsz)
142                 return error("Corrupted bitmap index (missing header data)");
143
144         if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0)
145                 return error("Corrupted bitmap index file (wrong header)");
146
147         index->version = ntohs(header->version);
148         if (index->version != 1)
149                 return error("Unsupported version for bitmap index file (%d)", index->version);
150
151         /* Parse known bitmap format options */
152         {
153                 uint32_t flags = ntohs(header->options);
154
155                 if ((flags & BITMAP_OPT_FULL_DAG) == 0)
156                         return error("Unsupported options for bitmap index file "
157                                 "(Git requires BITMAP_OPT_FULL_DAG)");
158
159                 if (flags & BITMAP_OPT_HASH_CACHE) {
160                         unsigned char *end = index->map + index->map_size - the_hash_algo->rawsz;
161                         index->hashes = ((uint32_t *)end) - index->pack->num_objects;
162                 }
163         }
164
165         index->entry_count = ntohl(header->entry_count);
166         index->map_pos += sizeof(*header) - GIT_MAX_RAWSZ + the_hash_algo->rawsz;
167         return 0;
168 }
169
170 static struct stored_bitmap *store_bitmap(struct bitmap_index *index,
171                                           struct ewah_bitmap *root,
172                                           const unsigned char *hash,
173                                           struct stored_bitmap *xor_with,
174                                           int flags)
175 {
176         struct stored_bitmap *stored;
177         khiter_t hash_pos;
178         int ret;
179
180         stored = xmalloc(sizeof(struct stored_bitmap));
181         stored->root = root;
182         stored->xor = xor_with;
183         stored->flags = flags;
184         oidread(&stored->oid, hash);
185
186         hash_pos = kh_put_oid_map(index->bitmaps, stored->oid, &ret);
187
188         /* a 0 return code means the insertion succeeded with no changes,
189          * because the SHA1 already existed on the map. this is bad, there
190          * shouldn't be duplicated commits in the index */
191         if (ret == 0) {
192                 error("Duplicate entry in bitmap index: %s", hash_to_hex(hash));
193                 return NULL;
194         }
195
196         kh_value(index->bitmaps, hash_pos) = stored;
197         return stored;
198 }
199
200 static inline uint32_t read_be32(const unsigned char *buffer, size_t *pos)
201 {
202         uint32_t result = get_be32(buffer + *pos);
203         (*pos) += sizeof(result);
204         return result;
205 }
206
207 static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
208 {
209         return buffer[(*pos)++];
210 }
211
212 #define MAX_XOR_OFFSET 160
213
214 static int load_bitmap_entries_v1(struct bitmap_index *index)
215 {
216         uint32_t i;
217         struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
218
219         for (i = 0; i < index->entry_count; ++i) {
220                 int xor_offset, flags;
221                 struct ewah_bitmap *bitmap = NULL;
222                 struct stored_bitmap *xor_bitmap = NULL;
223                 uint32_t commit_idx_pos;
224                 const unsigned char *sha1;
225
226                 commit_idx_pos = read_be32(index->map, &index->map_pos);
227                 xor_offset = read_u8(index->map, &index->map_pos);
228                 flags = read_u8(index->map, &index->map_pos);
229
230                 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
231
232                 bitmap = read_bitmap_1(index);
233                 if (!bitmap)
234                         return -1;
235
236                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
237                         return error("Corrupted bitmap pack index");
238
239                 if (xor_offset > 0) {
240                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
241
242                         if (xor_bitmap == NULL)
243                                 return error("Invalid XOR offset in bitmap pack index");
244                 }
245
246                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
247                         index, bitmap, sha1, xor_bitmap, flags);
248         }
249
250         return 0;
251 }
252
253 static char *pack_bitmap_filename(struct packed_git *p)
254 {
255         size_t len;
256
257         if (!strip_suffix(p->pack_name, ".pack", &len))
258                 BUG("pack_name does not end in .pack");
259         return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
260 }
261
262 static int open_pack_bitmap_1(struct bitmap_index *bitmap_git, struct packed_git *packfile)
263 {
264         int fd;
265         struct stat st;
266         char *idx_name;
267
268         if (open_pack_index(packfile))
269                 return -1;
270
271         idx_name = pack_bitmap_filename(packfile);
272         fd = git_open(idx_name);
273         free(idx_name);
274
275         if (fd < 0)
276                 return -1;
277
278         if (fstat(fd, &st)) {
279                 close(fd);
280                 return -1;
281         }
282
283         if (bitmap_git->pack) {
284                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
285                 close(fd);
286                 return -1;
287         }
288
289         bitmap_git->pack = packfile;
290         bitmap_git->map_size = xsize_t(st.st_size);
291         bitmap_git->map = xmmap(NULL, bitmap_git->map_size, PROT_READ, MAP_PRIVATE, fd, 0);
292         bitmap_git->map_pos = 0;
293         close(fd);
294
295         if (load_bitmap_header(bitmap_git) < 0) {
296                 munmap(bitmap_git->map, bitmap_git->map_size);
297                 bitmap_git->map = NULL;
298                 bitmap_git->map_size = 0;
299                 return -1;
300         }
301
302         return 0;
303 }
304
305 static int load_pack_bitmap(struct bitmap_index *bitmap_git)
306 {
307         assert(bitmap_git->map);
308
309         bitmap_git->bitmaps = kh_init_oid_map();
310         bitmap_git->ext_index.positions = kh_init_oid_pos();
311         if (load_pack_revindex(bitmap_git->pack))
312                 goto failed;
313
314         if (!(bitmap_git->commits = read_bitmap_1(bitmap_git)) ||
315                 !(bitmap_git->trees = read_bitmap_1(bitmap_git)) ||
316                 !(bitmap_git->blobs = read_bitmap_1(bitmap_git)) ||
317                 !(bitmap_git->tags = read_bitmap_1(bitmap_git)))
318                 goto failed;
319
320         if (load_bitmap_entries_v1(bitmap_git) < 0)
321                 goto failed;
322
323         return 0;
324
325 failed:
326         munmap(bitmap_git->map, bitmap_git->map_size);
327         bitmap_git->map = NULL;
328         bitmap_git->map_size = 0;
329         return -1;
330 }
331
332 static int open_pack_bitmap(struct repository *r,
333                             struct bitmap_index *bitmap_git)
334 {
335         struct packed_git *p;
336         int ret = -1;
337
338         assert(!bitmap_git->map);
339
340         for (p = get_all_packs(r); p; p = p->next) {
341                 if (open_pack_bitmap_1(bitmap_git, p) == 0)
342                         ret = 0;
343         }
344
345         return ret;
346 }
347
348 struct bitmap_index *prepare_bitmap_git(struct repository *r)
349 {
350         struct bitmap_index *bitmap_git = xcalloc(1, sizeof(*bitmap_git));
351
352         if (!open_pack_bitmap(r, bitmap_git) && !load_pack_bitmap(bitmap_git))
353                 return bitmap_git;
354
355         free_bitmap_index(bitmap_git);
356         return NULL;
357 }
358
359 struct include_data {
360         struct bitmap_index *bitmap_git;
361         struct bitmap *base;
362         struct bitmap *seen;
363 };
364
365 static inline int bitmap_position_extended(struct bitmap_index *bitmap_git,
366                                            const struct object_id *oid)
367 {
368         kh_oid_pos_t *positions = bitmap_git->ext_index.positions;
369         khiter_t pos = kh_get_oid_pos(positions, *oid);
370
371         if (pos < kh_end(positions)) {
372                 int bitmap_pos = kh_value(positions, pos);
373                 return bitmap_pos + bitmap_git->pack->num_objects;
374         }
375
376         return -1;
377 }
378
379 static inline int bitmap_position_packfile(struct bitmap_index *bitmap_git,
380                                            const struct object_id *oid)
381 {
382         off_t offset = find_pack_entry_one(oid->hash, bitmap_git->pack);
383         if (!offset)
384                 return -1;
385
386         return find_revindex_position(bitmap_git->pack, offset);
387 }
388
389 static int bitmap_position(struct bitmap_index *bitmap_git,
390                            const struct object_id *oid)
391 {
392         int pos = bitmap_position_packfile(bitmap_git, oid);
393         return (pos >= 0) ? pos : bitmap_position_extended(bitmap_git, oid);
394 }
395
396 static int ext_index_add_object(struct bitmap_index *bitmap_git,
397                                 struct object *object, const char *name)
398 {
399         struct eindex *eindex = &bitmap_git->ext_index;
400
401         khiter_t hash_pos;
402         int hash_ret;
403         int bitmap_pos;
404
405         hash_pos = kh_put_oid_pos(eindex->positions, object->oid, &hash_ret);
406         if (hash_ret > 0) {
407                 if (eindex->count >= eindex->alloc) {
408                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
409                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
410                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
411                 }
412
413                 bitmap_pos = eindex->count;
414                 eindex->objects[eindex->count] = object;
415                 eindex->hashes[eindex->count] = pack_name_hash(name);
416                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
417                 eindex->count++;
418         } else {
419                 bitmap_pos = kh_value(eindex->positions, hash_pos);
420         }
421
422         return bitmap_pos + bitmap_git->pack->num_objects;
423 }
424
425 struct bitmap_show_data {
426         struct bitmap_index *bitmap_git;
427         struct bitmap *base;
428 };
429
430 static void show_object(struct object *object, const char *name, void *data_)
431 {
432         struct bitmap_show_data *data = data_;
433         int bitmap_pos;
434
435         bitmap_pos = bitmap_position(data->bitmap_git, &object->oid);
436
437         if (bitmap_pos < 0)
438                 bitmap_pos = ext_index_add_object(data->bitmap_git, object,
439                                                   name);
440
441         bitmap_set(data->base, bitmap_pos);
442 }
443
444 static void show_commit(struct commit *commit, void *data)
445 {
446 }
447
448 static int add_to_include_set(struct bitmap_index *bitmap_git,
449                               struct include_data *data,
450                               const struct object_id *oid,
451                               int bitmap_pos)
452 {
453         khiter_t hash_pos;
454
455         if (data->seen && bitmap_get(data->seen, bitmap_pos))
456                 return 0;
457
458         if (bitmap_get(data->base, bitmap_pos))
459                 return 0;
460
461         hash_pos = kh_get_oid_map(bitmap_git->bitmaps, *oid);
462         if (hash_pos < kh_end(bitmap_git->bitmaps)) {
463                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, hash_pos);
464                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
465                 return 0;
466         }
467
468         bitmap_set(data->base, bitmap_pos);
469         return 1;
470 }
471
472 static int should_include(struct commit *commit, void *_data)
473 {
474         struct include_data *data = _data;
475         int bitmap_pos;
476
477         bitmap_pos = bitmap_position(data->bitmap_git, &commit->object.oid);
478         if (bitmap_pos < 0)
479                 bitmap_pos = ext_index_add_object(data->bitmap_git,
480                                                   (struct object *)commit,
481                                                   NULL);
482
483         if (!add_to_include_set(data->bitmap_git, data, &commit->object.oid,
484                                 bitmap_pos)) {
485                 struct commit_list *parent = commit->parents;
486
487                 while (parent) {
488                         parent->item->object.flags |= SEEN;
489                         parent = parent->next;
490                 }
491
492                 return 0;
493         }
494
495         return 1;
496 }
497
498 static struct bitmap *find_objects(struct bitmap_index *bitmap_git,
499                                    struct rev_info *revs,
500                                    struct object_list *roots,
501                                    struct bitmap *seen)
502 {
503         struct bitmap *base = NULL;
504         int needs_walk = 0;
505
506         struct object_list *not_mapped = NULL;
507
508         /*
509          * Go through all the roots for the walk. The ones that have bitmaps
510          * on the bitmap index will be `or`ed together to form an initial
511          * global reachability analysis.
512          *
513          * The ones without bitmaps in the index will be stored in the
514          * `not_mapped_list` for further processing.
515          */
516         while (roots) {
517                 struct object *object = roots->item;
518                 roots = roots->next;
519
520                 if (object->type == OBJ_COMMIT) {
521                         khiter_t pos = kh_get_oid_map(bitmap_git->bitmaps, object->oid);
522
523                         if (pos < kh_end(bitmap_git->bitmaps)) {
524                                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
525                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
526
527                                 if (base == NULL)
528                                         base = ewah_to_bitmap(or_with);
529                                 else
530                                         bitmap_or_ewah(base, or_with);
531
532                                 object->flags |= SEEN;
533                                 continue;
534                         }
535                 }
536
537                 object_list_insert(object, &not_mapped);
538         }
539
540         /*
541          * Best case scenario: We found bitmaps for all the roots,
542          * so the resulting `or` bitmap has the full reachability analysis
543          */
544         if (not_mapped == NULL)
545                 return base;
546
547         roots = not_mapped;
548
549         /*
550          * Let's iterate through all the roots that don't have bitmaps to
551          * check if we can determine them to be reachable from the existing
552          * global bitmap.
553          *
554          * If we cannot find them in the existing global bitmap, we'll need
555          * to push them to an actual walk and run it until we can confirm
556          * they are reachable
557          */
558         while (roots) {
559                 struct object *object = roots->item;
560                 int pos;
561
562                 roots = roots->next;
563                 pos = bitmap_position(bitmap_git, &object->oid);
564
565                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
566                         object->flags &= ~UNINTERESTING;
567                         add_pending_object(revs, object, "");
568                         needs_walk = 1;
569                 } else {
570                         object->flags |= SEEN;
571                 }
572         }
573
574         if (needs_walk) {
575                 struct include_data incdata;
576                 struct bitmap_show_data show_data;
577
578                 if (base == NULL)
579                         base = bitmap_new();
580
581                 incdata.bitmap_git = bitmap_git;
582                 incdata.base = base;
583                 incdata.seen = seen;
584
585                 revs->include_check = should_include;
586                 revs->include_check_data = &incdata;
587
588                 if (prepare_revision_walk(revs))
589                         die("revision walk setup failed");
590
591                 show_data.bitmap_git = bitmap_git;
592                 show_data.base = base;
593
594                 traverse_commit_list(revs, show_commit, show_object,
595                                      &show_data);
596         }
597
598         return base;
599 }
600
601 static void show_extended_objects(struct bitmap_index *bitmap_git,
602                                   show_reachable_fn show_reach)
603 {
604         struct bitmap *objects = bitmap_git->result;
605         struct eindex *eindex = &bitmap_git->ext_index;
606         uint32_t i;
607
608         for (i = 0; i < eindex->count; ++i) {
609                 struct object *obj;
610
611                 if (!bitmap_get(objects, bitmap_git->pack->num_objects + i))
612                         continue;
613
614                 obj = eindex->objects[i];
615                 show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
616         }
617 }
618
619 static void show_objects_for_type(
620         struct bitmap_index *bitmap_git,
621         struct ewah_bitmap *type_filter,
622         enum object_type object_type,
623         show_reachable_fn show_reach)
624 {
625         size_t pos = 0, i = 0;
626         uint32_t offset;
627
628         struct ewah_iterator it;
629         eword_t filter;
630
631         struct bitmap *objects = bitmap_git->result;
632
633         if (bitmap_git->reuse_objects == bitmap_git->pack->num_objects)
634                 return;
635
636         ewah_iterator_init(&it, type_filter);
637
638         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
639                 eword_t word = objects->words[i] & filter;
640
641                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
642                         struct object_id oid;
643                         struct revindex_entry *entry;
644                         uint32_t hash = 0;
645
646                         if ((word >> offset) == 0)
647                                 break;
648
649                         offset += ewah_bit_ctz64(word >> offset);
650
651                         if (pos + offset < bitmap_git->reuse_objects)
652                                 continue;
653
654                         entry = &bitmap_git->pack->revindex[pos + offset];
655                         nth_packed_object_oid(&oid, bitmap_git->pack, entry->nr);
656
657                         if (bitmap_git->hashes)
658                                 hash = get_be32(bitmap_git->hashes + entry->nr);
659
660                         show_reach(&oid, object_type, 0, hash, bitmap_git->pack, entry->offset);
661                 }
662
663                 pos += BITS_IN_EWORD;
664                 i++;
665         }
666 }
667
668 static int in_bitmapped_pack(struct bitmap_index *bitmap_git,
669                              struct object_list *roots)
670 {
671         while (roots) {
672                 struct object *object = roots->item;
673                 roots = roots->next;
674
675                 if (find_pack_entry_one(object->oid.hash, bitmap_git->pack) > 0)
676                         return 1;
677         }
678
679         return 0;
680 }
681
682 struct bitmap_index *prepare_bitmap_walk(struct rev_info *revs)
683 {
684         unsigned int i;
685
686         struct object_list *wants = NULL;
687         struct object_list *haves = NULL;
688
689         struct bitmap *wants_bitmap = NULL;
690         struct bitmap *haves_bitmap = NULL;
691
692         struct bitmap_index *bitmap_git = xcalloc(1, sizeof(*bitmap_git));
693         /* try to open a bitmapped pack, but don't parse it yet
694          * because we may not need to use it */
695         if (open_pack_bitmap(revs->repo, bitmap_git) < 0)
696                 goto cleanup;
697
698         for (i = 0; i < revs->pending.nr; ++i) {
699                 struct object *object = revs->pending.objects[i].item;
700
701                 if (object->type == OBJ_NONE)
702                         parse_object_or_die(&object->oid, NULL);
703
704                 while (object->type == OBJ_TAG) {
705                         struct tag *tag = (struct tag *) object;
706
707                         if (object->flags & UNINTERESTING)
708                                 object_list_insert(object, &haves);
709                         else
710                                 object_list_insert(object, &wants);
711
712                         object = parse_object_or_die(get_tagged_oid(tag), NULL);
713                 }
714
715                 if (object->flags & UNINTERESTING)
716                         object_list_insert(object, &haves);
717                 else
718                         object_list_insert(object, &wants);
719         }
720
721         /*
722          * if we have a HAVES list, but none of those haves is contained
723          * in the packfile that has a bitmap, we don't have anything to
724          * optimize here
725          */
726         if (haves && !in_bitmapped_pack(bitmap_git, haves))
727                 goto cleanup;
728
729         /* if we don't want anything, we're done here */
730         if (!wants)
731                 goto cleanup;
732
733         /*
734          * now we're going to use bitmaps, so load the actual bitmap entries
735          * from disk. this is the point of no return; after this the rev_list
736          * becomes invalidated and we must perform the revwalk through bitmaps
737          */
738         if (load_pack_bitmap(bitmap_git) < 0)
739                 goto cleanup;
740
741         object_array_clear(&revs->pending);
742
743         if (haves) {
744                 revs->ignore_missing_links = 1;
745                 haves_bitmap = find_objects(bitmap_git, revs, haves, NULL);
746                 reset_revision_walk();
747                 revs->ignore_missing_links = 0;
748
749                 if (haves_bitmap == NULL)
750                         BUG("failed to perform bitmap walk");
751         }
752
753         wants_bitmap = find_objects(bitmap_git, revs, wants, haves_bitmap);
754
755         if (!wants_bitmap)
756                 BUG("failed to perform bitmap walk");
757
758         if (haves_bitmap)
759                 bitmap_and_not(wants_bitmap, haves_bitmap);
760
761         bitmap_git->result = wants_bitmap;
762         bitmap_git->haves = haves_bitmap;
763
764         return bitmap_git;
765
766 cleanup:
767         free_bitmap_index(bitmap_git);
768         return NULL;
769 }
770
771 int reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
772                                        struct packed_git **packfile,
773                                        uint32_t *entries,
774                                        off_t *up_to)
775 {
776         /*
777          * Reuse the packfile content if we need more than
778          * 90% of its objects
779          */
780         static const double REUSE_PERCENT = 0.9;
781
782         struct bitmap *result = bitmap_git->result;
783         uint32_t reuse_threshold;
784         uint32_t i, reuse_objects = 0;
785
786         assert(result);
787
788         for (i = 0; i < result->word_alloc; ++i) {
789                 if (result->words[i] != (eword_t)~0) {
790                         reuse_objects += ewah_bit_ctz64(~result->words[i]);
791                         break;
792                 }
793
794                 reuse_objects += BITS_IN_EWORD;
795         }
796
797 #ifdef GIT_BITMAP_DEBUG
798         {
799                 const unsigned char *sha1;
800                 struct revindex_entry *entry;
801
802                 entry = &bitmap_git->reverse_index->revindex[reuse_objects];
803                 sha1 = nth_packed_object_sha1(bitmap_git->pack, entry->nr);
804
805                 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
806                         reuse_objects, result->words[i]);
807                 fprintf(stderr, " %s\n", hash_to_hex(sha1));
808         }
809 #endif
810
811         if (!reuse_objects)
812                 return -1;
813
814         if (reuse_objects >= bitmap_git->pack->num_objects) {
815                 bitmap_git->reuse_objects = *entries = bitmap_git->pack->num_objects;
816                 *up_to = -1; /* reuse the full pack */
817                 *packfile = bitmap_git->pack;
818                 return 0;
819         }
820
821         reuse_threshold = bitmap_popcount(bitmap_git->result) * REUSE_PERCENT;
822
823         if (reuse_objects < reuse_threshold)
824                 return -1;
825
826         bitmap_git->reuse_objects = *entries = reuse_objects;
827         *up_to = bitmap_git->pack->revindex[reuse_objects].offset;
828         *packfile = bitmap_git->pack;
829
830         return 0;
831 }
832
833 void traverse_bitmap_commit_list(struct bitmap_index *bitmap_git,
834                                  show_reachable_fn show_reachable)
835 {
836         assert(bitmap_git->result);
837
838         show_objects_for_type(bitmap_git, bitmap_git->commits,
839                 OBJ_COMMIT, show_reachable);
840         show_objects_for_type(bitmap_git, bitmap_git->trees,
841                 OBJ_TREE, show_reachable);
842         show_objects_for_type(bitmap_git, bitmap_git->blobs,
843                 OBJ_BLOB, show_reachable);
844         show_objects_for_type(bitmap_git, bitmap_git->tags,
845                 OBJ_TAG, show_reachable);
846
847         show_extended_objects(bitmap_git, show_reachable);
848 }
849
850 static uint32_t count_object_type(struct bitmap_index *bitmap_git,
851                                   enum object_type type)
852 {
853         struct bitmap *objects = bitmap_git->result;
854         struct eindex *eindex = &bitmap_git->ext_index;
855
856         uint32_t i = 0, count = 0;
857         struct ewah_iterator it;
858         eword_t filter;
859
860         switch (type) {
861         case OBJ_COMMIT:
862                 ewah_iterator_init(&it, bitmap_git->commits);
863                 break;
864
865         case OBJ_TREE:
866                 ewah_iterator_init(&it, bitmap_git->trees);
867                 break;
868
869         case OBJ_BLOB:
870                 ewah_iterator_init(&it, bitmap_git->blobs);
871                 break;
872
873         case OBJ_TAG:
874                 ewah_iterator_init(&it, bitmap_git->tags);
875                 break;
876
877         default:
878                 return 0;
879         }
880
881         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
882                 eword_t word = objects->words[i++] & filter;
883                 count += ewah_bit_popcount64(word);
884         }
885
886         for (i = 0; i < eindex->count; ++i) {
887                 if (eindex->objects[i]->type == type &&
888                         bitmap_get(objects, bitmap_git->pack->num_objects + i))
889                         count++;
890         }
891
892         return count;
893 }
894
895 void count_bitmap_commit_list(struct bitmap_index *bitmap_git,
896                               uint32_t *commits, uint32_t *trees,
897                               uint32_t *blobs, uint32_t *tags)
898 {
899         assert(bitmap_git->result);
900
901         if (commits)
902                 *commits = count_object_type(bitmap_git, OBJ_COMMIT);
903
904         if (trees)
905                 *trees = count_object_type(bitmap_git, OBJ_TREE);
906
907         if (blobs)
908                 *blobs = count_object_type(bitmap_git, OBJ_BLOB);
909
910         if (tags)
911                 *tags = count_object_type(bitmap_git, OBJ_TAG);
912 }
913
914 struct bitmap_test_data {
915         struct bitmap_index *bitmap_git;
916         struct bitmap *base;
917         struct progress *prg;
918         size_t seen;
919 };
920
921 static void test_show_object(struct object *object, const char *name,
922                              void *data)
923 {
924         struct bitmap_test_data *tdata = data;
925         int bitmap_pos;
926
927         bitmap_pos = bitmap_position(tdata->bitmap_git, &object->oid);
928         if (bitmap_pos < 0)
929                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
930
931         bitmap_set(tdata->base, bitmap_pos);
932         display_progress(tdata->prg, ++tdata->seen);
933 }
934
935 static void test_show_commit(struct commit *commit, void *data)
936 {
937         struct bitmap_test_data *tdata = data;
938         int bitmap_pos;
939
940         bitmap_pos = bitmap_position(tdata->bitmap_git,
941                                      &commit->object.oid);
942         if (bitmap_pos < 0)
943                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
944
945         bitmap_set(tdata->base, bitmap_pos);
946         display_progress(tdata->prg, ++tdata->seen);
947 }
948
949 void test_bitmap_walk(struct rev_info *revs)
950 {
951         struct object *root;
952         struct bitmap *result = NULL;
953         khiter_t pos;
954         size_t result_popcnt;
955         struct bitmap_test_data tdata;
956         struct bitmap_index *bitmap_git;
957
958         if (!(bitmap_git = prepare_bitmap_git(revs->repo)))
959                 die("failed to load bitmap indexes");
960
961         if (revs->pending.nr != 1)
962                 die("you must specify exactly one commit to test");
963
964         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
965                 bitmap_git->version, bitmap_git->entry_count);
966
967         root = revs->pending.objects[0].item;
968         pos = kh_get_oid_map(bitmap_git->bitmaps, root->oid);
969
970         if (pos < kh_end(bitmap_git->bitmaps)) {
971                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
972                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
973
974                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
975                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
976
977                 result = ewah_to_bitmap(bm);
978         }
979
980         if (result == NULL)
981                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
982
983         revs->tag_objects = 1;
984         revs->tree_objects = 1;
985         revs->blob_objects = 1;
986
987         result_popcnt = bitmap_popcount(result);
988
989         if (prepare_revision_walk(revs))
990                 die("revision walk setup failed");
991
992         tdata.bitmap_git = bitmap_git;
993         tdata.base = bitmap_new();
994         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
995         tdata.seen = 0;
996
997         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
998
999         stop_progress(&tdata.prg);
1000
1001         if (bitmap_equals(result, tdata.base))
1002                 fprintf(stderr, "OK!\n");
1003         else
1004                 fprintf(stderr, "Mismatch!\n");
1005
1006         free_bitmap_index(bitmap_git);
1007 }
1008
1009 static int rebuild_bitmap(uint32_t *reposition,
1010                           struct ewah_bitmap *source,
1011                           struct bitmap *dest)
1012 {
1013         uint32_t pos = 0;
1014         struct ewah_iterator it;
1015         eword_t word;
1016
1017         ewah_iterator_init(&it, source);
1018
1019         while (ewah_iterator_next(&word, &it)) {
1020                 uint32_t offset, bit_pos;
1021
1022                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1023                         if ((word >> offset) == 0)
1024                                 break;
1025
1026                         offset += ewah_bit_ctz64(word >> offset);
1027
1028                         bit_pos = reposition[pos + offset];
1029                         if (bit_pos > 0)
1030                                 bitmap_set(dest, bit_pos - 1);
1031                         else /* can't reuse, we don't have the object */
1032                                 return -1;
1033                 }
1034
1035                 pos += BITS_IN_EWORD;
1036         }
1037         return 0;
1038 }
1039
1040 int rebuild_existing_bitmaps(struct bitmap_index *bitmap_git,
1041                              struct packing_data *mapping,
1042                              kh_oid_map_t *reused_bitmaps,
1043                              int show_progress)
1044 {
1045         uint32_t i, num_objects;
1046         uint32_t *reposition;
1047         struct bitmap *rebuild;
1048         struct stored_bitmap *stored;
1049         struct progress *progress = NULL;
1050
1051         khiter_t hash_pos;
1052         int hash_ret;
1053
1054         num_objects = bitmap_git->pack->num_objects;
1055         reposition = xcalloc(num_objects, sizeof(uint32_t));
1056
1057         for (i = 0; i < num_objects; ++i) {
1058                 struct object_id oid;
1059                 struct revindex_entry *entry;
1060                 struct object_entry *oe;
1061
1062                 entry = &bitmap_git->pack->revindex[i];
1063                 nth_packed_object_oid(&oid, bitmap_git->pack, entry->nr);
1064                 oe = packlist_find(mapping, &oid);
1065
1066                 if (oe)
1067                         reposition[i] = oe_in_pack_pos(mapping, oe) + 1;
1068         }
1069
1070         rebuild = bitmap_new();
1071         i = 0;
1072
1073         if (show_progress)
1074                 progress = start_progress("Reusing bitmaps", 0);
1075
1076         kh_foreach_value(bitmap_git->bitmaps, stored, {
1077                 if (stored->flags & BITMAP_FLAG_REUSE) {
1078                         if (!rebuild_bitmap(reposition,
1079                                             lookup_stored_bitmap(stored),
1080                                             rebuild)) {
1081                                 hash_pos = kh_put_oid_map(reused_bitmaps,
1082                                                           stored->oid,
1083                                                           &hash_ret);
1084                                 kh_value(reused_bitmaps, hash_pos) =
1085                                         bitmap_to_ewah(rebuild);
1086                         }
1087                         bitmap_reset(rebuild);
1088                         display_progress(progress, ++i);
1089                 }
1090         });
1091
1092         stop_progress(&progress);
1093
1094         free(reposition);
1095         bitmap_free(rebuild);
1096         return 0;
1097 }
1098
1099 void free_bitmap_index(struct bitmap_index *b)
1100 {
1101         if (!b)
1102                 return;
1103
1104         if (b->map)
1105                 munmap(b->map, b->map_size);
1106         ewah_pool_free(b->commits);
1107         ewah_pool_free(b->trees);
1108         ewah_pool_free(b->blobs);
1109         ewah_pool_free(b->tags);
1110         kh_destroy_oid_map(b->bitmaps);
1111         free(b->ext_index.objects);
1112         free(b->ext_index.hashes);
1113         bitmap_free(b->result);
1114         bitmap_free(b->haves);
1115         free(b);
1116 }
1117
1118 int bitmap_has_oid_in_uninteresting(struct bitmap_index *bitmap_git,
1119                                     const struct object_id *oid)
1120 {
1121         int pos;
1122
1123         if (!bitmap_git)
1124                 return 0; /* no bitmap loaded */
1125         if (!bitmap_git->haves)
1126                 return 0; /* walk had no "haves" */
1127
1128         pos = bitmap_position_packfile(bitmap_git, oid);
1129         if (pos < 0)
1130                 return 0;
1131
1132         return bitmap_get(bitmap_git->haves, pos);
1133 }