Merge branch 'jk/ewah-bounds-check'
[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         unsigned char sha1[20];
22         struct ewah_bitmap *root;
23         struct stored_bitmap *xor;
24         int flags;
25 };
26
27 /*
28  * The currently active bitmap index. 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 static 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 SHA1 -> `stored_bitmap` for all the bitmapped commits */
64         khash_sha1 *bitmaps;
65
66         /* Number of bitmapped commits */
67         uint32_t entry_count;
68
69         /* Name-hash cache (or NULL if not present). */
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                 khash_sha1_pos *positions;
84         } ext_index;
85
86         /* Bitmap result of the last performed walk */
87         struct bitmap *result;
88
89         /* Version of the bitmap index */
90         unsigned int version;
91
92         unsigned loaded : 1;
93
94 } bitmap_git;
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) + 20)
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 - 20;
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);
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 *sha1,
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         hashcpy(stored->sha1, sha1);
185
186         hash_pos = kh_put_sha1(index->bitmaps, stored->sha1, &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", sha1_to_hex(sha1));
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 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(void)
306 {
307         assert(bitmap_git.map && !bitmap_git.loaded);
308
309         bitmap_git.bitmaps = kh_init_sha1();
310         bitmap_git.ext_index.positions = kh_init_sha1_pos();
311         load_pack_revindex(bitmap_git.pack);
312
313         if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
314                 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
315                 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
316                 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
317                 goto failed;
318
319         if (load_bitmap_entries_v1(&bitmap_git) < 0)
320                 goto failed;
321
322         bitmap_git.loaded = 1;
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(void)
333 {
334         struct packed_git *p;
335         int ret = -1;
336
337         assert(!bitmap_git.map && !bitmap_git.loaded);
338
339         for (p = get_packed_git(the_repository); p; p = p->next) {
340                 if (open_pack_bitmap_1(p) == 0)
341                         ret = 0;
342         }
343
344         return ret;
345 }
346
347 int prepare_bitmap_git(void)
348 {
349         if (bitmap_git.loaded)
350                 return 0;
351
352         if (!open_pack_bitmap())
353                 return load_pack_bitmap();
354
355         return -1;
356 }
357
358 struct include_data {
359         struct bitmap *base;
360         struct bitmap *seen;
361 };
362
363 static inline int bitmap_position_extended(const unsigned char *sha1)
364 {
365         khash_sha1_pos *positions = bitmap_git.ext_index.positions;
366         khiter_t pos = kh_get_sha1_pos(positions, sha1);
367
368         if (pos < kh_end(positions)) {
369                 int bitmap_pos = kh_value(positions, pos);
370                 return bitmap_pos + bitmap_git.pack->num_objects;
371         }
372
373         return -1;
374 }
375
376 static inline int bitmap_position_packfile(const unsigned char *sha1)
377 {
378         off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
379         if (!offset)
380                 return -1;
381
382         return find_revindex_position(bitmap_git.pack, offset);
383 }
384
385 static int bitmap_position(const unsigned char *sha1)
386 {
387         int pos = bitmap_position_packfile(sha1);
388         return (pos >= 0) ? pos : bitmap_position_extended(sha1);
389 }
390
391 static int ext_index_add_object(struct object *object, const char *name)
392 {
393         struct eindex *eindex = &bitmap_git.ext_index;
394
395         khiter_t hash_pos;
396         int hash_ret;
397         int bitmap_pos;
398
399         hash_pos = kh_put_sha1_pos(eindex->positions, object->oid.hash, &hash_ret);
400         if (hash_ret > 0) {
401                 if (eindex->count >= eindex->alloc) {
402                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
403                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
404                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
405                 }
406
407                 bitmap_pos = eindex->count;
408                 eindex->objects[eindex->count] = object;
409                 eindex->hashes[eindex->count] = pack_name_hash(name);
410                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
411                 eindex->count++;
412         } else {
413                 bitmap_pos = kh_value(eindex->positions, hash_pos);
414         }
415
416         return bitmap_pos + bitmap_git.pack->num_objects;
417 }
418
419 static void show_object(struct object *object, const char *name, void *data)
420 {
421         struct bitmap *base = data;
422         int bitmap_pos;
423
424         bitmap_pos = bitmap_position(object->oid.hash);
425
426         if (bitmap_pos < 0)
427                 bitmap_pos = ext_index_add_object(object, name);
428
429         bitmap_set(base, bitmap_pos);
430 }
431
432 static void show_commit(struct commit *commit, void *data)
433 {
434 }
435
436 static int add_to_include_set(struct include_data *data,
437                               const unsigned char *sha1,
438                               int bitmap_pos)
439 {
440         khiter_t hash_pos;
441
442         if (data->seen && bitmap_get(data->seen, bitmap_pos))
443                 return 0;
444
445         if (bitmap_get(data->base, bitmap_pos))
446                 return 0;
447
448         hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
449         if (hash_pos < kh_end(bitmap_git.bitmaps)) {
450                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
451                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
452                 return 0;
453         }
454
455         bitmap_set(data->base, bitmap_pos);
456         return 1;
457 }
458
459 static int should_include(struct commit *commit, void *_data)
460 {
461         struct include_data *data = _data;
462         int bitmap_pos;
463
464         bitmap_pos = bitmap_position(commit->object.oid.hash);
465         if (bitmap_pos < 0)
466                 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
467
468         if (!add_to_include_set(data, commit->object.oid.hash, bitmap_pos)) {
469                 struct commit_list *parent = commit->parents;
470
471                 while (parent) {
472                         parent->item->object.flags |= SEEN;
473                         parent = parent->next;
474                 }
475
476                 return 0;
477         }
478
479         return 1;
480 }
481
482 static struct bitmap *find_objects(struct rev_info *revs,
483                                    struct object_list *roots,
484                                    struct bitmap *seen)
485 {
486         struct bitmap *base = NULL;
487         int needs_walk = 0;
488
489         struct object_list *not_mapped = NULL;
490
491         /*
492          * Go through all the roots for the walk. The ones that have bitmaps
493          * on the bitmap index will be `or`ed together to form an initial
494          * global reachability analysis.
495          *
496          * The ones without bitmaps in the index will be stored in the
497          * `not_mapped_list` for further processing.
498          */
499         while (roots) {
500                 struct object *object = roots->item;
501                 roots = roots->next;
502
503                 if (object->type == OBJ_COMMIT) {
504                         khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->oid.hash);
505
506                         if (pos < kh_end(bitmap_git.bitmaps)) {
507                                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
508                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
509
510                                 if (base == NULL)
511                                         base = ewah_to_bitmap(or_with);
512                                 else
513                                         bitmap_or_ewah(base, or_with);
514
515                                 object->flags |= SEEN;
516                                 continue;
517                         }
518                 }
519
520                 object_list_insert(object, &not_mapped);
521         }
522
523         /*
524          * Best case scenario: We found bitmaps for all the roots,
525          * so the resulting `or` bitmap has the full reachability analysis
526          */
527         if (not_mapped == NULL)
528                 return base;
529
530         roots = not_mapped;
531
532         /*
533          * Let's iterate through all the roots that don't have bitmaps to
534          * check if we can determine them to be reachable from the existing
535          * global bitmap.
536          *
537          * If we cannot find them in the existing global bitmap, we'll need
538          * to push them to an actual walk and run it until we can confirm
539          * they are reachable
540          */
541         while (roots) {
542                 struct object *object = roots->item;
543                 int pos;
544
545                 roots = roots->next;
546                 pos = bitmap_position(object->oid.hash);
547
548                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
549                         object->flags &= ~UNINTERESTING;
550                         add_pending_object(revs, object, "");
551                         needs_walk = 1;
552                 } else {
553                         object->flags |= SEEN;
554                 }
555         }
556
557         if (needs_walk) {
558                 struct include_data incdata;
559
560                 if (base == NULL)
561                         base = bitmap_new();
562
563                 incdata.base = base;
564                 incdata.seen = seen;
565
566                 revs->include_check = should_include;
567                 revs->include_check_data = &incdata;
568
569                 if (prepare_revision_walk(revs))
570                         die("revision walk setup failed");
571
572                 traverse_commit_list(revs, show_commit, show_object, base);
573         }
574
575         return base;
576 }
577
578 static void show_extended_objects(struct bitmap *objects,
579                                   show_reachable_fn show_reach)
580 {
581         struct eindex *eindex = &bitmap_git.ext_index;
582         uint32_t i;
583
584         for (i = 0; i < eindex->count; ++i) {
585                 struct object *obj;
586
587                 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
588                         continue;
589
590                 obj = eindex->objects[i];
591                 show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
592         }
593 }
594
595 static void show_objects_for_type(
596         struct bitmap *objects,
597         struct ewah_bitmap *type_filter,
598         enum object_type object_type,
599         show_reachable_fn show_reach)
600 {
601         size_t pos = 0, i = 0;
602         uint32_t offset;
603
604         struct ewah_iterator it;
605         eword_t filter;
606
607         if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
608                 return;
609
610         ewah_iterator_init(&it, type_filter);
611
612         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
613                 eword_t word = objects->words[i] & filter;
614
615                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
616                         struct object_id oid;
617                         struct revindex_entry *entry;
618                         uint32_t hash = 0;
619
620                         if ((word >> offset) == 0)
621                                 break;
622
623                         offset += ewah_bit_ctz64(word >> offset);
624
625                         if (pos + offset < bitmap_git.reuse_objects)
626                                 continue;
627
628                         entry = &bitmap_git.pack->revindex[pos + offset];
629                         nth_packed_object_oid(&oid, bitmap_git.pack, entry->nr);
630
631                         if (bitmap_git.hashes)
632                                 hash = get_be32(bitmap_git.hashes + entry->nr);
633
634                         show_reach(&oid, object_type, 0, hash, bitmap_git.pack, entry->offset);
635                 }
636
637                 pos += BITS_IN_EWORD;
638                 i++;
639         }
640 }
641
642 static int in_bitmapped_pack(struct object_list *roots)
643 {
644         while (roots) {
645                 struct object *object = roots->item;
646                 roots = roots->next;
647
648                 if (find_pack_entry_one(object->oid.hash, bitmap_git.pack) > 0)
649                         return 1;
650         }
651
652         return 0;
653 }
654
655 int prepare_bitmap_walk(struct rev_info *revs)
656 {
657         unsigned int i;
658
659         struct object_list *wants = NULL;
660         struct object_list *haves = NULL;
661
662         struct bitmap *wants_bitmap = NULL;
663         struct bitmap *haves_bitmap = NULL;
664
665         if (!bitmap_git.loaded) {
666                 /* try to open a bitmapped pack, but don't parse it yet
667                  * because we may not need to use it */
668                 if (open_pack_bitmap() < 0)
669                         return -1;
670         }
671
672         for (i = 0; i < revs->pending.nr; ++i) {
673                 struct object *object = revs->pending.objects[i].item;
674
675                 if (object->type == OBJ_NONE)
676                         parse_object_or_die(&object->oid, NULL);
677
678                 while (object->type == OBJ_TAG) {
679                         struct tag *tag = (struct tag *) object;
680
681                         if (object->flags & UNINTERESTING)
682                                 object_list_insert(object, &haves);
683                         else
684                                 object_list_insert(object, &wants);
685
686                         if (!tag->tagged)
687                                 die("bad tag");
688                         object = parse_object_or_die(&tag->tagged->oid, NULL);
689                 }
690
691                 if (object->flags & UNINTERESTING)
692                         object_list_insert(object, &haves);
693                 else
694                         object_list_insert(object, &wants);
695         }
696
697         /*
698          * if we have a HAVES list, but none of those haves is contained
699          * in the packfile that has a bitmap, we don't have anything to
700          * optimize here
701          */
702         if (haves && !in_bitmapped_pack(haves))
703                 return -1;
704
705         /* if we don't want anything, we're done here */
706         if (!wants)
707                 return -1;
708
709         /*
710          * now we're going to use bitmaps, so load the actual bitmap entries
711          * from disk. this is the point of no return; after this the rev_list
712          * becomes invalidated and we must perform the revwalk through bitmaps
713          */
714         if (!bitmap_git.loaded && load_pack_bitmap() < 0)
715                 return -1;
716
717         object_array_clear(&revs->pending);
718
719         if (haves) {
720                 revs->ignore_missing_links = 1;
721                 haves_bitmap = find_objects(revs, haves, NULL);
722                 reset_revision_walk();
723                 revs->ignore_missing_links = 0;
724
725                 if (haves_bitmap == NULL)
726                         BUG("failed to perform bitmap walk");
727         }
728
729         wants_bitmap = find_objects(revs, wants, haves_bitmap);
730
731         if (!wants_bitmap)
732                 BUG("failed to perform bitmap walk");
733
734         if (haves_bitmap)
735                 bitmap_and_not(wants_bitmap, haves_bitmap);
736
737         bitmap_git.result = wants_bitmap;
738
739         bitmap_free(haves_bitmap);
740         return 0;
741 }
742
743 int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
744                                        uint32_t *entries,
745                                        off_t *up_to)
746 {
747         /*
748          * Reuse the packfile content if we need more than
749          * 90% of its objects
750          */
751         static const double REUSE_PERCENT = 0.9;
752
753         struct bitmap *result = bitmap_git.result;
754         uint32_t reuse_threshold;
755         uint32_t i, reuse_objects = 0;
756
757         assert(result);
758
759         for (i = 0; i < result->word_alloc; ++i) {
760                 if (result->words[i] != (eword_t)~0) {
761                         reuse_objects += ewah_bit_ctz64(~result->words[i]);
762                         break;
763                 }
764
765                 reuse_objects += BITS_IN_EWORD;
766         }
767
768 #ifdef GIT_BITMAP_DEBUG
769         {
770                 const unsigned char *sha1;
771                 struct revindex_entry *entry;
772
773                 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
774                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
775
776                 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
777                         reuse_objects, result->words[i]);
778                 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
779         }
780 #endif
781
782         if (!reuse_objects)
783                 return -1;
784
785         if (reuse_objects >= bitmap_git.pack->num_objects) {
786                 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
787                 *up_to = -1; /* reuse the full pack */
788                 *packfile = bitmap_git.pack;
789                 return 0;
790         }
791
792         reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
793
794         if (reuse_objects < reuse_threshold)
795                 return -1;
796
797         bitmap_git.reuse_objects = *entries = reuse_objects;
798         *up_to = bitmap_git.pack->revindex[reuse_objects].offset;
799         *packfile = bitmap_git.pack;
800
801         return 0;
802 }
803
804 void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
805 {
806         assert(bitmap_git.result);
807
808         show_objects_for_type(bitmap_git.result, bitmap_git.commits,
809                 OBJ_COMMIT, show_reachable);
810         show_objects_for_type(bitmap_git.result, bitmap_git.trees,
811                 OBJ_TREE, show_reachable);
812         show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
813                 OBJ_BLOB, show_reachable);
814         show_objects_for_type(bitmap_git.result, bitmap_git.tags,
815                 OBJ_TAG, show_reachable);
816
817         show_extended_objects(bitmap_git.result, show_reachable);
818
819         bitmap_free(bitmap_git.result);
820         bitmap_git.result = NULL;
821 }
822
823 static uint32_t count_object_type(struct bitmap *objects,
824                                   enum object_type type)
825 {
826         struct eindex *eindex = &bitmap_git.ext_index;
827
828         uint32_t i = 0, count = 0;
829         struct ewah_iterator it;
830         eword_t filter;
831
832         switch (type) {
833         case OBJ_COMMIT:
834                 ewah_iterator_init(&it, bitmap_git.commits);
835                 break;
836
837         case OBJ_TREE:
838                 ewah_iterator_init(&it, bitmap_git.trees);
839                 break;
840
841         case OBJ_BLOB:
842                 ewah_iterator_init(&it, bitmap_git.blobs);
843                 break;
844
845         case OBJ_TAG:
846                 ewah_iterator_init(&it, bitmap_git.tags);
847                 break;
848
849         default:
850                 return 0;
851         }
852
853         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
854                 eword_t word = objects->words[i++] & filter;
855                 count += ewah_bit_popcount64(word);
856         }
857
858         for (i = 0; i < eindex->count; ++i) {
859                 if (eindex->objects[i]->type == type &&
860                         bitmap_get(objects, bitmap_git.pack->num_objects + i))
861                         count++;
862         }
863
864         return count;
865 }
866
867 void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
868                               uint32_t *blobs, uint32_t *tags)
869 {
870         assert(bitmap_git.result);
871
872         if (commits)
873                 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
874
875         if (trees)
876                 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
877
878         if (blobs)
879                 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
880
881         if (tags)
882                 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
883 }
884
885 struct bitmap_test_data {
886         struct bitmap *base;
887         struct progress *prg;
888         size_t seen;
889 };
890
891 static void test_show_object(struct object *object, const char *name,
892                              void *data)
893 {
894         struct bitmap_test_data *tdata = data;
895         int bitmap_pos;
896
897         bitmap_pos = bitmap_position(object->oid.hash);
898         if (bitmap_pos < 0)
899                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
900
901         bitmap_set(tdata->base, bitmap_pos);
902         display_progress(tdata->prg, ++tdata->seen);
903 }
904
905 static void test_show_commit(struct commit *commit, void *data)
906 {
907         struct bitmap_test_data *tdata = data;
908         int bitmap_pos;
909
910         bitmap_pos = bitmap_position(commit->object.oid.hash);
911         if (bitmap_pos < 0)
912                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
913
914         bitmap_set(tdata->base, bitmap_pos);
915         display_progress(tdata->prg, ++tdata->seen);
916 }
917
918 void test_bitmap_walk(struct rev_info *revs)
919 {
920         struct object *root;
921         struct bitmap *result = NULL;
922         khiter_t pos;
923         size_t result_popcnt;
924         struct bitmap_test_data tdata;
925
926         if (prepare_bitmap_git())
927                 die("failed to load bitmap indexes");
928
929         if (revs->pending.nr != 1)
930                 die("you must specify exactly one commit to test");
931
932         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
933                 bitmap_git.version, bitmap_git.entry_count);
934
935         root = revs->pending.objects[0].item;
936         pos = kh_get_sha1(bitmap_git.bitmaps, root->oid.hash);
937
938         if (pos < kh_end(bitmap_git.bitmaps)) {
939                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
940                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
941
942                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
943                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
944
945                 result = ewah_to_bitmap(bm);
946         }
947
948         if (result == NULL)
949                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
950
951         revs->tag_objects = 1;
952         revs->tree_objects = 1;
953         revs->blob_objects = 1;
954
955         result_popcnt = bitmap_popcount(result);
956
957         if (prepare_revision_walk(revs))
958                 die("revision walk setup failed");
959
960         tdata.base = bitmap_new();
961         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
962         tdata.seen = 0;
963
964         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
965
966         stop_progress(&tdata.prg);
967
968         if (bitmap_equals(result, tdata.base))
969                 fprintf(stderr, "OK!\n");
970         else
971                 fprintf(stderr, "Mismatch!\n");
972
973         bitmap_free(result);
974 }
975
976 static int rebuild_bitmap(uint32_t *reposition,
977                           struct ewah_bitmap *source,
978                           struct bitmap *dest)
979 {
980         uint32_t pos = 0;
981         struct ewah_iterator it;
982         eword_t word;
983
984         ewah_iterator_init(&it, source);
985
986         while (ewah_iterator_next(&word, &it)) {
987                 uint32_t offset, bit_pos;
988
989                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
990                         if ((word >> offset) == 0)
991                                 break;
992
993                         offset += ewah_bit_ctz64(word >> offset);
994
995                         bit_pos = reposition[pos + offset];
996                         if (bit_pos > 0)
997                                 bitmap_set(dest, bit_pos - 1);
998                         else /* can't reuse, we don't have the object */
999                                 return -1;
1000                 }
1001
1002                 pos += BITS_IN_EWORD;
1003         }
1004         return 0;
1005 }
1006
1007 int rebuild_existing_bitmaps(struct packing_data *mapping,
1008                              khash_sha1 *reused_bitmaps,
1009                              int show_progress)
1010 {
1011         uint32_t i, num_objects;
1012         uint32_t *reposition;
1013         struct bitmap *rebuild;
1014         struct stored_bitmap *stored;
1015         struct progress *progress = NULL;
1016
1017         khiter_t hash_pos;
1018         int hash_ret;
1019
1020         if (prepare_bitmap_git() < 0)
1021                 return -1;
1022
1023         num_objects = bitmap_git.pack->num_objects;
1024         reposition = xcalloc(num_objects, sizeof(uint32_t));
1025
1026         for (i = 0; i < num_objects; ++i) {
1027                 const unsigned char *sha1;
1028                 struct revindex_entry *entry;
1029                 struct object_entry *oe;
1030
1031                 entry = &bitmap_git.pack->revindex[i];
1032                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1033                 oe = packlist_find(mapping, sha1, NULL);
1034
1035                 if (oe)
1036                         reposition[i] = oe_in_pack_pos(mapping, oe) + 1;
1037         }
1038
1039         rebuild = bitmap_new();
1040         i = 0;
1041
1042         if (show_progress)
1043                 progress = start_progress("Reusing bitmaps", 0);
1044
1045         kh_foreach_value(bitmap_git.bitmaps, stored, {
1046                 if (stored->flags & BITMAP_FLAG_REUSE) {
1047                         if (!rebuild_bitmap(reposition,
1048                                             lookup_stored_bitmap(stored),
1049                                             rebuild)) {
1050                                 hash_pos = kh_put_sha1(reused_bitmaps,
1051                                                        stored->sha1,
1052                                                        &hash_ret);
1053                                 kh_value(reused_bitmaps, hash_pos) =
1054                                         bitmap_to_ewah(rebuild);
1055                         }
1056                         bitmap_reset(rebuild);
1057                         display_progress(progress, ++i);
1058                 }
1059         });
1060
1061         stop_progress(&progress);
1062
1063         free(reposition);
1064         bitmap_free(rebuild);
1065         return 0;
1066 }