Merge branch 'jk/prune-with-corrupt-refs'
[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
13 /*
14  * An entry on the bitmap index, representing the bitmap for a given
15  * commit.
16  */
17 struct stored_bitmap {
18         unsigned char sha1[20];
19         struct ewah_bitmap *root;
20         struct stored_bitmap *xor;
21         int flags;
22 };
23
24 /*
25  * The currently active bitmap index. By design, repositories only have
26  * a single bitmap index available (the index for the biggest packfile in
27  * the repository), since bitmap indexes need full closure.
28  *
29  * If there is more than one bitmap index available (e.g. because of alternates),
30  * the active bitmap index is the largest one.
31  */
32 static struct bitmap_index {
33         /* Packfile to which this bitmap index belongs to */
34         struct packed_git *pack;
35
36         /* reverse index for the packfile */
37         struct pack_revindex *reverse_index;
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         int 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 static int load_bitmap_entries_v1(struct bitmap_index *index)
213 {
214         static const size_t MAX_XOR_OFFSET = 160;
215
216         uint32_t i;
217         struct stored_bitmap **recent_bitmaps;
218
219         recent_bitmaps = xcalloc(MAX_XOR_OFFSET, sizeof(struct stored_bitmap));
220
221         for (i = 0; i < index->entry_count; ++i) {
222                 int xor_offset, flags;
223                 struct ewah_bitmap *bitmap = NULL;
224                 struct stored_bitmap *xor_bitmap = NULL;
225                 uint32_t commit_idx_pos;
226                 const unsigned char *sha1;
227
228                 commit_idx_pos = read_be32(index->map, &index->map_pos);
229                 xor_offset = read_u8(index->map, &index->map_pos);
230                 flags = read_u8(index->map, &index->map_pos);
231
232                 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
233
234                 bitmap = read_bitmap_1(index);
235                 if (!bitmap)
236                         return -1;
237
238                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
239                         return error("Corrupted bitmap pack index");
240
241                 if (xor_offset > 0) {
242                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
243
244                         if (xor_bitmap == NULL)
245                                 return error("Invalid XOR offset in bitmap pack index");
246                 }
247
248                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
249                         index, bitmap, sha1, xor_bitmap, flags);
250         }
251
252         return 0;
253 }
254
255 static char *pack_bitmap_filename(struct packed_git *p)
256 {
257         char *idx_name;
258         int len;
259
260         len = strlen(p->pack_name) - strlen(".pack");
261         idx_name = xmalloc(len + strlen(".bitmap") + 1);
262
263         memcpy(idx_name, p->pack_name, len);
264         memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
265
266         return idx_name;
267 }
268
269 static int open_pack_bitmap_1(struct packed_git *packfile)
270 {
271         int fd;
272         struct stat st;
273         char *idx_name;
274
275         if (open_pack_index(packfile))
276                 return -1;
277
278         idx_name = pack_bitmap_filename(packfile);
279         fd = git_open_noatime(idx_name);
280         free(idx_name);
281
282         if (fd < 0)
283                 return -1;
284
285         if (fstat(fd, &st)) {
286                 close(fd);
287                 return -1;
288         }
289
290         if (bitmap_git.pack) {
291                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
292                 close(fd);
293                 return -1;
294         }
295
296         bitmap_git.pack = packfile;
297         bitmap_git.map_size = xsize_t(st.st_size);
298         bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
299         bitmap_git.map_pos = 0;
300         close(fd);
301
302         if (load_bitmap_header(&bitmap_git) < 0) {
303                 munmap(bitmap_git.map, bitmap_git.map_size);
304                 bitmap_git.map = NULL;
305                 bitmap_git.map_size = 0;
306                 return -1;
307         }
308
309         return 0;
310 }
311
312 static int load_pack_bitmap(void)
313 {
314         assert(bitmap_git.map && !bitmap_git.loaded);
315
316         bitmap_git.bitmaps = kh_init_sha1();
317         bitmap_git.ext_index.positions = kh_init_sha1_pos();
318         bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
319
320         if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
321                 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
322                 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
323                 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
324                 goto failed;
325
326         if (load_bitmap_entries_v1(&bitmap_git) < 0)
327                 goto failed;
328
329         bitmap_git.loaded = 1;
330         return 0;
331
332 failed:
333         munmap(bitmap_git.map, bitmap_git.map_size);
334         bitmap_git.map = NULL;
335         bitmap_git.map_size = 0;
336         return -1;
337 }
338
339 static int open_pack_bitmap(void)
340 {
341         struct packed_git *p;
342         int ret = -1;
343
344         assert(!bitmap_git.map && !bitmap_git.loaded);
345
346         prepare_packed_git();
347         for (p = packed_git; p; p = p->next) {
348                 if (open_pack_bitmap_1(p) == 0)
349                         ret = 0;
350         }
351
352         return ret;
353 }
354
355 int prepare_bitmap_git(void)
356 {
357         if (bitmap_git.loaded)
358                 return 0;
359
360         if (!open_pack_bitmap())
361                 return load_pack_bitmap();
362
363         return -1;
364 }
365
366 struct include_data {
367         struct bitmap *base;
368         struct bitmap *seen;
369 };
370
371 static inline int bitmap_position_extended(const unsigned char *sha1)
372 {
373         khash_sha1_pos *positions = bitmap_git.ext_index.positions;
374         khiter_t pos = kh_get_sha1_pos(positions, sha1);
375
376         if (pos < kh_end(positions)) {
377                 int bitmap_pos = kh_value(positions, pos);
378                 return bitmap_pos + bitmap_git.pack->num_objects;
379         }
380
381         return -1;
382 }
383
384 static inline int bitmap_position_packfile(const unsigned char *sha1)
385 {
386         off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
387         if (!offset)
388                 return -1;
389
390         return find_revindex_position(bitmap_git.reverse_index, offset);
391 }
392
393 static int bitmap_position(const unsigned char *sha1)
394 {
395         int pos = bitmap_position_packfile(sha1);
396         return (pos >= 0) ? pos : bitmap_position_extended(sha1);
397 }
398
399 static int ext_index_add_object(struct object *object, const char *name)
400 {
401         struct eindex *eindex = &bitmap_git.ext_index;
402
403         khiter_t hash_pos;
404         int hash_ret;
405         int bitmap_pos;
406
407         hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret);
408         if (hash_ret > 0) {
409                 if (eindex->count >= eindex->alloc) {
410                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
411                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
412                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
413                 }
414
415                 bitmap_pos = eindex->count;
416                 eindex->objects[eindex->count] = object;
417                 eindex->hashes[eindex->count] = pack_name_hash(name);
418                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
419                 eindex->count++;
420         } else {
421                 bitmap_pos = kh_value(eindex->positions, hash_pos);
422         }
423
424         return bitmap_pos + bitmap_git.pack->num_objects;
425 }
426
427 static void show_object(struct object *object, const struct name_path *path,
428                         const char *last, void *data)
429 {
430         struct bitmap *base = data;
431         int bitmap_pos;
432
433         bitmap_pos = bitmap_position(object->sha1);
434
435         if (bitmap_pos < 0) {
436                 char *name = path_name(path, last);
437                 bitmap_pos = ext_index_add_object(object, name);
438                 free(name);
439         }
440
441         bitmap_set(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 include_data *data,
449                               const unsigned char *sha1,
450                               int bitmap_pos)
451 {
452         khiter_t hash_pos;
453
454         if (data->seen && bitmap_get(data->seen, bitmap_pos))
455                 return 0;
456
457         if (bitmap_get(data->base, bitmap_pos))
458                 return 0;
459
460         hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
461         if (hash_pos < kh_end(bitmap_git.bitmaps)) {
462                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
463                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
464                 return 0;
465         }
466
467         bitmap_set(data->base, bitmap_pos);
468         return 1;
469 }
470
471 static int should_include(struct commit *commit, void *_data)
472 {
473         struct include_data *data = _data;
474         int bitmap_pos;
475
476         bitmap_pos = bitmap_position(commit->object.sha1);
477         if (bitmap_pos < 0)
478                 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
479
480         if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
481                 struct commit_list *parent = commit->parents;
482
483                 while (parent) {
484                         parent->item->object.flags |= SEEN;
485                         parent = parent->next;
486                 }
487
488                 return 0;
489         }
490
491         return 1;
492 }
493
494 static struct bitmap *find_objects(struct rev_info *revs,
495                                    struct object_list *roots,
496                                    struct bitmap *seen)
497 {
498         struct bitmap *base = NULL;
499         int needs_walk = 0;
500
501         struct object_list *not_mapped = NULL;
502
503         /*
504          * Go through all the roots for the walk. The ones that have bitmaps
505          * on the bitmap index will be `or`ed together to form an initial
506          * global reachability analysis.
507          *
508          * The ones without bitmaps in the index will be stored in the
509          * `not_mapped_list` for further processing.
510          */
511         while (roots) {
512                 struct object *object = roots->item;
513                 roots = roots->next;
514
515                 if (object->type == OBJ_COMMIT) {
516                         khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
517
518                         if (pos < kh_end(bitmap_git.bitmaps)) {
519                                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
520                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
521
522                                 if (base == NULL)
523                                         base = ewah_to_bitmap(or_with);
524                                 else
525                                         bitmap_or_ewah(base, or_with);
526
527                                 object->flags |= SEEN;
528                                 continue;
529                         }
530                 }
531
532                 object_list_insert(object, &not_mapped);
533         }
534
535         /*
536          * Best case scenario: We found bitmaps for all the roots,
537          * so the resulting `or` bitmap has the full reachability analysis
538          */
539         if (not_mapped == NULL)
540                 return base;
541
542         roots = not_mapped;
543
544         /*
545          * Let's iterate through all the roots that don't have bitmaps to
546          * check if we can determine them to be reachable from the existing
547          * global bitmap.
548          *
549          * If we cannot find them in the existing global bitmap, we'll need
550          * to push them to an actual walk and run it until we can confirm
551          * they are reachable
552          */
553         while (roots) {
554                 struct object *object = roots->item;
555                 int pos;
556
557                 roots = roots->next;
558                 pos = bitmap_position(object->sha1);
559
560                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
561                         object->flags &= ~UNINTERESTING;
562                         add_pending_object(revs, object, "");
563                         needs_walk = 1;
564                 } else {
565                         object->flags |= SEEN;
566                 }
567         }
568
569         if (needs_walk) {
570                 struct include_data incdata;
571
572                 if (base == NULL)
573                         base = bitmap_new();
574
575                 incdata.base = base;
576                 incdata.seen = seen;
577
578                 revs->include_check = should_include;
579                 revs->include_check_data = &incdata;
580
581                 if (prepare_revision_walk(revs))
582                         die("revision walk setup failed");
583
584                 traverse_commit_list(revs, show_commit, show_object, base);
585         }
586
587         return base;
588 }
589
590 static void show_extended_objects(struct bitmap *objects,
591                                   show_reachable_fn show_reach)
592 {
593         struct eindex *eindex = &bitmap_git.ext_index;
594         uint32_t i;
595
596         for (i = 0; i < eindex->count; ++i) {
597                 struct object *obj;
598
599                 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
600                         continue;
601
602                 obj = eindex->objects[i];
603                 show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
604         }
605 }
606
607 static void show_objects_for_type(
608         struct bitmap *objects,
609         struct ewah_bitmap *type_filter,
610         enum object_type object_type,
611         show_reachable_fn show_reach)
612 {
613         size_t pos = 0, i = 0;
614         uint32_t offset;
615
616         struct ewah_iterator it;
617         eword_t filter;
618
619         if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
620                 return;
621
622         ewah_iterator_init(&it, type_filter);
623
624         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
625                 eword_t word = objects->words[i] & filter;
626
627                 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
628                         const unsigned char *sha1;
629                         struct revindex_entry *entry;
630                         uint32_t hash = 0;
631
632                         if ((word >> offset) == 0)
633                                 break;
634
635                         offset += ewah_bit_ctz64(word >> offset);
636
637                         if (pos + offset < bitmap_git.reuse_objects)
638                                 continue;
639
640                         entry = &bitmap_git.reverse_index->revindex[pos + offset];
641                         sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
642
643                         if (bitmap_git.hashes)
644                                 hash = ntohl(bitmap_git.hashes[entry->nr]);
645
646                         show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
647                 }
648
649                 pos += BITS_IN_WORD;
650                 i++;
651         }
652 }
653
654 static int in_bitmapped_pack(struct object_list *roots)
655 {
656         while (roots) {
657                 struct object *object = roots->item;
658                 roots = roots->next;
659
660                 if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
661                         return 1;
662         }
663
664         return 0;
665 }
666
667 int prepare_bitmap_walk(struct rev_info *revs)
668 {
669         unsigned int i;
670         unsigned int pending_nr = revs->pending.nr;
671         struct object_array_entry *pending_e = revs->pending.objects;
672
673         struct object_list *wants = NULL;
674         struct object_list *haves = NULL;
675
676         struct bitmap *wants_bitmap = NULL;
677         struct bitmap *haves_bitmap = NULL;
678
679         if (!bitmap_git.loaded) {
680                 /* try to open a bitmapped pack, but don't parse it yet
681                  * because we may not need to use it */
682                 if (open_pack_bitmap() < 0)
683                         return -1;
684         }
685
686         for (i = 0; i < pending_nr; ++i) {
687                 struct object *object = pending_e[i].item;
688
689                 if (object->type == OBJ_NONE)
690                         parse_object_or_die(object->sha1, NULL);
691
692                 while (object->type == OBJ_TAG) {
693                         struct tag *tag = (struct tag *) object;
694
695                         if (object->flags & UNINTERESTING)
696                                 object_list_insert(object, &haves);
697                         else
698                                 object_list_insert(object, &wants);
699
700                         if (!tag->tagged)
701                                 die("bad tag");
702                         object = parse_object_or_die(tag->tagged->sha1, NULL);
703                 }
704
705                 if (object->flags & UNINTERESTING)
706                         object_list_insert(object, &haves);
707                 else
708                         object_list_insert(object, &wants);
709         }
710
711         /*
712          * if we have a HAVES list, but none of those haves is contained
713          * in the packfile that has a bitmap, we don't have anything to
714          * optimize here
715          */
716         if (haves && !in_bitmapped_pack(haves))
717                 return -1;
718
719         /* if we don't want anything, we're done here */
720         if (!wants)
721                 return -1;
722
723         /*
724          * now we're going to use bitmaps, so load the actual bitmap entries
725          * from disk. this is the point of no return; after this the rev_list
726          * becomes invalidated and we must perform the revwalk through bitmaps
727          */
728         if (!bitmap_git.loaded && load_pack_bitmap() < 0)
729                 return -1;
730
731         revs->pending.nr = 0;
732         revs->pending.alloc = 0;
733         revs->pending.objects = NULL;
734
735         if (haves) {
736                 revs->ignore_missing_links = 1;
737                 haves_bitmap = find_objects(revs, haves, NULL);
738                 reset_revision_walk();
739                 revs->ignore_missing_links = 0;
740
741                 if (haves_bitmap == NULL)
742                         die("BUG: failed to perform bitmap walk");
743         }
744
745         wants_bitmap = find_objects(revs, wants, haves_bitmap);
746
747         if (!wants_bitmap)
748                 die("BUG: failed to perform bitmap walk");
749
750         if (haves_bitmap)
751                 bitmap_and_not(wants_bitmap, haves_bitmap);
752
753         bitmap_git.result = wants_bitmap;
754
755         bitmap_free(haves_bitmap);
756         return 0;
757 }
758
759 int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
760                                        uint32_t *entries,
761                                        off_t *up_to)
762 {
763         /*
764          * Reuse the packfile content if we need more than
765          * 90% of its objects
766          */
767         static const double REUSE_PERCENT = 0.9;
768
769         struct bitmap *result = bitmap_git.result;
770         uint32_t reuse_threshold;
771         uint32_t i, reuse_objects = 0;
772
773         assert(result);
774
775         for (i = 0; i < result->word_alloc; ++i) {
776                 if (result->words[i] != (eword_t)~0) {
777                         reuse_objects += ewah_bit_ctz64(~result->words[i]);
778                         break;
779                 }
780
781                 reuse_objects += BITS_IN_WORD;
782         }
783
784 #ifdef GIT_BITMAP_DEBUG
785         {
786                 const unsigned char *sha1;
787                 struct revindex_entry *entry;
788
789                 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
790                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
791
792                 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
793                         reuse_objects, result->words[i]);
794                 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
795         }
796 #endif
797
798         if (!reuse_objects)
799                 return -1;
800
801         if (reuse_objects >= bitmap_git.pack->num_objects) {
802                 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
803                 *up_to = -1; /* reuse the full pack */
804                 *packfile = bitmap_git.pack;
805                 return 0;
806         }
807
808         reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
809
810         if (reuse_objects < reuse_threshold)
811                 return -1;
812
813         bitmap_git.reuse_objects = *entries = reuse_objects;
814         *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
815         *packfile = bitmap_git.pack;
816
817         return 0;
818 }
819
820 void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
821 {
822         assert(bitmap_git.result);
823
824         show_objects_for_type(bitmap_git.result, bitmap_git.commits,
825                 OBJ_COMMIT, show_reachable);
826         show_objects_for_type(bitmap_git.result, bitmap_git.trees,
827                 OBJ_TREE, show_reachable);
828         show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
829                 OBJ_BLOB, show_reachable);
830         show_objects_for_type(bitmap_git.result, bitmap_git.tags,
831                 OBJ_TAG, show_reachable);
832
833         show_extended_objects(bitmap_git.result, show_reachable);
834
835         bitmap_free(bitmap_git.result);
836         bitmap_git.result = NULL;
837 }
838
839 static uint32_t count_object_type(struct bitmap *objects,
840                                   enum object_type type)
841 {
842         struct eindex *eindex = &bitmap_git.ext_index;
843
844         uint32_t i = 0, count = 0;
845         struct ewah_iterator it;
846         eword_t filter;
847
848         switch (type) {
849         case OBJ_COMMIT:
850                 ewah_iterator_init(&it, bitmap_git.commits);
851                 break;
852
853         case OBJ_TREE:
854                 ewah_iterator_init(&it, bitmap_git.trees);
855                 break;
856
857         case OBJ_BLOB:
858                 ewah_iterator_init(&it, bitmap_git.blobs);
859                 break;
860
861         case OBJ_TAG:
862                 ewah_iterator_init(&it, bitmap_git.tags);
863                 break;
864
865         default:
866                 return 0;
867         }
868
869         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
870                 eword_t word = objects->words[i++] & filter;
871                 count += ewah_bit_popcount64(word);
872         }
873
874         for (i = 0; i < eindex->count; ++i) {
875                 if (eindex->objects[i]->type == type &&
876                         bitmap_get(objects, bitmap_git.pack->num_objects + i))
877                         count++;
878         }
879
880         return count;
881 }
882
883 void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
884                               uint32_t *blobs, uint32_t *tags)
885 {
886         assert(bitmap_git.result);
887
888         if (commits)
889                 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
890
891         if (trees)
892                 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
893
894         if (blobs)
895                 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
896
897         if (tags)
898                 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
899 }
900
901 struct bitmap_test_data {
902         struct bitmap *base;
903         struct progress *prg;
904         size_t seen;
905 };
906
907 static void test_show_object(struct object *object,
908                              const struct name_path *path,
909                              const char *last, void *data)
910 {
911         struct bitmap_test_data *tdata = data;
912         int bitmap_pos;
913
914         bitmap_pos = bitmap_position(object->sha1);
915         if (bitmap_pos < 0)
916                 die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
917
918         bitmap_set(tdata->base, bitmap_pos);
919         display_progress(tdata->prg, ++tdata->seen);
920 }
921
922 static void test_show_commit(struct commit *commit, void *data)
923 {
924         struct bitmap_test_data *tdata = data;
925         int bitmap_pos;
926
927         bitmap_pos = bitmap_position(commit->object.sha1);
928         if (bitmap_pos < 0)
929                 die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
930
931         bitmap_set(tdata->base, bitmap_pos);
932         display_progress(tdata->prg, ++tdata->seen);
933 }
934
935 void test_bitmap_walk(struct rev_info *revs)
936 {
937         struct object *root;
938         struct bitmap *result = NULL;
939         khiter_t pos;
940         size_t result_popcnt;
941         struct bitmap_test_data tdata;
942
943         if (prepare_bitmap_git())
944                 die("failed to load bitmap indexes");
945
946         if (revs->pending.nr != 1)
947                 die("you must specify exactly one commit to test");
948
949         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
950                 bitmap_git.version, bitmap_git.entry_count);
951
952         root = revs->pending.objects[0].item;
953         pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
954
955         if (pos < kh_end(bitmap_git.bitmaps)) {
956                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
957                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
958
959                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
960                         sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
961
962                 result = ewah_to_bitmap(bm);
963         }
964
965         if (result == NULL)
966                 die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
967
968         revs->tag_objects = 1;
969         revs->tree_objects = 1;
970         revs->blob_objects = 1;
971
972         result_popcnt = bitmap_popcount(result);
973
974         if (prepare_revision_walk(revs))
975                 die("revision walk setup failed");
976
977         tdata.base = bitmap_new();
978         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
979         tdata.seen = 0;
980
981         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
982
983         stop_progress(&tdata.prg);
984
985         if (bitmap_equals(result, tdata.base))
986                 fprintf(stderr, "OK!\n");
987         else
988                 fprintf(stderr, "Mismatch!\n");
989 }
990
991 static int rebuild_bitmap(uint32_t *reposition,
992                           struct ewah_bitmap *source,
993                           struct bitmap *dest)
994 {
995         uint32_t pos = 0;
996         struct ewah_iterator it;
997         eword_t word;
998
999         ewah_iterator_init(&it, source);
1000
1001         while (ewah_iterator_next(&word, &it)) {
1002                 uint32_t offset, bit_pos;
1003
1004                 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
1005                         if ((word >> offset) == 0)
1006                                 break;
1007
1008                         offset += ewah_bit_ctz64(word >> offset);
1009
1010                         bit_pos = reposition[pos + offset];
1011                         if (bit_pos > 0)
1012                                 bitmap_set(dest, bit_pos - 1);
1013                         else /* can't reuse, we don't have the object */
1014                                 return -1;
1015                 }
1016
1017                 pos += BITS_IN_WORD;
1018         }
1019         return 0;
1020 }
1021
1022 int rebuild_existing_bitmaps(struct packing_data *mapping,
1023                              khash_sha1 *reused_bitmaps,
1024                              int show_progress)
1025 {
1026         uint32_t i, num_objects;
1027         uint32_t *reposition;
1028         struct bitmap *rebuild;
1029         struct stored_bitmap *stored;
1030         struct progress *progress = NULL;
1031
1032         khiter_t hash_pos;
1033         int hash_ret;
1034
1035         if (prepare_bitmap_git() < 0)
1036                 return -1;
1037
1038         num_objects = bitmap_git.pack->num_objects;
1039         reposition = xcalloc(num_objects, sizeof(uint32_t));
1040
1041         for (i = 0; i < num_objects; ++i) {
1042                 const unsigned char *sha1;
1043                 struct revindex_entry *entry;
1044                 struct object_entry *oe;
1045
1046                 entry = &bitmap_git.reverse_index->revindex[i];
1047                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1048                 oe = packlist_find(mapping, sha1, NULL);
1049
1050                 if (oe)
1051                         reposition[i] = oe->in_pack_pos + 1;
1052         }
1053
1054         rebuild = bitmap_new();
1055         i = 0;
1056
1057         if (show_progress)
1058                 progress = start_progress("Reusing bitmaps", 0);
1059
1060         kh_foreach_value(bitmap_git.bitmaps, stored, {
1061                 if (stored->flags & BITMAP_FLAG_REUSE) {
1062                         if (!rebuild_bitmap(reposition,
1063                                             lookup_stored_bitmap(stored),
1064                                             rebuild)) {
1065                                 hash_pos = kh_put_sha1(reused_bitmaps,
1066                                                        stored->sha1,
1067                                                        &hash_ret);
1068                                 kh_value(reused_bitmaps, hash_pos) =
1069                                         bitmap_to_ewah(rebuild);
1070                         }
1071                         bitmap_reset(rebuild);
1072                         display_progress(progress, ++i);
1073                 }
1074         });
1075
1076         stop_progress(&progress);
1077
1078         free(reposition);
1079         bitmap_free(rebuild);
1080         return 0;
1081 }