Merge branch 'sg/doc-pretty-formats'
[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         /*
37          * Mark the first `reuse_objects` in the packfile as reused:
38          * they will be sent as-is without using them for repacking
39          * calculations
40          */
41         uint32_t reuse_objects;
42
43         /* mmapped buffer of the whole bitmap index */
44         unsigned char *map;
45         size_t map_size; /* size of the mmaped buffer */
46         size_t map_pos; /* current position when loading the index */
47
48         /*
49          * Type indexes.
50          *
51          * Each bitmap marks which objects in the packfile  are of the given
52          * type. This provides type information when yielding the objects from
53          * the packfile during a walk, which allows for better delta bases.
54          */
55         struct ewah_bitmap *commits;
56         struct ewah_bitmap *trees;
57         struct ewah_bitmap *blobs;
58         struct ewah_bitmap *tags;
59
60         /* Map from SHA1 -> `stored_bitmap` for all the bitmapped commits */
61         khash_sha1 *bitmaps;
62
63         /* Number of bitmapped commits */
64         uint32_t entry_count;
65
66         /* Name-hash cache (or NULL if not present). */
67         uint32_t *hashes;
68
69         /*
70          * Extended index.
71          *
72          * When trying to perform bitmap operations with objects that are not
73          * packed in `pack`, these objects are added to this "fake index" and
74          * are assumed to appear at the end of the packfile for all operations
75          */
76         struct eindex {
77                 struct object **objects;
78                 uint32_t *hashes;
79                 uint32_t count, alloc;
80                 khash_sha1_pos *positions;
81         } ext_index;
82
83         /* Bitmap result of the last performed walk */
84         struct bitmap *result;
85
86         /* Version of the bitmap index */
87         unsigned int version;
88
89         unsigned loaded : 1;
90
91 } bitmap_git;
92
93 static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st)
94 {
95         struct ewah_bitmap *parent;
96         struct ewah_bitmap *composed;
97
98         if (st->xor == NULL)
99                 return st->root;
100
101         composed = ewah_pool_new();
102         parent = lookup_stored_bitmap(st->xor);
103         ewah_xor(st->root, parent, composed);
104
105         ewah_pool_free(st->root);
106         st->root = composed;
107         st->xor = NULL;
108
109         return composed;
110 }
111
112 /*
113  * Read a bitmap from the current read position on the mmaped
114  * index, and increase the read position accordingly
115  */
116 static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index)
117 {
118         struct ewah_bitmap *b = ewah_pool_new();
119
120         int bitmap_size = ewah_read_mmap(b,
121                 index->map + index->map_pos,
122                 index->map_size - index->map_pos);
123
124         if (bitmap_size < 0) {
125                 error("Failed to load bitmap index (corrupted?)");
126                 ewah_pool_free(b);
127                 return NULL;
128         }
129
130         index->map_pos += bitmap_size;
131         return b;
132 }
133
134 static int load_bitmap_header(struct bitmap_index *index)
135 {
136         struct bitmap_disk_header *header = (void *)index->map;
137
138         if (index->map_size < sizeof(*header) + 20)
139                 return error("Corrupted bitmap index (missing header data)");
140
141         if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0)
142                 return error("Corrupted bitmap index file (wrong header)");
143
144         index->version = ntohs(header->version);
145         if (index->version != 1)
146                 return error("Unsupported version for bitmap index file (%d)", index->version);
147
148         /* Parse known bitmap format options */
149         {
150                 uint32_t flags = ntohs(header->options);
151
152                 if ((flags & BITMAP_OPT_FULL_DAG) == 0)
153                         return error("Unsupported options for bitmap index file "
154                                 "(Git requires BITMAP_OPT_FULL_DAG)");
155
156                 if (flags & BITMAP_OPT_HASH_CACHE) {
157                         unsigned char *end = index->map + index->map_size - 20;
158                         index->hashes = ((uint32_t *)end) - index->pack->num_objects;
159                 }
160         }
161
162         index->entry_count = ntohl(header->entry_count);
163         index->map_pos += sizeof(*header);
164         return 0;
165 }
166
167 static struct stored_bitmap *store_bitmap(struct bitmap_index *index,
168                                           struct ewah_bitmap *root,
169                                           const unsigned char *sha1,
170                                           struct stored_bitmap *xor_with,
171                                           int flags)
172 {
173         struct stored_bitmap *stored;
174         khiter_t hash_pos;
175         int ret;
176
177         stored = xmalloc(sizeof(struct stored_bitmap));
178         stored->root = root;
179         stored->xor = xor_with;
180         stored->flags = flags;
181         hashcpy(stored->sha1, sha1);
182
183         hash_pos = kh_put_sha1(index->bitmaps, stored->sha1, &ret);
184
185         /* a 0 return code means the insertion succeeded with no changes,
186          * because the SHA1 already existed on the map. this is bad, there
187          * shouldn't be duplicated commits in the index */
188         if (ret == 0) {
189                 error("Duplicate entry in bitmap index: %s", sha1_to_hex(sha1));
190                 return NULL;
191         }
192
193         kh_value(index->bitmaps, hash_pos) = stored;
194         return stored;
195 }
196
197 static inline uint32_t read_be32(const unsigned char *buffer, size_t *pos)
198 {
199         uint32_t result = get_be32(buffer + *pos);
200         (*pos) += sizeof(result);
201         return result;
202 }
203
204 static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
205 {
206         return buffer[(*pos)++];
207 }
208
209 #define MAX_XOR_OFFSET 160
210
211 static int load_bitmap_entries_v1(struct bitmap_index *index)
212 {
213         uint32_t i;
214         struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
215
216         for (i = 0; i < index->entry_count; ++i) {
217                 int xor_offset, flags;
218                 struct ewah_bitmap *bitmap = NULL;
219                 struct stored_bitmap *xor_bitmap = NULL;
220                 uint32_t commit_idx_pos;
221                 const unsigned char *sha1;
222
223                 commit_idx_pos = read_be32(index->map, &index->map_pos);
224                 xor_offset = read_u8(index->map, &index->map_pos);
225                 flags = read_u8(index->map, &index->map_pos);
226
227                 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
228
229                 bitmap = read_bitmap_1(index);
230                 if (!bitmap)
231                         return -1;
232
233                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
234                         return error("Corrupted bitmap pack index");
235
236                 if (xor_offset > 0) {
237                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
238
239                         if (xor_bitmap == NULL)
240                                 return error("Invalid XOR offset in bitmap pack index");
241                 }
242
243                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
244                         index, bitmap, sha1, xor_bitmap, flags);
245         }
246
247         return 0;
248 }
249
250 static char *pack_bitmap_filename(struct packed_git *p)
251 {
252         size_t len;
253
254         if (!strip_suffix(p->pack_name, ".pack", &len))
255                 die("BUG: pack_name does not end in .pack");
256         return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
257 }
258
259 static int open_pack_bitmap_1(struct packed_git *packfile)
260 {
261         int fd;
262         struct stat st;
263         char *idx_name;
264
265         if (open_pack_index(packfile))
266                 return -1;
267
268         idx_name = pack_bitmap_filename(packfile);
269         fd = git_open(idx_name);
270         free(idx_name);
271
272         if (fd < 0)
273                 return -1;
274
275         if (fstat(fd, &st)) {
276                 close(fd);
277                 return -1;
278         }
279
280         if (bitmap_git.pack) {
281                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
282                 close(fd);
283                 return -1;
284         }
285
286         bitmap_git.pack = packfile;
287         bitmap_git.map_size = xsize_t(st.st_size);
288         bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
289         bitmap_git.map_pos = 0;
290         close(fd);
291
292         if (load_bitmap_header(&bitmap_git) < 0) {
293                 munmap(bitmap_git.map, bitmap_git.map_size);
294                 bitmap_git.map = NULL;
295                 bitmap_git.map_size = 0;
296                 return -1;
297         }
298
299         return 0;
300 }
301
302 static int load_pack_bitmap(void)
303 {
304         assert(bitmap_git.map && !bitmap_git.loaded);
305
306         bitmap_git.bitmaps = kh_init_sha1();
307         bitmap_git.ext_index.positions = kh_init_sha1_pos();
308         load_pack_revindex(bitmap_git.pack);
309
310         if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
311                 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
312                 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
313                 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
314                 goto failed;
315
316         if (load_bitmap_entries_v1(&bitmap_git) < 0)
317                 goto failed;
318
319         bitmap_git.loaded = 1;
320         return 0;
321
322 failed:
323         munmap(bitmap_git.map, bitmap_git.map_size);
324         bitmap_git.map = NULL;
325         bitmap_git.map_size = 0;
326         return -1;
327 }
328
329 static int open_pack_bitmap(void)
330 {
331         struct packed_git *p;
332         int ret = -1;
333
334         assert(!bitmap_git.map && !bitmap_git.loaded);
335
336         prepare_packed_git();
337         for (p = packed_git; p; p = p->next) {
338                 if (open_pack_bitmap_1(p) == 0)
339                         ret = 0;
340         }
341
342         return ret;
343 }
344
345 int prepare_bitmap_git(void)
346 {
347         if (bitmap_git.loaded)
348                 return 0;
349
350         if (!open_pack_bitmap())
351                 return load_pack_bitmap();
352
353         return -1;
354 }
355
356 struct include_data {
357         struct bitmap *base;
358         struct bitmap *seen;
359 };
360
361 static inline int bitmap_position_extended(const unsigned char *sha1)
362 {
363         khash_sha1_pos *positions = bitmap_git.ext_index.positions;
364         khiter_t pos = kh_get_sha1_pos(positions, sha1);
365
366         if (pos < kh_end(positions)) {
367                 int bitmap_pos = kh_value(positions, pos);
368                 return bitmap_pos + bitmap_git.pack->num_objects;
369         }
370
371         return -1;
372 }
373
374 static inline int bitmap_position_packfile(const unsigned char *sha1)
375 {
376         off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
377         if (!offset)
378                 return -1;
379
380         return find_revindex_position(bitmap_git.pack, offset);
381 }
382
383 static int bitmap_position(const unsigned char *sha1)
384 {
385         int pos = bitmap_position_packfile(sha1);
386         return (pos >= 0) ? pos : bitmap_position_extended(sha1);
387 }
388
389 static int ext_index_add_object(struct object *object, const char *name)
390 {
391         struct eindex *eindex = &bitmap_git.ext_index;
392
393         khiter_t hash_pos;
394         int hash_ret;
395         int bitmap_pos;
396
397         hash_pos = kh_put_sha1_pos(eindex->positions, object->oid.hash, &hash_ret);
398         if (hash_ret > 0) {
399                 if (eindex->count >= eindex->alloc) {
400                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
401                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
402                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
403                 }
404
405                 bitmap_pos = eindex->count;
406                 eindex->objects[eindex->count] = object;
407                 eindex->hashes[eindex->count] = pack_name_hash(name);
408                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
409                 eindex->count++;
410         } else {
411                 bitmap_pos = kh_value(eindex->positions, hash_pos);
412         }
413
414         return bitmap_pos + bitmap_git.pack->num_objects;
415 }
416
417 static void show_object(struct object *object, const char *name, void *data)
418 {
419         struct bitmap *base = data;
420         int bitmap_pos;
421
422         bitmap_pos = bitmap_position(object->oid.hash);
423
424         if (bitmap_pos < 0)
425                 bitmap_pos = ext_index_add_object(object, name);
426
427         bitmap_set(base, bitmap_pos);
428 }
429
430 static void show_commit(struct commit *commit, void *data)
431 {
432 }
433
434 static int add_to_include_set(struct include_data *data,
435                               const unsigned char *sha1,
436                               int bitmap_pos)
437 {
438         khiter_t hash_pos;
439
440         if (data->seen && bitmap_get(data->seen, bitmap_pos))
441                 return 0;
442
443         if (bitmap_get(data->base, bitmap_pos))
444                 return 0;
445
446         hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
447         if (hash_pos < kh_end(bitmap_git.bitmaps)) {
448                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
449                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
450                 return 0;
451         }
452
453         bitmap_set(data->base, bitmap_pos);
454         return 1;
455 }
456
457 static int should_include(struct commit *commit, void *_data)
458 {
459         struct include_data *data = _data;
460         int bitmap_pos;
461
462         bitmap_pos = bitmap_position(commit->object.oid.hash);
463         if (bitmap_pos < 0)
464                 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
465
466         if (!add_to_include_set(data, commit->object.oid.hash, bitmap_pos)) {
467                 struct commit_list *parent = commit->parents;
468
469                 while (parent) {
470                         parent->item->object.flags |= SEEN;
471                         parent = parent->next;
472                 }
473
474                 return 0;
475         }
476
477         return 1;
478 }
479
480 static struct bitmap *find_objects(struct rev_info *revs,
481                                    struct object_list *roots,
482                                    struct bitmap *seen)
483 {
484         struct bitmap *base = NULL;
485         int needs_walk = 0;
486
487         struct object_list *not_mapped = NULL;
488
489         /*
490          * Go through all the roots for the walk. The ones that have bitmaps
491          * on the bitmap index will be `or`ed together to form an initial
492          * global reachability analysis.
493          *
494          * The ones without bitmaps in the index will be stored in the
495          * `not_mapped_list` for further processing.
496          */
497         while (roots) {
498                 struct object *object = roots->item;
499                 roots = roots->next;
500
501                 if (object->type == OBJ_COMMIT) {
502                         khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->oid.hash);
503
504                         if (pos < kh_end(bitmap_git.bitmaps)) {
505                                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
506                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
507
508                                 if (base == NULL)
509                                         base = ewah_to_bitmap(or_with);
510                                 else
511                                         bitmap_or_ewah(base, or_with);
512
513                                 object->flags |= SEEN;
514                                 continue;
515                         }
516                 }
517
518                 object_list_insert(object, &not_mapped);
519         }
520
521         /*
522          * Best case scenario: We found bitmaps for all the roots,
523          * so the resulting `or` bitmap has the full reachability analysis
524          */
525         if (not_mapped == NULL)
526                 return base;
527
528         roots = not_mapped;
529
530         /*
531          * Let's iterate through all the roots that don't have bitmaps to
532          * check if we can determine them to be reachable from the existing
533          * global bitmap.
534          *
535          * If we cannot find them in the existing global bitmap, we'll need
536          * to push them to an actual walk and run it until we can confirm
537          * they are reachable
538          */
539         while (roots) {
540                 struct object *object = roots->item;
541                 int pos;
542
543                 roots = roots->next;
544                 pos = bitmap_position(object->oid.hash);
545
546                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
547                         object->flags &= ~UNINTERESTING;
548                         add_pending_object(revs, object, "");
549                         needs_walk = 1;
550                 } else {
551                         object->flags |= SEEN;
552                 }
553         }
554
555         if (needs_walk) {
556                 struct include_data incdata;
557
558                 if (base == NULL)
559                         base = bitmap_new();
560
561                 incdata.base = base;
562                 incdata.seen = seen;
563
564                 revs->include_check = should_include;
565                 revs->include_check_data = &incdata;
566
567                 if (prepare_revision_walk(revs))
568                         die("revision walk setup failed");
569
570                 traverse_commit_list(revs, show_commit, show_object, base);
571         }
572
573         return base;
574 }
575
576 static void show_extended_objects(struct bitmap *objects,
577                                   show_reachable_fn show_reach)
578 {
579         struct eindex *eindex = &bitmap_git.ext_index;
580         uint32_t i;
581
582         for (i = 0; i < eindex->count; ++i) {
583                 struct object *obj;
584
585                 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
586                         continue;
587
588                 obj = eindex->objects[i];
589                 show_reach(obj->oid.hash, obj->type, 0, eindex->hashes[i], NULL, 0);
590         }
591 }
592
593 static void show_objects_for_type(
594         struct bitmap *objects,
595         struct ewah_bitmap *type_filter,
596         enum object_type object_type,
597         show_reachable_fn show_reach)
598 {
599         size_t pos = 0, i = 0;
600         uint32_t offset;
601
602         struct ewah_iterator it;
603         eword_t filter;
604
605         if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
606                 return;
607
608         ewah_iterator_init(&it, type_filter);
609
610         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
611                 eword_t word = objects->words[i] & filter;
612
613                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
614                         const unsigned char *sha1;
615                         struct revindex_entry *entry;
616                         uint32_t hash = 0;
617
618                         if ((word >> offset) == 0)
619                                 break;
620
621                         offset += ewah_bit_ctz64(word >> offset);
622
623                         if (pos + offset < bitmap_git.reuse_objects)
624                                 continue;
625
626                         entry = &bitmap_git.pack->revindex[pos + offset];
627                         sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
628
629                         if (bitmap_git.hashes)
630                                 hash = ntohl(bitmap_git.hashes[entry->nr]);
631
632                         show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
633                 }
634
635                 pos += BITS_IN_EWORD;
636                 i++;
637         }
638 }
639
640 static int in_bitmapped_pack(struct object_list *roots)
641 {
642         while (roots) {
643                 struct object *object = roots->item;
644                 roots = roots->next;
645
646                 if (find_pack_entry_one(object->oid.hash, bitmap_git.pack) > 0)
647                         return 1;
648         }
649
650         return 0;
651 }
652
653 int prepare_bitmap_walk(struct rev_info *revs)
654 {
655         unsigned int i;
656         unsigned int pending_nr = revs->pending.nr;
657         struct object_array_entry *pending_e = revs->pending.objects;
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 < pending_nr; ++i) {
673                 struct object *object = pending_e[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         revs->pending.nr = 0;
718         revs->pending.alloc = 0;
719         revs->pending.objects = NULL;
720
721         if (haves) {
722                 revs->ignore_missing_links = 1;
723                 haves_bitmap = find_objects(revs, haves, NULL);
724                 reset_revision_walk();
725                 revs->ignore_missing_links = 0;
726
727                 if (haves_bitmap == NULL)
728                         die("BUG: failed to perform bitmap walk");
729         }
730
731         wants_bitmap = find_objects(revs, wants, haves_bitmap);
732
733         if (!wants_bitmap)
734                 die("BUG: failed to perform bitmap walk");
735
736         if (haves_bitmap)
737                 bitmap_and_not(wants_bitmap, haves_bitmap);
738
739         bitmap_git.result = wants_bitmap;
740
741         bitmap_free(haves_bitmap);
742         return 0;
743 }
744
745 int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
746                                        uint32_t *entries,
747                                        off_t *up_to)
748 {
749         /*
750          * Reuse the packfile content if we need more than
751          * 90% of its objects
752          */
753         static const double REUSE_PERCENT = 0.9;
754
755         struct bitmap *result = bitmap_git.result;
756         uint32_t reuse_threshold;
757         uint32_t i, reuse_objects = 0;
758
759         assert(result);
760
761         for (i = 0; i < result->word_alloc; ++i) {
762                 if (result->words[i] != (eword_t)~0) {
763                         reuse_objects += ewah_bit_ctz64(~result->words[i]);
764                         break;
765                 }
766
767                 reuse_objects += BITS_IN_EWORD;
768         }
769
770 #ifdef GIT_BITMAP_DEBUG
771         {
772                 const unsigned char *sha1;
773                 struct revindex_entry *entry;
774
775                 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
776                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
777
778                 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
779                         reuse_objects, result->words[i]);
780                 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
781         }
782 #endif
783
784         if (!reuse_objects)
785                 return -1;
786
787         if (reuse_objects >= bitmap_git.pack->num_objects) {
788                 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
789                 *up_to = -1; /* reuse the full pack */
790                 *packfile = bitmap_git.pack;
791                 return 0;
792         }
793
794         reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
795
796         if (reuse_objects < reuse_threshold)
797                 return -1;
798
799         bitmap_git.reuse_objects = *entries = reuse_objects;
800         *up_to = bitmap_git.pack->revindex[reuse_objects].offset;
801         *packfile = bitmap_git.pack;
802
803         return 0;
804 }
805
806 void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
807 {
808         assert(bitmap_git.result);
809
810         show_objects_for_type(bitmap_git.result, bitmap_git.commits,
811                 OBJ_COMMIT, show_reachable);
812         show_objects_for_type(bitmap_git.result, bitmap_git.trees,
813                 OBJ_TREE, show_reachable);
814         show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
815                 OBJ_BLOB, show_reachable);
816         show_objects_for_type(bitmap_git.result, bitmap_git.tags,
817                 OBJ_TAG, show_reachable);
818
819         show_extended_objects(bitmap_git.result, show_reachable);
820
821         bitmap_free(bitmap_git.result);
822         bitmap_git.result = NULL;
823 }
824
825 static uint32_t count_object_type(struct bitmap *objects,
826                                   enum object_type type)
827 {
828         struct eindex *eindex = &bitmap_git.ext_index;
829
830         uint32_t i = 0, count = 0;
831         struct ewah_iterator it;
832         eword_t filter;
833
834         switch (type) {
835         case OBJ_COMMIT:
836                 ewah_iterator_init(&it, bitmap_git.commits);
837                 break;
838
839         case OBJ_TREE:
840                 ewah_iterator_init(&it, bitmap_git.trees);
841                 break;
842
843         case OBJ_BLOB:
844                 ewah_iterator_init(&it, bitmap_git.blobs);
845                 break;
846
847         case OBJ_TAG:
848                 ewah_iterator_init(&it, bitmap_git.tags);
849                 break;
850
851         default:
852                 return 0;
853         }
854
855         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
856                 eword_t word = objects->words[i++] & filter;
857                 count += ewah_bit_popcount64(word);
858         }
859
860         for (i = 0; i < eindex->count; ++i) {
861                 if (eindex->objects[i]->type == type &&
862                         bitmap_get(objects, bitmap_git.pack->num_objects + i))
863                         count++;
864         }
865
866         return count;
867 }
868
869 void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
870                               uint32_t *blobs, uint32_t *tags)
871 {
872         assert(bitmap_git.result);
873
874         if (commits)
875                 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
876
877         if (trees)
878                 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
879
880         if (blobs)
881                 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
882
883         if (tags)
884                 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
885 }
886
887 struct bitmap_test_data {
888         struct bitmap *base;
889         struct progress *prg;
890         size_t seen;
891 };
892
893 static void test_show_object(struct object *object, const char *name,
894                              void *data)
895 {
896         struct bitmap_test_data *tdata = data;
897         int bitmap_pos;
898
899         bitmap_pos = bitmap_position(object->oid.hash);
900         if (bitmap_pos < 0)
901                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
902
903         bitmap_set(tdata->base, bitmap_pos);
904         display_progress(tdata->prg, ++tdata->seen);
905 }
906
907 static void test_show_commit(struct commit *commit, void *data)
908 {
909         struct bitmap_test_data *tdata = data;
910         int bitmap_pos;
911
912         bitmap_pos = bitmap_position(commit->object.oid.hash);
913         if (bitmap_pos < 0)
914                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
915
916         bitmap_set(tdata->base, bitmap_pos);
917         display_progress(tdata->prg, ++tdata->seen);
918 }
919
920 void test_bitmap_walk(struct rev_info *revs)
921 {
922         struct object *root;
923         struct bitmap *result = NULL;
924         khiter_t pos;
925         size_t result_popcnt;
926         struct bitmap_test_data tdata;
927
928         if (prepare_bitmap_git())
929                 die("failed to load bitmap indexes");
930
931         if (revs->pending.nr != 1)
932                 die("you must specify exactly one commit to test");
933
934         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
935                 bitmap_git.version, bitmap_git.entry_count);
936
937         root = revs->pending.objects[0].item;
938         pos = kh_get_sha1(bitmap_git.bitmaps, root->oid.hash);
939
940         if (pos < kh_end(bitmap_git.bitmaps)) {
941                 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
942                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
943
944                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
945                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
946
947                 result = ewah_to_bitmap(bm);
948         }
949
950         if (result == NULL)
951                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
952
953         revs->tag_objects = 1;
954         revs->tree_objects = 1;
955         revs->blob_objects = 1;
956
957         result_popcnt = bitmap_popcount(result);
958
959         if (prepare_revision_walk(revs))
960                 die("revision walk setup failed");
961
962         tdata.base = bitmap_new();
963         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
964         tdata.seen = 0;
965
966         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
967
968         stop_progress(&tdata.prg);
969
970         if (bitmap_equals(result, tdata.base))
971                 fprintf(stderr, "OK!\n");
972         else
973                 fprintf(stderr, "Mismatch!\n");
974
975         bitmap_free(result);
976 }
977
978 static int rebuild_bitmap(uint32_t *reposition,
979                           struct ewah_bitmap *source,
980                           struct bitmap *dest)
981 {
982         uint32_t pos = 0;
983         struct ewah_iterator it;
984         eword_t word;
985
986         ewah_iterator_init(&it, source);
987
988         while (ewah_iterator_next(&word, &it)) {
989                 uint32_t offset, bit_pos;
990
991                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
992                         if ((word >> offset) == 0)
993                                 break;
994
995                         offset += ewah_bit_ctz64(word >> offset);
996
997                         bit_pos = reposition[pos + offset];
998                         if (bit_pos > 0)
999                                 bitmap_set(dest, bit_pos - 1);
1000                         else /* can't reuse, we don't have the object */
1001                                 return -1;
1002                 }
1003
1004                 pos += BITS_IN_EWORD;
1005         }
1006         return 0;
1007 }
1008
1009 int rebuild_existing_bitmaps(struct packing_data *mapping,
1010                              khash_sha1 *reused_bitmaps,
1011                              int show_progress)
1012 {
1013         uint32_t i, num_objects;
1014         uint32_t *reposition;
1015         struct bitmap *rebuild;
1016         struct stored_bitmap *stored;
1017         struct progress *progress = NULL;
1018
1019         khiter_t hash_pos;
1020         int hash_ret;
1021
1022         if (prepare_bitmap_git() < 0)
1023                 return -1;
1024
1025         num_objects = bitmap_git.pack->num_objects;
1026         reposition = xcalloc(num_objects, sizeof(uint32_t));
1027
1028         for (i = 0; i < num_objects; ++i) {
1029                 const unsigned char *sha1;
1030                 struct revindex_entry *entry;
1031                 struct object_entry *oe;
1032
1033                 entry = &bitmap_git.pack->revindex[i];
1034                 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1035                 oe = packlist_find(mapping, sha1, NULL);
1036
1037                 if (oe)
1038                         reposition[i] = oe->in_pack_pos + 1;
1039         }
1040
1041         rebuild = bitmap_new();
1042         i = 0;
1043
1044         if (show_progress)
1045                 progress = start_progress("Reusing bitmaps", 0);
1046
1047         kh_foreach_value(bitmap_git.bitmaps, stored, {
1048                 if (stored->flags & BITMAP_FLAG_REUSE) {
1049                         if (!rebuild_bitmap(reposition,
1050                                             lookup_stored_bitmap(stored),
1051                                             rebuild)) {
1052                                 hash_pos = kh_put_sha1(reused_bitmaps,
1053                                                        stored->sha1,
1054                                                        &hash_ret);
1055                                 kh_value(reused_bitmaps, hash_pos) =
1056                                         bitmap_to_ewah(rebuild);
1057                         }
1058                         bitmap_reset(rebuild);
1059                         display_progress(progress, ++i);
1060                 }
1061         });
1062
1063         stop_progress(&progress);
1064
1065         free(reposition);
1066         bitmap_free(rebuild);
1067         return 0;
1068 }