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 */
17struct 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 */
32static 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 comits */
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
96static 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 */
119static 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
137static 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
170static 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
200static 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
207static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
208{
209 return buffer[(*pos)++];
210}
211
212static 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
255static int open_pack_bitmap_1(struct packed_git *packfile)
256{
257 int fd;
258 struct stat st;
259 char *idx_name;
260
261 if (open_pack_index(packfile))
262 return -1;
263
264 idx_name = pack_bitmap_filename(packfile);
265 fd = git_open_noatime(idx_name);
266 free(idx_name);
267
268 if (fd < 0)
269 return -1;
270
271 if (fstat(fd, &st)) {
272 close(fd);
273 return -1;
274 }
275
276 if (bitmap_git.pack) {
277 warning("ignoring extra bitmap file: %s", packfile->pack_name);
278 close(fd);
279 return -1;
280 }
281
282 bitmap_git.pack = packfile;
283 bitmap_git.map_size = xsize_t(st.st_size);
284 bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
285 bitmap_git.map_pos = 0;
286 close(fd);
287
288 if (load_bitmap_header(&bitmap_git) < 0) {
289 munmap(bitmap_git.map, bitmap_git.map_size);
290 bitmap_git.map = NULL;
291 bitmap_git.map_size = 0;
292 return -1;
293 }
294
295 return 0;
296}
297
298static int load_pack_bitmap(void)
299{
300 assert(bitmap_git.map && !bitmap_git.loaded);
301
302 bitmap_git.bitmaps = kh_init_sha1();
303 bitmap_git.ext_index.positions = kh_init_sha1_pos();
304 bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
305
306 if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
307 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
308 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
309 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
310 goto failed;
311
312 if (load_bitmap_entries_v1(&bitmap_git) < 0)
313 goto failed;
314
315 bitmap_git.loaded = 1;
316 return 0;
317
318failed:
319 munmap(bitmap_git.map, bitmap_git.map_size);
320 bitmap_git.map = NULL;
321 bitmap_git.map_size = 0;
322 return -1;
323}
324
325char *pack_bitmap_filename(struct packed_git *p)
326{
327 char *idx_name;
328 int len;
329
330 len = strlen(p->pack_name) - strlen(".pack");
331 idx_name = xmalloc(len + strlen(".bitmap") + 1);
332
333 memcpy(idx_name, p->pack_name, len);
334 memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
335
336 return idx_name;
337}
338
339static 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
355int 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
366struct include_data {
367 struct bitmap *base;
368 struct bitmap *seen;
369};
370
371static 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
384static 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
393static 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
399static 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 eindex->objects = xrealloc(eindex->objects,
412 eindex->alloc * sizeof(struct object *));
413 eindex->hashes = xrealloc(eindex->hashes,
414 eindex->alloc * sizeof(uint32_t));
415 }
416
417 bitmap_pos = eindex->count;
418 eindex->objects[eindex->count] = object;
419 eindex->hashes[eindex->count] = pack_name_hash(name);
420 kh_value(eindex->positions, hash_pos) = bitmap_pos;
421 eindex->count++;
422 } else {
423 bitmap_pos = kh_value(eindex->positions, hash_pos);
424 }
425
426 return bitmap_pos + bitmap_git.pack->num_objects;
427}
428
429static void show_object(struct object *object, const struct name_path *path,
430 const char *last, void *data)
431{
432 struct bitmap *base = data;
433 int bitmap_pos;
434
435 bitmap_pos = bitmap_position(object->sha1);
436
437 if (bitmap_pos < 0) {
438 char *name = path_name(path, last);
439 bitmap_pos = ext_index_add_object(object, name);
440 free(name);
441 }
442
443 bitmap_set(base, bitmap_pos);
444}
445
446static void show_commit(struct commit *commit, void *data)
447{
448}
449
450static int add_to_include_set(struct include_data *data,
451 const unsigned char *sha1,
452 int bitmap_pos)
453{
454 khiter_t hash_pos;
455
456 if (data->seen && bitmap_get(data->seen, bitmap_pos))
457 return 0;
458
459 if (bitmap_get(data->base, bitmap_pos))
460 return 0;
461
462 hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
463 if (hash_pos < kh_end(bitmap_git.bitmaps)) {
464 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
465 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
466 return 0;
467 }
468
469 bitmap_set(data->base, bitmap_pos);
470 return 1;
471}
472
473static int should_include(struct commit *commit, void *_data)
474{
475 struct include_data *data = _data;
476 int bitmap_pos;
477
478 bitmap_pos = bitmap_position(commit->object.sha1);
479 if (bitmap_pos < 0)
480 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
481
482 if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
483 struct commit_list *parent = commit->parents;
484
485 while (parent) {
486 parent->item->object.flags |= SEEN;
487 parent = parent->next;
488 }
489
490 return 0;
491 }
492
493 return 1;
494}
495
496static struct bitmap *find_objects(struct rev_info *revs,
497 struct object_list *roots,
498 struct bitmap *seen)
499{
500 struct bitmap *base = NULL;
501 int needs_walk = 0;
502
503 struct object_list *not_mapped = NULL;
504
505 /*
506 * Go through all the roots for the walk. The ones that have bitmaps
507 * on the bitmap index will be `or`ed together to form an initial
508 * global reachability analysis.
509 *
510 * The ones without bitmaps in the index will be stored in the
511 * `not_mapped_list` for further processing.
512 */
513 while (roots) {
514 struct object *object = roots->item;
515 roots = roots->next;
516
517 if (object->type == OBJ_COMMIT) {
518 khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
519
520 if (pos < kh_end(bitmap_git.bitmaps)) {
521 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
522 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
523
524 if (base == NULL)
525 base = ewah_to_bitmap(or_with);
526 else
527 bitmap_or_ewah(base, or_with);
528
529 object->flags |= SEEN;
530 continue;
531 }
532 }
533
534 object_list_insert(object, ¬_mapped);
535 }
536
537 /*
538 * Best case scenario: We found bitmaps for all the roots,
539 * so the resulting `or` bitmap has the full reachability analysis
540 */
541 if (not_mapped == NULL)
542 return base;
543
544 roots = not_mapped;
545
546 /*
547 * Let's iterate through all the roots that don't have bitmaps to
548 * check if we can determine them to be reachable from the existing
549 * global bitmap.
550 *
551 * If we cannot find them in the existing global bitmap, we'll need
552 * to push them to an actual walk and run it until we can confirm
553 * they are reachable
554 */
555 while (roots) {
556 struct object *object = roots->item;
557 int pos;
558
559 roots = roots->next;
560 pos = bitmap_position(object->sha1);
561
562 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
563 object->flags &= ~UNINTERESTING;
564 add_pending_object(revs, object, "");
565 needs_walk = 1;
566 } else {
567 object->flags |= SEEN;
568 }
569 }
570
571 if (needs_walk) {
572 struct include_data incdata;
573
574 if (base == NULL)
575 base = bitmap_new();
576
577 incdata.base = base;
578 incdata.seen = seen;
579
580 revs->include_check = should_include;
581 revs->include_check_data = &incdata;
582
583 if (prepare_revision_walk(revs))
584 die("revision walk setup failed");
585
586 traverse_commit_list(revs, show_commit, show_object, base);
587 }
588
589 return base;
590}
591
592static void show_extended_objects(struct bitmap *objects,
593 show_reachable_fn show_reach)
594{
595 struct eindex *eindex = &bitmap_git.ext_index;
596 uint32_t i;
597
598 for (i = 0; i < eindex->count; ++i) {
599 struct object *obj;
600
601 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
602 continue;
603
604 obj = eindex->objects[i];
605 show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
606 }
607}
608
609static void show_objects_for_type(
610 struct bitmap *objects,
611 struct ewah_bitmap *type_filter,
612 enum object_type object_type,
613 show_reachable_fn show_reach)
614{
615 size_t pos = 0, i = 0;
616 uint32_t offset;
617
618 struct ewah_iterator it;
619 eword_t filter;
620
621 if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
622 return;
623
624 ewah_iterator_init(&it, type_filter);
625
626 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
627 eword_t word = objects->words[i] & filter;
628
629 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
630 const unsigned char *sha1;
631 struct revindex_entry *entry;
632 uint32_t hash = 0;
633
634 if ((word >> offset) == 0)
635 break;
636
637 offset += ewah_bit_ctz64(word >> offset);
638
639 if (pos + offset < bitmap_git.reuse_objects)
640 continue;
641
642 entry = &bitmap_git.reverse_index->revindex[pos + offset];
643 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
644
645 if (bitmap_git.hashes)
646 hash = ntohl(bitmap_git.hashes[entry->nr]);
647
648 show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
649 }
650
651 pos += BITS_IN_WORD;
652 i++;
653 }
654}
655
656static int in_bitmapped_pack(struct object_list *roots)
657{
658 while (roots) {
659 struct object *object = roots->item;
660 roots = roots->next;
661
662 if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
663 return 1;
664 }
665
666 return 0;
667}
668
669int prepare_bitmap_walk(struct rev_info *revs)
670{
671 unsigned int i;
672 unsigned int pending_nr = revs->pending.nr;
673 struct object_array_entry *pending_e = revs->pending.objects;
674
675 struct object_list *wants = NULL;
676 struct object_list *haves = NULL;
677
678 struct bitmap *wants_bitmap = NULL;
679 struct bitmap *haves_bitmap = NULL;
680
681 if (!bitmap_git.loaded) {
682 /* try to open a bitmapped pack, but don't parse it yet
683 * because we may not need to use it */
684 if (open_pack_bitmap() < 0)
685 return -1;
686 }
687
688 for (i = 0; i < pending_nr; ++i) {
689 struct object *object = pending_e[i].item;
690
691 if (object->type == OBJ_NONE)
692 parse_object_or_die(object->sha1, NULL);
693
694 while (object->type == OBJ_TAG) {
695 struct tag *tag = (struct tag *) object;
696
697 if (object->flags & UNINTERESTING)
698 object_list_insert(object, &haves);
699 else
700 object_list_insert(object, &wants);
701
702 if (!tag->tagged)
703 die("bad tag");
704 object = parse_object_or_die(tag->tagged->sha1, NULL);
705 }
706
707 if (object->flags & UNINTERESTING)
708 object_list_insert(object, &haves);
709 else
710 object_list_insert(object, &wants);
711 }
712
713 /*
714 * if we have a HAVES list, but none of those haves is contained
715 * in the packfile that has a bitmap, we don't have anything to
716 * optimize here
717 */
718 if (haves && !in_bitmapped_pack(haves))
719 return -1;
720
721 /* if we don't want anything, we're done here */
722 if (!wants)
723 return -1;
724
725 /*
726 * now we're going to use bitmaps, so load the actual bitmap entries
727 * from disk. this is the point of no return; after this the rev_list
728 * becomes invalidated and we must perform the revwalk through bitmaps
729 */
730 if (!bitmap_git.loaded && load_pack_bitmap() < 0)
731 return -1;
732
733 revs->pending.nr = 0;
734 revs->pending.alloc = 0;
735 revs->pending.objects = NULL;
736
737 if (haves) {
738 revs->ignore_missing_links = 1;
739 haves_bitmap = find_objects(revs, haves, NULL);
740 reset_revision_walk();
741 revs->ignore_missing_links = 0;
742
743 if (haves_bitmap == NULL)
744 die("BUG: failed to perform bitmap walk");
745 }
746
747 wants_bitmap = find_objects(revs, wants, haves_bitmap);
748
749 if (!wants_bitmap)
750 die("BUG: failed to perform bitmap walk");
751
752 if (haves_bitmap)
753 bitmap_and_not(wants_bitmap, haves_bitmap);
754
755 bitmap_git.result = wants_bitmap;
756
757 bitmap_free(haves_bitmap);
758 return 0;
759}
760
761int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
762 uint32_t *entries,
763 off_t *up_to)
764{
765 /*
766 * Reuse the packfile content if we need more than
767 * 90% of its objects
768 */
769 static const double REUSE_PERCENT = 0.9;
770
771 struct bitmap *result = bitmap_git.result;
772 uint32_t reuse_threshold;
773 uint32_t i, reuse_objects = 0;
774
775 assert(result);
776
777 for (i = 0; i < result->word_alloc; ++i) {
778 if (result->words[i] != (eword_t)~0) {
779 reuse_objects += ewah_bit_ctz64(~result->words[i]);
780 break;
781 }
782
783 reuse_objects += BITS_IN_WORD;
784 }
785
786#ifdef GIT_BITMAP_DEBUG
787 {
788 const unsigned char *sha1;
789 struct revindex_entry *entry;
790
791 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
792 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
793
794 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
795 reuse_objects, result->words[i]);
796 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
797 }
798#endif
799
800 if (!reuse_objects)
801 return -1;
802
803 if (reuse_objects >= bitmap_git.pack->num_objects) {
804 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
805 *up_to = -1; /* reuse the full pack */
806 *packfile = bitmap_git.pack;
807 return 0;
808 }
809
810 reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
811
812 if (reuse_objects < reuse_threshold)
813 return -1;
814
815 bitmap_git.reuse_objects = *entries = reuse_objects;
816 *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
817 *packfile = bitmap_git.pack;
818
819 return 0;
820}
821
822void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
823{
824 assert(bitmap_git.result);
825
826 show_objects_for_type(bitmap_git.result, bitmap_git.commits,
827 OBJ_COMMIT, show_reachable);
828 show_objects_for_type(bitmap_git.result, bitmap_git.trees,
829 OBJ_TREE, show_reachable);
830 show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
831 OBJ_BLOB, show_reachable);
832 show_objects_for_type(bitmap_git.result, bitmap_git.tags,
833 OBJ_TAG, show_reachable);
834
835 show_extended_objects(bitmap_git.result, show_reachable);
836
837 bitmap_free(bitmap_git.result);
838 bitmap_git.result = NULL;
839}
840
841static uint32_t count_object_type(struct bitmap *objects,
842 enum object_type type)
843{
844 struct eindex *eindex = &bitmap_git.ext_index;
845
846 uint32_t i = 0, count = 0;
847 struct ewah_iterator it;
848 eword_t filter;
849
850 switch (type) {
851 case OBJ_COMMIT:
852 ewah_iterator_init(&it, bitmap_git.commits);
853 break;
854
855 case OBJ_TREE:
856 ewah_iterator_init(&it, bitmap_git.trees);
857 break;
858
859 case OBJ_BLOB:
860 ewah_iterator_init(&it, bitmap_git.blobs);
861 break;
862
863 case OBJ_TAG:
864 ewah_iterator_init(&it, bitmap_git.tags);
865 break;
866
867 default:
868 return 0;
869 }
870
871 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
872 eword_t word = objects->words[i++] & filter;
873 count += ewah_bit_popcount64(word);
874 }
875
876 for (i = 0; i < eindex->count; ++i) {
877 if (eindex->objects[i]->type == type &&
878 bitmap_get(objects, bitmap_git.pack->num_objects + i))
879 count++;
880 }
881
882 return count;
883}
884
885void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
886 uint32_t *blobs, uint32_t *tags)
887{
888 assert(bitmap_git.result);
889
890 if (commits)
891 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
892
893 if (trees)
894 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
895
896 if (blobs)
897 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
898
899 if (tags)
900 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
901}
902
903struct bitmap_test_data {
904 struct bitmap *base;
905 struct progress *prg;
906 size_t seen;
907};
908
909static void test_show_object(struct object *object,
910 const struct name_path *path,
911 const char *last, void *data)
912{
913 struct bitmap_test_data *tdata = data;
914 int bitmap_pos;
915
916 bitmap_pos = bitmap_position(object->sha1);
917 if (bitmap_pos < 0)
918 die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
919
920 bitmap_set(tdata->base, bitmap_pos);
921 display_progress(tdata->prg, ++tdata->seen);
922}
923
924static void test_show_commit(struct commit *commit, void *data)
925{
926 struct bitmap_test_data *tdata = data;
927 int bitmap_pos;
928
929 bitmap_pos = bitmap_position(commit->object.sha1);
930 if (bitmap_pos < 0)
931 die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
932
933 bitmap_set(tdata->base, bitmap_pos);
934 display_progress(tdata->prg, ++tdata->seen);
935}
936
937void test_bitmap_walk(struct rev_info *revs)
938{
939 struct object *root;
940 struct bitmap *result = NULL;
941 khiter_t pos;
942 size_t result_popcnt;
943 struct bitmap_test_data tdata;
944
945 if (prepare_bitmap_git())
946 die("failed to load bitmap indexes");
947
948 if (revs->pending.nr != 1)
949 die("you must specify exactly one commit to test");
950
951 fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
952 bitmap_git.version, bitmap_git.entry_count);
953
954 root = revs->pending.objects[0].item;
955 pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
956
957 if (pos < kh_end(bitmap_git.bitmaps)) {
958 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
959 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
960
961 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
962 sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
963
964 result = ewah_to_bitmap(bm);
965 }
966
967 if (result == NULL)
968 die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
969
970 revs->tag_objects = 1;
971 revs->tree_objects = 1;
972 revs->blob_objects = 1;
973
974 result_popcnt = bitmap_popcount(result);
975
976 if (prepare_revision_walk(revs))
977 die("revision walk setup failed");
978
979 tdata.base = bitmap_new();
980 tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
981 tdata.seen = 0;
982
983 traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
984
985 stop_progress(&tdata.prg);
986
987 if (bitmap_equals(result, tdata.base))
988 fprintf(stderr, "OK!\n");
989 else
990 fprintf(stderr, "Mismatch!\n");
991}
992
993static int rebuild_bitmap(uint32_t *reposition,
994 struct ewah_bitmap *source,
995 struct bitmap *dest)
996{
997 uint32_t pos = 0;
998 struct ewah_iterator it;
999 eword_t word;
1000
1001 ewah_iterator_init(&it, source);
1002
1003 while (ewah_iterator_next(&word, &it)) {
1004 uint32_t offset, bit_pos;
1005
1006 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
1007 if ((word >> offset) == 0)
1008 break;
1009
1010 offset += ewah_bit_ctz64(word >> offset);
1011
1012 bit_pos = reposition[pos + offset];
1013 if (bit_pos > 0)
1014 bitmap_set(dest, bit_pos - 1);
1015 else /* can't reuse, we don't have the object */
1016 return -1;
1017 }
1018
1019 pos += BITS_IN_WORD;
1020 }
1021 return 0;
1022}
1023
1024int rebuild_existing_bitmaps(struct packing_data *mapping,
1025 khash_sha1 *reused_bitmaps,
1026 int show_progress)
1027{
1028 uint32_t i, num_objects;
1029 uint32_t *reposition;
1030 struct bitmap *rebuild;
1031 struct stored_bitmap *stored;
1032 struct progress *progress = NULL;
1033
1034 khiter_t hash_pos;
1035 int hash_ret;
1036
1037 if (prepare_bitmap_git() < 0)
1038 return -1;
1039
1040 num_objects = bitmap_git.pack->num_objects;
1041 reposition = xcalloc(num_objects, sizeof(uint32_t));
1042
1043 for (i = 0; i < num_objects; ++i) {
1044 const unsigned char *sha1;
1045 struct revindex_entry *entry;
1046 struct object_entry *oe;
1047
1048 entry = &bitmap_git.reverse_index->revindex[i];
1049 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1050 oe = packlist_find(mapping, sha1, NULL);
1051
1052 if (oe)
1053 reposition[i] = oe->in_pack_pos + 1;
1054 }
1055
1056 rebuild = bitmap_new();
1057 i = 0;
1058
1059 if (show_progress)
1060 progress = start_progress("Reusing bitmaps", 0);
1061
1062 kh_foreach_value(bitmap_git.bitmaps, stored, {
1063 if (stored->flags & BITMAP_FLAG_REUSE) {
1064 if (!rebuild_bitmap(reposition,
1065 lookup_stored_bitmap(stored),
1066 rebuild)) {
1067 hash_pos = kh_put_sha1(reused_bitmaps,
1068 stored->sha1,
1069 &hash_ret);
1070 kh_value(reused_bitmaps, hash_pos) =
1071 bitmap_to_ewah(rebuild);
1072 }
1073 bitmap_reset(rebuild);
1074 display_progress(progress, ++i);
1075 }
1076 });
1077
1078 stop_progress(&progress);
1079
1080 free(reposition);
1081 bitmap_free(rebuild);
1082 return 0;
1083}