Merge branch 'es/utf8-stupid-compiler-workaround' into maint
[git/git.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 #define MAX_XOR_OFFSET 160
213
214 static int load_bitmap_entries_v1(struct bitmap_index *index)
215 {
216 uint32_t i;
217 struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
218
219 for (i = 0; i < index->entry_count; ++i) {
220 int xor_offset, flags;
221 struct ewah_bitmap *bitmap = NULL;
222 struct stored_bitmap *xor_bitmap = NULL;
223 uint32_t commit_idx_pos;
224 const unsigned char *sha1;
225
226 commit_idx_pos = read_be32(index->map, &index->map_pos);
227 xor_offset = read_u8(index->map, &index->map_pos);
228 flags = read_u8(index->map, &index->map_pos);
229
230 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
231
232 bitmap = read_bitmap_1(index);
233 if (!bitmap)
234 return -1;
235
236 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
237 return error("Corrupted bitmap pack index");
238
239 if (xor_offset > 0) {
240 xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
241
242 if (xor_bitmap == NULL)
243 return error("Invalid XOR offset in bitmap pack index");
244 }
245
246 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
247 index, bitmap, sha1, xor_bitmap, flags);
248 }
249
250 return 0;
251 }
252
253 static char *pack_bitmap_filename(struct packed_git *p)
254 {
255 char *idx_name;
256 int len;
257
258 len = strlen(p->pack_name) - strlen(".pack");
259 idx_name = xmalloc(len + strlen(".bitmap") + 1);
260
261 memcpy(idx_name, p->pack_name, len);
262 memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
263
264 return idx_name;
265 }
266
267 static int open_pack_bitmap_1(struct packed_git *packfile)
268 {
269 int fd;
270 struct stat st;
271 char *idx_name;
272
273 if (open_pack_index(packfile))
274 return -1;
275
276 idx_name = pack_bitmap_filename(packfile);
277 fd = git_open_noatime(idx_name);
278 free(idx_name);
279
280 if (fd < 0)
281 return -1;
282
283 if (fstat(fd, &st)) {
284 close(fd);
285 return -1;
286 }
287
288 if (bitmap_git.pack) {
289 warning("ignoring extra bitmap file: %s", packfile->pack_name);
290 close(fd);
291 return -1;
292 }
293
294 bitmap_git.pack = packfile;
295 bitmap_git.map_size = xsize_t(st.st_size);
296 bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
297 bitmap_git.map_pos = 0;
298 close(fd);
299
300 if (load_bitmap_header(&bitmap_git) < 0) {
301 munmap(bitmap_git.map, bitmap_git.map_size);
302 bitmap_git.map = NULL;
303 bitmap_git.map_size = 0;
304 return -1;
305 }
306
307 return 0;
308 }
309
310 static int load_pack_bitmap(void)
311 {
312 assert(bitmap_git.map && !bitmap_git.loaded);
313
314 bitmap_git.bitmaps = kh_init_sha1();
315 bitmap_git.ext_index.positions = kh_init_sha1_pos();
316 bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
317
318 if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
319 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
320 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
321 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
322 goto failed;
323
324 if (load_bitmap_entries_v1(&bitmap_git) < 0)
325 goto failed;
326
327 bitmap_git.loaded = 1;
328 return 0;
329
330 failed:
331 munmap(bitmap_git.map, bitmap_git.map_size);
332 bitmap_git.map = NULL;
333 bitmap_git.map_size = 0;
334 return -1;
335 }
336
337 static int open_pack_bitmap(void)
338 {
339 struct packed_git *p;
340 int ret = -1;
341
342 assert(!bitmap_git.map && !bitmap_git.loaded);
343
344 prepare_packed_git();
345 for (p = packed_git; p; p = p->next) {
346 if (open_pack_bitmap_1(p) == 0)
347 ret = 0;
348 }
349
350 return ret;
351 }
352
353 int prepare_bitmap_git(void)
354 {
355 if (bitmap_git.loaded)
356 return 0;
357
358 if (!open_pack_bitmap())
359 return load_pack_bitmap();
360
361 return -1;
362 }
363
364 struct include_data {
365 struct bitmap *base;
366 struct bitmap *seen;
367 };
368
369 static inline int bitmap_position_extended(const unsigned char *sha1)
370 {
371 khash_sha1_pos *positions = bitmap_git.ext_index.positions;
372 khiter_t pos = kh_get_sha1_pos(positions, sha1);
373
374 if (pos < kh_end(positions)) {
375 int bitmap_pos = kh_value(positions, pos);
376 return bitmap_pos + bitmap_git.pack->num_objects;
377 }
378
379 return -1;
380 }
381
382 static inline int bitmap_position_packfile(const unsigned char *sha1)
383 {
384 off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
385 if (!offset)
386 return -1;
387
388 return find_revindex_position(bitmap_git.reverse_index, offset);
389 }
390
391 static int bitmap_position(const unsigned char *sha1)
392 {
393 int pos = bitmap_position_packfile(sha1);
394 return (pos >= 0) ? pos : bitmap_position_extended(sha1);
395 }
396
397 static int ext_index_add_object(struct object *object, const char *name)
398 {
399 struct eindex *eindex = &bitmap_git.ext_index;
400
401 khiter_t hash_pos;
402 int hash_ret;
403 int bitmap_pos;
404
405 hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret);
406 if (hash_ret > 0) {
407 if (eindex->count >= eindex->alloc) {
408 eindex->alloc = (eindex->alloc + 16) * 3 / 2;
409 REALLOC_ARRAY(eindex->objects, eindex->alloc);
410 REALLOC_ARRAY(eindex->hashes, eindex->alloc);
411 }
412
413 bitmap_pos = eindex->count;
414 eindex->objects[eindex->count] = object;
415 eindex->hashes[eindex->count] = pack_name_hash(name);
416 kh_value(eindex->positions, hash_pos) = bitmap_pos;
417 eindex->count++;
418 } else {
419 bitmap_pos = kh_value(eindex->positions, hash_pos);
420 }
421
422 return bitmap_pos + bitmap_git.pack->num_objects;
423 }
424
425 static void show_object(struct object *object, const struct name_path *path,
426 const char *last, void *data)
427 {
428 struct bitmap *base = data;
429 int bitmap_pos;
430
431 bitmap_pos = bitmap_position(object->sha1);
432
433 if (bitmap_pos < 0) {
434 char *name = path_name(path, last);
435 bitmap_pos = ext_index_add_object(object, name);
436 free(name);
437 }
438
439 bitmap_set(base, bitmap_pos);
440 }
441
442 static void show_commit(struct commit *commit, void *data)
443 {
444 }
445
446 static int add_to_include_set(struct include_data *data,
447 const unsigned char *sha1,
448 int bitmap_pos)
449 {
450 khiter_t hash_pos;
451
452 if (data->seen && bitmap_get(data->seen, bitmap_pos))
453 return 0;
454
455 if (bitmap_get(data->base, bitmap_pos))
456 return 0;
457
458 hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
459 if (hash_pos < kh_end(bitmap_git.bitmaps)) {
460 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
461 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
462 return 0;
463 }
464
465 bitmap_set(data->base, bitmap_pos);
466 return 1;
467 }
468
469 static int should_include(struct commit *commit, void *_data)
470 {
471 struct include_data *data = _data;
472 int bitmap_pos;
473
474 bitmap_pos = bitmap_position(commit->object.sha1);
475 if (bitmap_pos < 0)
476 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
477
478 if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
479 struct commit_list *parent = commit->parents;
480
481 while (parent) {
482 parent->item->object.flags |= SEEN;
483 parent = parent->next;
484 }
485
486 return 0;
487 }
488
489 return 1;
490 }
491
492 static struct bitmap *find_objects(struct rev_info *revs,
493 struct object_list *roots,
494 struct bitmap *seen)
495 {
496 struct bitmap *base = NULL;
497 int needs_walk = 0;
498
499 struct object_list *not_mapped = NULL;
500
501 /*
502 * Go through all the roots for the walk. The ones that have bitmaps
503 * on the bitmap index will be `or`ed together to form an initial
504 * global reachability analysis.
505 *
506 * The ones without bitmaps in the index will be stored in the
507 * `not_mapped_list` for further processing.
508 */
509 while (roots) {
510 struct object *object = roots->item;
511 roots = roots->next;
512
513 if (object->type == OBJ_COMMIT) {
514 khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
515
516 if (pos < kh_end(bitmap_git.bitmaps)) {
517 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
518 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
519
520 if (base == NULL)
521 base = ewah_to_bitmap(or_with);
522 else
523 bitmap_or_ewah(base, or_with);
524
525 object->flags |= SEEN;
526 continue;
527 }
528 }
529
530 object_list_insert(object, &not_mapped);
531 }
532
533 /*
534 * Best case scenario: We found bitmaps for all the roots,
535 * so the resulting `or` bitmap has the full reachability analysis
536 */
537 if (not_mapped == NULL)
538 return base;
539
540 roots = not_mapped;
541
542 /*
543 * Let's iterate through all the roots that don't have bitmaps to
544 * check if we can determine them to be reachable from the existing
545 * global bitmap.
546 *
547 * If we cannot find them in the existing global bitmap, we'll need
548 * to push them to an actual walk and run it until we can confirm
549 * they are reachable
550 */
551 while (roots) {
552 struct object *object = roots->item;
553 int pos;
554
555 roots = roots->next;
556 pos = bitmap_position(object->sha1);
557
558 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
559 object->flags &= ~UNINTERESTING;
560 add_pending_object(revs, object, "");
561 needs_walk = 1;
562 } else {
563 object->flags |= SEEN;
564 }
565 }
566
567 if (needs_walk) {
568 struct include_data incdata;
569
570 if (base == NULL)
571 base = bitmap_new();
572
573 incdata.base = base;
574 incdata.seen = seen;
575
576 revs->include_check = should_include;
577 revs->include_check_data = &incdata;
578
579 if (prepare_revision_walk(revs))
580 die("revision walk setup failed");
581
582 traverse_commit_list(revs, show_commit, show_object, base);
583 }
584
585 return base;
586 }
587
588 static void show_extended_objects(struct bitmap *objects,
589 show_reachable_fn show_reach)
590 {
591 struct eindex *eindex = &bitmap_git.ext_index;
592 uint32_t i;
593
594 for (i = 0; i < eindex->count; ++i) {
595 struct object *obj;
596
597 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
598 continue;
599
600 obj = eindex->objects[i];
601 show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
602 }
603 }
604
605 static void show_objects_for_type(
606 struct bitmap *objects,
607 struct ewah_bitmap *type_filter,
608 enum object_type object_type,
609 show_reachable_fn show_reach)
610 {
611 size_t pos = 0, i = 0;
612 uint32_t offset;
613
614 struct ewah_iterator it;
615 eword_t filter;
616
617 if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
618 return;
619
620 ewah_iterator_init(&it, type_filter);
621
622 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
623 eword_t word = objects->words[i] & filter;
624
625 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
626 const unsigned char *sha1;
627 struct revindex_entry *entry;
628 uint32_t hash = 0;
629
630 if ((word >> offset) == 0)
631 break;
632
633 offset += ewah_bit_ctz64(word >> offset);
634
635 if (pos + offset < bitmap_git.reuse_objects)
636 continue;
637
638 entry = &bitmap_git.reverse_index->revindex[pos + offset];
639 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
640
641 if (bitmap_git.hashes)
642 hash = ntohl(bitmap_git.hashes[entry->nr]);
643
644 show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
645 }
646
647 pos += BITS_IN_WORD;
648 i++;
649 }
650 }
651
652 static int in_bitmapped_pack(struct object_list *roots)
653 {
654 while (roots) {
655 struct object *object = roots->item;
656 roots = roots->next;
657
658 if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
659 return 1;
660 }
661
662 return 0;
663 }
664
665 int prepare_bitmap_walk(struct rev_info *revs)
666 {
667 unsigned int i;
668 unsigned int pending_nr = revs->pending.nr;
669 struct object_array_entry *pending_e = revs->pending.objects;
670
671 struct object_list *wants = NULL;
672 struct object_list *haves = NULL;
673
674 struct bitmap *wants_bitmap = NULL;
675 struct bitmap *haves_bitmap = NULL;
676
677 if (!bitmap_git.loaded) {
678 /* try to open a bitmapped pack, but don't parse it yet
679 * because we may not need to use it */
680 if (open_pack_bitmap() < 0)
681 return -1;
682 }
683
684 for (i = 0; i < pending_nr; ++i) {
685 struct object *object = pending_e[i].item;
686
687 if (object->type == OBJ_NONE)
688 parse_object_or_die(object->sha1, NULL);
689
690 while (object->type == OBJ_TAG) {
691 struct tag *tag = (struct tag *) object;
692
693 if (object->flags & UNINTERESTING)
694 object_list_insert(object, &haves);
695 else
696 object_list_insert(object, &wants);
697
698 if (!tag->tagged)
699 die("bad tag");
700 object = parse_object_or_die(tag->tagged->sha1, NULL);
701 }
702
703 if (object->flags & UNINTERESTING)
704 object_list_insert(object, &haves);
705 else
706 object_list_insert(object, &wants);
707 }
708
709 /*
710 * if we have a HAVES list, but none of those haves is contained
711 * in the packfile that has a bitmap, we don't have anything to
712 * optimize here
713 */
714 if (haves && !in_bitmapped_pack(haves))
715 return -1;
716
717 /* if we don't want anything, we're done here */
718 if (!wants)
719 return -1;
720
721 /*
722 * now we're going to use bitmaps, so load the actual bitmap entries
723 * from disk. this is the point of no return; after this the rev_list
724 * becomes invalidated and we must perform the revwalk through bitmaps
725 */
726 if (!bitmap_git.loaded && load_pack_bitmap() < 0)
727 return -1;
728
729 revs->pending.nr = 0;
730 revs->pending.alloc = 0;
731 revs->pending.objects = NULL;
732
733 if (haves) {
734 revs->ignore_missing_links = 1;
735 haves_bitmap = find_objects(revs, haves, NULL);
736 reset_revision_walk();
737 revs->ignore_missing_links = 0;
738
739 if (haves_bitmap == NULL)
740 die("BUG: failed to perform bitmap walk");
741 }
742
743 wants_bitmap = find_objects(revs, wants, haves_bitmap);
744
745 if (!wants_bitmap)
746 die("BUG: failed to perform bitmap walk");
747
748 if (haves_bitmap)
749 bitmap_and_not(wants_bitmap, haves_bitmap);
750
751 bitmap_git.result = wants_bitmap;
752
753 bitmap_free(haves_bitmap);
754 return 0;
755 }
756
757 int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
758 uint32_t *entries,
759 off_t *up_to)
760 {
761 /*
762 * Reuse the packfile content if we need more than
763 * 90% of its objects
764 */
765 static const double REUSE_PERCENT = 0.9;
766
767 struct bitmap *result = bitmap_git.result;
768 uint32_t reuse_threshold;
769 uint32_t i, reuse_objects = 0;
770
771 assert(result);
772
773 for (i = 0; i < result->word_alloc; ++i) {
774 if (result->words[i] != (eword_t)~0) {
775 reuse_objects += ewah_bit_ctz64(~result->words[i]);
776 break;
777 }
778
779 reuse_objects += BITS_IN_WORD;
780 }
781
782 #ifdef GIT_BITMAP_DEBUG
783 {
784 const unsigned char *sha1;
785 struct revindex_entry *entry;
786
787 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
788 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
789
790 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
791 reuse_objects, result->words[i]);
792 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
793 }
794 #endif
795
796 if (!reuse_objects)
797 return -1;
798
799 if (reuse_objects >= bitmap_git.pack->num_objects) {
800 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
801 *up_to = -1; /* reuse the full pack */
802 *packfile = bitmap_git.pack;
803 return 0;
804 }
805
806 reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
807
808 if (reuse_objects < reuse_threshold)
809 return -1;
810
811 bitmap_git.reuse_objects = *entries = reuse_objects;
812 *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
813 *packfile = bitmap_git.pack;
814
815 return 0;
816 }
817
818 void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
819 {
820 assert(bitmap_git.result);
821
822 show_objects_for_type(bitmap_git.result, bitmap_git.commits,
823 OBJ_COMMIT, show_reachable);
824 show_objects_for_type(bitmap_git.result, bitmap_git.trees,
825 OBJ_TREE, show_reachable);
826 show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
827 OBJ_BLOB, show_reachable);
828 show_objects_for_type(bitmap_git.result, bitmap_git.tags,
829 OBJ_TAG, show_reachable);
830
831 show_extended_objects(bitmap_git.result, show_reachable);
832
833 bitmap_free(bitmap_git.result);
834 bitmap_git.result = NULL;
835 }
836
837 static uint32_t count_object_type(struct bitmap *objects,
838 enum object_type type)
839 {
840 struct eindex *eindex = &bitmap_git.ext_index;
841
842 uint32_t i = 0, count = 0;
843 struct ewah_iterator it;
844 eword_t filter;
845
846 switch (type) {
847 case OBJ_COMMIT:
848 ewah_iterator_init(&it, bitmap_git.commits);
849 break;
850
851 case OBJ_TREE:
852 ewah_iterator_init(&it, bitmap_git.trees);
853 break;
854
855 case OBJ_BLOB:
856 ewah_iterator_init(&it, bitmap_git.blobs);
857 break;
858
859 case OBJ_TAG:
860 ewah_iterator_init(&it, bitmap_git.tags);
861 break;
862
863 default:
864 return 0;
865 }
866
867 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
868 eword_t word = objects->words[i++] & filter;
869 count += ewah_bit_popcount64(word);
870 }
871
872 for (i = 0; i < eindex->count; ++i) {
873 if (eindex->objects[i]->type == type &&
874 bitmap_get(objects, bitmap_git.pack->num_objects + i))
875 count++;
876 }
877
878 return count;
879 }
880
881 void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
882 uint32_t *blobs, uint32_t *tags)
883 {
884 assert(bitmap_git.result);
885
886 if (commits)
887 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
888
889 if (trees)
890 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
891
892 if (blobs)
893 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
894
895 if (tags)
896 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
897 }
898
899 struct bitmap_test_data {
900 struct bitmap *base;
901 struct progress *prg;
902 size_t seen;
903 };
904
905 static void test_show_object(struct object *object,
906 const struct name_path *path,
907 const char *last, void *data)
908 {
909 struct bitmap_test_data *tdata = data;
910 int bitmap_pos;
911
912 bitmap_pos = bitmap_position(object->sha1);
913 if (bitmap_pos < 0)
914 die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
915
916 bitmap_set(tdata->base, bitmap_pos);
917 display_progress(tdata->prg, ++tdata->seen);
918 }
919
920 static void test_show_commit(struct commit *commit, void *data)
921 {
922 struct bitmap_test_data *tdata = data;
923 int bitmap_pos;
924
925 bitmap_pos = bitmap_position(commit->object.sha1);
926 if (bitmap_pos < 0)
927 die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
928
929 bitmap_set(tdata->base, bitmap_pos);
930 display_progress(tdata->prg, ++tdata->seen);
931 }
932
933 void test_bitmap_walk(struct rev_info *revs)
934 {
935 struct object *root;
936 struct bitmap *result = NULL;
937 khiter_t pos;
938 size_t result_popcnt;
939 struct bitmap_test_data tdata;
940
941 if (prepare_bitmap_git())
942 die("failed to load bitmap indexes");
943
944 if (revs->pending.nr != 1)
945 die("you must specify exactly one commit to test");
946
947 fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
948 bitmap_git.version, bitmap_git.entry_count);
949
950 root = revs->pending.objects[0].item;
951 pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
952
953 if (pos < kh_end(bitmap_git.bitmaps)) {
954 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
955 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
956
957 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
958 sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
959
960 result = ewah_to_bitmap(bm);
961 }
962
963 if (result == NULL)
964 die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
965
966 revs->tag_objects = 1;
967 revs->tree_objects = 1;
968 revs->blob_objects = 1;
969
970 result_popcnt = bitmap_popcount(result);
971
972 if (prepare_revision_walk(revs))
973 die("revision walk setup failed");
974
975 tdata.base = bitmap_new();
976 tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
977 tdata.seen = 0;
978
979 traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
980
981 stop_progress(&tdata.prg);
982
983 if (bitmap_equals(result, tdata.base))
984 fprintf(stderr, "OK!\n");
985 else
986 fprintf(stderr, "Mismatch!\n");
987
988 bitmap_free(result);
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 }