pack-revindex: store entries directly in packed_git
[git/git.git] / pack-bitmap.c
CommitLineData
fff42755
VM
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
fff42755
VM
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 comits */
61 khash_sha1 *bitmaps;
62
63 /* Number of bitmapped commits */
64 uint32_t entry_count;
65
ae4f07fb
VM
66 /* Name-hash cache (or NULL if not present). */
67 uint32_t *hashes;
68
fff42755
VM
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
93static 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 */
116static 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
134static 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)");
ae4f07fb
VM
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 }
fff42755
VM
160 }
161
162 index->entry_count = ntohl(header->entry_count);
163 index->map_pos += sizeof(*header);
164 return 0;
165}
166
167static 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
197static int load_bitmap_entries_v1(struct bitmap_index *index)
198{
199 static const size_t MAX_XOR_OFFSET = 160;
200
201 uint32_t i;
202 struct stored_bitmap **recent_bitmaps;
203 struct bitmap_disk_entry *entry;
204
205 recent_bitmaps = xcalloc(MAX_XOR_OFFSET, sizeof(struct stored_bitmap));
206
207 for (i = 0; i < index->entry_count; ++i) {
208 int xor_offset, flags;
209 struct ewah_bitmap *bitmap = NULL;
210 struct stored_bitmap *xor_bitmap = NULL;
211 uint32_t commit_idx_pos;
212 const unsigned char *sha1;
213
214 entry = (struct bitmap_disk_entry *)(index->map + index->map_pos);
215 index->map_pos += sizeof(struct bitmap_disk_entry);
216
217 commit_idx_pos = ntohl(entry->object_pos);
218 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
219
220 xor_offset = (int)entry->xor_offset;
221 flags = (int)entry->flags;
222
223 bitmap = read_bitmap_1(index);
224 if (!bitmap)
225 return -1;
226
227 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
228 return error("Corrupted bitmap pack index");
229
230 if (xor_offset > 0) {
231 xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
232
233 if (xor_bitmap == NULL)
234 return error("Invalid XOR offset in bitmap pack index");
235 }
236
237 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
238 index, bitmap, sha1, xor_bitmap, flags);
239 }
240
241 return 0;
242}
243
244static int open_pack_bitmap_1(struct packed_git *packfile)
245{
246 int fd;
247 struct stat st;
248 char *idx_name;
249
250 if (open_pack_index(packfile))
251 return -1;
252
253 idx_name = pack_bitmap_filename(packfile);
254 fd = git_open_noatime(idx_name);
255 free(idx_name);
256
257 if (fd < 0)
258 return -1;
259
260 if (fstat(fd, &st)) {
261 close(fd);
262 return -1;
263 }
264
265 if (bitmap_git.pack) {
266 warning("ignoring extra bitmap file: %s", packfile->pack_name);
267 close(fd);
268 return -1;
269 }
270
271 bitmap_git.pack = packfile;
272 bitmap_git.map_size = xsize_t(st.st_size);
273 bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
274 bitmap_git.map_pos = 0;
275 close(fd);
276
277 if (load_bitmap_header(&bitmap_git) < 0) {
278 munmap(bitmap_git.map, bitmap_git.map_size);
279 bitmap_git.map = NULL;
280 bitmap_git.map_size = 0;
281 return -1;
282 }
283
284 return 0;
285}
286
287static int load_pack_bitmap(void)
288{
289 assert(bitmap_git.map && !bitmap_git.loaded);
290
291 bitmap_git.bitmaps = kh_init_sha1();
292 bitmap_git.ext_index.positions = kh_init_sha1_pos();
9d98bbf5 293 load_pack_revindex(bitmap_git.pack);
fff42755
VM
294
295 if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
296 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
297 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
298 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
299 goto failed;
300
301 if (load_bitmap_entries_v1(&bitmap_git) < 0)
302 goto failed;
303
304 bitmap_git.loaded = 1;
305 return 0;
306
307failed:
308 munmap(bitmap_git.map, bitmap_git.map_size);
309 bitmap_git.map = NULL;
310 bitmap_git.map_size = 0;
311 return -1;
312}
313
314char *pack_bitmap_filename(struct packed_git *p)
315{
316 char *idx_name;
317 int len;
318
319 len = strlen(p->pack_name) - strlen(".pack");
320 idx_name = xmalloc(len + strlen(".bitmap") + 1);
321
322 memcpy(idx_name, p->pack_name, len);
323 memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
324
325 return idx_name;
326}
327
328static int open_pack_bitmap(void)
329{
330 struct packed_git *p;
331 int ret = -1;
332
333 assert(!bitmap_git.map && !bitmap_git.loaded);
334
335 prepare_packed_git();
336 for (p = packed_git; p; p = p->next) {
337 if (open_pack_bitmap_1(p) == 0)
338 ret = 0;
339 }
340
341 return ret;
342}
343
344int prepare_bitmap_git(void)
345{
346 if (bitmap_git.loaded)
347 return 0;
348
349 if (!open_pack_bitmap())
350 return load_pack_bitmap();
351
352 return -1;
353}
354
355struct include_data {
356 struct bitmap *base;
357 struct bitmap *seen;
358};
359
360static inline int bitmap_position_extended(const unsigned char *sha1)
361{
362 khash_sha1_pos *positions = bitmap_git.ext_index.positions;
363 khiter_t pos = kh_get_sha1_pos(positions, sha1);
364
365 if (pos < kh_end(positions)) {
366 int bitmap_pos = kh_value(positions, pos);
367 return bitmap_pos + bitmap_git.pack->num_objects;
368 }
369
370 return -1;
371}
372
373static inline int bitmap_position_packfile(const unsigned char *sha1)
374{
375 off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
376 if (!offset)
377 return -1;
378
9d98bbf5 379 return find_revindex_position(bitmap_git.pack, offset);
fff42755
VM
380}
381
382static int bitmap_position(const unsigned char *sha1)
383{
384 int pos = bitmap_position_packfile(sha1);
385 return (pos >= 0) ? pos : bitmap_position_extended(sha1);
386}
387
388static int ext_index_add_object(struct object *object, const char *name)
389{
390 struct eindex *eindex = &bitmap_git.ext_index;
391
392 khiter_t hash_pos;
393 int hash_ret;
394 int bitmap_pos;
395
396 hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret);
397 if (hash_ret > 0) {
398 if (eindex->count >= eindex->alloc) {
399 eindex->alloc = (eindex->alloc + 16) * 3 / 2;
400 eindex->objects = xrealloc(eindex->objects,
401 eindex->alloc * sizeof(struct object *));
402 eindex->hashes = xrealloc(eindex->hashes,
403 eindex->alloc * sizeof(uint32_t));
404 }
405
406 bitmap_pos = eindex->count;
407 eindex->objects[eindex->count] = object;
408 eindex->hashes[eindex->count] = pack_name_hash(name);
409 kh_value(eindex->positions, hash_pos) = bitmap_pos;
410 eindex->count++;
411 } else {
412 bitmap_pos = kh_value(eindex->positions, hash_pos);
413 }
414
415 return bitmap_pos + bitmap_git.pack->num_objects;
416}
417
418static void show_object(struct object *object, const struct name_path *path,
419 const char *last, void *data)
420{
421 struct bitmap *base = data;
422 int bitmap_pos;
423
424 bitmap_pos = bitmap_position(object->sha1);
425
426 if (bitmap_pos < 0) {
427 char *name = path_name(path, last);
428 bitmap_pos = ext_index_add_object(object, name);
429 free(name);
430 }
431
432 bitmap_set(base, bitmap_pos);
433}
434
435static void show_commit(struct commit *commit, void *data)
436{
437}
438
439static int add_to_include_set(struct include_data *data,
440 const unsigned char *sha1,
441 int bitmap_pos)
442{
443 khiter_t hash_pos;
444
445 if (data->seen && bitmap_get(data->seen, bitmap_pos))
446 return 0;
447
448 if (bitmap_get(data->base, bitmap_pos))
449 return 0;
450
451 hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
452 if (hash_pos < kh_end(bitmap_git.bitmaps)) {
453 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
454 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
455 return 0;
456 }
457
458 bitmap_set(data->base, bitmap_pos);
459 return 1;
460}
461
462static int should_include(struct commit *commit, void *_data)
463{
464 struct include_data *data = _data;
465 int bitmap_pos;
466
467 bitmap_pos = bitmap_position(commit->object.sha1);
468 if (bitmap_pos < 0)
469 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
470
471 if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
472 struct commit_list *parent = commit->parents;
473
474 while (parent) {
475 parent->item->object.flags |= SEEN;
476 parent = parent->next;
477 }
478
479 return 0;
480 }
481
482 return 1;
483}
484
485static struct bitmap *find_objects(struct rev_info *revs,
486 struct object_list *roots,
487 struct bitmap *seen)
488{
489 struct bitmap *base = NULL;
490 int needs_walk = 0;
491
492 struct object_list *not_mapped = NULL;
493
494 /*
495 * Go through all the roots for the walk. The ones that have bitmaps
496 * on the bitmap index will be `or`ed together to form an initial
497 * global reachability analysis.
498 *
499 * The ones without bitmaps in the index will be stored in the
500 * `not_mapped_list` for further processing.
501 */
502 while (roots) {
503 struct object *object = roots->item;
504 roots = roots->next;
505
506 if (object->type == OBJ_COMMIT) {
507 khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
508
509 if (pos < kh_end(bitmap_git.bitmaps)) {
510 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
511 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
512
513 if (base == NULL)
514 base = ewah_to_bitmap(or_with);
515 else
516 bitmap_or_ewah(base, or_with);
517
518 object->flags |= SEEN;
519 continue;
520 }
521 }
522
523 object_list_insert(object, &not_mapped);
524 }
525
526 /*
527 * Best case scenario: We found bitmaps for all the roots,
528 * so the resulting `or` bitmap has the full reachability analysis
529 */
530 if (not_mapped == NULL)
531 return base;
532
533 roots = not_mapped;
534
535 /*
536 * Let's iterate through all the roots that don't have bitmaps to
537 * check if we can determine them to be reachable from the existing
538 * global bitmap.
539 *
540 * If we cannot find them in the existing global bitmap, we'll need
541 * to push them to an actual walk and run it until we can confirm
542 * they are reachable
543 */
544 while (roots) {
545 struct object *object = roots->item;
546 int pos;
547
548 roots = roots->next;
549 pos = bitmap_position(object->sha1);
550
551 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
552 object->flags &= ~UNINTERESTING;
553 add_pending_object(revs, object, "");
554 needs_walk = 1;
555 } else {
556 object->flags |= SEEN;
557 }
558 }
559
560 if (needs_walk) {
561 struct include_data incdata;
562
563 if (base == NULL)
564 base = bitmap_new();
565
566 incdata.base = base;
567 incdata.seen = seen;
568
569 revs->include_check = should_include;
570 revs->include_check_data = &incdata;
571
572 if (prepare_revision_walk(revs))
573 die("revision walk setup failed");
574
575 traverse_commit_list(revs, show_commit, show_object, base);
576 }
577
578 return base;
579}
580
581static void show_extended_objects(struct bitmap *objects,
582 show_reachable_fn show_reach)
583{
584 struct eindex *eindex = &bitmap_git.ext_index;
585 uint32_t i;
586
587 for (i = 0; i < eindex->count; ++i) {
588 struct object *obj;
589
590 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
591 continue;
592
593 obj = eindex->objects[i];
594 show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
595 }
596}
597
598static void show_objects_for_type(
599 struct bitmap *objects,
600 struct ewah_bitmap *type_filter,
601 enum object_type object_type,
602 show_reachable_fn show_reach)
603{
604 size_t pos = 0, i = 0;
605 uint32_t offset;
606
607 struct ewah_iterator it;
608 eword_t filter;
609
610 if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
611 return;
612
613 ewah_iterator_init(&it, type_filter);
614
615 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
616 eword_t word = objects->words[i] & filter;
617
618 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
619 const unsigned char *sha1;
620 struct revindex_entry *entry;
621 uint32_t hash = 0;
622
623 if ((word >> offset) == 0)
624 break;
625
626 offset += ewah_bit_ctz64(word >> offset);
627
628 if (pos + offset < bitmap_git.reuse_objects)
629 continue;
630
9d98bbf5 631 entry = &bitmap_git.pack->revindex[pos + offset];
fff42755
VM
632 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
633
ae4f07fb
VM
634 if (bitmap_git.hashes)
635 hash = ntohl(bitmap_git.hashes[entry->nr]);
636
fff42755
VM
637 show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
638 }
639
640 pos += BITS_IN_WORD;
641 i++;
642 }
643}
644
645static int in_bitmapped_pack(struct object_list *roots)
646{
647 while (roots) {
648 struct object *object = roots->item;
649 roots = roots->next;
650
651 if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
652 return 1;
653 }
654
655 return 0;
656}
657
658int prepare_bitmap_walk(struct rev_info *revs)
659{
660 unsigned int i;
661 unsigned int pending_nr = revs->pending.nr;
662 struct object_array_entry *pending_e = revs->pending.objects;
663
664 struct object_list *wants = NULL;
665 struct object_list *haves = NULL;
666
667 struct bitmap *wants_bitmap = NULL;
668 struct bitmap *haves_bitmap = NULL;
669
670 if (!bitmap_git.loaded) {
671 /* try to open a bitmapped pack, but don't parse it yet
672 * because we may not need to use it */
673 if (open_pack_bitmap() < 0)
674 return -1;
675 }
676
677 for (i = 0; i < pending_nr; ++i) {
678 struct object *object = pending_e[i].item;
679
680 if (object->type == OBJ_NONE)
681 parse_object_or_die(object->sha1, NULL);
682
683 while (object->type == OBJ_TAG) {
684 struct tag *tag = (struct tag *) object;
685
686 if (object->flags & UNINTERESTING)
687 object_list_insert(object, &haves);
688 else
689 object_list_insert(object, &wants);
690
691 if (!tag->tagged)
692 die("bad tag");
693 object = parse_object_or_die(tag->tagged->sha1, NULL);
694 }
695
696 if (object->flags & UNINTERESTING)
697 object_list_insert(object, &haves);
698 else
699 object_list_insert(object, &wants);
700 }
701
702 /*
703 * if we have a HAVES list, but none of those haves is contained
704 * in the packfile that has a bitmap, we don't have anything to
705 * optimize here
706 */
707 if (haves && !in_bitmapped_pack(haves))
708 return -1;
709
710 /* if we don't want anything, we're done here */
711 if (!wants)
712 return -1;
713
714 /*
715 * now we're going to use bitmaps, so load the actual bitmap entries
716 * from disk. this is the point of no return; after this the rev_list
717 * becomes invalidated and we must perform the revwalk through bitmaps
718 */
719 if (!bitmap_git.loaded && load_pack_bitmap() < 0)
720 return -1;
721
722 revs->pending.nr = 0;
723 revs->pending.alloc = 0;
724 revs->pending.objects = NULL;
725
726 if (haves) {
2db1a43f 727 revs->ignore_missing_links = 1;
fff42755
VM
728 haves_bitmap = find_objects(revs, haves, NULL);
729 reset_revision_walk();
2db1a43f 730 revs->ignore_missing_links = 0;
fff42755
VM
731
732 if (haves_bitmap == NULL)
733 die("BUG: failed to perform bitmap walk");
734 }
735
736 wants_bitmap = find_objects(revs, wants, haves_bitmap);
737
738 if (!wants_bitmap)
739 die("BUG: failed to perform bitmap walk");
740
741 if (haves_bitmap)
742 bitmap_and_not(wants_bitmap, haves_bitmap);
743
744 bitmap_git.result = wants_bitmap;
745
746 bitmap_free(haves_bitmap);
747 return 0;
748}
749
750int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
751 uint32_t *entries,
752 off_t *up_to)
753{
754 /*
755 * Reuse the packfile content if we need more than
756 * 90% of its objects
757 */
758 static const double REUSE_PERCENT = 0.9;
759
760 struct bitmap *result = bitmap_git.result;
761 uint32_t reuse_threshold;
762 uint32_t i, reuse_objects = 0;
763
764 assert(result);
765
766 for (i = 0; i < result->word_alloc; ++i) {
767 if (result->words[i] != (eword_t)~0) {
768 reuse_objects += ewah_bit_ctz64(~result->words[i]);
769 break;
770 }
771
772 reuse_objects += BITS_IN_WORD;
773 }
774
775#ifdef GIT_BITMAP_DEBUG
776 {
777 const unsigned char *sha1;
778 struct revindex_entry *entry;
779
780 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
781 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
782
783 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
784 reuse_objects, result->words[i]);
785 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
786 }
787#endif
788
789 if (!reuse_objects)
790 return -1;
791
792 if (reuse_objects >= bitmap_git.pack->num_objects) {
793 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
794 *up_to = -1; /* reuse the full pack */
795 *packfile = bitmap_git.pack;
796 return 0;
797 }
798
799 reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
800
801 if (reuse_objects < reuse_threshold)
802 return -1;
803
804 bitmap_git.reuse_objects = *entries = reuse_objects;
9d98bbf5 805 *up_to = bitmap_git.pack->revindex[reuse_objects].offset;
fff42755
VM
806 *packfile = bitmap_git.pack;
807
808 return 0;
809}
810
811void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
812{
813 assert(bitmap_git.result);
814
815 show_objects_for_type(bitmap_git.result, bitmap_git.commits,
816 OBJ_COMMIT, show_reachable);
817 show_objects_for_type(bitmap_git.result, bitmap_git.trees,
818 OBJ_TREE, show_reachable);
819 show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
820 OBJ_BLOB, show_reachable);
821 show_objects_for_type(bitmap_git.result, bitmap_git.tags,
822 OBJ_TAG, show_reachable);
823
824 show_extended_objects(bitmap_git.result, show_reachable);
825
826 bitmap_free(bitmap_git.result);
827 bitmap_git.result = NULL;
828}
829
830static uint32_t count_object_type(struct bitmap *objects,
831 enum object_type type)
832{
833 struct eindex *eindex = &bitmap_git.ext_index;
834
835 uint32_t i = 0, count = 0;
836 struct ewah_iterator it;
837 eword_t filter;
838
839 switch (type) {
840 case OBJ_COMMIT:
841 ewah_iterator_init(&it, bitmap_git.commits);
842 break;
843
844 case OBJ_TREE:
845 ewah_iterator_init(&it, bitmap_git.trees);
846 break;
847
848 case OBJ_BLOB:
849 ewah_iterator_init(&it, bitmap_git.blobs);
850 break;
851
852 case OBJ_TAG:
853 ewah_iterator_init(&it, bitmap_git.tags);
854 break;
855
856 default:
857 return 0;
858 }
859
860 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
861 eword_t word = objects->words[i++] & filter;
862 count += ewah_bit_popcount64(word);
863 }
864
865 for (i = 0; i < eindex->count; ++i) {
866 if (eindex->objects[i]->type == type &&
867 bitmap_get(objects, bitmap_git.pack->num_objects + i))
868 count++;
869 }
870
871 return count;
872}
873
874void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
875 uint32_t *blobs, uint32_t *tags)
876{
877 assert(bitmap_git.result);
878
879 if (commits)
880 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
881
882 if (trees)
883 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
884
885 if (blobs)
886 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
887
888 if (tags)
889 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
890}
891
892struct bitmap_test_data {
893 struct bitmap *base;
894 struct progress *prg;
895 size_t seen;
896};
897
898static void test_show_object(struct object *object,
899 const struct name_path *path,
900 const char *last, void *data)
901{
902 struct bitmap_test_data *tdata = data;
903 int bitmap_pos;
904
905 bitmap_pos = bitmap_position(object->sha1);
906 if (bitmap_pos < 0)
907 die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
908
909 bitmap_set(tdata->base, bitmap_pos);
910 display_progress(tdata->prg, ++tdata->seen);
911}
912
913static void test_show_commit(struct commit *commit, void *data)
914{
915 struct bitmap_test_data *tdata = data;
916 int bitmap_pos;
917
918 bitmap_pos = bitmap_position(commit->object.sha1);
919 if (bitmap_pos < 0)
920 die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
921
922 bitmap_set(tdata->base, bitmap_pos);
923 display_progress(tdata->prg, ++tdata->seen);
924}
925
926void test_bitmap_walk(struct rev_info *revs)
927{
928 struct object *root;
929 struct bitmap *result = NULL;
930 khiter_t pos;
931 size_t result_popcnt;
932 struct bitmap_test_data tdata;
933
934 if (prepare_bitmap_git())
935 die("failed to load bitmap indexes");
936
937 if (revs->pending.nr != 1)
938 die("you must specify exactly one commit to test");
939
940 fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
941 bitmap_git.version, bitmap_git.entry_count);
942
943 root = revs->pending.objects[0].item;
944 pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
945
946 if (pos < kh_end(bitmap_git.bitmaps)) {
947 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
948 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
949
950 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
951 sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
952
953 result = ewah_to_bitmap(bm);
954 }
955
956 if (result == NULL)
957 die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
958
959 revs->tag_objects = 1;
960 revs->tree_objects = 1;
961 revs->blob_objects = 1;
962
963 result_popcnt = bitmap_popcount(result);
964
965 if (prepare_revision_walk(revs))
966 die("revision walk setup failed");
967
968 tdata.base = bitmap_new();
969 tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
970 tdata.seen = 0;
971
972 traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
973
974 stop_progress(&tdata.prg);
975
976 if (bitmap_equals(result, tdata.base))
977 fprintf(stderr, "OK!\n");
978 else
979 fprintf(stderr, "Mismatch!\n");
980}
7cc8f971
VM
981
982static int rebuild_bitmap(uint32_t *reposition,
983 struct ewah_bitmap *source,
984 struct bitmap *dest)
985{
986 uint32_t pos = 0;
987 struct ewah_iterator it;
988 eword_t word;
989
990 ewah_iterator_init(&it, source);
991
992 while (ewah_iterator_next(&word, &it)) {
993 uint32_t offset, bit_pos;
994
995 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
996 if ((word >> offset) == 0)
997 break;
998
999 offset += ewah_bit_ctz64(word >> offset);
1000
1001 bit_pos = reposition[pos + offset];
1002 if (bit_pos > 0)
1003 bitmap_set(dest, bit_pos - 1);
1004 else /* can't reuse, we don't have the object */
1005 return -1;
1006 }
1007
1008 pos += BITS_IN_WORD;
1009 }
1010 return 0;
1011}
1012
1013int rebuild_existing_bitmaps(struct packing_data *mapping,
1014 khash_sha1 *reused_bitmaps,
1015 int show_progress)
1016{
1017 uint32_t i, num_objects;
1018 uint32_t *reposition;
1019 struct bitmap *rebuild;
1020 struct stored_bitmap *stored;
1021 struct progress *progress = NULL;
1022
1023 khiter_t hash_pos;
1024 int hash_ret;
1025
1026 if (prepare_bitmap_git() < 0)
1027 return -1;
1028
1029 num_objects = bitmap_git.pack->num_objects;
1030 reposition = xcalloc(num_objects, sizeof(uint32_t));
1031
1032 for (i = 0; i < num_objects; ++i) {
1033 const unsigned char *sha1;
1034 struct revindex_entry *entry;
1035 struct object_entry *oe;
1036
9d98bbf5 1037 entry = &bitmap_git.pack->revindex[i];
7cc8f971
VM
1038 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1039 oe = packlist_find(mapping, sha1, NULL);
1040
1041 if (oe)
1042 reposition[i] = oe->in_pack_pos + 1;
1043 }
1044
1045 rebuild = bitmap_new();
1046 i = 0;
1047
1048 if (show_progress)
1049 progress = start_progress("Reusing bitmaps", 0);
1050
1051 kh_foreach_value(bitmap_git.bitmaps, stored, {
1052 if (stored->flags & BITMAP_FLAG_REUSE) {
1053 if (!rebuild_bitmap(reposition,
1054 lookup_stored_bitmap(stored),
1055 rebuild)) {
1056 hash_pos = kh_put_sha1(reused_bitmaps,
1057 stored->sha1,
1058 &hash_ret);
1059 kh_value(reused_bitmaps, hash_pos) =
1060 bitmap_to_ewah(rebuild);
1061 }
1062 bitmap_reset(rebuild);
1063 display_progress(progress, ++i);
1064 }
1065 });
1066
1067 stop_progress(&progress);
1068
1069 free(reposition);
1070 bitmap_free(rebuild);
1071 return 0;
1072}