ewah_bitmap.c: do not assume size_t and eword_t are the same size
[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
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
ae4f07fb
VM
69 /* Name-hash cache (or NULL if not present). */
70 uint32_t *hashes;
71
fff42755
VM
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)");
ae4f07fb
VM
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 }
fff42755
VM
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 int load_bitmap_entries_v1(struct bitmap_index *index)
201{
202 static const size_t MAX_XOR_OFFSET = 160;
203
204 uint32_t i;
205 struct stored_bitmap **recent_bitmaps;
206 struct bitmap_disk_entry *entry;
207
208 recent_bitmaps = xcalloc(MAX_XOR_OFFSET, sizeof(struct stored_bitmap));
209
210 for (i = 0; i < index->entry_count; ++i) {
211 int xor_offset, flags;
212 struct ewah_bitmap *bitmap = NULL;
213 struct stored_bitmap *xor_bitmap = NULL;
214 uint32_t commit_idx_pos;
215 const unsigned char *sha1;
216
217 entry = (struct bitmap_disk_entry *)(index->map + index->map_pos);
218 index->map_pos += sizeof(struct bitmap_disk_entry);
219
220 commit_idx_pos = ntohl(entry->object_pos);
221 sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
222
223 xor_offset = (int)entry->xor_offset;
224 flags = (int)entry->flags;
225
226 bitmap = read_bitmap_1(index);
227 if (!bitmap)
228 return -1;
229
230 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
231 return error("Corrupted bitmap pack index");
232
233 if (xor_offset > 0) {
234 xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
235
236 if (xor_bitmap == NULL)
237 return error("Invalid XOR offset in bitmap pack index");
238 }
239
240 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
241 index, bitmap, sha1, xor_bitmap, flags);
242 }
243
244 return 0;
245}
246
247static int open_pack_bitmap_1(struct packed_git *packfile)
248{
249 int fd;
250 struct stat st;
251 char *idx_name;
252
253 if (open_pack_index(packfile))
254 return -1;
255
256 idx_name = pack_bitmap_filename(packfile);
257 fd = git_open_noatime(idx_name);
258 free(idx_name);
259
260 if (fd < 0)
261 return -1;
262
263 if (fstat(fd, &st)) {
264 close(fd);
265 return -1;
266 }
267
268 if (bitmap_git.pack) {
269 warning("ignoring extra bitmap file: %s", packfile->pack_name);
270 close(fd);
271 return -1;
272 }
273
274 bitmap_git.pack = packfile;
275 bitmap_git.map_size = xsize_t(st.st_size);
276 bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
277 bitmap_git.map_pos = 0;
278 close(fd);
279
280 if (load_bitmap_header(&bitmap_git) < 0) {
281 munmap(bitmap_git.map, bitmap_git.map_size);
282 bitmap_git.map = NULL;
283 bitmap_git.map_size = 0;
284 return -1;
285 }
286
287 return 0;
288}
289
290static int load_pack_bitmap(void)
291{
292 assert(bitmap_git.map && !bitmap_git.loaded);
293
294 bitmap_git.bitmaps = kh_init_sha1();
295 bitmap_git.ext_index.positions = kh_init_sha1_pos();
296 bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
297
298 if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
299 !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
300 !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
301 !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
302 goto failed;
303
304 if (load_bitmap_entries_v1(&bitmap_git) < 0)
305 goto failed;
306
307 bitmap_git.loaded = 1;
308 return 0;
309
310failed:
311 munmap(bitmap_git.map, bitmap_git.map_size);
312 bitmap_git.map = NULL;
313 bitmap_git.map_size = 0;
314 return -1;
315}
316
317char *pack_bitmap_filename(struct packed_git *p)
318{
319 char *idx_name;
320 int len;
321
322 len = strlen(p->pack_name) - strlen(".pack");
323 idx_name = xmalloc(len + strlen(".bitmap") + 1);
324
325 memcpy(idx_name, p->pack_name, len);
326 memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
327
328 return idx_name;
329}
330
331static int open_pack_bitmap(void)
332{
333 struct packed_git *p;
334 int ret = -1;
335
336 assert(!bitmap_git.map && !bitmap_git.loaded);
337
338 prepare_packed_git();
339 for (p = packed_git; p; p = p->next) {
340 if (open_pack_bitmap_1(p) == 0)
341 ret = 0;
342 }
343
344 return ret;
345}
346
347int prepare_bitmap_git(void)
348{
349 if (bitmap_git.loaded)
350 return 0;
351
352 if (!open_pack_bitmap())
353 return load_pack_bitmap();
354
355 return -1;
356}
357
358struct include_data {
359 struct bitmap *base;
360 struct bitmap *seen;
361};
362
363static inline int bitmap_position_extended(const unsigned char *sha1)
364{
365 khash_sha1_pos *positions = bitmap_git.ext_index.positions;
366 khiter_t pos = kh_get_sha1_pos(positions, sha1);
367
368 if (pos < kh_end(positions)) {
369 int bitmap_pos = kh_value(positions, pos);
370 return bitmap_pos + bitmap_git.pack->num_objects;
371 }
372
373 return -1;
374}
375
376static inline int bitmap_position_packfile(const unsigned char *sha1)
377{
378 off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
379 if (!offset)
380 return -1;
381
382 return find_revindex_position(bitmap_git.reverse_index, offset);
383}
384
385static int bitmap_position(const unsigned char *sha1)
386{
387 int pos = bitmap_position_packfile(sha1);
388 return (pos >= 0) ? pos : bitmap_position_extended(sha1);
389}
390
391static int ext_index_add_object(struct object *object, const char *name)
392{
393 struct eindex *eindex = &bitmap_git.ext_index;
394
395 khiter_t hash_pos;
396 int hash_ret;
397 int bitmap_pos;
398
399 hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret);
400 if (hash_ret > 0) {
401 if (eindex->count >= eindex->alloc) {
402 eindex->alloc = (eindex->alloc + 16) * 3 / 2;
403 eindex->objects = xrealloc(eindex->objects,
404 eindex->alloc * sizeof(struct object *));
405 eindex->hashes = xrealloc(eindex->hashes,
406 eindex->alloc * sizeof(uint32_t));
407 }
408
409 bitmap_pos = eindex->count;
410 eindex->objects[eindex->count] = object;
411 eindex->hashes[eindex->count] = pack_name_hash(name);
412 kh_value(eindex->positions, hash_pos) = bitmap_pos;
413 eindex->count++;
414 } else {
415 bitmap_pos = kh_value(eindex->positions, hash_pos);
416 }
417
418 return bitmap_pos + bitmap_git.pack->num_objects;
419}
420
421static void show_object(struct object *object, const struct name_path *path,
422 const char *last, void *data)
423{
424 struct bitmap *base = data;
425 int bitmap_pos;
426
427 bitmap_pos = bitmap_position(object->sha1);
428
429 if (bitmap_pos < 0) {
430 char *name = path_name(path, last);
431 bitmap_pos = ext_index_add_object(object, name);
432 free(name);
433 }
434
435 bitmap_set(base, bitmap_pos);
436}
437
438static void show_commit(struct commit *commit, void *data)
439{
440}
441
442static int add_to_include_set(struct include_data *data,
443 const unsigned char *sha1,
444 int bitmap_pos)
445{
446 khiter_t hash_pos;
447
448 if (data->seen && bitmap_get(data->seen, bitmap_pos))
449 return 0;
450
451 if (bitmap_get(data->base, bitmap_pos))
452 return 0;
453
454 hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
455 if (hash_pos < kh_end(bitmap_git.bitmaps)) {
456 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
457 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
458 return 0;
459 }
460
461 bitmap_set(data->base, bitmap_pos);
462 return 1;
463}
464
465static int should_include(struct commit *commit, void *_data)
466{
467 struct include_data *data = _data;
468 int bitmap_pos;
469
470 bitmap_pos = bitmap_position(commit->object.sha1);
471 if (bitmap_pos < 0)
472 bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
473
474 if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
475 struct commit_list *parent = commit->parents;
476
477 while (parent) {
478 parent->item->object.flags |= SEEN;
479 parent = parent->next;
480 }
481
482 return 0;
483 }
484
485 return 1;
486}
487
488static struct bitmap *find_objects(struct rev_info *revs,
489 struct object_list *roots,
490 struct bitmap *seen)
491{
492 struct bitmap *base = NULL;
493 int needs_walk = 0;
494
495 struct object_list *not_mapped = NULL;
496
497 /*
498 * Go through all the roots for the walk. The ones that have bitmaps
499 * on the bitmap index will be `or`ed together to form an initial
500 * global reachability analysis.
501 *
502 * The ones without bitmaps in the index will be stored in the
503 * `not_mapped_list` for further processing.
504 */
505 while (roots) {
506 struct object *object = roots->item;
507 roots = roots->next;
508
509 if (object->type == OBJ_COMMIT) {
510 khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
511
512 if (pos < kh_end(bitmap_git.bitmaps)) {
513 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
514 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
515
516 if (base == NULL)
517 base = ewah_to_bitmap(or_with);
518 else
519 bitmap_or_ewah(base, or_with);
520
521 object->flags |= SEEN;
522 continue;
523 }
524 }
525
526 object_list_insert(object, &not_mapped);
527 }
528
529 /*
530 * Best case scenario: We found bitmaps for all the roots,
531 * so the resulting `or` bitmap has the full reachability analysis
532 */
533 if (not_mapped == NULL)
534 return base;
535
536 roots = not_mapped;
537
538 /*
539 * Let's iterate through all the roots that don't have bitmaps to
540 * check if we can determine them to be reachable from the existing
541 * global bitmap.
542 *
543 * If we cannot find them in the existing global bitmap, we'll need
544 * to push them to an actual walk and run it until we can confirm
545 * they are reachable
546 */
547 while (roots) {
548 struct object *object = roots->item;
549 int pos;
550
551 roots = roots->next;
552 pos = bitmap_position(object->sha1);
553
554 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
555 object->flags &= ~UNINTERESTING;
556 add_pending_object(revs, object, "");
557 needs_walk = 1;
558 } else {
559 object->flags |= SEEN;
560 }
561 }
562
563 if (needs_walk) {
564 struct include_data incdata;
565
566 if (base == NULL)
567 base = bitmap_new();
568
569 incdata.base = base;
570 incdata.seen = seen;
571
572 revs->include_check = should_include;
573 revs->include_check_data = &incdata;
574
575 if (prepare_revision_walk(revs))
576 die("revision walk setup failed");
577
578 traverse_commit_list(revs, show_commit, show_object, base);
579 }
580
581 return base;
582}
583
584static void show_extended_objects(struct bitmap *objects,
585 show_reachable_fn show_reach)
586{
587 struct eindex *eindex = &bitmap_git.ext_index;
588 uint32_t i;
589
590 for (i = 0; i < eindex->count; ++i) {
591 struct object *obj;
592
593 if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
594 continue;
595
596 obj = eindex->objects[i];
597 show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
598 }
599}
600
601static void show_objects_for_type(
602 struct bitmap *objects,
603 struct ewah_bitmap *type_filter,
604 enum object_type object_type,
605 show_reachable_fn show_reach)
606{
607 size_t pos = 0, i = 0;
608 uint32_t offset;
609
610 struct ewah_iterator it;
611 eword_t filter;
612
613 if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
614 return;
615
616 ewah_iterator_init(&it, type_filter);
617
618 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
619 eword_t word = objects->words[i] & filter;
620
621 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
622 const unsigned char *sha1;
623 struct revindex_entry *entry;
624 uint32_t hash = 0;
625
626 if ((word >> offset) == 0)
627 break;
628
629 offset += ewah_bit_ctz64(word >> offset);
630
631 if (pos + offset < bitmap_git.reuse_objects)
632 continue;
633
634 entry = &bitmap_git.reverse_index->revindex[pos + offset];
635 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
636
ae4f07fb
VM
637 if (bitmap_git.hashes)
638 hash = ntohl(bitmap_git.hashes[entry->nr]);
639
fff42755
VM
640 show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
641 }
642
643 pos += BITS_IN_WORD;
644 i++;
645 }
646}
647
648static int in_bitmapped_pack(struct object_list *roots)
649{
650 while (roots) {
651 struct object *object = roots->item;
652 roots = roots->next;
653
654 if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
655 return 1;
656 }
657
658 return 0;
659}
660
661int prepare_bitmap_walk(struct rev_info *revs)
662{
663 unsigned int i;
664 unsigned int pending_nr = revs->pending.nr;
665 struct object_array_entry *pending_e = revs->pending.objects;
666
667 struct object_list *wants = NULL;
668 struct object_list *haves = NULL;
669
670 struct bitmap *wants_bitmap = NULL;
671 struct bitmap *haves_bitmap = NULL;
672
673 if (!bitmap_git.loaded) {
674 /* try to open a bitmapped pack, but don't parse it yet
675 * because we may not need to use it */
676 if (open_pack_bitmap() < 0)
677 return -1;
678 }
679
680 for (i = 0; i < pending_nr; ++i) {
681 struct object *object = pending_e[i].item;
682
683 if (object->type == OBJ_NONE)
684 parse_object_or_die(object->sha1, NULL);
685
686 while (object->type == OBJ_TAG) {
687 struct tag *tag = (struct tag *) object;
688
689 if (object->flags & UNINTERESTING)
690 object_list_insert(object, &haves);
691 else
692 object_list_insert(object, &wants);
693
694 if (!tag->tagged)
695 die("bad tag");
696 object = parse_object_or_die(tag->tagged->sha1, NULL);
697 }
698
699 if (object->flags & UNINTERESTING)
700 object_list_insert(object, &haves);
701 else
702 object_list_insert(object, &wants);
703 }
704
705 /*
706 * if we have a HAVES list, but none of those haves is contained
707 * in the packfile that has a bitmap, we don't have anything to
708 * optimize here
709 */
710 if (haves && !in_bitmapped_pack(haves))
711 return -1;
712
713 /* if we don't want anything, we're done here */
714 if (!wants)
715 return -1;
716
717 /*
718 * now we're going to use bitmaps, so load the actual bitmap entries
719 * from disk. this is the point of no return; after this the rev_list
720 * becomes invalidated and we must perform the revwalk through bitmaps
721 */
722 if (!bitmap_git.loaded && load_pack_bitmap() < 0)
723 return -1;
724
725 revs->pending.nr = 0;
726 revs->pending.alloc = 0;
727 revs->pending.objects = NULL;
728
729 if (haves) {
2db1a43f 730 revs->ignore_missing_links = 1;
fff42755
VM
731 haves_bitmap = find_objects(revs, haves, NULL);
732 reset_revision_walk();
2db1a43f 733 revs->ignore_missing_links = 0;
fff42755
VM
734
735 if (haves_bitmap == NULL)
736 die("BUG: failed to perform bitmap walk");
737 }
738
739 wants_bitmap = find_objects(revs, wants, haves_bitmap);
740
741 if (!wants_bitmap)
742 die("BUG: failed to perform bitmap walk");
743
744 if (haves_bitmap)
745 bitmap_and_not(wants_bitmap, haves_bitmap);
746
747 bitmap_git.result = wants_bitmap;
748
749 bitmap_free(haves_bitmap);
750 return 0;
751}
752
753int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
754 uint32_t *entries,
755 off_t *up_to)
756{
757 /*
758 * Reuse the packfile content if we need more than
759 * 90% of its objects
760 */
761 static const double REUSE_PERCENT = 0.9;
762
763 struct bitmap *result = bitmap_git.result;
764 uint32_t reuse_threshold;
765 uint32_t i, reuse_objects = 0;
766
767 assert(result);
768
769 for (i = 0; i < result->word_alloc; ++i) {
770 if (result->words[i] != (eword_t)~0) {
771 reuse_objects += ewah_bit_ctz64(~result->words[i]);
772 break;
773 }
774
775 reuse_objects += BITS_IN_WORD;
776 }
777
778#ifdef GIT_BITMAP_DEBUG
779 {
780 const unsigned char *sha1;
781 struct revindex_entry *entry;
782
783 entry = &bitmap_git.reverse_index->revindex[reuse_objects];
784 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
785
786 fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
787 reuse_objects, result->words[i]);
788 fprintf(stderr, " %s\n", sha1_to_hex(sha1));
789 }
790#endif
791
792 if (!reuse_objects)
793 return -1;
794
795 if (reuse_objects >= bitmap_git.pack->num_objects) {
796 bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
797 *up_to = -1; /* reuse the full pack */
798 *packfile = bitmap_git.pack;
799 return 0;
800 }
801
802 reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
803
804 if (reuse_objects < reuse_threshold)
805 return -1;
806
807 bitmap_git.reuse_objects = *entries = reuse_objects;
808 *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
809 *packfile = bitmap_git.pack;
810
811 return 0;
812}
813
814void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
815{
816 assert(bitmap_git.result);
817
818 show_objects_for_type(bitmap_git.result, bitmap_git.commits,
819 OBJ_COMMIT, show_reachable);
820 show_objects_for_type(bitmap_git.result, bitmap_git.trees,
821 OBJ_TREE, show_reachable);
822 show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
823 OBJ_BLOB, show_reachable);
824 show_objects_for_type(bitmap_git.result, bitmap_git.tags,
825 OBJ_TAG, show_reachable);
826
827 show_extended_objects(bitmap_git.result, show_reachable);
828
829 bitmap_free(bitmap_git.result);
830 bitmap_git.result = NULL;
831}
832
833static uint32_t count_object_type(struct bitmap *objects,
834 enum object_type type)
835{
836 struct eindex *eindex = &bitmap_git.ext_index;
837
838 uint32_t i = 0, count = 0;
839 struct ewah_iterator it;
840 eword_t filter;
841
842 switch (type) {
843 case OBJ_COMMIT:
844 ewah_iterator_init(&it, bitmap_git.commits);
845 break;
846
847 case OBJ_TREE:
848 ewah_iterator_init(&it, bitmap_git.trees);
849 break;
850
851 case OBJ_BLOB:
852 ewah_iterator_init(&it, bitmap_git.blobs);
853 break;
854
855 case OBJ_TAG:
856 ewah_iterator_init(&it, bitmap_git.tags);
857 break;
858
859 default:
860 return 0;
861 }
862
863 while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
864 eword_t word = objects->words[i++] & filter;
865 count += ewah_bit_popcount64(word);
866 }
867
868 for (i = 0; i < eindex->count; ++i) {
869 if (eindex->objects[i]->type == type &&
870 bitmap_get(objects, bitmap_git.pack->num_objects + i))
871 count++;
872 }
873
874 return count;
875}
876
877void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
878 uint32_t *blobs, uint32_t *tags)
879{
880 assert(bitmap_git.result);
881
882 if (commits)
883 *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
884
885 if (trees)
886 *trees = count_object_type(bitmap_git.result, OBJ_TREE);
887
888 if (blobs)
889 *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
890
891 if (tags)
892 *tags = count_object_type(bitmap_git.result, OBJ_TAG);
893}
894
895struct bitmap_test_data {
896 struct bitmap *base;
897 struct progress *prg;
898 size_t seen;
899};
900
901static void test_show_object(struct object *object,
902 const struct name_path *path,
903 const char *last, void *data)
904{
905 struct bitmap_test_data *tdata = data;
906 int bitmap_pos;
907
908 bitmap_pos = bitmap_position(object->sha1);
909 if (bitmap_pos < 0)
910 die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
911
912 bitmap_set(tdata->base, bitmap_pos);
913 display_progress(tdata->prg, ++tdata->seen);
914}
915
916static void test_show_commit(struct commit *commit, void *data)
917{
918 struct bitmap_test_data *tdata = data;
919 int bitmap_pos;
920
921 bitmap_pos = bitmap_position(commit->object.sha1);
922 if (bitmap_pos < 0)
923 die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
924
925 bitmap_set(tdata->base, bitmap_pos);
926 display_progress(tdata->prg, ++tdata->seen);
927}
928
929void test_bitmap_walk(struct rev_info *revs)
930{
931 struct object *root;
932 struct bitmap *result = NULL;
933 khiter_t pos;
934 size_t result_popcnt;
935 struct bitmap_test_data tdata;
936
937 if (prepare_bitmap_git())
938 die("failed to load bitmap indexes");
939
940 if (revs->pending.nr != 1)
941 die("you must specify exactly one commit to test");
942
943 fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
944 bitmap_git.version, bitmap_git.entry_count);
945
946 root = revs->pending.objects[0].item;
947 pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
948
949 if (pos < kh_end(bitmap_git.bitmaps)) {
950 struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
951 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
952
953 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
954 sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
955
956 result = ewah_to_bitmap(bm);
957 }
958
959 if (result == NULL)
960 die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
961
962 revs->tag_objects = 1;
963 revs->tree_objects = 1;
964 revs->blob_objects = 1;
965
966 result_popcnt = bitmap_popcount(result);
967
968 if (prepare_revision_walk(revs))
969 die("revision walk setup failed");
970
971 tdata.base = bitmap_new();
972 tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
973 tdata.seen = 0;
974
975 traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
976
977 stop_progress(&tdata.prg);
978
979 if (bitmap_equals(result, tdata.base))
980 fprintf(stderr, "OK!\n");
981 else
982 fprintf(stderr, "Mismatch!\n");
983}
7cc8f971
VM
984
985static int rebuild_bitmap(uint32_t *reposition,
986 struct ewah_bitmap *source,
987 struct bitmap *dest)
988{
989 uint32_t pos = 0;
990 struct ewah_iterator it;
991 eword_t word;
992
993 ewah_iterator_init(&it, source);
994
995 while (ewah_iterator_next(&word, &it)) {
996 uint32_t offset, bit_pos;
997
998 for (offset = 0; offset < BITS_IN_WORD; ++offset) {
999 if ((word >> offset) == 0)
1000 break;
1001
1002 offset += ewah_bit_ctz64(word >> offset);
1003
1004 bit_pos = reposition[pos + offset];
1005 if (bit_pos > 0)
1006 bitmap_set(dest, bit_pos - 1);
1007 else /* can't reuse, we don't have the object */
1008 return -1;
1009 }
1010
1011 pos += BITS_IN_WORD;
1012 }
1013 return 0;
1014}
1015
1016int rebuild_existing_bitmaps(struct packing_data *mapping,
1017 khash_sha1 *reused_bitmaps,
1018 int show_progress)
1019{
1020 uint32_t i, num_objects;
1021 uint32_t *reposition;
1022 struct bitmap *rebuild;
1023 struct stored_bitmap *stored;
1024 struct progress *progress = NULL;
1025
1026 khiter_t hash_pos;
1027 int hash_ret;
1028
1029 if (prepare_bitmap_git() < 0)
1030 return -1;
1031
1032 num_objects = bitmap_git.pack->num_objects;
1033 reposition = xcalloc(num_objects, sizeof(uint32_t));
1034
1035 for (i = 0; i < num_objects; ++i) {
1036 const unsigned char *sha1;
1037 struct revindex_entry *entry;
1038 struct object_entry *oe;
1039
1040 entry = &bitmap_git.reverse_index->revindex[i];
1041 sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1042 oe = packlist_find(mapping, sha1, NULL);
1043
1044 if (oe)
1045 reposition[i] = oe->in_pack_pos + 1;
1046 }
1047
1048 rebuild = bitmap_new();
1049 i = 0;
1050
1051 if (show_progress)
1052 progress = start_progress("Reusing bitmaps", 0);
1053
1054 kh_foreach_value(bitmap_git.bitmaps, stored, {
1055 if (stored->flags & BITMAP_FLAG_REUSE) {
1056 if (!rebuild_bitmap(reposition,
1057 lookup_stored_bitmap(stored),
1058 rebuild)) {
1059 hash_pos = kh_put_sha1(reused_bitmaps,
1060 stored->sha1,
1061 &hash_ret);
1062 kh_value(reused_bitmaps, hash_pos) =
1063 bitmap_to_ewah(rebuild);
1064 }
1065 bitmap_reset(rebuild);
1066 display_progress(progress, ++i);
1067 }
1068 });
1069
1070 stop_progress(&progress);
1071
1072 free(reposition);
1073 bitmap_free(rebuild);
1074 return 0;
1075}