t0000: consistently use single quotes for outer tests
[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 #include "packfile.h"
13 #include "repository.h"
14 #include "object-store.h"
15 #include "list-objects-filter-options.h"
16
17 /*
18  * An entry on the bitmap index, representing the bitmap for a given
19  * commit.
20  */
21 struct stored_bitmap {
22         struct object_id oid;
23         struct ewah_bitmap *root;
24         struct stored_bitmap *xor;
25         int flags;
26 };
27
28 /*
29  * The active bitmap index for a repository. By design, repositories only have
30  * a single bitmap index available (the index for the biggest packfile in
31  * the repository), since bitmap indexes need full closure.
32  *
33  * If there is more than one bitmap index available (e.g. because of alternates),
34  * the active bitmap index is the largest one.
35  */
36 struct bitmap_index {
37         /* Packfile to which this bitmap index belongs to */
38         struct packed_git *pack;
39
40         /*
41          * Mark the first `reuse_objects` in the packfile as reused:
42          * they will be sent as-is without using them for repacking
43          * calculations
44          */
45         uint32_t reuse_objects;
46
47         /* mmapped buffer of the whole bitmap index */
48         unsigned char *map;
49         size_t map_size; /* size of the mmaped buffer */
50         size_t map_pos; /* current position when loading the index */
51
52         /*
53          * Type indexes.
54          *
55          * Each bitmap marks which objects in the packfile  are of the given
56          * type. This provides type information when yielding the objects from
57          * the packfile during a walk, which allows for better delta bases.
58          */
59         struct ewah_bitmap *commits;
60         struct ewah_bitmap *trees;
61         struct ewah_bitmap *blobs;
62         struct ewah_bitmap *tags;
63
64         /* Map from object ID -> `stored_bitmap` for all the bitmapped commits */
65         kh_oid_map_t *bitmaps;
66
67         /* Number of bitmapped commits */
68         uint32_t entry_count;
69
70         /* If not NULL, this is a name-hash cache pointing into map. */
71         uint32_t *hashes;
72
73         /*
74          * Extended index.
75          *
76          * When trying to perform bitmap operations with objects that are not
77          * packed in `pack`, these objects are added to this "fake index" and
78          * are assumed to appear at the end of the packfile for all operations
79          */
80         struct eindex {
81                 struct object **objects;
82                 uint32_t *hashes;
83                 uint32_t count, alloc;
84                 kh_oid_pos_t *positions;
85         } ext_index;
86
87         /* Bitmap result of the last performed walk */
88         struct bitmap *result;
89
90         /* "have" bitmap from the last performed walk */
91         struct bitmap *haves;
92
93         /* Version of the bitmap index */
94         unsigned int version;
95 };
96
97 static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st)
98 {
99         struct ewah_bitmap *parent;
100         struct ewah_bitmap *composed;
101
102         if (st->xor == NULL)
103                 return st->root;
104
105         composed = ewah_pool_new();
106         parent = lookup_stored_bitmap(st->xor);
107         ewah_xor(st->root, parent, composed);
108
109         ewah_pool_free(st->root);
110         st->root = composed;
111         st->xor = NULL;
112
113         return composed;
114 }
115
116 /*
117  * Read a bitmap from the current read position on the mmaped
118  * index, and increase the read position accordingly
119  */
120 static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index)
121 {
122         struct ewah_bitmap *b = ewah_pool_new();
123
124         ssize_t bitmap_size = ewah_read_mmap(b,
125                 index->map + index->map_pos,
126                 index->map_size - index->map_pos);
127
128         if (bitmap_size < 0) {
129                 error("Failed to load bitmap index (corrupted?)");
130                 ewah_pool_free(b);
131                 return NULL;
132         }
133
134         index->map_pos += bitmap_size;
135         return b;
136 }
137
138 static int load_bitmap_header(struct bitmap_index *index)
139 {
140         struct bitmap_disk_header *header = (void *)index->map;
141         size_t header_size = sizeof(*header) - GIT_MAX_RAWSZ + the_hash_algo->rawsz;
142
143         if (index->map_size < header_size + the_hash_algo->rawsz)
144                 return error("Corrupted bitmap index (too small)");
145
146         if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0)
147                 return error("Corrupted bitmap index file (wrong header)");
148
149         index->version = ntohs(header->version);
150         if (index->version != 1)
151                 return error("Unsupported version for bitmap index file (%d)", index->version);
152
153         /* Parse known bitmap format options */
154         {
155                 uint32_t flags = ntohs(header->options);
156                 size_t cache_size = st_mult(index->pack->num_objects, sizeof(uint32_t));
157                 unsigned char *index_end = index->map + index->map_size - the_hash_algo->rawsz;
158
159                 if ((flags & BITMAP_OPT_FULL_DAG) == 0)
160                         return error("Unsupported options for bitmap index file "
161                                 "(Git requires BITMAP_OPT_FULL_DAG)");
162
163                 if (flags & BITMAP_OPT_HASH_CACHE) {
164                         if (cache_size > index_end - index->map - header_size)
165                                 return error("corrupted bitmap index file (too short to fit hash cache)");
166                         index->hashes = (void *)(index_end - cache_size);
167                         index_end -= cache_size;
168                 }
169         }
170
171         index->entry_count = ntohl(header->entry_count);
172         index->map_pos += header_size;
173         return 0;
174 }
175
176 static struct stored_bitmap *store_bitmap(struct bitmap_index *index,
177                                           struct ewah_bitmap *root,
178                                           const struct object_id *oid,
179                                           struct stored_bitmap *xor_with,
180                                           int flags)
181 {
182         struct stored_bitmap *stored;
183         khiter_t hash_pos;
184         int ret;
185
186         stored = xmalloc(sizeof(struct stored_bitmap));
187         stored->root = root;
188         stored->xor = xor_with;
189         stored->flags = flags;
190         oidcpy(&stored->oid, oid);
191
192         hash_pos = kh_put_oid_map(index->bitmaps, stored->oid, &ret);
193
194         /* a 0 return code means the insertion succeeded with no changes,
195          * because the SHA1 already existed on the map. this is bad, there
196          * shouldn't be duplicated commits in the index */
197         if (ret == 0) {
198                 error("Duplicate entry in bitmap index: %s", oid_to_hex(oid));
199                 return NULL;
200         }
201
202         kh_value(index->bitmaps, hash_pos) = stored;
203         return stored;
204 }
205
206 static inline uint32_t read_be32(const unsigned char *buffer, size_t *pos)
207 {
208         uint32_t result = get_be32(buffer + *pos);
209         (*pos) += sizeof(result);
210         return result;
211 }
212
213 static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
214 {
215         return buffer[(*pos)++];
216 }
217
218 #define MAX_XOR_OFFSET 160
219
220 static int load_bitmap_entries_v1(struct bitmap_index *index)
221 {
222         uint32_t i;
223         struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
224
225         for (i = 0; i < index->entry_count; ++i) {
226                 int xor_offset, flags;
227                 struct ewah_bitmap *bitmap = NULL;
228                 struct stored_bitmap *xor_bitmap = NULL;
229                 uint32_t commit_idx_pos;
230                 struct object_id oid;
231
232                 if (index->map_size - index->map_pos < 6)
233                         return error("corrupt ewah bitmap: truncated header for entry %d", i);
234
235                 commit_idx_pos = read_be32(index->map, &index->map_pos);
236                 xor_offset = read_u8(index->map, &index->map_pos);
237                 flags = read_u8(index->map, &index->map_pos);
238
239                 if (nth_packed_object_id(&oid, index->pack, commit_idx_pos) < 0)
240                         return error("corrupt ewah bitmap: commit index %u out of range",
241                                      (unsigned)commit_idx_pos);
242
243                 bitmap = read_bitmap_1(index);
244                 if (!bitmap)
245                         return -1;
246
247                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
248                         return error("Corrupted bitmap pack index");
249
250                 if (xor_offset > 0) {
251                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
252
253                         if (xor_bitmap == NULL)
254                                 return error("Invalid XOR offset in bitmap pack index");
255                 }
256
257                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
258                         index, bitmap, &oid, xor_bitmap, flags);
259         }
260
261         return 0;
262 }
263
264 static char *pack_bitmap_filename(struct packed_git *p)
265 {
266         size_t len;
267
268         if (!strip_suffix(p->pack_name, ".pack", &len))
269                 BUG("pack_name does not end in .pack");
270         return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
271 }
272
273 static int open_pack_bitmap_1(struct bitmap_index *bitmap_git, struct packed_git *packfile)
274 {
275         int fd;
276         struct stat st;
277         char *idx_name;
278
279         if (open_pack_index(packfile))
280                 return -1;
281
282         idx_name = pack_bitmap_filename(packfile);
283         fd = git_open(idx_name);
284         free(idx_name);
285
286         if (fd < 0)
287                 return -1;
288
289         if (fstat(fd, &st)) {
290                 close(fd);
291                 return -1;
292         }
293
294         if (bitmap_git->pack) {
295                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
296                 close(fd);
297                 return -1;
298         }
299
300         bitmap_git->pack = packfile;
301         bitmap_git->map_size = xsize_t(st.st_size);
302         bitmap_git->map = xmmap(NULL, bitmap_git->map_size, PROT_READ, MAP_PRIVATE, fd, 0);
303         bitmap_git->map_pos = 0;
304         close(fd);
305
306         if (load_bitmap_header(bitmap_git) < 0) {
307                 munmap(bitmap_git->map, bitmap_git->map_size);
308                 bitmap_git->map = NULL;
309                 bitmap_git->map_size = 0;
310                 return -1;
311         }
312
313         return 0;
314 }
315
316 static int load_pack_bitmap(struct bitmap_index *bitmap_git)
317 {
318         assert(bitmap_git->map);
319
320         bitmap_git->bitmaps = kh_init_oid_map();
321         bitmap_git->ext_index.positions = kh_init_oid_pos();
322         if (load_pack_revindex(bitmap_git->pack))
323                 goto failed;
324
325         if (!(bitmap_git->commits = read_bitmap_1(bitmap_git)) ||
326                 !(bitmap_git->trees = read_bitmap_1(bitmap_git)) ||
327                 !(bitmap_git->blobs = read_bitmap_1(bitmap_git)) ||
328                 !(bitmap_git->tags = read_bitmap_1(bitmap_git)))
329                 goto failed;
330
331         if (load_bitmap_entries_v1(bitmap_git) < 0)
332                 goto failed;
333
334         return 0;
335
336 failed:
337         munmap(bitmap_git->map, bitmap_git->map_size);
338         bitmap_git->map = NULL;
339         bitmap_git->map_size = 0;
340
341         kh_destroy_oid_map(bitmap_git->bitmaps);
342         bitmap_git->bitmaps = NULL;
343
344         kh_destroy_oid_pos(bitmap_git->ext_index.positions);
345         bitmap_git->ext_index.positions = NULL;
346
347         return -1;
348 }
349
350 static int open_pack_bitmap(struct repository *r,
351                             struct bitmap_index *bitmap_git)
352 {
353         struct packed_git *p;
354         int ret = -1;
355
356         assert(!bitmap_git->map);
357
358         for (p = get_all_packs(r); p; p = p->next) {
359                 if (open_pack_bitmap_1(bitmap_git, p) == 0)
360                         ret = 0;
361         }
362
363         return ret;
364 }
365
366 struct bitmap_index *prepare_bitmap_git(struct repository *r)
367 {
368         struct bitmap_index *bitmap_git = xcalloc(1, sizeof(*bitmap_git));
369
370         if (!open_pack_bitmap(r, bitmap_git) && !load_pack_bitmap(bitmap_git))
371                 return bitmap_git;
372
373         free_bitmap_index(bitmap_git);
374         return NULL;
375 }
376
377 struct include_data {
378         struct bitmap_index *bitmap_git;
379         struct bitmap *base;
380         struct bitmap *seen;
381 };
382
383 struct ewah_bitmap *bitmap_for_commit(struct bitmap_index *bitmap_git,
384                                       struct commit *commit)
385 {
386         khiter_t hash_pos = kh_get_oid_map(bitmap_git->bitmaps,
387                                            commit->object.oid);
388         if (hash_pos >= kh_end(bitmap_git->bitmaps))
389                 return NULL;
390         return lookup_stored_bitmap(kh_value(bitmap_git->bitmaps, hash_pos));
391 }
392
393 static inline int bitmap_position_extended(struct bitmap_index *bitmap_git,
394                                            const struct object_id *oid)
395 {
396         kh_oid_pos_t *positions = bitmap_git->ext_index.positions;
397         khiter_t pos = kh_get_oid_pos(positions, *oid);
398
399         if (pos < kh_end(positions)) {
400                 int bitmap_pos = kh_value(positions, pos);
401                 return bitmap_pos + bitmap_git->pack->num_objects;
402         }
403
404         return -1;
405 }
406
407 static inline int bitmap_position_packfile(struct bitmap_index *bitmap_git,
408                                            const struct object_id *oid)
409 {
410         uint32_t pos;
411         off_t offset = find_pack_entry_one(oid->hash, bitmap_git->pack);
412         if (!offset)
413                 return -1;
414
415         if (offset_to_pack_pos(bitmap_git->pack, offset, &pos) < 0)
416                 return -1;
417         return pos;
418 }
419
420 static int bitmap_position(struct bitmap_index *bitmap_git,
421                            const struct object_id *oid)
422 {
423         int pos = bitmap_position_packfile(bitmap_git, oid);
424         return (pos >= 0) ? pos : bitmap_position_extended(bitmap_git, oid);
425 }
426
427 static int ext_index_add_object(struct bitmap_index *bitmap_git,
428                                 struct object *object, const char *name)
429 {
430         struct eindex *eindex = &bitmap_git->ext_index;
431
432         khiter_t hash_pos;
433         int hash_ret;
434         int bitmap_pos;
435
436         hash_pos = kh_put_oid_pos(eindex->positions, object->oid, &hash_ret);
437         if (hash_ret > 0) {
438                 if (eindex->count >= eindex->alloc) {
439                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
440                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
441                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
442                 }
443
444                 bitmap_pos = eindex->count;
445                 eindex->objects[eindex->count] = object;
446                 eindex->hashes[eindex->count] = pack_name_hash(name);
447                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
448                 eindex->count++;
449         } else {
450                 bitmap_pos = kh_value(eindex->positions, hash_pos);
451         }
452
453         return bitmap_pos + bitmap_git->pack->num_objects;
454 }
455
456 struct bitmap_show_data {
457         struct bitmap_index *bitmap_git;
458         struct bitmap *base;
459 };
460
461 static void show_object(struct object *object, const char *name, void *data_)
462 {
463         struct bitmap_show_data *data = data_;
464         int bitmap_pos;
465
466         bitmap_pos = bitmap_position(data->bitmap_git, &object->oid);
467
468         if (bitmap_pos < 0)
469                 bitmap_pos = ext_index_add_object(data->bitmap_git, object,
470                                                   name);
471
472         bitmap_set(data->base, bitmap_pos);
473 }
474
475 static void show_commit(struct commit *commit, void *data)
476 {
477 }
478
479 static int add_to_include_set(struct bitmap_index *bitmap_git,
480                               struct include_data *data,
481                               struct commit *commit,
482                               int bitmap_pos)
483 {
484         struct ewah_bitmap *partial;
485
486         if (data->seen && bitmap_get(data->seen, bitmap_pos))
487                 return 0;
488
489         if (bitmap_get(data->base, bitmap_pos))
490                 return 0;
491
492         partial = bitmap_for_commit(bitmap_git, commit);
493         if (partial) {
494                 bitmap_or_ewah(data->base, partial);
495                 return 0;
496         }
497
498         bitmap_set(data->base, bitmap_pos);
499         return 1;
500 }
501
502 static int should_include(struct commit *commit, void *_data)
503 {
504         struct include_data *data = _data;
505         int bitmap_pos;
506
507         bitmap_pos = bitmap_position(data->bitmap_git, &commit->object.oid);
508         if (bitmap_pos < 0)
509                 bitmap_pos = ext_index_add_object(data->bitmap_git,
510                                                   (struct object *)commit,
511                                                   NULL);
512
513         if (!add_to_include_set(data->bitmap_git, data, commit, bitmap_pos)) {
514                 struct commit_list *parent = commit->parents;
515
516                 while (parent) {
517                         parent->item->object.flags |= SEEN;
518                         parent = parent->next;
519                 }
520
521                 return 0;
522         }
523
524         return 1;
525 }
526
527 static int add_commit_to_bitmap(struct bitmap_index *bitmap_git,
528                                 struct bitmap **base,
529                                 struct commit *commit)
530 {
531         struct ewah_bitmap *or_with = bitmap_for_commit(bitmap_git, commit);
532
533         if (!or_with)
534                 return 0;
535
536         if (*base == NULL)
537                 *base = ewah_to_bitmap(or_with);
538         else
539                 bitmap_or_ewah(*base, or_with);
540
541         return 1;
542 }
543
544 static struct bitmap *find_objects(struct bitmap_index *bitmap_git,
545                                    struct rev_info *revs,
546                                    struct object_list *roots,
547                                    struct bitmap *seen,
548                                    struct list_objects_filter_options *filter)
549 {
550         struct bitmap *base = NULL;
551         int needs_walk = 0;
552
553         struct object_list *not_mapped = NULL;
554
555         /*
556          * Go through all the roots for the walk. The ones that have bitmaps
557          * on the bitmap index will be `or`ed together to form an initial
558          * global reachability analysis.
559          *
560          * The ones without bitmaps in the index will be stored in the
561          * `not_mapped_list` for further processing.
562          */
563         while (roots) {
564                 struct object *object = roots->item;
565                 roots = roots->next;
566
567                 if (object->type == OBJ_COMMIT &&
568                     add_commit_to_bitmap(bitmap_git, &base, (struct commit *)object)) {
569                         object->flags |= SEEN;
570                         continue;
571                 }
572
573                 object_list_insert(object, &not_mapped);
574         }
575
576         /*
577          * Best case scenario: We found bitmaps for all the roots,
578          * so the resulting `or` bitmap has the full reachability analysis
579          */
580         if (not_mapped == NULL)
581                 return base;
582
583         roots = not_mapped;
584
585         /*
586          * Let's iterate through all the roots that don't have bitmaps to
587          * check if we can determine them to be reachable from the existing
588          * global bitmap.
589          *
590          * If we cannot find them in the existing global bitmap, we'll need
591          * to push them to an actual walk and run it until we can confirm
592          * they are reachable
593          */
594         while (roots) {
595                 struct object *object = roots->item;
596                 int pos;
597
598                 roots = roots->next;
599                 pos = bitmap_position(bitmap_git, &object->oid);
600
601                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
602                         object->flags &= ~UNINTERESTING;
603                         add_pending_object(revs, object, "");
604                         needs_walk = 1;
605                 } else {
606                         object->flags |= SEEN;
607                 }
608         }
609
610         if (needs_walk) {
611                 struct include_data incdata;
612                 struct bitmap_show_data show_data;
613
614                 if (base == NULL)
615                         base = bitmap_new();
616
617                 incdata.bitmap_git = bitmap_git;
618                 incdata.base = base;
619                 incdata.seen = seen;
620
621                 revs->include_check = should_include;
622                 revs->include_check_data = &incdata;
623
624                 if (prepare_revision_walk(revs))
625                         die("revision walk setup failed");
626
627                 show_data.bitmap_git = bitmap_git;
628                 show_data.base = base;
629
630                 traverse_commit_list_filtered(filter, revs,
631                                               show_commit, show_object,
632                                               &show_data, NULL);
633         }
634
635         return base;
636 }
637
638 static void show_extended_objects(struct bitmap_index *bitmap_git,
639                                   struct rev_info *revs,
640                                   show_reachable_fn show_reach)
641 {
642         struct bitmap *objects = bitmap_git->result;
643         struct eindex *eindex = &bitmap_git->ext_index;
644         uint32_t i;
645
646         for (i = 0; i < eindex->count; ++i) {
647                 struct object *obj;
648
649                 if (!bitmap_get(objects, bitmap_git->pack->num_objects + i))
650                         continue;
651
652                 obj = eindex->objects[i];
653                 if ((obj->type == OBJ_BLOB && !revs->blob_objects) ||
654                     (obj->type == OBJ_TREE && !revs->tree_objects) ||
655                     (obj->type == OBJ_TAG && !revs->tag_objects))
656                         continue;
657
658                 show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
659         }
660 }
661
662 static void init_type_iterator(struct ewah_iterator *it,
663                                struct bitmap_index *bitmap_git,
664                                enum object_type type)
665 {
666         switch (type) {
667         case OBJ_COMMIT:
668                 ewah_iterator_init(it, bitmap_git->commits);
669                 break;
670
671         case OBJ_TREE:
672                 ewah_iterator_init(it, bitmap_git->trees);
673                 break;
674
675         case OBJ_BLOB:
676                 ewah_iterator_init(it, bitmap_git->blobs);
677                 break;
678
679         case OBJ_TAG:
680                 ewah_iterator_init(it, bitmap_git->tags);
681                 break;
682
683         default:
684                 BUG("object type %d not stored by bitmap type index", type);
685                 break;
686         }
687 }
688
689 static void show_objects_for_type(
690         struct bitmap_index *bitmap_git,
691         enum object_type object_type,
692         show_reachable_fn show_reach)
693 {
694         size_t i = 0;
695         uint32_t offset;
696
697         struct ewah_iterator it;
698         eword_t filter;
699
700         struct bitmap *objects = bitmap_git->result;
701
702         init_type_iterator(&it, bitmap_git, object_type);
703
704         for (i = 0; i < objects->word_alloc &&
705                         ewah_iterator_next(&filter, &it); i++) {
706                 eword_t word = objects->words[i] & filter;
707                 size_t pos = (i * BITS_IN_EWORD);
708
709                 if (!word)
710                         continue;
711
712                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
713                         struct object_id oid;
714                         uint32_t hash = 0, index_pos;
715                         off_t ofs;
716
717                         if ((word >> offset) == 0)
718                                 break;
719
720                         offset += ewah_bit_ctz64(word >> offset);
721
722                         index_pos = pack_pos_to_index(bitmap_git->pack, pos + offset);
723                         ofs = pack_pos_to_offset(bitmap_git->pack, pos + offset);
724                         nth_packed_object_id(&oid, bitmap_git->pack, index_pos);
725
726                         if (bitmap_git->hashes)
727                                 hash = get_be32(bitmap_git->hashes + index_pos);
728
729                         show_reach(&oid, object_type, 0, hash, bitmap_git->pack, ofs);
730                 }
731         }
732 }
733
734 static int in_bitmapped_pack(struct bitmap_index *bitmap_git,
735                              struct object_list *roots)
736 {
737         while (roots) {
738                 struct object *object = roots->item;
739                 roots = roots->next;
740
741                 if (find_pack_entry_one(object->oid.hash, bitmap_git->pack) > 0)
742                         return 1;
743         }
744
745         return 0;
746 }
747
748 static struct bitmap *find_tip_objects(struct bitmap_index *bitmap_git,
749                                        struct object_list *tip_objects,
750                                        enum object_type type)
751 {
752         struct bitmap *result = bitmap_new();
753         struct object_list *p;
754
755         for (p = tip_objects; p; p = p->next) {
756                 int pos;
757
758                 if (p->item->type != type)
759                         continue;
760
761                 pos = bitmap_position(bitmap_git, &p->item->oid);
762                 if (pos < 0)
763                         continue;
764
765                 bitmap_set(result, pos);
766         }
767
768         return result;
769 }
770
771 static void filter_bitmap_exclude_type(struct bitmap_index *bitmap_git,
772                                        struct object_list *tip_objects,
773                                        struct bitmap *to_filter,
774                                        enum object_type type)
775 {
776         struct eindex *eindex = &bitmap_git->ext_index;
777         struct bitmap *tips;
778         struct ewah_iterator it;
779         eword_t mask;
780         uint32_t i;
781
782         if (type != OBJ_BLOB && type != OBJ_TREE)
783                 BUG("filter_bitmap_exclude_type: unsupported type '%d'", type);
784
785         /*
786          * The non-bitmap version of this filter never removes
787          * objects which the other side specifically asked for,
788          * so we must match that behavior.
789          */
790         tips = find_tip_objects(bitmap_git, tip_objects, type);
791
792         /*
793          * We can use the blob type-bitmap to work in whole words
794          * for the objects that are actually in the bitmapped packfile.
795          */
796         for (i = 0, init_type_iterator(&it, bitmap_git, type);
797              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
798              i++) {
799                 if (i < tips->word_alloc)
800                         mask &= ~tips->words[i];
801                 to_filter->words[i] &= ~mask;
802         }
803
804         /*
805          * Clear any blobs that weren't in the packfile (and so would not have
806          * been caught by the loop above. We'll have to check them
807          * individually.
808          */
809         for (i = 0; i < eindex->count; i++) {
810                 uint32_t pos = i + bitmap_git->pack->num_objects;
811                 if (eindex->objects[i]->type == type &&
812                     bitmap_get(to_filter, pos) &&
813                     !bitmap_get(tips, pos))
814                         bitmap_unset(to_filter, pos);
815         }
816
817         bitmap_free(tips);
818 }
819
820 static void filter_bitmap_blob_none(struct bitmap_index *bitmap_git,
821                                     struct object_list *tip_objects,
822                                     struct bitmap *to_filter)
823 {
824         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
825                                    OBJ_BLOB);
826 }
827
828 static unsigned long get_size_by_pos(struct bitmap_index *bitmap_git,
829                                      uint32_t pos)
830 {
831         struct packed_git *pack = bitmap_git->pack;
832         unsigned long size;
833         struct object_info oi = OBJECT_INFO_INIT;
834
835         oi.sizep = &size;
836
837         if (pos < pack->num_objects) {
838                 off_t ofs = pack_pos_to_offset(pack, pos);
839                 if (packed_object_info(the_repository, pack, ofs, &oi) < 0) {
840                         struct object_id oid;
841                         nth_packed_object_id(&oid, pack,
842                                              pack_pos_to_index(pack, pos));
843                         die(_("unable to get size of %s"), oid_to_hex(&oid));
844                 }
845         } else {
846                 struct eindex *eindex = &bitmap_git->ext_index;
847                 struct object *obj = eindex->objects[pos - pack->num_objects];
848                 if (oid_object_info_extended(the_repository, &obj->oid, &oi, 0) < 0)
849                         die(_("unable to get size of %s"), oid_to_hex(&obj->oid));
850         }
851
852         return size;
853 }
854
855 static void filter_bitmap_blob_limit(struct bitmap_index *bitmap_git,
856                                      struct object_list *tip_objects,
857                                      struct bitmap *to_filter,
858                                      unsigned long limit)
859 {
860         struct eindex *eindex = &bitmap_git->ext_index;
861         struct bitmap *tips;
862         struct ewah_iterator it;
863         eword_t mask;
864         uint32_t i;
865
866         tips = find_tip_objects(bitmap_git, tip_objects, OBJ_BLOB);
867
868         for (i = 0, init_type_iterator(&it, bitmap_git, OBJ_BLOB);
869              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
870              i++) {
871                 eword_t word = to_filter->words[i] & mask;
872                 unsigned offset;
873
874                 for (offset = 0; offset < BITS_IN_EWORD; offset++) {
875                         uint32_t pos;
876
877                         if ((word >> offset) == 0)
878                                 break;
879                         offset += ewah_bit_ctz64(word >> offset);
880                         pos = i * BITS_IN_EWORD + offset;
881
882                         if (!bitmap_get(tips, pos) &&
883                             get_size_by_pos(bitmap_git, pos) >= limit)
884                                 bitmap_unset(to_filter, pos);
885                 }
886         }
887
888         for (i = 0; i < eindex->count; i++) {
889                 uint32_t pos = i + bitmap_git->pack->num_objects;
890                 if (eindex->objects[i]->type == OBJ_BLOB &&
891                     bitmap_get(to_filter, pos) &&
892                     !bitmap_get(tips, pos) &&
893                     get_size_by_pos(bitmap_git, pos) >= limit)
894                         bitmap_unset(to_filter, pos);
895         }
896
897         bitmap_free(tips);
898 }
899
900 static void filter_bitmap_tree_depth(struct bitmap_index *bitmap_git,
901                                      struct object_list *tip_objects,
902                                      struct bitmap *to_filter,
903                                      unsigned long limit)
904 {
905         if (limit)
906                 BUG("filter_bitmap_tree_depth given non-zero limit");
907
908         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
909                                    OBJ_TREE);
910         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
911                                    OBJ_BLOB);
912 }
913
914 static int filter_bitmap(struct bitmap_index *bitmap_git,
915                          struct object_list *tip_objects,
916                          struct bitmap *to_filter,
917                          struct list_objects_filter_options *filter)
918 {
919         if (!filter || filter->choice == LOFC_DISABLED)
920                 return 0;
921
922         if (filter->choice == LOFC_BLOB_NONE) {
923                 if (bitmap_git)
924                         filter_bitmap_blob_none(bitmap_git, tip_objects,
925                                                 to_filter);
926                 return 0;
927         }
928
929         if (filter->choice == LOFC_BLOB_LIMIT) {
930                 if (bitmap_git)
931                         filter_bitmap_blob_limit(bitmap_git, tip_objects,
932                                                  to_filter,
933                                                  filter->blob_limit_value);
934                 return 0;
935         }
936
937         if (filter->choice == LOFC_TREE_DEPTH &&
938             filter->tree_exclude_depth == 0) {
939                 if (bitmap_git)
940                         filter_bitmap_tree_depth(bitmap_git, tip_objects,
941                                                  to_filter,
942                                                  filter->tree_exclude_depth);
943                 return 0;
944         }
945
946         /* filter choice not handled */
947         return -1;
948 }
949
950 static int can_filter_bitmap(struct list_objects_filter_options *filter)
951 {
952         return !filter_bitmap(NULL, NULL, NULL, filter);
953 }
954
955 struct bitmap_index *prepare_bitmap_walk(struct rev_info *revs,
956                                          struct list_objects_filter_options *filter)
957 {
958         unsigned int i;
959
960         struct object_list *wants = NULL;
961         struct object_list *haves = NULL;
962
963         struct bitmap *wants_bitmap = NULL;
964         struct bitmap *haves_bitmap = NULL;
965
966         struct bitmap_index *bitmap_git;
967
968         /*
969          * We can't do pathspec limiting with bitmaps, because we don't know
970          * which commits are associated with which object changes (let alone
971          * even which objects are associated with which paths).
972          */
973         if (revs->prune)
974                 return NULL;
975
976         if (!can_filter_bitmap(filter))
977                 return NULL;
978
979         /* try to open a bitmapped pack, but don't parse it yet
980          * because we may not need to use it */
981         bitmap_git = xcalloc(1, sizeof(*bitmap_git));
982         if (open_pack_bitmap(revs->repo, bitmap_git) < 0)
983                 goto cleanup;
984
985         for (i = 0; i < revs->pending.nr; ++i) {
986                 struct object *object = revs->pending.objects[i].item;
987
988                 if (object->type == OBJ_NONE)
989                         parse_object_or_die(&object->oid, NULL);
990
991                 while (object->type == OBJ_TAG) {
992                         struct tag *tag = (struct tag *) object;
993
994                         if (object->flags & UNINTERESTING)
995                                 object_list_insert(object, &haves);
996                         else
997                                 object_list_insert(object, &wants);
998
999                         object = parse_object_or_die(get_tagged_oid(tag), NULL);
1000                 }
1001
1002                 if (object->flags & UNINTERESTING)
1003                         object_list_insert(object, &haves);
1004                 else
1005                         object_list_insert(object, &wants);
1006         }
1007
1008         /*
1009          * if we have a HAVES list, but none of those haves is contained
1010          * in the packfile that has a bitmap, we don't have anything to
1011          * optimize here
1012          */
1013         if (haves && !in_bitmapped_pack(bitmap_git, haves))
1014                 goto cleanup;
1015
1016         /* if we don't want anything, we're done here */
1017         if (!wants)
1018                 goto cleanup;
1019
1020         /*
1021          * now we're going to use bitmaps, so load the actual bitmap entries
1022          * from disk. this is the point of no return; after this the rev_list
1023          * becomes invalidated and we must perform the revwalk through bitmaps
1024          */
1025         if (load_pack_bitmap(bitmap_git) < 0)
1026                 goto cleanup;
1027
1028         object_array_clear(&revs->pending);
1029
1030         if (haves) {
1031                 revs->ignore_missing_links = 1;
1032                 haves_bitmap = find_objects(bitmap_git, revs, haves, NULL,
1033                                             filter);
1034                 reset_revision_walk();
1035                 revs->ignore_missing_links = 0;
1036
1037                 if (haves_bitmap == NULL)
1038                         BUG("failed to perform bitmap walk");
1039         }
1040
1041         wants_bitmap = find_objects(bitmap_git, revs, wants, haves_bitmap,
1042                                     filter);
1043
1044         if (!wants_bitmap)
1045                 BUG("failed to perform bitmap walk");
1046
1047         if (haves_bitmap)
1048                 bitmap_and_not(wants_bitmap, haves_bitmap);
1049
1050         filter_bitmap(bitmap_git, wants, wants_bitmap, filter);
1051
1052         bitmap_git->result = wants_bitmap;
1053         bitmap_git->haves = haves_bitmap;
1054
1055         object_list_free(&wants);
1056         object_list_free(&haves);
1057
1058         return bitmap_git;
1059
1060 cleanup:
1061         free_bitmap_index(bitmap_git);
1062         object_list_free(&wants);
1063         object_list_free(&haves);
1064         return NULL;
1065 }
1066
1067 static void try_partial_reuse(struct bitmap_index *bitmap_git,
1068                               size_t pos,
1069                               struct bitmap *reuse,
1070                               struct pack_window **w_curs)
1071 {
1072         off_t offset, header;
1073         enum object_type type;
1074         unsigned long size;
1075
1076         if (pos >= bitmap_git->pack->num_objects)
1077                 return; /* not actually in the pack */
1078
1079         offset = header = pack_pos_to_offset(bitmap_git->pack, pos);
1080         type = unpack_object_header(bitmap_git->pack, w_curs, &offset, &size);
1081         if (type < 0)
1082                 return; /* broken packfile, punt */
1083
1084         if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA) {
1085                 off_t base_offset;
1086                 uint32_t base_pos;
1087
1088                 /*
1089                  * Find the position of the base object so we can look it up
1090                  * in our bitmaps. If we can't come up with an offset, or if
1091                  * that offset is not in the revidx, the pack is corrupt.
1092                  * There's nothing we can do, so just punt on this object,
1093                  * and the normal slow path will complain about it in
1094                  * more detail.
1095                  */
1096                 base_offset = get_delta_base(bitmap_git->pack, w_curs,
1097                                              &offset, type, header);
1098                 if (!base_offset)
1099                         return;
1100                 if (offset_to_pack_pos(bitmap_git->pack, base_offset, &base_pos) < 0)
1101                         return;
1102
1103                 /*
1104                  * We assume delta dependencies always point backwards. This
1105                  * lets us do a single pass, and is basically always true
1106                  * due to the way OFS_DELTAs work. You would not typically
1107                  * find REF_DELTA in a bitmapped pack, since we only bitmap
1108                  * packs we write fresh, and OFS_DELTA is the default). But
1109                  * let's double check to make sure the pack wasn't written with
1110                  * odd parameters.
1111                  */
1112                 if (base_pos >= pos)
1113                         return;
1114
1115                 /*
1116                  * And finally, if we're not sending the base as part of our
1117                  * reuse chunk, then don't send this object either. The base
1118                  * would come after us, along with other objects not
1119                  * necessarily in the pack, which means we'd need to convert
1120                  * to REF_DELTA on the fly. Better to just let the normal
1121                  * object_entry code path handle it.
1122                  */
1123                 if (!bitmap_get(reuse, base_pos))
1124                         return;
1125         }
1126
1127         /*
1128          * If we got here, then the object is OK to reuse. Mark it.
1129          */
1130         bitmap_set(reuse, pos);
1131 }
1132
1133 int reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
1134                                        struct packed_git **packfile_out,
1135                                        uint32_t *entries,
1136                                        struct bitmap **reuse_out)
1137 {
1138         struct bitmap *result = bitmap_git->result;
1139         struct bitmap *reuse;
1140         struct pack_window *w_curs = NULL;
1141         size_t i = 0;
1142         uint32_t offset;
1143
1144         assert(result);
1145
1146         while (i < result->word_alloc && result->words[i] == (eword_t)~0)
1147                 i++;
1148
1149         /* Don't mark objects not in the packfile */
1150         if (i > bitmap_git->pack->num_objects / BITS_IN_EWORD)
1151                 i = bitmap_git->pack->num_objects / BITS_IN_EWORD;
1152
1153         reuse = bitmap_word_alloc(i);
1154         memset(reuse->words, 0xFF, i * sizeof(eword_t));
1155
1156         for (; i < result->word_alloc; ++i) {
1157                 eword_t word = result->words[i];
1158                 size_t pos = (i * BITS_IN_EWORD);
1159
1160                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1161                         if ((word >> offset) == 0)
1162                                 break;
1163
1164                         offset += ewah_bit_ctz64(word >> offset);
1165                         try_partial_reuse(bitmap_git, pos + offset, reuse, &w_curs);
1166                 }
1167         }
1168
1169         unuse_pack(&w_curs);
1170
1171         *entries = bitmap_popcount(reuse);
1172         if (!*entries) {
1173                 bitmap_free(reuse);
1174                 return -1;
1175         }
1176
1177         /*
1178          * Drop any reused objects from the result, since they will not
1179          * need to be handled separately.
1180          */
1181         bitmap_and_not(result, reuse);
1182         *packfile_out = bitmap_git->pack;
1183         *reuse_out = reuse;
1184         return 0;
1185 }
1186
1187 int bitmap_walk_contains(struct bitmap_index *bitmap_git,
1188                          struct bitmap *bitmap, const struct object_id *oid)
1189 {
1190         int idx;
1191
1192         if (!bitmap)
1193                 return 0;
1194
1195         idx = bitmap_position(bitmap_git, oid);
1196         return idx >= 0 && bitmap_get(bitmap, idx);
1197 }
1198
1199 void traverse_bitmap_commit_list(struct bitmap_index *bitmap_git,
1200                                  struct rev_info *revs,
1201                                  show_reachable_fn show_reachable)
1202 {
1203         assert(bitmap_git->result);
1204
1205         show_objects_for_type(bitmap_git, OBJ_COMMIT, show_reachable);
1206         if (revs->tree_objects)
1207                 show_objects_for_type(bitmap_git, OBJ_TREE, show_reachable);
1208         if (revs->blob_objects)
1209                 show_objects_for_type(bitmap_git, OBJ_BLOB, show_reachable);
1210         if (revs->tag_objects)
1211                 show_objects_for_type(bitmap_git, OBJ_TAG, show_reachable);
1212
1213         show_extended_objects(bitmap_git, revs, show_reachable);
1214 }
1215
1216 static uint32_t count_object_type(struct bitmap_index *bitmap_git,
1217                                   enum object_type type)
1218 {
1219         struct bitmap *objects = bitmap_git->result;
1220         struct eindex *eindex = &bitmap_git->ext_index;
1221
1222         uint32_t i = 0, count = 0;
1223         struct ewah_iterator it;
1224         eword_t filter;
1225
1226         init_type_iterator(&it, bitmap_git, type);
1227
1228         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
1229                 eword_t word = objects->words[i++] & filter;
1230                 count += ewah_bit_popcount64(word);
1231         }
1232
1233         for (i = 0; i < eindex->count; ++i) {
1234                 if (eindex->objects[i]->type == type &&
1235                         bitmap_get(objects, bitmap_git->pack->num_objects + i))
1236                         count++;
1237         }
1238
1239         return count;
1240 }
1241
1242 void count_bitmap_commit_list(struct bitmap_index *bitmap_git,
1243                               uint32_t *commits, uint32_t *trees,
1244                               uint32_t *blobs, uint32_t *tags)
1245 {
1246         assert(bitmap_git->result);
1247
1248         if (commits)
1249                 *commits = count_object_type(bitmap_git, OBJ_COMMIT);
1250
1251         if (trees)
1252                 *trees = count_object_type(bitmap_git, OBJ_TREE);
1253
1254         if (blobs)
1255                 *blobs = count_object_type(bitmap_git, OBJ_BLOB);
1256
1257         if (tags)
1258                 *tags = count_object_type(bitmap_git, OBJ_TAG);
1259 }
1260
1261 struct bitmap_test_data {
1262         struct bitmap_index *bitmap_git;
1263         struct bitmap *base;
1264         struct progress *prg;
1265         size_t seen;
1266 };
1267
1268 static void test_show_object(struct object *object, const char *name,
1269                              void *data)
1270 {
1271         struct bitmap_test_data *tdata = data;
1272         int bitmap_pos;
1273
1274         bitmap_pos = bitmap_position(tdata->bitmap_git, &object->oid);
1275         if (bitmap_pos < 0)
1276                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
1277
1278         bitmap_set(tdata->base, bitmap_pos);
1279         display_progress(tdata->prg, ++tdata->seen);
1280 }
1281
1282 static void test_show_commit(struct commit *commit, void *data)
1283 {
1284         struct bitmap_test_data *tdata = data;
1285         int bitmap_pos;
1286
1287         bitmap_pos = bitmap_position(tdata->bitmap_git,
1288                                      &commit->object.oid);
1289         if (bitmap_pos < 0)
1290                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
1291
1292         bitmap_set(tdata->base, bitmap_pos);
1293         display_progress(tdata->prg, ++tdata->seen);
1294 }
1295
1296 void test_bitmap_walk(struct rev_info *revs)
1297 {
1298         struct object *root;
1299         struct bitmap *result = NULL;
1300         size_t result_popcnt;
1301         struct bitmap_test_data tdata;
1302         struct bitmap_index *bitmap_git;
1303         struct ewah_bitmap *bm;
1304
1305         if (!(bitmap_git = prepare_bitmap_git(revs->repo)))
1306                 die("failed to load bitmap indexes");
1307
1308         if (revs->pending.nr != 1)
1309                 die("you must specify exactly one commit to test");
1310
1311         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
1312                 bitmap_git->version, bitmap_git->entry_count);
1313
1314         root = revs->pending.objects[0].item;
1315         bm = bitmap_for_commit(bitmap_git, (struct commit *)root);
1316
1317         if (bm) {
1318                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
1319                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
1320
1321                 result = ewah_to_bitmap(bm);
1322         }
1323
1324         if (result == NULL)
1325                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
1326
1327         revs->tag_objects = 1;
1328         revs->tree_objects = 1;
1329         revs->blob_objects = 1;
1330
1331         result_popcnt = bitmap_popcount(result);
1332
1333         if (prepare_revision_walk(revs))
1334                 die("revision walk setup failed");
1335
1336         tdata.bitmap_git = bitmap_git;
1337         tdata.base = bitmap_new();
1338         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
1339         tdata.seen = 0;
1340
1341         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
1342
1343         stop_progress(&tdata.prg);
1344
1345         if (bitmap_equals(result, tdata.base))
1346                 fprintf(stderr, "OK!\n");
1347         else
1348                 die("mismatch in bitmap results");
1349
1350         free_bitmap_index(bitmap_git);
1351 }
1352
1353 int rebuild_bitmap(const uint32_t *reposition,
1354                    struct ewah_bitmap *source,
1355                    struct bitmap *dest)
1356 {
1357         uint32_t pos = 0;
1358         struct ewah_iterator it;
1359         eword_t word;
1360
1361         ewah_iterator_init(&it, source);
1362
1363         while (ewah_iterator_next(&word, &it)) {
1364                 uint32_t offset, bit_pos;
1365
1366                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1367                         if ((word >> offset) == 0)
1368                                 break;
1369
1370                         offset += ewah_bit_ctz64(word >> offset);
1371
1372                         bit_pos = reposition[pos + offset];
1373                         if (bit_pos > 0)
1374                                 bitmap_set(dest, bit_pos - 1);
1375                         else /* can't reuse, we don't have the object */
1376                                 return -1;
1377                 }
1378
1379                 pos += BITS_IN_EWORD;
1380         }
1381         return 0;
1382 }
1383
1384 uint32_t *create_bitmap_mapping(struct bitmap_index *bitmap_git,
1385                                 struct packing_data *mapping)
1386 {
1387         uint32_t i, num_objects;
1388         uint32_t *reposition;
1389
1390         num_objects = bitmap_git->pack->num_objects;
1391         reposition = xcalloc(num_objects, sizeof(uint32_t));
1392
1393         for (i = 0; i < num_objects; ++i) {
1394                 struct object_id oid;
1395                 struct object_entry *oe;
1396
1397                 nth_packed_object_id(&oid, bitmap_git->pack,
1398                                      pack_pos_to_index(bitmap_git->pack, i));
1399                 oe = packlist_find(mapping, &oid);
1400
1401                 if (oe)
1402                         reposition[i] = oe_in_pack_pos(mapping, oe) + 1;
1403         }
1404
1405         return reposition;
1406 }
1407
1408 void free_bitmap_index(struct bitmap_index *b)
1409 {
1410         if (!b)
1411                 return;
1412
1413         if (b->map)
1414                 munmap(b->map, b->map_size);
1415         ewah_pool_free(b->commits);
1416         ewah_pool_free(b->trees);
1417         ewah_pool_free(b->blobs);
1418         ewah_pool_free(b->tags);
1419         kh_destroy_oid_map(b->bitmaps);
1420         free(b->ext_index.objects);
1421         free(b->ext_index.hashes);
1422         bitmap_free(b->result);
1423         bitmap_free(b->haves);
1424         free(b);
1425 }
1426
1427 int bitmap_has_oid_in_uninteresting(struct bitmap_index *bitmap_git,
1428                                     const struct object_id *oid)
1429 {
1430         return bitmap_git &&
1431                 bitmap_walk_contains(bitmap_git, bitmap_git->haves, oid);
1432 }