t5310: add branch-based checks
[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                 commit_idx_pos = read_be32(index->map, &index->map_pos);
233                 xor_offset = read_u8(index->map, &index->map_pos);
234                 flags = read_u8(index->map, &index->map_pos);
235
236                 nth_packed_object_id(&oid, index->pack, commit_idx_pos);
237
238                 bitmap = read_bitmap_1(index);
239                 if (!bitmap)
240                         return -1;
241
242                 if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
243                         return error("Corrupted bitmap pack index");
244
245                 if (xor_offset > 0) {
246                         xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
247
248                         if (xor_bitmap == NULL)
249                                 return error("Invalid XOR offset in bitmap pack index");
250                 }
251
252                 recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
253                         index, bitmap, &oid, xor_bitmap, flags);
254         }
255
256         return 0;
257 }
258
259 static char *pack_bitmap_filename(struct packed_git *p)
260 {
261         size_t len;
262
263         if (!strip_suffix(p->pack_name, ".pack", &len))
264                 BUG("pack_name does not end in .pack");
265         return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
266 }
267
268 static int open_pack_bitmap_1(struct bitmap_index *bitmap_git, struct packed_git *packfile)
269 {
270         int fd;
271         struct stat st;
272         char *idx_name;
273
274         if (open_pack_index(packfile))
275                 return -1;
276
277         idx_name = pack_bitmap_filename(packfile);
278         fd = git_open(idx_name);
279         free(idx_name);
280
281         if (fd < 0)
282                 return -1;
283
284         if (fstat(fd, &st)) {
285                 close(fd);
286                 return -1;
287         }
288
289         if (bitmap_git->pack) {
290                 warning("ignoring extra bitmap file: %s", packfile->pack_name);
291                 close(fd);
292                 return -1;
293         }
294
295         bitmap_git->pack = packfile;
296         bitmap_git->map_size = xsize_t(st.st_size);
297         bitmap_git->map = xmmap(NULL, bitmap_git->map_size, PROT_READ, MAP_PRIVATE, fd, 0);
298         bitmap_git->map_pos = 0;
299         close(fd);
300
301         if (load_bitmap_header(bitmap_git) < 0) {
302                 munmap(bitmap_git->map, bitmap_git->map_size);
303                 bitmap_git->map = NULL;
304                 bitmap_git->map_size = 0;
305                 return -1;
306         }
307
308         return 0;
309 }
310
311 static int load_pack_bitmap(struct bitmap_index *bitmap_git)
312 {
313         assert(bitmap_git->map);
314
315         bitmap_git->bitmaps = kh_init_oid_map();
316         bitmap_git->ext_index.positions = kh_init_oid_pos();
317         if (load_pack_revindex(bitmap_git->pack))
318                 goto failed;
319
320         if (!(bitmap_git->commits = read_bitmap_1(bitmap_git)) ||
321                 !(bitmap_git->trees = read_bitmap_1(bitmap_git)) ||
322                 !(bitmap_git->blobs = read_bitmap_1(bitmap_git)) ||
323                 !(bitmap_git->tags = read_bitmap_1(bitmap_git)))
324                 goto failed;
325
326         if (load_bitmap_entries_v1(bitmap_git) < 0)
327                 goto failed;
328
329         return 0;
330
331 failed:
332         munmap(bitmap_git->map, bitmap_git->map_size);
333         bitmap_git->map = NULL;
334         bitmap_git->map_size = 0;
335
336         kh_destroy_oid_map(bitmap_git->bitmaps);
337         bitmap_git->bitmaps = NULL;
338
339         kh_destroy_oid_pos(bitmap_git->ext_index.positions);
340         bitmap_git->ext_index.positions = NULL;
341
342         return -1;
343 }
344
345 static int open_pack_bitmap(struct repository *r,
346                             struct bitmap_index *bitmap_git)
347 {
348         struct packed_git *p;
349         int ret = -1;
350
351         assert(!bitmap_git->map);
352
353         for (p = get_all_packs(r); p; p = p->next) {
354                 if (open_pack_bitmap_1(bitmap_git, p) == 0)
355                         ret = 0;
356         }
357
358         return ret;
359 }
360
361 struct bitmap_index *prepare_bitmap_git(struct repository *r)
362 {
363         struct bitmap_index *bitmap_git = xcalloc(1, sizeof(*bitmap_git));
364
365         if (!open_pack_bitmap(r, bitmap_git) && !load_pack_bitmap(bitmap_git))
366                 return bitmap_git;
367
368         free_bitmap_index(bitmap_git);
369         return NULL;
370 }
371
372 struct include_data {
373         struct bitmap_index *bitmap_git;
374         struct bitmap *base;
375         struct bitmap *seen;
376 };
377
378 static inline int bitmap_position_extended(struct bitmap_index *bitmap_git,
379                                            const struct object_id *oid)
380 {
381         kh_oid_pos_t *positions = bitmap_git->ext_index.positions;
382         khiter_t pos = kh_get_oid_pos(positions, *oid);
383
384         if (pos < kh_end(positions)) {
385                 int bitmap_pos = kh_value(positions, pos);
386                 return bitmap_pos + bitmap_git->pack->num_objects;
387         }
388
389         return -1;
390 }
391
392 static inline int bitmap_position_packfile(struct bitmap_index *bitmap_git,
393                                            const struct object_id *oid)
394 {
395         off_t offset = find_pack_entry_one(oid->hash, bitmap_git->pack);
396         if (!offset)
397                 return -1;
398
399         return find_revindex_position(bitmap_git->pack, offset);
400 }
401
402 static int bitmap_position(struct bitmap_index *bitmap_git,
403                            const struct object_id *oid)
404 {
405         int pos = bitmap_position_packfile(bitmap_git, oid);
406         return (pos >= 0) ? pos : bitmap_position_extended(bitmap_git, oid);
407 }
408
409 static int ext_index_add_object(struct bitmap_index *bitmap_git,
410                                 struct object *object, const char *name)
411 {
412         struct eindex *eindex = &bitmap_git->ext_index;
413
414         khiter_t hash_pos;
415         int hash_ret;
416         int bitmap_pos;
417
418         hash_pos = kh_put_oid_pos(eindex->positions, object->oid, &hash_ret);
419         if (hash_ret > 0) {
420                 if (eindex->count >= eindex->alloc) {
421                         eindex->alloc = (eindex->alloc + 16) * 3 / 2;
422                         REALLOC_ARRAY(eindex->objects, eindex->alloc);
423                         REALLOC_ARRAY(eindex->hashes, eindex->alloc);
424                 }
425
426                 bitmap_pos = eindex->count;
427                 eindex->objects[eindex->count] = object;
428                 eindex->hashes[eindex->count] = pack_name_hash(name);
429                 kh_value(eindex->positions, hash_pos) = bitmap_pos;
430                 eindex->count++;
431         } else {
432                 bitmap_pos = kh_value(eindex->positions, hash_pos);
433         }
434
435         return bitmap_pos + bitmap_git->pack->num_objects;
436 }
437
438 struct bitmap_show_data {
439         struct bitmap_index *bitmap_git;
440         struct bitmap *base;
441 };
442
443 static void show_object(struct object *object, const char *name, void *data_)
444 {
445         struct bitmap_show_data *data = data_;
446         int bitmap_pos;
447
448         bitmap_pos = bitmap_position(data->bitmap_git, &object->oid);
449
450         if (bitmap_pos < 0)
451                 bitmap_pos = ext_index_add_object(data->bitmap_git, object,
452                                                   name);
453
454         bitmap_set(data->base, bitmap_pos);
455 }
456
457 static void show_commit(struct commit *commit, void *data)
458 {
459 }
460
461 static int add_to_include_set(struct bitmap_index *bitmap_git,
462                               struct include_data *data,
463                               const struct object_id *oid,
464                               int bitmap_pos)
465 {
466         khiter_t hash_pos;
467
468         if (data->seen && bitmap_get(data->seen, bitmap_pos))
469                 return 0;
470
471         if (bitmap_get(data->base, bitmap_pos))
472                 return 0;
473
474         hash_pos = kh_get_oid_map(bitmap_git->bitmaps, *oid);
475         if (hash_pos < kh_end(bitmap_git->bitmaps)) {
476                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, hash_pos);
477                 bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
478                 return 0;
479         }
480
481         bitmap_set(data->base, bitmap_pos);
482         return 1;
483 }
484
485 static int should_include(struct commit *commit, void *_data)
486 {
487         struct include_data *data = _data;
488         int bitmap_pos;
489
490         bitmap_pos = bitmap_position(data->bitmap_git, &commit->object.oid);
491         if (bitmap_pos < 0)
492                 bitmap_pos = ext_index_add_object(data->bitmap_git,
493                                                   (struct object *)commit,
494                                                   NULL);
495
496         if (!add_to_include_set(data->bitmap_git, data, &commit->object.oid,
497                                 bitmap_pos)) {
498                 struct commit_list *parent = commit->parents;
499
500                 while (parent) {
501                         parent->item->object.flags |= SEEN;
502                         parent = parent->next;
503                 }
504
505                 return 0;
506         }
507
508         return 1;
509 }
510
511 static struct bitmap *find_objects(struct bitmap_index *bitmap_git,
512                                    struct rev_info *revs,
513                                    struct object_list *roots,
514                                    struct bitmap *seen,
515                                    struct list_objects_filter_options *filter)
516 {
517         struct bitmap *base = NULL;
518         int needs_walk = 0;
519
520         struct object_list *not_mapped = NULL;
521
522         /*
523          * Go through all the roots for the walk. The ones that have bitmaps
524          * on the bitmap index will be `or`ed together to form an initial
525          * global reachability analysis.
526          *
527          * The ones without bitmaps in the index will be stored in the
528          * `not_mapped_list` for further processing.
529          */
530         while (roots) {
531                 struct object *object = roots->item;
532                 roots = roots->next;
533
534                 if (object->type == OBJ_COMMIT) {
535                         khiter_t pos = kh_get_oid_map(bitmap_git->bitmaps, object->oid);
536
537                         if (pos < kh_end(bitmap_git->bitmaps)) {
538                                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
539                                 struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
540
541                                 if (base == NULL)
542                                         base = ewah_to_bitmap(or_with);
543                                 else
544                                         bitmap_or_ewah(base, or_with);
545
546                                 object->flags |= SEEN;
547                                 continue;
548                         }
549                 }
550
551                 object_list_insert(object, &not_mapped);
552         }
553
554         /*
555          * Best case scenario: We found bitmaps for all the roots,
556          * so the resulting `or` bitmap has the full reachability analysis
557          */
558         if (not_mapped == NULL)
559                 return base;
560
561         roots = not_mapped;
562
563         /*
564          * Let's iterate through all the roots that don't have bitmaps to
565          * check if we can determine them to be reachable from the existing
566          * global bitmap.
567          *
568          * If we cannot find them in the existing global bitmap, we'll need
569          * to push them to an actual walk and run it until we can confirm
570          * they are reachable
571          */
572         while (roots) {
573                 struct object *object = roots->item;
574                 int pos;
575
576                 roots = roots->next;
577                 pos = bitmap_position(bitmap_git, &object->oid);
578
579                 if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
580                         object->flags &= ~UNINTERESTING;
581                         add_pending_object(revs, object, "");
582                         needs_walk = 1;
583                 } else {
584                         object->flags |= SEEN;
585                 }
586         }
587
588         if (needs_walk) {
589                 struct include_data incdata;
590                 struct bitmap_show_data show_data;
591
592                 if (base == NULL)
593                         base = bitmap_new();
594
595                 incdata.bitmap_git = bitmap_git;
596                 incdata.base = base;
597                 incdata.seen = seen;
598
599                 revs->include_check = should_include;
600                 revs->include_check_data = &incdata;
601
602                 if (prepare_revision_walk(revs))
603                         die("revision walk setup failed");
604
605                 show_data.bitmap_git = bitmap_git;
606                 show_data.base = base;
607
608                 traverse_commit_list_filtered(filter, revs,
609                                               show_commit, show_object,
610                                               &show_data, NULL);
611         }
612
613         return base;
614 }
615
616 static void show_extended_objects(struct bitmap_index *bitmap_git,
617                                   struct rev_info *revs,
618                                   show_reachable_fn show_reach)
619 {
620         struct bitmap *objects = bitmap_git->result;
621         struct eindex *eindex = &bitmap_git->ext_index;
622         uint32_t i;
623
624         for (i = 0; i < eindex->count; ++i) {
625                 struct object *obj;
626
627                 if (!bitmap_get(objects, bitmap_git->pack->num_objects + i))
628                         continue;
629
630                 obj = eindex->objects[i];
631                 if ((obj->type == OBJ_BLOB && !revs->blob_objects) ||
632                     (obj->type == OBJ_TREE && !revs->tree_objects) ||
633                     (obj->type == OBJ_TAG && !revs->tag_objects))
634                         continue;
635
636                 show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
637         }
638 }
639
640 static void init_type_iterator(struct ewah_iterator *it,
641                                struct bitmap_index *bitmap_git,
642                                enum object_type type)
643 {
644         switch (type) {
645         case OBJ_COMMIT:
646                 ewah_iterator_init(it, bitmap_git->commits);
647                 break;
648
649         case OBJ_TREE:
650                 ewah_iterator_init(it, bitmap_git->trees);
651                 break;
652
653         case OBJ_BLOB:
654                 ewah_iterator_init(it, bitmap_git->blobs);
655                 break;
656
657         case OBJ_TAG:
658                 ewah_iterator_init(it, bitmap_git->tags);
659                 break;
660
661         default:
662                 BUG("object type %d not stored by bitmap type index", type);
663                 break;
664         }
665 }
666
667 static void show_objects_for_type(
668         struct bitmap_index *bitmap_git,
669         enum object_type object_type,
670         show_reachable_fn show_reach)
671 {
672         size_t i = 0;
673         uint32_t offset;
674
675         struct ewah_iterator it;
676         eword_t filter;
677
678         struct bitmap *objects = bitmap_git->result;
679
680         init_type_iterator(&it, bitmap_git, object_type);
681
682         for (i = 0; i < objects->word_alloc &&
683                         ewah_iterator_next(&filter, &it); i++) {
684                 eword_t word = objects->words[i] & filter;
685                 size_t pos = (i * BITS_IN_EWORD);
686
687                 if (!word)
688                         continue;
689
690                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
691                         struct object_id oid;
692                         struct revindex_entry *entry;
693                         uint32_t hash = 0;
694
695                         if ((word >> offset) == 0)
696                                 break;
697
698                         offset += ewah_bit_ctz64(word >> offset);
699
700                         entry = &bitmap_git->pack->revindex[pos + offset];
701                         nth_packed_object_id(&oid, bitmap_git->pack, entry->nr);
702
703                         if (bitmap_git->hashes)
704                                 hash = get_be32(bitmap_git->hashes + entry->nr);
705
706                         show_reach(&oid, object_type, 0, hash, bitmap_git->pack, entry->offset);
707                 }
708         }
709 }
710
711 static int in_bitmapped_pack(struct bitmap_index *bitmap_git,
712                              struct object_list *roots)
713 {
714         while (roots) {
715                 struct object *object = roots->item;
716                 roots = roots->next;
717
718                 if (find_pack_entry_one(object->oid.hash, bitmap_git->pack) > 0)
719                         return 1;
720         }
721
722         return 0;
723 }
724
725 static struct bitmap *find_tip_objects(struct bitmap_index *bitmap_git,
726                                        struct object_list *tip_objects,
727                                        enum object_type type)
728 {
729         struct bitmap *result = bitmap_new();
730         struct object_list *p;
731
732         for (p = tip_objects; p; p = p->next) {
733                 int pos;
734
735                 if (p->item->type != type)
736                         continue;
737
738                 pos = bitmap_position(bitmap_git, &p->item->oid);
739                 if (pos < 0)
740                         continue;
741
742                 bitmap_set(result, pos);
743         }
744
745         return result;
746 }
747
748 static void filter_bitmap_exclude_type(struct bitmap_index *bitmap_git,
749                                        struct object_list *tip_objects,
750                                        struct bitmap *to_filter,
751                                        enum object_type type)
752 {
753         struct eindex *eindex = &bitmap_git->ext_index;
754         struct bitmap *tips;
755         struct ewah_iterator it;
756         eword_t mask;
757         uint32_t i;
758
759         if (type != OBJ_BLOB && type != OBJ_TREE)
760                 BUG("filter_bitmap_exclude_type: unsupported type '%d'", type);
761
762         /*
763          * The non-bitmap version of this filter never removes
764          * objects which the other side specifically asked for,
765          * so we must match that behavior.
766          */
767         tips = find_tip_objects(bitmap_git, tip_objects, type);
768
769         /*
770          * We can use the blob type-bitmap to work in whole words
771          * for the objects that are actually in the bitmapped packfile.
772          */
773         for (i = 0, init_type_iterator(&it, bitmap_git, type);
774              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
775              i++) {
776                 if (i < tips->word_alloc)
777                         mask &= ~tips->words[i];
778                 to_filter->words[i] &= ~mask;
779         }
780
781         /*
782          * Clear any blobs that weren't in the packfile (and so would not have
783          * been caught by the loop above. We'll have to check them
784          * individually.
785          */
786         for (i = 0; i < eindex->count; i++) {
787                 uint32_t pos = i + bitmap_git->pack->num_objects;
788                 if (eindex->objects[i]->type == type &&
789                     bitmap_get(to_filter, pos) &&
790                     !bitmap_get(tips, pos))
791                         bitmap_unset(to_filter, pos);
792         }
793
794         bitmap_free(tips);
795 }
796
797 static void filter_bitmap_blob_none(struct bitmap_index *bitmap_git,
798                                     struct object_list *tip_objects,
799                                     struct bitmap *to_filter)
800 {
801         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
802                                    OBJ_BLOB);
803 }
804
805 static unsigned long get_size_by_pos(struct bitmap_index *bitmap_git,
806                                      uint32_t pos)
807 {
808         struct packed_git *pack = bitmap_git->pack;
809         unsigned long size;
810         struct object_info oi = OBJECT_INFO_INIT;
811
812         oi.sizep = &size;
813
814         if (pos < pack->num_objects) {
815                 struct revindex_entry *entry = &pack->revindex[pos];
816                 if (packed_object_info(the_repository, pack,
817                                        entry->offset, &oi) < 0) {
818                         struct object_id oid;
819                         nth_packed_object_id(&oid, pack, entry->nr);
820                         die(_("unable to get size of %s"), oid_to_hex(&oid));
821                 }
822         } else {
823                 struct eindex *eindex = &bitmap_git->ext_index;
824                 struct object *obj = eindex->objects[pos - pack->num_objects];
825                 if (oid_object_info_extended(the_repository, &obj->oid, &oi, 0) < 0)
826                         die(_("unable to get size of %s"), oid_to_hex(&obj->oid));
827         }
828
829         return size;
830 }
831
832 static void filter_bitmap_blob_limit(struct bitmap_index *bitmap_git,
833                                      struct object_list *tip_objects,
834                                      struct bitmap *to_filter,
835                                      unsigned long limit)
836 {
837         struct eindex *eindex = &bitmap_git->ext_index;
838         struct bitmap *tips;
839         struct ewah_iterator it;
840         eword_t mask;
841         uint32_t i;
842
843         tips = find_tip_objects(bitmap_git, tip_objects, OBJ_BLOB);
844
845         for (i = 0, init_type_iterator(&it, bitmap_git, OBJ_BLOB);
846              i < to_filter->word_alloc && ewah_iterator_next(&mask, &it);
847              i++) {
848                 eword_t word = to_filter->words[i] & mask;
849                 unsigned offset;
850
851                 for (offset = 0; offset < BITS_IN_EWORD; offset++) {
852                         uint32_t pos;
853
854                         if ((word >> offset) == 0)
855                                 break;
856                         offset += ewah_bit_ctz64(word >> offset);
857                         pos = i * BITS_IN_EWORD + offset;
858
859                         if (!bitmap_get(tips, pos) &&
860                             get_size_by_pos(bitmap_git, pos) >= limit)
861                                 bitmap_unset(to_filter, pos);
862                 }
863         }
864
865         for (i = 0; i < eindex->count; i++) {
866                 uint32_t pos = i + bitmap_git->pack->num_objects;
867                 if (eindex->objects[i]->type == OBJ_BLOB &&
868                     bitmap_get(to_filter, pos) &&
869                     !bitmap_get(tips, pos) &&
870                     get_size_by_pos(bitmap_git, pos) >= limit)
871                         bitmap_unset(to_filter, pos);
872         }
873
874         bitmap_free(tips);
875 }
876
877 static void filter_bitmap_tree_depth(struct bitmap_index *bitmap_git,
878                                      struct object_list *tip_objects,
879                                      struct bitmap *to_filter,
880                                      unsigned long limit)
881 {
882         if (limit)
883                 BUG("filter_bitmap_tree_depth given non-zero limit");
884
885         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
886                                    OBJ_TREE);
887         filter_bitmap_exclude_type(bitmap_git, tip_objects, to_filter,
888                                    OBJ_BLOB);
889 }
890
891 static int filter_bitmap(struct bitmap_index *bitmap_git,
892                          struct object_list *tip_objects,
893                          struct bitmap *to_filter,
894                          struct list_objects_filter_options *filter)
895 {
896         if (!filter || filter->choice == LOFC_DISABLED)
897                 return 0;
898
899         if (filter->choice == LOFC_BLOB_NONE) {
900                 if (bitmap_git)
901                         filter_bitmap_blob_none(bitmap_git, tip_objects,
902                                                 to_filter);
903                 return 0;
904         }
905
906         if (filter->choice == LOFC_BLOB_LIMIT) {
907                 if (bitmap_git)
908                         filter_bitmap_blob_limit(bitmap_git, tip_objects,
909                                                  to_filter,
910                                                  filter->blob_limit_value);
911                 return 0;
912         }
913
914         if (filter->choice == LOFC_TREE_DEPTH &&
915             filter->tree_exclude_depth == 0) {
916                 if (bitmap_git)
917                         filter_bitmap_tree_depth(bitmap_git, tip_objects,
918                                                  to_filter,
919                                                  filter->tree_exclude_depth);
920                 return 0;
921         }
922
923         /* filter choice not handled */
924         return -1;
925 }
926
927 static int can_filter_bitmap(struct list_objects_filter_options *filter)
928 {
929         return !filter_bitmap(NULL, NULL, NULL, filter);
930 }
931
932 struct bitmap_index *prepare_bitmap_walk(struct rev_info *revs,
933                                          struct list_objects_filter_options *filter)
934 {
935         unsigned int i;
936
937         struct object_list *wants = NULL;
938         struct object_list *haves = NULL;
939
940         struct bitmap *wants_bitmap = NULL;
941         struct bitmap *haves_bitmap = NULL;
942
943         struct bitmap_index *bitmap_git;
944
945         /*
946          * We can't do pathspec limiting with bitmaps, because we don't know
947          * which commits are associated with which object changes (let alone
948          * even which objects are associated with which paths).
949          */
950         if (revs->prune)
951                 return NULL;
952
953         if (!can_filter_bitmap(filter))
954                 return NULL;
955
956         /* try to open a bitmapped pack, but don't parse it yet
957          * because we may not need to use it */
958         bitmap_git = xcalloc(1, sizeof(*bitmap_git));
959         if (open_pack_bitmap(revs->repo, bitmap_git) < 0)
960                 goto cleanup;
961
962         for (i = 0; i < revs->pending.nr; ++i) {
963                 struct object *object = revs->pending.objects[i].item;
964
965                 if (object->type == OBJ_NONE)
966                         parse_object_or_die(&object->oid, NULL);
967
968                 while (object->type == OBJ_TAG) {
969                         struct tag *tag = (struct tag *) object;
970
971                         if (object->flags & UNINTERESTING)
972                                 object_list_insert(object, &haves);
973                         else
974                                 object_list_insert(object, &wants);
975
976                         object = parse_object_or_die(get_tagged_oid(tag), NULL);
977                 }
978
979                 if (object->flags & UNINTERESTING)
980                         object_list_insert(object, &haves);
981                 else
982                         object_list_insert(object, &wants);
983         }
984
985         /*
986          * if we have a HAVES list, but none of those haves is contained
987          * in the packfile that has a bitmap, we don't have anything to
988          * optimize here
989          */
990         if (haves && !in_bitmapped_pack(bitmap_git, haves))
991                 goto cleanup;
992
993         /* if we don't want anything, we're done here */
994         if (!wants)
995                 goto cleanup;
996
997         /*
998          * now we're going to use bitmaps, so load the actual bitmap entries
999          * from disk. this is the point of no return; after this the rev_list
1000          * becomes invalidated and we must perform the revwalk through bitmaps
1001          */
1002         if (load_pack_bitmap(bitmap_git) < 0)
1003                 goto cleanup;
1004
1005         object_array_clear(&revs->pending);
1006
1007         if (haves) {
1008                 revs->ignore_missing_links = 1;
1009                 haves_bitmap = find_objects(bitmap_git, revs, haves, NULL,
1010                                             filter);
1011                 reset_revision_walk();
1012                 revs->ignore_missing_links = 0;
1013
1014                 if (haves_bitmap == NULL)
1015                         BUG("failed to perform bitmap walk");
1016         }
1017
1018         wants_bitmap = find_objects(bitmap_git, revs, wants, haves_bitmap,
1019                                     filter);
1020
1021         if (!wants_bitmap)
1022                 BUG("failed to perform bitmap walk");
1023
1024         if (haves_bitmap)
1025                 bitmap_and_not(wants_bitmap, haves_bitmap);
1026
1027         filter_bitmap(bitmap_git, wants, wants_bitmap, filter);
1028
1029         bitmap_git->result = wants_bitmap;
1030         bitmap_git->haves = haves_bitmap;
1031
1032         object_list_free(&wants);
1033         object_list_free(&haves);
1034
1035         return bitmap_git;
1036
1037 cleanup:
1038         free_bitmap_index(bitmap_git);
1039         object_list_free(&wants);
1040         object_list_free(&haves);
1041         return NULL;
1042 }
1043
1044 static void try_partial_reuse(struct bitmap_index *bitmap_git,
1045                               size_t pos,
1046                               struct bitmap *reuse,
1047                               struct pack_window **w_curs)
1048 {
1049         struct revindex_entry *revidx;
1050         off_t offset;
1051         enum object_type type;
1052         unsigned long size;
1053
1054         if (pos >= bitmap_git->pack->num_objects)
1055                 return; /* not actually in the pack */
1056
1057         revidx = &bitmap_git->pack->revindex[pos];
1058         offset = revidx->offset;
1059         type = unpack_object_header(bitmap_git->pack, w_curs, &offset, &size);
1060         if (type < 0)
1061                 return; /* broken packfile, punt */
1062
1063         if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA) {
1064                 off_t base_offset;
1065                 int base_pos;
1066
1067                 /*
1068                  * Find the position of the base object so we can look it up
1069                  * in our bitmaps. If we can't come up with an offset, or if
1070                  * that offset is not in the revidx, the pack is corrupt.
1071                  * There's nothing we can do, so just punt on this object,
1072                  * and the normal slow path will complain about it in
1073                  * more detail.
1074                  */
1075                 base_offset = get_delta_base(bitmap_git->pack, w_curs,
1076                                              &offset, type, revidx->offset);
1077                 if (!base_offset)
1078                         return;
1079                 base_pos = find_revindex_position(bitmap_git->pack, base_offset);
1080                 if (base_pos < 0)
1081                         return;
1082
1083                 /*
1084                  * We assume delta dependencies always point backwards. This
1085                  * lets us do a single pass, and is basically always true
1086                  * due to the way OFS_DELTAs work. You would not typically
1087                  * find REF_DELTA in a bitmapped pack, since we only bitmap
1088                  * packs we write fresh, and OFS_DELTA is the default). But
1089                  * let's double check to make sure the pack wasn't written with
1090                  * odd parameters.
1091                  */
1092                 if (base_pos >= pos)
1093                         return;
1094
1095                 /*
1096                  * And finally, if we're not sending the base as part of our
1097                  * reuse chunk, then don't send this object either. The base
1098                  * would come after us, along with other objects not
1099                  * necessarily in the pack, which means we'd need to convert
1100                  * to REF_DELTA on the fly. Better to just let the normal
1101                  * object_entry code path handle it.
1102                  */
1103                 if (!bitmap_get(reuse, base_pos))
1104                         return;
1105         }
1106
1107         /*
1108          * If we got here, then the object is OK to reuse. Mark it.
1109          */
1110         bitmap_set(reuse, pos);
1111 }
1112
1113 int reuse_partial_packfile_from_bitmap(struct bitmap_index *bitmap_git,
1114                                        struct packed_git **packfile_out,
1115                                        uint32_t *entries,
1116                                        struct bitmap **reuse_out)
1117 {
1118         struct bitmap *result = bitmap_git->result;
1119         struct bitmap *reuse;
1120         struct pack_window *w_curs = NULL;
1121         size_t i = 0;
1122         uint32_t offset;
1123
1124         assert(result);
1125
1126         while (i < result->word_alloc && result->words[i] == (eword_t)~0)
1127                 i++;
1128
1129         /* Don't mark objects not in the packfile */
1130         if (i > bitmap_git->pack->num_objects / BITS_IN_EWORD)
1131                 i = bitmap_git->pack->num_objects / BITS_IN_EWORD;
1132
1133         reuse = bitmap_word_alloc(i);
1134         memset(reuse->words, 0xFF, i * sizeof(eword_t));
1135
1136         for (; i < result->word_alloc; ++i) {
1137                 eword_t word = result->words[i];
1138                 size_t pos = (i * BITS_IN_EWORD);
1139
1140                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1141                         if ((word >> offset) == 0)
1142                                 break;
1143
1144                         offset += ewah_bit_ctz64(word >> offset);
1145                         try_partial_reuse(bitmap_git, pos + offset, reuse, &w_curs);
1146                 }
1147         }
1148
1149         unuse_pack(&w_curs);
1150
1151         *entries = bitmap_popcount(reuse);
1152         if (!*entries) {
1153                 bitmap_free(reuse);
1154                 return -1;
1155         }
1156
1157         /*
1158          * Drop any reused objects from the result, since they will not
1159          * need to be handled separately.
1160          */
1161         bitmap_and_not(result, reuse);
1162         *packfile_out = bitmap_git->pack;
1163         *reuse_out = reuse;
1164         return 0;
1165 }
1166
1167 int bitmap_walk_contains(struct bitmap_index *bitmap_git,
1168                          struct bitmap *bitmap, const struct object_id *oid)
1169 {
1170         int idx;
1171
1172         if (!bitmap)
1173                 return 0;
1174
1175         idx = bitmap_position(bitmap_git, oid);
1176         return idx >= 0 && bitmap_get(bitmap, idx);
1177 }
1178
1179 void traverse_bitmap_commit_list(struct bitmap_index *bitmap_git,
1180                                  struct rev_info *revs,
1181                                  show_reachable_fn show_reachable)
1182 {
1183         assert(bitmap_git->result);
1184
1185         show_objects_for_type(bitmap_git, OBJ_COMMIT, show_reachable);
1186         if (revs->tree_objects)
1187                 show_objects_for_type(bitmap_git, OBJ_TREE, show_reachable);
1188         if (revs->blob_objects)
1189                 show_objects_for_type(bitmap_git, OBJ_BLOB, show_reachable);
1190         if (revs->tag_objects)
1191                 show_objects_for_type(bitmap_git, OBJ_TAG, show_reachable);
1192
1193         show_extended_objects(bitmap_git, revs, show_reachable);
1194 }
1195
1196 static uint32_t count_object_type(struct bitmap_index *bitmap_git,
1197                                   enum object_type type)
1198 {
1199         struct bitmap *objects = bitmap_git->result;
1200         struct eindex *eindex = &bitmap_git->ext_index;
1201
1202         uint32_t i = 0, count = 0;
1203         struct ewah_iterator it;
1204         eword_t filter;
1205
1206         init_type_iterator(&it, bitmap_git, type);
1207
1208         while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
1209                 eword_t word = objects->words[i++] & filter;
1210                 count += ewah_bit_popcount64(word);
1211         }
1212
1213         for (i = 0; i < eindex->count; ++i) {
1214                 if (eindex->objects[i]->type == type &&
1215                         bitmap_get(objects, bitmap_git->pack->num_objects + i))
1216                         count++;
1217         }
1218
1219         return count;
1220 }
1221
1222 void count_bitmap_commit_list(struct bitmap_index *bitmap_git,
1223                               uint32_t *commits, uint32_t *trees,
1224                               uint32_t *blobs, uint32_t *tags)
1225 {
1226         assert(bitmap_git->result);
1227
1228         if (commits)
1229                 *commits = count_object_type(bitmap_git, OBJ_COMMIT);
1230
1231         if (trees)
1232                 *trees = count_object_type(bitmap_git, OBJ_TREE);
1233
1234         if (blobs)
1235                 *blobs = count_object_type(bitmap_git, OBJ_BLOB);
1236
1237         if (tags)
1238                 *tags = count_object_type(bitmap_git, OBJ_TAG);
1239 }
1240
1241 struct bitmap_test_data {
1242         struct bitmap_index *bitmap_git;
1243         struct bitmap *base;
1244         struct progress *prg;
1245         size_t seen;
1246 };
1247
1248 static void test_show_object(struct object *object, const char *name,
1249                              void *data)
1250 {
1251         struct bitmap_test_data *tdata = data;
1252         int bitmap_pos;
1253
1254         bitmap_pos = bitmap_position(tdata->bitmap_git, &object->oid);
1255         if (bitmap_pos < 0)
1256                 die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
1257
1258         bitmap_set(tdata->base, bitmap_pos);
1259         display_progress(tdata->prg, ++tdata->seen);
1260 }
1261
1262 static void test_show_commit(struct commit *commit, void *data)
1263 {
1264         struct bitmap_test_data *tdata = data;
1265         int bitmap_pos;
1266
1267         bitmap_pos = bitmap_position(tdata->bitmap_git,
1268                                      &commit->object.oid);
1269         if (bitmap_pos < 0)
1270                 die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
1271
1272         bitmap_set(tdata->base, bitmap_pos);
1273         display_progress(tdata->prg, ++tdata->seen);
1274 }
1275
1276 void test_bitmap_walk(struct rev_info *revs)
1277 {
1278         struct object *root;
1279         struct bitmap *result = NULL;
1280         khiter_t pos;
1281         size_t result_popcnt;
1282         struct bitmap_test_data tdata;
1283         struct bitmap_index *bitmap_git;
1284
1285         if (!(bitmap_git = prepare_bitmap_git(revs->repo)))
1286                 die("failed to load bitmap indexes");
1287
1288         if (revs->pending.nr != 1)
1289                 die("you must specify exactly one commit to test");
1290
1291         fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
1292                 bitmap_git->version, bitmap_git->entry_count);
1293
1294         root = revs->pending.objects[0].item;
1295         pos = kh_get_oid_map(bitmap_git->bitmaps, root->oid);
1296
1297         if (pos < kh_end(bitmap_git->bitmaps)) {
1298                 struct stored_bitmap *st = kh_value(bitmap_git->bitmaps, pos);
1299                 struct ewah_bitmap *bm = lookup_stored_bitmap(st);
1300
1301                 fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
1302                         oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
1303
1304                 result = ewah_to_bitmap(bm);
1305         }
1306
1307         if (result == NULL)
1308                 die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
1309
1310         revs->tag_objects = 1;
1311         revs->tree_objects = 1;
1312         revs->blob_objects = 1;
1313
1314         result_popcnt = bitmap_popcount(result);
1315
1316         if (prepare_revision_walk(revs))
1317                 die("revision walk setup failed");
1318
1319         tdata.bitmap_git = bitmap_git;
1320         tdata.base = bitmap_new();
1321         tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
1322         tdata.seen = 0;
1323
1324         traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
1325
1326         stop_progress(&tdata.prg);
1327
1328         if (bitmap_equals(result, tdata.base))
1329                 fprintf(stderr, "OK!\n");
1330         else
1331                 die("mismatch in bitmap results");
1332
1333         free_bitmap_index(bitmap_git);
1334 }
1335
1336 static int rebuild_bitmap(uint32_t *reposition,
1337                           struct ewah_bitmap *source,
1338                           struct bitmap *dest)
1339 {
1340         uint32_t pos = 0;
1341         struct ewah_iterator it;
1342         eword_t word;
1343
1344         ewah_iterator_init(&it, source);
1345
1346         while (ewah_iterator_next(&word, &it)) {
1347                 uint32_t offset, bit_pos;
1348
1349                 for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
1350                         if ((word >> offset) == 0)
1351                                 break;
1352
1353                         offset += ewah_bit_ctz64(word >> offset);
1354
1355                         bit_pos = reposition[pos + offset];
1356                         if (bit_pos > 0)
1357                                 bitmap_set(dest, bit_pos - 1);
1358                         else /* can't reuse, we don't have the object */
1359                                 return -1;
1360                 }
1361
1362                 pos += BITS_IN_EWORD;
1363         }
1364         return 0;
1365 }
1366
1367 int rebuild_existing_bitmaps(struct bitmap_index *bitmap_git,
1368                              struct packing_data *mapping,
1369                              kh_oid_map_t *reused_bitmaps,
1370                              int show_progress)
1371 {
1372         uint32_t i, num_objects;
1373         uint32_t *reposition;
1374         struct bitmap *rebuild;
1375         struct stored_bitmap *stored;
1376         struct progress *progress = NULL;
1377
1378         khiter_t hash_pos;
1379         int hash_ret;
1380
1381         num_objects = bitmap_git->pack->num_objects;
1382         reposition = xcalloc(num_objects, sizeof(uint32_t));
1383
1384         for (i = 0; i < num_objects; ++i) {
1385                 struct object_id oid;
1386                 struct revindex_entry *entry;
1387                 struct object_entry *oe;
1388
1389                 entry = &bitmap_git->pack->revindex[i];
1390                 nth_packed_object_id(&oid, bitmap_git->pack, entry->nr);
1391                 oe = packlist_find(mapping, &oid);
1392
1393                 if (oe)
1394                         reposition[i] = oe_in_pack_pos(mapping, oe) + 1;
1395         }
1396
1397         rebuild = bitmap_new();
1398         i = 0;
1399
1400         if (show_progress)
1401                 progress = start_progress("Reusing bitmaps", 0);
1402
1403         kh_foreach_value(bitmap_git->bitmaps, stored, {
1404                 if (stored->flags & BITMAP_FLAG_REUSE) {
1405                         if (!rebuild_bitmap(reposition,
1406                                             lookup_stored_bitmap(stored),
1407                                             rebuild)) {
1408                                 hash_pos = kh_put_oid_map(reused_bitmaps,
1409                                                           stored->oid,
1410                                                           &hash_ret);
1411                                 kh_value(reused_bitmaps, hash_pos) =
1412                                         bitmap_to_ewah(rebuild);
1413                         }
1414                         bitmap_reset(rebuild);
1415                         display_progress(progress, ++i);
1416                 }
1417         });
1418
1419         stop_progress(&progress);
1420
1421         free(reposition);
1422         bitmap_free(rebuild);
1423         return 0;
1424 }
1425
1426 void free_bitmap_index(struct bitmap_index *b)
1427 {
1428         if (!b)
1429                 return;
1430
1431         if (b->map)
1432                 munmap(b->map, b->map_size);
1433         ewah_pool_free(b->commits);
1434         ewah_pool_free(b->trees);
1435         ewah_pool_free(b->blobs);
1436         ewah_pool_free(b->tags);
1437         kh_destroy_oid_map(b->bitmaps);
1438         free(b->ext_index.objects);
1439         free(b->ext_index.hashes);
1440         bitmap_free(b->result);
1441         bitmap_free(b->haves);
1442         free(b);
1443 }
1444
1445 int bitmap_has_oid_in_uninteresting(struct bitmap_index *bitmap_git,
1446                                     const struct object_id *oid)
1447 {
1448         return bitmap_git &&
1449                 bitmap_walk_contains(bitmap_git, bitmap_git->haves, oid);
1450 }