]>
Commit | Line | Data |
---|---|---|
fff42755 VM |
1 | #include "cache.h" |
2 | #include "commit.h" | |
3 | #include "tag.h" | |
4 | #include "diff.h" | |
5 | #include "revision.h" | |
6 | #include "progress.h" | |
7 | #include "list-objects.h" | |
8 | #include "pack.h" | |
9 | #include "pack-bitmap.h" | |
10 | #include "pack-revindex.h" | |
11 | #include "pack-objects.h" | |
12 | ||
13 | /* | |
14 | * An entry on the bitmap index, representing the bitmap for a given | |
15 | * commit. | |
16 | */ | |
17 | struct stored_bitmap { | |
18 | unsigned char sha1[20]; | |
19 | struct ewah_bitmap *root; | |
20 | struct stored_bitmap *xor; | |
21 | int flags; | |
22 | }; | |
23 | ||
24 | /* | |
25 | * The currently active bitmap index. By design, repositories only have | |
26 | * a single bitmap index available (the index for the biggest packfile in | |
27 | * the repository), since bitmap indexes need full closure. | |
28 | * | |
29 | * If there is more than one bitmap index available (e.g. because of alternates), | |
30 | * the active bitmap index is the largest one. | |
31 | */ | |
32 | static struct bitmap_index { | |
33 | /* Packfile to which this bitmap index belongs to */ | |
34 | struct packed_git *pack; | |
35 | ||
36 | /* reverse index for the packfile */ | |
37 | struct pack_revindex *reverse_index; | |
38 | ||
39 | /* | |
40 | * Mark the first `reuse_objects` in the packfile as reused: | |
41 | * they will be sent as-is without using them for repacking | |
42 | * calculations | |
43 | */ | |
44 | uint32_t reuse_objects; | |
45 | ||
46 | /* mmapped buffer of the whole bitmap index */ | |
47 | unsigned char *map; | |
48 | size_t map_size; /* size of the mmaped buffer */ | |
49 | size_t map_pos; /* current position when loading the index */ | |
50 | ||
51 | /* | |
52 | * Type indexes. | |
53 | * | |
54 | * Each bitmap marks which objects in the packfile are of the given | |
55 | * type. This provides type information when yielding the objects from | |
56 | * the packfile during a walk, which allows for better delta bases. | |
57 | */ | |
58 | struct ewah_bitmap *commits; | |
59 | struct ewah_bitmap *trees; | |
60 | struct ewah_bitmap *blobs; | |
61 | struct ewah_bitmap *tags; | |
62 | ||
63 | /* Map from SHA1 -> `stored_bitmap` for all the bitmapped comits */ | |
64 | khash_sha1 *bitmaps; | |
65 | ||
66 | /* Number of bitmapped commits */ | |
67 | uint32_t entry_count; | |
68 | ||
69 | /* | |
70 | * Extended index. | |
71 | * | |
72 | * When trying to perform bitmap operations with objects that are not | |
73 | * packed in `pack`, these objects are added to this "fake index" and | |
74 | * are assumed to appear at the end of the packfile for all operations | |
75 | */ | |
76 | struct eindex { | |
77 | struct object **objects; | |
78 | uint32_t *hashes; | |
79 | uint32_t count, alloc; | |
80 | khash_sha1_pos *positions; | |
81 | } ext_index; | |
82 | ||
83 | /* Bitmap result of the last performed walk */ | |
84 | struct bitmap *result; | |
85 | ||
86 | /* Version of the bitmap index */ | |
87 | unsigned int version; | |
88 | ||
89 | unsigned loaded : 1; | |
90 | ||
91 | } bitmap_git; | |
92 | ||
93 | static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st) | |
94 | { | |
95 | struct ewah_bitmap *parent; | |
96 | struct ewah_bitmap *composed; | |
97 | ||
98 | if (st->xor == NULL) | |
99 | return st->root; | |
100 | ||
101 | composed = ewah_pool_new(); | |
102 | parent = lookup_stored_bitmap(st->xor); | |
103 | ewah_xor(st->root, parent, composed); | |
104 | ||
105 | ewah_pool_free(st->root); | |
106 | st->root = composed; | |
107 | st->xor = NULL; | |
108 | ||
109 | return composed; | |
110 | } | |
111 | ||
112 | /* | |
113 | * Read a bitmap from the current read position on the mmaped | |
114 | * index, and increase the read position accordingly | |
115 | */ | |
116 | static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index) | |
117 | { | |
118 | struct ewah_bitmap *b = ewah_pool_new(); | |
119 | ||
120 | int bitmap_size = ewah_read_mmap(b, | |
121 | index->map + index->map_pos, | |
122 | index->map_size - index->map_pos); | |
123 | ||
124 | if (bitmap_size < 0) { | |
125 | error("Failed to load bitmap index (corrupted?)"); | |
126 | ewah_pool_free(b); | |
127 | return NULL; | |
128 | } | |
129 | ||
130 | index->map_pos += bitmap_size; | |
131 | return b; | |
132 | } | |
133 | ||
134 | static int load_bitmap_header(struct bitmap_index *index) | |
135 | { | |
136 | struct bitmap_disk_header *header = (void *)index->map; | |
137 | ||
138 | if (index->map_size < sizeof(*header) + 20) | |
139 | return error("Corrupted bitmap index (missing header data)"); | |
140 | ||
141 | if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0) | |
142 | return error("Corrupted bitmap index file (wrong header)"); | |
143 | ||
144 | index->version = ntohs(header->version); | |
145 | if (index->version != 1) | |
146 | return error("Unsupported version for bitmap index file (%d)", index->version); | |
147 | ||
148 | /* Parse known bitmap format options */ | |
149 | { | |
150 | uint32_t flags = ntohs(header->options); | |
151 | ||
152 | if ((flags & BITMAP_OPT_FULL_DAG) == 0) | |
153 | return error("Unsupported options for bitmap index file " | |
154 | "(Git requires BITMAP_OPT_FULL_DAG)"); | |
155 | } | |
156 | ||
157 | index->entry_count = ntohl(header->entry_count); | |
158 | index->map_pos += sizeof(*header); | |
159 | return 0; | |
160 | } | |
161 | ||
162 | static struct stored_bitmap *store_bitmap(struct bitmap_index *index, | |
163 | struct ewah_bitmap *root, | |
164 | const unsigned char *sha1, | |
165 | struct stored_bitmap *xor_with, | |
166 | int flags) | |
167 | { | |
168 | struct stored_bitmap *stored; | |
169 | khiter_t hash_pos; | |
170 | int ret; | |
171 | ||
172 | stored = xmalloc(sizeof(struct stored_bitmap)); | |
173 | stored->root = root; | |
174 | stored->xor = xor_with; | |
175 | stored->flags = flags; | |
176 | hashcpy(stored->sha1, sha1); | |
177 | ||
178 | hash_pos = kh_put_sha1(index->bitmaps, stored->sha1, &ret); | |
179 | ||
180 | /* a 0 return code means the insertion succeeded with no changes, | |
181 | * because the SHA1 already existed on the map. this is bad, there | |
182 | * shouldn't be duplicated commits in the index */ | |
183 | if (ret == 0) { | |
184 | error("Duplicate entry in bitmap index: %s", sha1_to_hex(sha1)); | |
185 | return NULL; | |
186 | } | |
187 | ||
188 | kh_value(index->bitmaps, hash_pos) = stored; | |
189 | return stored; | |
190 | } | |
191 | ||
192 | static int load_bitmap_entries_v1(struct bitmap_index *index) | |
193 | { | |
194 | static const size_t MAX_XOR_OFFSET = 160; | |
195 | ||
196 | uint32_t i; | |
197 | struct stored_bitmap **recent_bitmaps; | |
198 | struct bitmap_disk_entry *entry; | |
199 | ||
200 | recent_bitmaps = xcalloc(MAX_XOR_OFFSET, sizeof(struct stored_bitmap)); | |
201 | ||
202 | for (i = 0; i < index->entry_count; ++i) { | |
203 | int xor_offset, flags; | |
204 | struct ewah_bitmap *bitmap = NULL; | |
205 | struct stored_bitmap *xor_bitmap = NULL; | |
206 | uint32_t commit_idx_pos; | |
207 | const unsigned char *sha1; | |
208 | ||
209 | entry = (struct bitmap_disk_entry *)(index->map + index->map_pos); | |
210 | index->map_pos += sizeof(struct bitmap_disk_entry); | |
211 | ||
212 | commit_idx_pos = ntohl(entry->object_pos); | |
213 | sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos); | |
214 | ||
215 | xor_offset = (int)entry->xor_offset; | |
216 | flags = (int)entry->flags; | |
217 | ||
218 | bitmap = read_bitmap_1(index); | |
219 | if (!bitmap) | |
220 | return -1; | |
221 | ||
222 | if (xor_offset > MAX_XOR_OFFSET || xor_offset > i) | |
223 | return error("Corrupted bitmap pack index"); | |
224 | ||
225 | if (xor_offset > 0) { | |
226 | xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET]; | |
227 | ||
228 | if (xor_bitmap == NULL) | |
229 | return error("Invalid XOR offset in bitmap pack index"); | |
230 | } | |
231 | ||
232 | recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap( | |
233 | index, bitmap, sha1, xor_bitmap, flags); | |
234 | } | |
235 | ||
236 | return 0; | |
237 | } | |
238 | ||
239 | static int open_pack_bitmap_1(struct packed_git *packfile) | |
240 | { | |
241 | int fd; | |
242 | struct stat st; | |
243 | char *idx_name; | |
244 | ||
245 | if (open_pack_index(packfile)) | |
246 | return -1; | |
247 | ||
248 | idx_name = pack_bitmap_filename(packfile); | |
249 | fd = git_open_noatime(idx_name); | |
250 | free(idx_name); | |
251 | ||
252 | if (fd < 0) | |
253 | return -1; | |
254 | ||
255 | if (fstat(fd, &st)) { | |
256 | close(fd); | |
257 | return -1; | |
258 | } | |
259 | ||
260 | if (bitmap_git.pack) { | |
261 | warning("ignoring extra bitmap file: %s", packfile->pack_name); | |
262 | close(fd); | |
263 | return -1; | |
264 | } | |
265 | ||
266 | bitmap_git.pack = packfile; | |
267 | bitmap_git.map_size = xsize_t(st.st_size); | |
268 | bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0); | |
269 | bitmap_git.map_pos = 0; | |
270 | close(fd); | |
271 | ||
272 | if (load_bitmap_header(&bitmap_git) < 0) { | |
273 | munmap(bitmap_git.map, bitmap_git.map_size); | |
274 | bitmap_git.map = NULL; | |
275 | bitmap_git.map_size = 0; | |
276 | return -1; | |
277 | } | |
278 | ||
279 | return 0; | |
280 | } | |
281 | ||
282 | static int load_pack_bitmap(void) | |
283 | { | |
284 | assert(bitmap_git.map && !bitmap_git.loaded); | |
285 | ||
286 | bitmap_git.bitmaps = kh_init_sha1(); | |
287 | bitmap_git.ext_index.positions = kh_init_sha1_pos(); | |
288 | bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack); | |
289 | ||
290 | if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) || | |
291 | !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) || | |
292 | !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) || | |
293 | !(bitmap_git.tags = read_bitmap_1(&bitmap_git))) | |
294 | goto failed; | |
295 | ||
296 | if (load_bitmap_entries_v1(&bitmap_git) < 0) | |
297 | goto failed; | |
298 | ||
299 | bitmap_git.loaded = 1; | |
300 | return 0; | |
301 | ||
302 | failed: | |
303 | munmap(bitmap_git.map, bitmap_git.map_size); | |
304 | bitmap_git.map = NULL; | |
305 | bitmap_git.map_size = 0; | |
306 | return -1; | |
307 | } | |
308 | ||
309 | char *pack_bitmap_filename(struct packed_git *p) | |
310 | { | |
311 | char *idx_name; | |
312 | int len; | |
313 | ||
314 | len = strlen(p->pack_name) - strlen(".pack"); | |
315 | idx_name = xmalloc(len + strlen(".bitmap") + 1); | |
316 | ||
317 | memcpy(idx_name, p->pack_name, len); | |
318 | memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1); | |
319 | ||
320 | return idx_name; | |
321 | } | |
322 | ||
323 | static int open_pack_bitmap(void) | |
324 | { | |
325 | struct packed_git *p; | |
326 | int ret = -1; | |
327 | ||
328 | assert(!bitmap_git.map && !bitmap_git.loaded); | |
329 | ||
330 | prepare_packed_git(); | |
331 | for (p = packed_git; p; p = p->next) { | |
332 | if (open_pack_bitmap_1(p) == 0) | |
333 | ret = 0; | |
334 | } | |
335 | ||
336 | return ret; | |
337 | } | |
338 | ||
339 | int prepare_bitmap_git(void) | |
340 | { | |
341 | if (bitmap_git.loaded) | |
342 | return 0; | |
343 | ||
344 | if (!open_pack_bitmap()) | |
345 | return load_pack_bitmap(); | |
346 | ||
347 | return -1; | |
348 | } | |
349 | ||
350 | struct include_data { | |
351 | struct bitmap *base; | |
352 | struct bitmap *seen; | |
353 | }; | |
354 | ||
355 | static inline int bitmap_position_extended(const unsigned char *sha1) | |
356 | { | |
357 | khash_sha1_pos *positions = bitmap_git.ext_index.positions; | |
358 | khiter_t pos = kh_get_sha1_pos(positions, sha1); | |
359 | ||
360 | if (pos < kh_end(positions)) { | |
361 | int bitmap_pos = kh_value(positions, pos); | |
362 | return bitmap_pos + bitmap_git.pack->num_objects; | |
363 | } | |
364 | ||
365 | return -1; | |
366 | } | |
367 | ||
368 | static inline int bitmap_position_packfile(const unsigned char *sha1) | |
369 | { | |
370 | off_t offset = find_pack_entry_one(sha1, bitmap_git.pack); | |
371 | if (!offset) | |
372 | return -1; | |
373 | ||
374 | return find_revindex_position(bitmap_git.reverse_index, offset); | |
375 | } | |
376 | ||
377 | static int bitmap_position(const unsigned char *sha1) | |
378 | { | |
379 | int pos = bitmap_position_packfile(sha1); | |
380 | return (pos >= 0) ? pos : bitmap_position_extended(sha1); | |
381 | } | |
382 | ||
383 | static int ext_index_add_object(struct object *object, const char *name) | |
384 | { | |
385 | struct eindex *eindex = &bitmap_git.ext_index; | |
386 | ||
387 | khiter_t hash_pos; | |
388 | int hash_ret; | |
389 | int bitmap_pos; | |
390 | ||
391 | hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret); | |
392 | if (hash_ret > 0) { | |
393 | if (eindex->count >= eindex->alloc) { | |
394 | eindex->alloc = (eindex->alloc + 16) * 3 / 2; | |
395 | eindex->objects = xrealloc(eindex->objects, | |
396 | eindex->alloc * sizeof(struct object *)); | |
397 | eindex->hashes = xrealloc(eindex->hashes, | |
398 | eindex->alloc * sizeof(uint32_t)); | |
399 | } | |
400 | ||
401 | bitmap_pos = eindex->count; | |
402 | eindex->objects[eindex->count] = object; | |
403 | eindex->hashes[eindex->count] = pack_name_hash(name); | |
404 | kh_value(eindex->positions, hash_pos) = bitmap_pos; | |
405 | eindex->count++; | |
406 | } else { | |
407 | bitmap_pos = kh_value(eindex->positions, hash_pos); | |
408 | } | |
409 | ||
410 | return bitmap_pos + bitmap_git.pack->num_objects; | |
411 | } | |
412 | ||
413 | static void show_object(struct object *object, const struct name_path *path, | |
414 | const char *last, void *data) | |
415 | { | |
416 | struct bitmap *base = data; | |
417 | int bitmap_pos; | |
418 | ||
419 | bitmap_pos = bitmap_position(object->sha1); | |
420 | ||
421 | if (bitmap_pos < 0) { | |
422 | char *name = path_name(path, last); | |
423 | bitmap_pos = ext_index_add_object(object, name); | |
424 | free(name); | |
425 | } | |
426 | ||
427 | bitmap_set(base, bitmap_pos); | |
428 | } | |
429 | ||
430 | static void show_commit(struct commit *commit, void *data) | |
431 | { | |
432 | } | |
433 | ||
434 | static int add_to_include_set(struct include_data *data, | |
435 | const unsigned char *sha1, | |
436 | int bitmap_pos) | |
437 | { | |
438 | khiter_t hash_pos; | |
439 | ||
440 | if (data->seen && bitmap_get(data->seen, bitmap_pos)) | |
441 | return 0; | |
442 | ||
443 | if (bitmap_get(data->base, bitmap_pos)) | |
444 | return 0; | |
445 | ||
446 | hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1); | |
447 | if (hash_pos < kh_end(bitmap_git.bitmaps)) { | |
448 | struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos); | |
449 | bitmap_or_ewah(data->base, lookup_stored_bitmap(st)); | |
450 | return 0; | |
451 | } | |
452 | ||
453 | bitmap_set(data->base, bitmap_pos); | |
454 | return 1; | |
455 | } | |
456 | ||
457 | static int should_include(struct commit *commit, void *_data) | |
458 | { | |
459 | struct include_data *data = _data; | |
460 | int bitmap_pos; | |
461 | ||
462 | bitmap_pos = bitmap_position(commit->object.sha1); | |
463 | if (bitmap_pos < 0) | |
464 | bitmap_pos = ext_index_add_object((struct object *)commit, NULL); | |
465 | ||
466 | if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) { | |
467 | struct commit_list *parent = commit->parents; | |
468 | ||
469 | while (parent) { | |
470 | parent->item->object.flags |= SEEN; | |
471 | parent = parent->next; | |
472 | } | |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | return 1; | |
478 | } | |
479 | ||
480 | static struct bitmap *find_objects(struct rev_info *revs, | |
481 | struct object_list *roots, | |
482 | struct bitmap *seen) | |
483 | { | |
484 | struct bitmap *base = NULL; | |
485 | int needs_walk = 0; | |
486 | ||
487 | struct object_list *not_mapped = NULL; | |
488 | ||
489 | /* | |
490 | * Go through all the roots for the walk. The ones that have bitmaps | |
491 | * on the bitmap index will be `or`ed together to form an initial | |
492 | * global reachability analysis. | |
493 | * | |
494 | * The ones without bitmaps in the index will be stored in the | |
495 | * `not_mapped_list` for further processing. | |
496 | */ | |
497 | while (roots) { | |
498 | struct object *object = roots->item; | |
499 | roots = roots->next; | |
500 | ||
501 | if (object->type == OBJ_COMMIT) { | |
502 | khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1); | |
503 | ||
504 | if (pos < kh_end(bitmap_git.bitmaps)) { | |
505 | struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos); | |
506 | struct ewah_bitmap *or_with = lookup_stored_bitmap(st); | |
507 | ||
508 | if (base == NULL) | |
509 | base = ewah_to_bitmap(or_with); | |
510 | else | |
511 | bitmap_or_ewah(base, or_with); | |
512 | ||
513 | object->flags |= SEEN; | |
514 | continue; | |
515 | } | |
516 | } | |
517 | ||
518 | object_list_insert(object, ¬_mapped); | |
519 | } | |
520 | ||
521 | /* | |
522 | * Best case scenario: We found bitmaps for all the roots, | |
523 | * so the resulting `or` bitmap has the full reachability analysis | |
524 | */ | |
525 | if (not_mapped == NULL) | |
526 | return base; | |
527 | ||
528 | roots = not_mapped; | |
529 | ||
530 | /* | |
531 | * Let's iterate through all the roots that don't have bitmaps to | |
532 | * check if we can determine them to be reachable from the existing | |
533 | * global bitmap. | |
534 | * | |
535 | * If we cannot find them in the existing global bitmap, we'll need | |
536 | * to push them to an actual walk and run it until we can confirm | |
537 | * they are reachable | |
538 | */ | |
539 | while (roots) { | |
540 | struct object *object = roots->item; | |
541 | int pos; | |
542 | ||
543 | roots = roots->next; | |
544 | pos = bitmap_position(object->sha1); | |
545 | ||
546 | if (pos < 0 || base == NULL || !bitmap_get(base, pos)) { | |
547 | object->flags &= ~UNINTERESTING; | |
548 | add_pending_object(revs, object, ""); | |
549 | needs_walk = 1; | |
550 | } else { | |
551 | object->flags |= SEEN; | |
552 | } | |
553 | } | |
554 | ||
555 | if (needs_walk) { | |
556 | struct include_data incdata; | |
557 | ||
558 | if (base == NULL) | |
559 | base = bitmap_new(); | |
560 | ||
561 | incdata.base = base; | |
562 | incdata.seen = seen; | |
563 | ||
564 | revs->include_check = should_include; | |
565 | revs->include_check_data = &incdata; | |
566 | ||
567 | if (prepare_revision_walk(revs)) | |
568 | die("revision walk setup failed"); | |
569 | ||
570 | traverse_commit_list(revs, show_commit, show_object, base); | |
571 | } | |
572 | ||
573 | return base; | |
574 | } | |
575 | ||
576 | static void show_extended_objects(struct bitmap *objects, | |
577 | show_reachable_fn show_reach) | |
578 | { | |
579 | struct eindex *eindex = &bitmap_git.ext_index; | |
580 | uint32_t i; | |
581 | ||
582 | for (i = 0; i < eindex->count; ++i) { | |
583 | struct object *obj; | |
584 | ||
585 | if (!bitmap_get(objects, bitmap_git.pack->num_objects + i)) | |
586 | continue; | |
587 | ||
588 | obj = eindex->objects[i]; | |
589 | show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0); | |
590 | } | |
591 | } | |
592 | ||
593 | static void show_objects_for_type( | |
594 | struct bitmap *objects, | |
595 | struct ewah_bitmap *type_filter, | |
596 | enum object_type object_type, | |
597 | show_reachable_fn show_reach) | |
598 | { | |
599 | size_t pos = 0, i = 0; | |
600 | uint32_t offset; | |
601 | ||
602 | struct ewah_iterator it; | |
603 | eword_t filter; | |
604 | ||
605 | if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects) | |
606 | return; | |
607 | ||
608 | ewah_iterator_init(&it, type_filter); | |
609 | ||
610 | while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) { | |
611 | eword_t word = objects->words[i] & filter; | |
612 | ||
613 | for (offset = 0; offset < BITS_IN_WORD; ++offset) { | |
614 | const unsigned char *sha1; | |
615 | struct revindex_entry *entry; | |
616 | uint32_t hash = 0; | |
617 | ||
618 | if ((word >> offset) == 0) | |
619 | break; | |
620 | ||
621 | offset += ewah_bit_ctz64(word >> offset); | |
622 | ||
623 | if (pos + offset < bitmap_git.reuse_objects) | |
624 | continue; | |
625 | ||
626 | entry = &bitmap_git.reverse_index->revindex[pos + offset]; | |
627 | sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr); | |
628 | ||
629 | show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset); | |
630 | } | |
631 | ||
632 | pos += BITS_IN_WORD; | |
633 | i++; | |
634 | } | |
635 | } | |
636 | ||
637 | static int in_bitmapped_pack(struct object_list *roots) | |
638 | { | |
639 | while (roots) { | |
640 | struct object *object = roots->item; | |
641 | roots = roots->next; | |
642 | ||
643 | if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0) | |
644 | return 1; | |
645 | } | |
646 | ||
647 | return 0; | |
648 | } | |
649 | ||
650 | int prepare_bitmap_walk(struct rev_info *revs) | |
651 | { | |
652 | unsigned int i; | |
653 | unsigned int pending_nr = revs->pending.nr; | |
654 | struct object_array_entry *pending_e = revs->pending.objects; | |
655 | ||
656 | struct object_list *wants = NULL; | |
657 | struct object_list *haves = NULL; | |
658 | ||
659 | struct bitmap *wants_bitmap = NULL; | |
660 | struct bitmap *haves_bitmap = NULL; | |
661 | ||
662 | if (!bitmap_git.loaded) { | |
663 | /* try to open a bitmapped pack, but don't parse it yet | |
664 | * because we may not need to use it */ | |
665 | if (open_pack_bitmap() < 0) | |
666 | return -1; | |
667 | } | |
668 | ||
669 | for (i = 0; i < pending_nr; ++i) { | |
670 | struct object *object = pending_e[i].item; | |
671 | ||
672 | if (object->type == OBJ_NONE) | |
673 | parse_object_or_die(object->sha1, NULL); | |
674 | ||
675 | while (object->type == OBJ_TAG) { | |
676 | struct tag *tag = (struct tag *) object; | |
677 | ||
678 | if (object->flags & UNINTERESTING) | |
679 | object_list_insert(object, &haves); | |
680 | else | |
681 | object_list_insert(object, &wants); | |
682 | ||
683 | if (!tag->tagged) | |
684 | die("bad tag"); | |
685 | object = parse_object_or_die(tag->tagged->sha1, NULL); | |
686 | } | |
687 | ||
688 | if (object->flags & UNINTERESTING) | |
689 | object_list_insert(object, &haves); | |
690 | else | |
691 | object_list_insert(object, &wants); | |
692 | } | |
693 | ||
694 | /* | |
695 | * if we have a HAVES list, but none of those haves is contained | |
696 | * in the packfile that has a bitmap, we don't have anything to | |
697 | * optimize here | |
698 | */ | |
699 | if (haves && !in_bitmapped_pack(haves)) | |
700 | return -1; | |
701 | ||
702 | /* if we don't want anything, we're done here */ | |
703 | if (!wants) | |
704 | return -1; | |
705 | ||
706 | /* | |
707 | * now we're going to use bitmaps, so load the actual bitmap entries | |
708 | * from disk. this is the point of no return; after this the rev_list | |
709 | * becomes invalidated and we must perform the revwalk through bitmaps | |
710 | */ | |
711 | if (!bitmap_git.loaded && load_pack_bitmap() < 0) | |
712 | return -1; | |
713 | ||
714 | revs->pending.nr = 0; | |
715 | revs->pending.alloc = 0; | |
716 | revs->pending.objects = NULL; | |
717 | ||
718 | if (haves) { | |
719 | haves_bitmap = find_objects(revs, haves, NULL); | |
720 | reset_revision_walk(); | |
721 | ||
722 | if (haves_bitmap == NULL) | |
723 | die("BUG: failed to perform bitmap walk"); | |
724 | } | |
725 | ||
726 | wants_bitmap = find_objects(revs, wants, haves_bitmap); | |
727 | ||
728 | if (!wants_bitmap) | |
729 | die("BUG: failed to perform bitmap walk"); | |
730 | ||
731 | if (haves_bitmap) | |
732 | bitmap_and_not(wants_bitmap, haves_bitmap); | |
733 | ||
734 | bitmap_git.result = wants_bitmap; | |
735 | ||
736 | bitmap_free(haves_bitmap); | |
737 | return 0; | |
738 | } | |
739 | ||
740 | int reuse_partial_packfile_from_bitmap(struct packed_git **packfile, | |
741 | uint32_t *entries, | |
742 | off_t *up_to) | |
743 | { | |
744 | /* | |
745 | * Reuse the packfile content if we need more than | |
746 | * 90% of its objects | |
747 | */ | |
748 | static const double REUSE_PERCENT = 0.9; | |
749 | ||
750 | struct bitmap *result = bitmap_git.result; | |
751 | uint32_t reuse_threshold; | |
752 | uint32_t i, reuse_objects = 0; | |
753 | ||
754 | assert(result); | |
755 | ||
756 | for (i = 0; i < result->word_alloc; ++i) { | |
757 | if (result->words[i] != (eword_t)~0) { | |
758 | reuse_objects += ewah_bit_ctz64(~result->words[i]); | |
759 | break; | |
760 | } | |
761 | ||
762 | reuse_objects += BITS_IN_WORD; | |
763 | } | |
764 | ||
765 | #ifdef GIT_BITMAP_DEBUG | |
766 | { | |
767 | const unsigned char *sha1; | |
768 | struct revindex_entry *entry; | |
769 | ||
770 | entry = &bitmap_git.reverse_index->revindex[reuse_objects]; | |
771 | sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr); | |
772 | ||
773 | fprintf(stderr, "Failed to reuse at %d (%016llx)\n", | |
774 | reuse_objects, result->words[i]); | |
775 | fprintf(stderr, " %s\n", sha1_to_hex(sha1)); | |
776 | } | |
777 | #endif | |
778 | ||
779 | if (!reuse_objects) | |
780 | return -1; | |
781 | ||
782 | if (reuse_objects >= bitmap_git.pack->num_objects) { | |
783 | bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects; | |
784 | *up_to = -1; /* reuse the full pack */ | |
785 | *packfile = bitmap_git.pack; | |
786 | return 0; | |
787 | } | |
788 | ||
789 | reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT; | |
790 | ||
791 | if (reuse_objects < reuse_threshold) | |
792 | return -1; | |
793 | ||
794 | bitmap_git.reuse_objects = *entries = reuse_objects; | |
795 | *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset; | |
796 | *packfile = bitmap_git.pack; | |
797 | ||
798 | return 0; | |
799 | } | |
800 | ||
801 | void traverse_bitmap_commit_list(show_reachable_fn show_reachable) | |
802 | { | |
803 | assert(bitmap_git.result); | |
804 | ||
805 | show_objects_for_type(bitmap_git.result, bitmap_git.commits, | |
806 | OBJ_COMMIT, show_reachable); | |
807 | show_objects_for_type(bitmap_git.result, bitmap_git.trees, | |
808 | OBJ_TREE, show_reachable); | |
809 | show_objects_for_type(bitmap_git.result, bitmap_git.blobs, | |
810 | OBJ_BLOB, show_reachable); | |
811 | show_objects_for_type(bitmap_git.result, bitmap_git.tags, | |
812 | OBJ_TAG, show_reachable); | |
813 | ||
814 | show_extended_objects(bitmap_git.result, show_reachable); | |
815 | ||
816 | bitmap_free(bitmap_git.result); | |
817 | bitmap_git.result = NULL; | |
818 | } | |
819 | ||
820 | static uint32_t count_object_type(struct bitmap *objects, | |
821 | enum object_type type) | |
822 | { | |
823 | struct eindex *eindex = &bitmap_git.ext_index; | |
824 | ||
825 | uint32_t i = 0, count = 0; | |
826 | struct ewah_iterator it; | |
827 | eword_t filter; | |
828 | ||
829 | switch (type) { | |
830 | case OBJ_COMMIT: | |
831 | ewah_iterator_init(&it, bitmap_git.commits); | |
832 | break; | |
833 | ||
834 | case OBJ_TREE: | |
835 | ewah_iterator_init(&it, bitmap_git.trees); | |
836 | break; | |
837 | ||
838 | case OBJ_BLOB: | |
839 | ewah_iterator_init(&it, bitmap_git.blobs); | |
840 | break; | |
841 | ||
842 | case OBJ_TAG: | |
843 | ewah_iterator_init(&it, bitmap_git.tags); | |
844 | break; | |
845 | ||
846 | default: | |
847 | return 0; | |
848 | } | |
849 | ||
850 | while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) { | |
851 | eword_t word = objects->words[i++] & filter; | |
852 | count += ewah_bit_popcount64(word); | |
853 | } | |
854 | ||
855 | for (i = 0; i < eindex->count; ++i) { | |
856 | if (eindex->objects[i]->type == type && | |
857 | bitmap_get(objects, bitmap_git.pack->num_objects + i)) | |
858 | count++; | |
859 | } | |
860 | ||
861 | return count; | |
862 | } | |
863 | ||
864 | void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees, | |
865 | uint32_t *blobs, uint32_t *tags) | |
866 | { | |
867 | assert(bitmap_git.result); | |
868 | ||
869 | if (commits) | |
870 | *commits = count_object_type(bitmap_git.result, OBJ_COMMIT); | |
871 | ||
872 | if (trees) | |
873 | *trees = count_object_type(bitmap_git.result, OBJ_TREE); | |
874 | ||
875 | if (blobs) | |
876 | *blobs = count_object_type(bitmap_git.result, OBJ_BLOB); | |
877 | ||
878 | if (tags) | |
879 | *tags = count_object_type(bitmap_git.result, OBJ_TAG); | |
880 | } | |
881 | ||
882 | struct bitmap_test_data { | |
883 | struct bitmap *base; | |
884 | struct progress *prg; | |
885 | size_t seen; | |
886 | }; | |
887 | ||
888 | static void test_show_object(struct object *object, | |
889 | const struct name_path *path, | |
890 | const char *last, void *data) | |
891 | { | |
892 | struct bitmap_test_data *tdata = data; | |
893 | int bitmap_pos; | |
894 | ||
895 | bitmap_pos = bitmap_position(object->sha1); | |
896 | if (bitmap_pos < 0) | |
897 | die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1)); | |
898 | ||
899 | bitmap_set(tdata->base, bitmap_pos); | |
900 | display_progress(tdata->prg, ++tdata->seen); | |
901 | } | |
902 | ||
903 | static void test_show_commit(struct commit *commit, void *data) | |
904 | { | |
905 | struct bitmap_test_data *tdata = data; | |
906 | int bitmap_pos; | |
907 | ||
908 | bitmap_pos = bitmap_position(commit->object.sha1); | |
909 | if (bitmap_pos < 0) | |
910 | die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1)); | |
911 | ||
912 | bitmap_set(tdata->base, bitmap_pos); | |
913 | display_progress(tdata->prg, ++tdata->seen); | |
914 | } | |
915 | ||
916 | void test_bitmap_walk(struct rev_info *revs) | |
917 | { | |
918 | struct object *root; | |
919 | struct bitmap *result = NULL; | |
920 | khiter_t pos; | |
921 | size_t result_popcnt; | |
922 | struct bitmap_test_data tdata; | |
923 | ||
924 | if (prepare_bitmap_git()) | |
925 | die("failed to load bitmap indexes"); | |
926 | ||
927 | if (revs->pending.nr != 1) | |
928 | die("you must specify exactly one commit to test"); | |
929 | ||
930 | fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n", | |
931 | bitmap_git.version, bitmap_git.entry_count); | |
932 | ||
933 | root = revs->pending.objects[0].item; | |
934 | pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1); | |
935 | ||
936 | if (pos < kh_end(bitmap_git.bitmaps)) { | |
937 | struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos); | |
938 | struct ewah_bitmap *bm = lookup_stored_bitmap(st); | |
939 | ||
940 | fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n", | |
941 | sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm)); | |
942 | ||
943 | result = ewah_to_bitmap(bm); | |
944 | } | |
945 | ||
946 | if (result == NULL) | |
947 | die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1)); | |
948 | ||
949 | revs->tag_objects = 1; | |
950 | revs->tree_objects = 1; | |
951 | revs->blob_objects = 1; | |
952 | ||
953 | result_popcnt = bitmap_popcount(result); | |
954 | ||
955 | if (prepare_revision_walk(revs)) | |
956 | die("revision walk setup failed"); | |
957 | ||
958 | tdata.base = bitmap_new(); | |
959 | tdata.prg = start_progress("Verifying bitmap entries", result_popcnt); | |
960 | tdata.seen = 0; | |
961 | ||
962 | traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata); | |
963 | ||
964 | stop_progress(&tdata.prg); | |
965 | ||
966 | if (bitmap_equals(result, tdata.base)) | |
967 | fprintf(stderr, "OK!\n"); | |
968 | else | |
969 | fprintf(stderr, "Mismatch!\n"); | |
970 | } |