]> git.ipfire.org Git - thirdparty/git.git/blob - midx.c
Merge git-gui into ml/git-gui-exec-path-fix
[thirdparty/git.git] / midx.c
1 #include "git-compat-util.h"
2 #include "abspath.h"
3 #include "config.h"
4 #include "csum-file.h"
5 #include "dir.h"
6 #include "gettext.h"
7 #include "hex.h"
8 #include "lockfile.h"
9 #include "packfile.h"
10 #include "object-file.h"
11 #include "object-store-ll.h"
12 #include "hash-lookup.h"
13 #include "midx.h"
14 #include "progress.h"
15 #include "trace2.h"
16 #include "run-command.h"
17 #include "repository.h"
18 #include "chunk-format.h"
19 #include "pack.h"
20 #include "pack-bitmap.h"
21 #include "refs.h"
22 #include "revision.h"
23 #include "list-objects.h"
24
25 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
26 #define MIDX_VERSION 1
27 #define MIDX_BYTE_FILE_VERSION 4
28 #define MIDX_BYTE_HASH_VERSION 5
29 #define MIDX_BYTE_NUM_CHUNKS 6
30 #define MIDX_BYTE_NUM_PACKS 8
31 #define MIDX_HEADER_SIZE 12
32 #define MIDX_MIN_SIZE (MIDX_HEADER_SIZE + the_hash_algo->rawsz)
33
34 #define MIDX_CHUNK_ALIGNMENT 4
35 #define MIDX_CHUNKID_PACKNAMES 0x504e414d /* "PNAM" */
36 #define MIDX_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
37 #define MIDX_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
38 #define MIDX_CHUNKID_OBJECTOFFSETS 0x4f4f4646 /* "OOFF" */
39 #define MIDX_CHUNKID_LARGEOFFSETS 0x4c4f4646 /* "LOFF" */
40 #define MIDX_CHUNKID_REVINDEX 0x52494458 /* "RIDX" */
41 #define MIDX_CHUNK_FANOUT_SIZE (sizeof(uint32_t) * 256)
42 #define MIDX_CHUNK_OFFSET_WIDTH (2 * sizeof(uint32_t))
43 #define MIDX_CHUNK_LARGE_OFFSET_WIDTH (sizeof(uint64_t))
44 #define MIDX_LARGE_OFFSET_NEEDED 0x80000000
45
46 #define PACK_EXPIRED UINT_MAX
47
48 const unsigned char *get_midx_checksum(struct multi_pack_index *m)
49 {
50 return m->data + m->data_len - the_hash_algo->rawsz;
51 }
52
53 void get_midx_filename(struct strbuf *out, const char *object_dir)
54 {
55 strbuf_addf(out, "%s/pack/multi-pack-index", object_dir);
56 }
57
58 void get_midx_rev_filename(struct strbuf *out, struct multi_pack_index *m)
59 {
60 get_midx_filename(out, m->object_dir);
61 strbuf_addf(out, "-%s.rev", hash_to_hex(get_midx_checksum(m)));
62 }
63
64 static int midx_read_oid_fanout(const unsigned char *chunk_start,
65 size_t chunk_size, void *data)
66 {
67 struct multi_pack_index *m = data;
68 m->chunk_oid_fanout = (uint32_t *)chunk_start;
69
70 if (chunk_size != 4 * 256) {
71 error(_("multi-pack-index OID fanout is of the wrong size"));
72 return 1;
73 }
74 return 0;
75 }
76
77 struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
78 {
79 struct multi_pack_index *m = NULL;
80 int fd;
81 struct stat st;
82 size_t midx_size;
83 void *midx_map = NULL;
84 uint32_t hash_version;
85 struct strbuf midx_name = STRBUF_INIT;
86 uint32_t i;
87 const char *cur_pack_name;
88 struct chunkfile *cf = NULL;
89
90 get_midx_filename(&midx_name, object_dir);
91
92 fd = git_open(midx_name.buf);
93
94 if (fd < 0)
95 goto cleanup_fail;
96 if (fstat(fd, &st)) {
97 error_errno(_("failed to read %s"), midx_name.buf);
98 goto cleanup_fail;
99 }
100
101 midx_size = xsize_t(st.st_size);
102
103 if (midx_size < MIDX_MIN_SIZE) {
104 error(_("multi-pack-index file %s is too small"), midx_name.buf);
105 goto cleanup_fail;
106 }
107
108 strbuf_release(&midx_name);
109
110 midx_map = xmmap(NULL, midx_size, PROT_READ, MAP_PRIVATE, fd, 0);
111 close(fd);
112
113 FLEX_ALLOC_STR(m, object_dir, object_dir);
114 m->data = midx_map;
115 m->data_len = midx_size;
116 m->local = local;
117
118 m->signature = get_be32(m->data);
119 if (m->signature != MIDX_SIGNATURE)
120 die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
121 m->signature, MIDX_SIGNATURE);
122
123 m->version = m->data[MIDX_BYTE_FILE_VERSION];
124 if (m->version != MIDX_VERSION)
125 die(_("multi-pack-index version %d not recognized"),
126 m->version);
127
128 hash_version = m->data[MIDX_BYTE_HASH_VERSION];
129 if (hash_version != oid_version(the_hash_algo)) {
130 error(_("multi-pack-index hash version %u does not match version %u"),
131 hash_version, oid_version(the_hash_algo));
132 goto cleanup_fail;
133 }
134 m->hash_len = the_hash_algo->rawsz;
135
136 m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
137
138 m->num_packs = get_be32(m->data + MIDX_BYTE_NUM_PACKS);
139
140 cf = init_chunkfile(NULL);
141
142 if (read_table_of_contents(cf, m->data, midx_size,
143 MIDX_HEADER_SIZE, m->num_chunks))
144 goto cleanup_fail;
145
146 if (pair_chunk(cf, MIDX_CHUNKID_PACKNAMES, &m->chunk_pack_names) == CHUNK_NOT_FOUND)
147 die(_("multi-pack-index missing required pack-name chunk"));
148 if (read_chunk(cf, MIDX_CHUNKID_OIDFANOUT, midx_read_oid_fanout, m) == CHUNK_NOT_FOUND)
149 die(_("multi-pack-index missing required OID fanout chunk"));
150 if (pair_chunk(cf, MIDX_CHUNKID_OIDLOOKUP, &m->chunk_oid_lookup) == CHUNK_NOT_FOUND)
151 die(_("multi-pack-index missing required OID lookup chunk"));
152 if (pair_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS, &m->chunk_object_offsets) == CHUNK_NOT_FOUND)
153 die(_("multi-pack-index missing required object offsets chunk"));
154
155 pair_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS, &m->chunk_large_offsets);
156
157 if (git_env_bool("GIT_TEST_MIDX_READ_RIDX", 1))
158 pair_chunk(cf, MIDX_CHUNKID_REVINDEX, &m->chunk_revindex);
159
160 m->num_objects = ntohl(m->chunk_oid_fanout[255]);
161
162 CALLOC_ARRAY(m->pack_names, m->num_packs);
163 CALLOC_ARRAY(m->packs, m->num_packs);
164
165 cur_pack_name = (const char *)m->chunk_pack_names;
166 for (i = 0; i < m->num_packs; i++) {
167 m->pack_names[i] = cur_pack_name;
168
169 cur_pack_name += strlen(cur_pack_name) + 1;
170
171 if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
172 die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
173 m->pack_names[i - 1],
174 m->pack_names[i]);
175 }
176
177 trace2_data_intmax("midx", the_repository, "load/num_packs", m->num_packs);
178 trace2_data_intmax("midx", the_repository, "load/num_objects", m->num_objects);
179
180 free_chunkfile(cf);
181 return m;
182
183 cleanup_fail:
184 free(m);
185 strbuf_release(&midx_name);
186 free_chunkfile(cf);
187 if (midx_map)
188 munmap(midx_map, midx_size);
189 if (0 <= fd)
190 close(fd);
191 return NULL;
192 }
193
194 void close_midx(struct multi_pack_index *m)
195 {
196 uint32_t i;
197
198 if (!m)
199 return;
200
201 close_midx(m->next);
202
203 munmap((unsigned char *)m->data, m->data_len);
204
205 for (i = 0; i < m->num_packs; i++) {
206 if (m->packs[i])
207 m->packs[i]->multi_pack_index = 0;
208 }
209 FREE_AND_NULL(m->packs);
210 FREE_AND_NULL(m->pack_names);
211 free(m);
212 }
213
214 int prepare_midx_pack(struct repository *r, struct multi_pack_index *m, uint32_t pack_int_id)
215 {
216 struct strbuf pack_name = STRBUF_INIT;
217 struct packed_git *p;
218
219 if (pack_int_id >= m->num_packs)
220 die(_("bad pack-int-id: %u (%u total packs)"),
221 pack_int_id, m->num_packs);
222
223 if (m->packs[pack_int_id])
224 return 0;
225
226 strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
227 m->pack_names[pack_int_id]);
228
229 p = add_packed_git(pack_name.buf, pack_name.len, m->local);
230 strbuf_release(&pack_name);
231
232 if (!p)
233 return 1;
234
235 p->multi_pack_index = 1;
236 m->packs[pack_int_id] = p;
237 install_packed_git(r, p);
238 list_add_tail(&p->mru, &r->objects->packed_git_mru);
239
240 return 0;
241 }
242
243 int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
244 {
245 return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
246 the_hash_algo->rawsz, result);
247 }
248
249 struct object_id *nth_midxed_object_oid(struct object_id *oid,
250 struct multi_pack_index *m,
251 uint32_t n)
252 {
253 if (n >= m->num_objects)
254 return NULL;
255
256 oidread(oid, m->chunk_oid_lookup + st_mult(m->hash_len, n));
257 return oid;
258 }
259
260 off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
261 {
262 const unsigned char *offset_data;
263 uint32_t offset32;
264
265 offset_data = m->chunk_object_offsets + (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH;
266 offset32 = get_be32(offset_data + sizeof(uint32_t));
267
268 if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
269 if (sizeof(off_t) < sizeof(uint64_t))
270 die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
271
272 offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
273 return get_be64(m->chunk_large_offsets +
274 st_mult(sizeof(uint64_t), offset32));
275 }
276
277 return offset32;
278 }
279
280 uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
281 {
282 return get_be32(m->chunk_object_offsets +
283 (off_t)pos * MIDX_CHUNK_OFFSET_WIDTH);
284 }
285
286 int fill_midx_entry(struct repository *r,
287 const struct object_id *oid,
288 struct pack_entry *e,
289 struct multi_pack_index *m)
290 {
291 uint32_t pos;
292 uint32_t pack_int_id;
293 struct packed_git *p;
294
295 if (!bsearch_midx(oid, m, &pos))
296 return 0;
297
298 if (pos >= m->num_objects)
299 return 0;
300
301 pack_int_id = nth_midxed_pack_int_id(m, pos);
302
303 if (prepare_midx_pack(r, m, pack_int_id))
304 return 0;
305 p = m->packs[pack_int_id];
306
307 /*
308 * We are about to tell the caller where they can locate the
309 * requested object. We better make sure the packfile is
310 * still here and can be accessed before supplying that
311 * answer, as it may have been deleted since the MIDX was
312 * loaded!
313 */
314 if (!is_pack_valid(p))
315 return 0;
316
317 if (oidset_size(&p->bad_objects) &&
318 oidset_contains(&p->bad_objects, oid))
319 return 0;
320
321 e->offset = nth_midxed_offset(m, pos);
322 e->p = p;
323
324 return 1;
325 }
326
327 /* Match "foo.idx" against either "foo.pack" _or_ "foo.idx". */
328 static int cmp_idx_or_pack_name(const char *idx_or_pack_name,
329 const char *idx_name)
330 {
331 /* Skip past any initial matching prefix. */
332 while (*idx_name && *idx_name == *idx_or_pack_name) {
333 idx_name++;
334 idx_or_pack_name++;
335 }
336
337 /*
338 * If we didn't match completely, we may have matched "pack-1234." and
339 * be left with "idx" and "pack" respectively, which is also OK. We do
340 * not have to check for "idx" and "idx", because that would have been
341 * a complete match (and in that case these strcmps will be false, but
342 * we'll correctly return 0 from the final strcmp() below.
343 *
344 * Technically this matches "fooidx" and "foopack", but we'd never have
345 * such names in the first place.
346 */
347 if (!strcmp(idx_name, "idx") && !strcmp(idx_or_pack_name, "pack"))
348 return 0;
349
350 /*
351 * This not only checks for a complete match, but also orders based on
352 * the first non-identical character, which means our ordering will
353 * match a raw strcmp(). That makes it OK to use this to binary search
354 * a naively-sorted list.
355 */
356 return strcmp(idx_or_pack_name, idx_name);
357 }
358
359 int midx_contains_pack(struct multi_pack_index *m, const char *idx_or_pack_name)
360 {
361 uint32_t first = 0, last = m->num_packs;
362
363 while (first < last) {
364 uint32_t mid = first + (last - first) / 2;
365 const char *current;
366 int cmp;
367
368 current = m->pack_names[mid];
369 cmp = cmp_idx_or_pack_name(idx_or_pack_name, current);
370 if (!cmp)
371 return 1;
372 if (cmp > 0) {
373 first = mid + 1;
374 continue;
375 }
376 last = mid;
377 }
378
379 return 0;
380 }
381
382 int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
383 {
384 struct multi_pack_index *m;
385 struct multi_pack_index *m_search;
386
387 prepare_repo_settings(r);
388 if (!r->settings.core_multi_pack_index)
389 return 0;
390
391 for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
392 if (!strcmp(object_dir, m_search->object_dir))
393 return 1;
394
395 m = load_multi_pack_index(object_dir, local);
396
397 if (m) {
398 struct multi_pack_index *mp = r->objects->multi_pack_index;
399 if (mp) {
400 m->next = mp->next;
401 mp->next = m;
402 } else
403 r->objects->multi_pack_index = m;
404 return 1;
405 }
406
407 return 0;
408 }
409
410 static size_t write_midx_header(struct hashfile *f,
411 unsigned char num_chunks,
412 uint32_t num_packs)
413 {
414 hashwrite_be32(f, MIDX_SIGNATURE);
415 hashwrite_u8(f, MIDX_VERSION);
416 hashwrite_u8(f, oid_version(the_hash_algo));
417 hashwrite_u8(f, num_chunks);
418 hashwrite_u8(f, 0); /* unused */
419 hashwrite_be32(f, num_packs);
420
421 return MIDX_HEADER_SIZE;
422 }
423
424 struct pack_info {
425 uint32_t orig_pack_int_id;
426 char *pack_name;
427 struct packed_git *p;
428 unsigned expired : 1;
429 };
430
431 static int pack_info_compare(const void *_a, const void *_b)
432 {
433 struct pack_info *a = (struct pack_info *)_a;
434 struct pack_info *b = (struct pack_info *)_b;
435 return strcmp(a->pack_name, b->pack_name);
436 }
437
438 static int idx_or_pack_name_cmp(const void *_va, const void *_vb)
439 {
440 const char *pack_name = _va;
441 const struct pack_info *compar = _vb;
442
443 return cmp_idx_or_pack_name(pack_name, compar->pack_name);
444 }
445
446 struct write_midx_context {
447 struct pack_info *info;
448 size_t nr;
449 size_t alloc;
450 struct multi_pack_index *m;
451 struct progress *progress;
452 unsigned pack_paths_checked;
453
454 struct pack_midx_entry *entries;
455 size_t entries_nr;
456
457 uint32_t *pack_perm;
458 uint32_t *pack_order;
459 unsigned large_offsets_needed:1;
460 uint32_t num_large_offsets;
461
462 int preferred_pack_idx;
463
464 struct string_list *to_include;
465 };
466
467 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
468 const char *file_name, void *data)
469 {
470 struct write_midx_context *ctx = data;
471
472 if (ends_with(file_name, ".idx")) {
473 display_progress(ctx->progress, ++ctx->pack_paths_checked);
474 /*
475 * Note that at most one of ctx->m and ctx->to_include are set,
476 * so we are testing midx_contains_pack() and
477 * string_list_has_string() independently (guarded by the
478 * appropriate NULL checks).
479 *
480 * We could support passing to_include while reusing an existing
481 * MIDX, but don't currently since the reuse process drags
482 * forward all packs from an existing MIDX (without checking
483 * whether or not they appear in the to_include list).
484 *
485 * If we added support for that, these next two conditional
486 * should be performed independently (likely checking
487 * to_include before the existing MIDX).
488 */
489 if (ctx->m && midx_contains_pack(ctx->m, file_name))
490 return;
491 else if (ctx->to_include &&
492 !string_list_has_string(ctx->to_include, file_name))
493 return;
494
495 ALLOC_GROW(ctx->info, ctx->nr + 1, ctx->alloc);
496
497 ctx->info[ctx->nr].p = add_packed_git(full_path,
498 full_path_len,
499 0);
500
501 if (!ctx->info[ctx->nr].p) {
502 warning(_("failed to add packfile '%s'"),
503 full_path);
504 return;
505 }
506
507 if (open_pack_index(ctx->info[ctx->nr].p)) {
508 warning(_("failed to open pack-index '%s'"),
509 full_path);
510 close_pack(ctx->info[ctx->nr].p);
511 FREE_AND_NULL(ctx->info[ctx->nr].p);
512 return;
513 }
514
515 ctx->info[ctx->nr].pack_name = xstrdup(file_name);
516 ctx->info[ctx->nr].orig_pack_int_id = ctx->nr;
517 ctx->info[ctx->nr].expired = 0;
518 ctx->nr++;
519 }
520 }
521
522 struct pack_midx_entry {
523 struct object_id oid;
524 uint32_t pack_int_id;
525 time_t pack_mtime;
526 uint64_t offset;
527 unsigned preferred : 1;
528 };
529
530 static int midx_oid_compare(const void *_a, const void *_b)
531 {
532 const struct pack_midx_entry *a = (const struct pack_midx_entry *)_a;
533 const struct pack_midx_entry *b = (const struct pack_midx_entry *)_b;
534 int cmp = oidcmp(&a->oid, &b->oid);
535
536 if (cmp)
537 return cmp;
538
539 /* Sort objects in a preferred pack first when multiple copies exist. */
540 if (a->preferred > b->preferred)
541 return -1;
542 if (a->preferred < b->preferred)
543 return 1;
544
545 if (a->pack_mtime > b->pack_mtime)
546 return -1;
547 else if (a->pack_mtime < b->pack_mtime)
548 return 1;
549
550 return a->pack_int_id - b->pack_int_id;
551 }
552
553 static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
554 struct pack_midx_entry *e,
555 uint32_t pos)
556 {
557 if (pos >= m->num_objects)
558 return 1;
559
560 nth_midxed_object_oid(&e->oid, m, pos);
561 e->pack_int_id = nth_midxed_pack_int_id(m, pos);
562 e->offset = nth_midxed_offset(m, pos);
563
564 /* consider objects in midx to be from "old" packs */
565 e->pack_mtime = 0;
566 return 0;
567 }
568
569 static void fill_pack_entry(uint32_t pack_int_id,
570 struct packed_git *p,
571 uint32_t cur_object,
572 struct pack_midx_entry *entry,
573 int preferred)
574 {
575 if (nth_packed_object_id(&entry->oid, p, cur_object) < 0)
576 die(_("failed to locate object %d in packfile"), cur_object);
577
578 entry->pack_int_id = pack_int_id;
579 entry->pack_mtime = p->mtime;
580
581 entry->offset = nth_packed_object_offset(p, cur_object);
582 entry->preferred = !!preferred;
583 }
584
585 struct midx_fanout {
586 struct pack_midx_entry *entries;
587 size_t nr, alloc;
588 };
589
590 static void midx_fanout_grow(struct midx_fanout *fanout, size_t nr)
591 {
592 if (nr < fanout->nr)
593 BUG("negative growth in midx_fanout_grow() (%"PRIuMAX" < %"PRIuMAX")",
594 (uintmax_t)nr, (uintmax_t)fanout->nr);
595 ALLOC_GROW(fanout->entries, nr, fanout->alloc);
596 }
597
598 static void midx_fanout_sort(struct midx_fanout *fanout)
599 {
600 QSORT(fanout->entries, fanout->nr, midx_oid_compare);
601 }
602
603 static void midx_fanout_add_midx_fanout(struct midx_fanout *fanout,
604 struct multi_pack_index *m,
605 uint32_t cur_fanout,
606 int preferred_pack)
607 {
608 uint32_t start = 0, end;
609 uint32_t cur_object;
610
611 if (cur_fanout)
612 start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
613 end = ntohl(m->chunk_oid_fanout[cur_fanout]);
614
615 for (cur_object = start; cur_object < end; cur_object++) {
616 if ((preferred_pack > -1) &&
617 (preferred_pack == nth_midxed_pack_int_id(m, cur_object))) {
618 /*
619 * Objects from preferred packs are added
620 * separately.
621 */
622 continue;
623 }
624
625 midx_fanout_grow(fanout, fanout->nr + 1);
626 nth_midxed_pack_midx_entry(m,
627 &fanout->entries[fanout->nr],
628 cur_object);
629 fanout->entries[fanout->nr].preferred = 0;
630 fanout->nr++;
631 }
632 }
633
634 static void midx_fanout_add_pack_fanout(struct midx_fanout *fanout,
635 struct pack_info *info,
636 uint32_t cur_pack,
637 int preferred,
638 uint32_t cur_fanout)
639 {
640 struct packed_git *pack = info[cur_pack].p;
641 uint32_t start = 0, end;
642 uint32_t cur_object;
643
644 if (cur_fanout)
645 start = get_pack_fanout(pack, cur_fanout - 1);
646 end = get_pack_fanout(pack, cur_fanout);
647
648 for (cur_object = start; cur_object < end; cur_object++) {
649 midx_fanout_grow(fanout, fanout->nr + 1);
650 fill_pack_entry(cur_pack,
651 info[cur_pack].p,
652 cur_object,
653 &fanout->entries[fanout->nr],
654 preferred);
655 fanout->nr++;
656 }
657 }
658
659 /*
660 * It is possible to artificially get into a state where there are many
661 * duplicate copies of objects. That can create high memory pressure if
662 * we are to create a list of all objects before de-duplication. To reduce
663 * this memory pressure without a significant performance drop, automatically
664 * group objects by the first byte of their object id. Use the IDX fanout
665 * tables to group the data, copy to a local array, then sort.
666 *
667 * Copy only the de-duplicated entries (selected by most-recent modified time
668 * of a packfile containing the object).
669 */
670 static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
671 struct pack_info *info,
672 uint32_t nr_packs,
673 size_t *nr_objects,
674 int preferred_pack)
675 {
676 uint32_t cur_fanout, cur_pack, cur_object;
677 size_t alloc_objects, total_objects = 0;
678 struct midx_fanout fanout = { 0 };
679 struct pack_midx_entry *deduplicated_entries = NULL;
680 uint32_t start_pack = m ? m->num_packs : 0;
681
682 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
683 total_objects = st_add(total_objects,
684 info[cur_pack].p->num_objects);
685
686 /*
687 * As we de-duplicate by fanout value, we expect the fanout
688 * slices to be evenly distributed, with some noise. Hence,
689 * allocate slightly more than one 256th.
690 */
691 alloc_objects = fanout.alloc = total_objects > 3200 ? total_objects / 200 : 16;
692
693 ALLOC_ARRAY(fanout.entries, fanout.alloc);
694 ALLOC_ARRAY(deduplicated_entries, alloc_objects);
695 *nr_objects = 0;
696
697 for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
698 fanout.nr = 0;
699
700 if (m)
701 midx_fanout_add_midx_fanout(&fanout, m, cur_fanout,
702 preferred_pack);
703
704 for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
705 int preferred = cur_pack == preferred_pack;
706 midx_fanout_add_pack_fanout(&fanout,
707 info, cur_pack,
708 preferred, cur_fanout);
709 }
710
711 if (-1 < preferred_pack && preferred_pack < start_pack)
712 midx_fanout_add_pack_fanout(&fanout, info,
713 preferred_pack, 1,
714 cur_fanout);
715
716 midx_fanout_sort(&fanout);
717
718 /*
719 * The batch is now sorted by OID and then mtime (descending).
720 * Take only the first duplicate.
721 */
722 for (cur_object = 0; cur_object < fanout.nr; cur_object++) {
723 if (cur_object && oideq(&fanout.entries[cur_object - 1].oid,
724 &fanout.entries[cur_object].oid))
725 continue;
726
727 ALLOC_GROW(deduplicated_entries, st_add(*nr_objects, 1),
728 alloc_objects);
729 memcpy(&deduplicated_entries[*nr_objects],
730 &fanout.entries[cur_object],
731 sizeof(struct pack_midx_entry));
732 (*nr_objects)++;
733 }
734 }
735
736 free(fanout.entries);
737 return deduplicated_entries;
738 }
739
740 static int write_midx_pack_names(struct hashfile *f, void *data)
741 {
742 struct write_midx_context *ctx = data;
743 uint32_t i;
744 unsigned char padding[MIDX_CHUNK_ALIGNMENT];
745 size_t written = 0;
746
747 for (i = 0; i < ctx->nr; i++) {
748 size_t writelen;
749
750 if (ctx->info[i].expired)
751 continue;
752
753 if (i && strcmp(ctx->info[i].pack_name, ctx->info[i - 1].pack_name) <= 0)
754 BUG("incorrect pack-file order: %s before %s",
755 ctx->info[i - 1].pack_name,
756 ctx->info[i].pack_name);
757
758 writelen = strlen(ctx->info[i].pack_name) + 1;
759 hashwrite(f, ctx->info[i].pack_name, writelen);
760 written += writelen;
761 }
762
763 /* add padding to be aligned */
764 i = MIDX_CHUNK_ALIGNMENT - (written % MIDX_CHUNK_ALIGNMENT);
765 if (i < MIDX_CHUNK_ALIGNMENT) {
766 memset(padding, 0, sizeof(padding));
767 hashwrite(f, padding, i);
768 }
769
770 return 0;
771 }
772
773 static int write_midx_oid_fanout(struct hashfile *f,
774 void *data)
775 {
776 struct write_midx_context *ctx = data;
777 struct pack_midx_entry *list = ctx->entries;
778 struct pack_midx_entry *last = ctx->entries + ctx->entries_nr;
779 uint32_t count = 0;
780 uint32_t i;
781
782 /*
783 * Write the first-level table (the list is sorted,
784 * but we use a 256-entry lookup to be able to avoid
785 * having to do eight extra binary search iterations).
786 */
787 for (i = 0; i < 256; i++) {
788 struct pack_midx_entry *next = list;
789
790 while (next < last && next->oid.hash[0] == i) {
791 count++;
792 next++;
793 }
794
795 hashwrite_be32(f, count);
796 list = next;
797 }
798
799 return 0;
800 }
801
802 static int write_midx_oid_lookup(struct hashfile *f,
803 void *data)
804 {
805 struct write_midx_context *ctx = data;
806 unsigned char hash_len = the_hash_algo->rawsz;
807 struct pack_midx_entry *list = ctx->entries;
808 uint32_t i;
809
810 for (i = 0; i < ctx->entries_nr; i++) {
811 struct pack_midx_entry *obj = list++;
812
813 if (i < ctx->entries_nr - 1) {
814 struct pack_midx_entry *next = list;
815 if (oidcmp(&obj->oid, &next->oid) >= 0)
816 BUG("OIDs not in order: %s >= %s",
817 oid_to_hex(&obj->oid),
818 oid_to_hex(&next->oid));
819 }
820
821 hashwrite(f, obj->oid.hash, (int)hash_len);
822 }
823
824 return 0;
825 }
826
827 static int write_midx_object_offsets(struct hashfile *f,
828 void *data)
829 {
830 struct write_midx_context *ctx = data;
831 struct pack_midx_entry *list = ctx->entries;
832 uint32_t i, nr_large_offset = 0;
833
834 for (i = 0; i < ctx->entries_nr; i++) {
835 struct pack_midx_entry *obj = list++;
836
837 if (ctx->pack_perm[obj->pack_int_id] == PACK_EXPIRED)
838 BUG("object %s is in an expired pack with int-id %d",
839 oid_to_hex(&obj->oid),
840 obj->pack_int_id);
841
842 hashwrite_be32(f, ctx->pack_perm[obj->pack_int_id]);
843
844 if (ctx->large_offsets_needed && obj->offset >> 31)
845 hashwrite_be32(f, MIDX_LARGE_OFFSET_NEEDED | nr_large_offset++);
846 else if (!ctx->large_offsets_needed && obj->offset >> 32)
847 BUG("object %s requires a large offset (%"PRIx64") but the MIDX is not writing large offsets!",
848 oid_to_hex(&obj->oid),
849 obj->offset);
850 else
851 hashwrite_be32(f, (uint32_t)obj->offset);
852 }
853
854 return 0;
855 }
856
857 static int write_midx_large_offsets(struct hashfile *f,
858 void *data)
859 {
860 struct write_midx_context *ctx = data;
861 struct pack_midx_entry *list = ctx->entries;
862 struct pack_midx_entry *end = ctx->entries + ctx->entries_nr;
863 uint32_t nr_large_offset = ctx->num_large_offsets;
864
865 while (nr_large_offset) {
866 struct pack_midx_entry *obj;
867 uint64_t offset;
868
869 if (list >= end)
870 BUG("too many large-offset objects");
871
872 obj = list++;
873 offset = obj->offset;
874
875 if (!(offset >> 31))
876 continue;
877
878 hashwrite_be64(f, offset);
879
880 nr_large_offset--;
881 }
882
883 return 0;
884 }
885
886 static int write_midx_revindex(struct hashfile *f,
887 void *data)
888 {
889 struct write_midx_context *ctx = data;
890 uint32_t i;
891
892 for (i = 0; i < ctx->entries_nr; i++)
893 hashwrite_be32(f, ctx->pack_order[i]);
894
895 return 0;
896 }
897
898 struct midx_pack_order_data {
899 uint32_t nr;
900 uint32_t pack;
901 off_t offset;
902 };
903
904 static int midx_pack_order_cmp(const void *va, const void *vb)
905 {
906 const struct midx_pack_order_data *a = va, *b = vb;
907 if (a->pack < b->pack)
908 return -1;
909 else if (a->pack > b->pack)
910 return 1;
911 else if (a->offset < b->offset)
912 return -1;
913 else if (a->offset > b->offset)
914 return 1;
915 else
916 return 0;
917 }
918
919 static uint32_t *midx_pack_order(struct write_midx_context *ctx)
920 {
921 struct midx_pack_order_data *data;
922 uint32_t *pack_order;
923 uint32_t i;
924
925 trace2_region_enter("midx", "midx_pack_order", the_repository);
926
927 ALLOC_ARRAY(data, ctx->entries_nr);
928 for (i = 0; i < ctx->entries_nr; i++) {
929 struct pack_midx_entry *e = &ctx->entries[i];
930 data[i].nr = i;
931 data[i].pack = ctx->pack_perm[e->pack_int_id];
932 if (!e->preferred)
933 data[i].pack |= (1U << 31);
934 data[i].offset = e->offset;
935 }
936
937 QSORT(data, ctx->entries_nr, midx_pack_order_cmp);
938
939 ALLOC_ARRAY(pack_order, ctx->entries_nr);
940 for (i = 0; i < ctx->entries_nr; i++)
941 pack_order[i] = data[i].nr;
942 free(data);
943
944 trace2_region_leave("midx", "midx_pack_order", the_repository);
945
946 return pack_order;
947 }
948
949 static void write_midx_reverse_index(char *midx_name, unsigned char *midx_hash,
950 struct write_midx_context *ctx)
951 {
952 struct strbuf buf = STRBUF_INIT;
953 const char *tmp_file;
954
955 trace2_region_enter("midx", "write_midx_reverse_index", the_repository);
956
957 strbuf_addf(&buf, "%s-%s.rev", midx_name, hash_to_hex(midx_hash));
958
959 tmp_file = write_rev_file_order(NULL, ctx->pack_order, ctx->entries_nr,
960 midx_hash, WRITE_REV);
961
962 if (finalize_object_file(tmp_file, buf.buf))
963 die(_("cannot store reverse index file"));
964
965 strbuf_release(&buf);
966
967 trace2_region_leave("midx", "write_midx_reverse_index", the_repository);
968 }
969
970 static void clear_midx_files_ext(const char *object_dir, const char *ext,
971 unsigned char *keep_hash);
972
973 static int midx_checksum_valid(struct multi_pack_index *m)
974 {
975 return hashfile_checksum_valid(m->data, m->data_len);
976 }
977
978 static void prepare_midx_packing_data(struct packing_data *pdata,
979 struct write_midx_context *ctx)
980 {
981 uint32_t i;
982
983 trace2_region_enter("midx", "prepare_midx_packing_data", the_repository);
984
985 memset(pdata, 0, sizeof(struct packing_data));
986 prepare_packing_data(the_repository, pdata);
987
988 for (i = 0; i < ctx->entries_nr; i++) {
989 struct pack_midx_entry *from = &ctx->entries[ctx->pack_order[i]];
990 struct object_entry *to = packlist_alloc(pdata, &from->oid);
991
992 oe_set_in_pack(pdata, to,
993 ctx->info[ctx->pack_perm[from->pack_int_id]].p);
994 }
995
996 trace2_region_leave("midx", "prepare_midx_packing_data", the_repository);
997 }
998
999 static int add_ref_to_pending(const char *refname,
1000 const struct object_id *oid,
1001 int flag, void *cb_data)
1002 {
1003 struct rev_info *revs = (struct rev_info*)cb_data;
1004 struct object_id peeled;
1005 struct object *object;
1006
1007 if ((flag & REF_ISSYMREF) && (flag & REF_ISBROKEN)) {
1008 warning("symbolic ref is dangling: %s", refname);
1009 return 0;
1010 }
1011
1012 if (!peel_iterated_oid(oid, &peeled))
1013 oid = &peeled;
1014
1015 object = parse_object_or_die(oid, refname);
1016 if (object->type != OBJ_COMMIT)
1017 return 0;
1018
1019 add_pending_object(revs, object, "");
1020 if (bitmap_is_preferred_refname(revs->repo, refname))
1021 object->flags |= NEEDS_BITMAP;
1022 return 0;
1023 }
1024
1025 struct bitmap_commit_cb {
1026 struct commit **commits;
1027 size_t commits_nr, commits_alloc;
1028
1029 struct write_midx_context *ctx;
1030 };
1031
1032 static const struct object_id *bitmap_oid_access(size_t index,
1033 const void *_entries)
1034 {
1035 const struct pack_midx_entry *entries = _entries;
1036 return &entries[index].oid;
1037 }
1038
1039 static void bitmap_show_commit(struct commit *commit, void *_data)
1040 {
1041 struct bitmap_commit_cb *data = _data;
1042 int pos = oid_pos(&commit->object.oid, data->ctx->entries,
1043 data->ctx->entries_nr,
1044 bitmap_oid_access);
1045 if (pos < 0)
1046 return;
1047
1048 ALLOC_GROW(data->commits, data->commits_nr + 1, data->commits_alloc);
1049 data->commits[data->commits_nr++] = commit;
1050 }
1051
1052 static int read_refs_snapshot(const char *refs_snapshot,
1053 struct rev_info *revs)
1054 {
1055 struct strbuf buf = STRBUF_INIT;
1056 struct object_id oid;
1057 FILE *f = xfopen(refs_snapshot, "r");
1058
1059 while (strbuf_getline(&buf, f) != EOF) {
1060 struct object *object;
1061 int preferred = 0;
1062 char *hex = buf.buf;
1063 const char *end = NULL;
1064
1065 if (buf.len && *buf.buf == '+') {
1066 preferred = 1;
1067 hex = &buf.buf[1];
1068 }
1069
1070 if (parse_oid_hex(hex, &oid, &end) < 0)
1071 die(_("could not parse line: %s"), buf.buf);
1072 if (*end)
1073 die(_("malformed line: %s"), buf.buf);
1074
1075 object = parse_object_or_die(&oid, NULL);
1076 if (preferred)
1077 object->flags |= NEEDS_BITMAP;
1078
1079 add_pending_object(revs, object, "");
1080 }
1081
1082 fclose(f);
1083 strbuf_release(&buf);
1084 return 0;
1085 }
1086
1087 static struct commit **find_commits_for_midx_bitmap(uint32_t *indexed_commits_nr_p,
1088 const char *refs_snapshot,
1089 struct write_midx_context *ctx)
1090 {
1091 struct rev_info revs;
1092 struct bitmap_commit_cb cb = {0};
1093
1094 trace2_region_enter("midx", "find_commits_for_midx_bitmap",
1095 the_repository);
1096
1097 cb.ctx = ctx;
1098
1099 repo_init_revisions(the_repository, &revs, NULL);
1100 if (refs_snapshot) {
1101 read_refs_snapshot(refs_snapshot, &revs);
1102 } else {
1103 setup_revisions(0, NULL, &revs, NULL);
1104 for_each_ref(add_ref_to_pending, &revs);
1105 }
1106
1107 /*
1108 * Skipping promisor objects here is intentional, since it only excludes
1109 * them from the list of reachable commits that we want to select from
1110 * when computing the selection of MIDX'd commits to receive bitmaps.
1111 *
1112 * Reachability bitmaps do require that their objects be closed under
1113 * reachability, but fetching any objects missing from promisors at this
1114 * point is too late. But, if one of those objects can be reached from
1115 * an another object that is included in the bitmap, then we will
1116 * complain later that we don't have reachability closure (and fail
1117 * appropriately).
1118 */
1119 fetch_if_missing = 0;
1120 revs.exclude_promisor_objects = 1;
1121
1122 if (prepare_revision_walk(&revs))
1123 die(_("revision walk setup failed"));
1124
1125 traverse_commit_list(&revs, bitmap_show_commit, NULL, &cb);
1126 if (indexed_commits_nr_p)
1127 *indexed_commits_nr_p = cb.commits_nr;
1128
1129 release_revisions(&revs);
1130
1131 trace2_region_leave("midx", "find_commits_for_midx_bitmap",
1132 the_repository);
1133
1134 return cb.commits;
1135 }
1136
1137 static int write_midx_bitmap(const char *midx_name,
1138 const unsigned char *midx_hash,
1139 struct packing_data *pdata,
1140 struct commit **commits,
1141 uint32_t commits_nr,
1142 uint32_t *pack_order,
1143 unsigned flags)
1144 {
1145 int ret, i;
1146 uint16_t options = 0;
1147 struct pack_idx_entry **index;
1148 char *bitmap_name = xstrfmt("%s-%s.bitmap", midx_name,
1149 hash_to_hex(midx_hash));
1150
1151 trace2_region_enter("midx", "write_midx_bitmap", the_repository);
1152
1153 if (flags & MIDX_WRITE_BITMAP_HASH_CACHE)
1154 options |= BITMAP_OPT_HASH_CACHE;
1155
1156 if (flags & MIDX_WRITE_BITMAP_LOOKUP_TABLE)
1157 options |= BITMAP_OPT_LOOKUP_TABLE;
1158
1159 /*
1160 * Build the MIDX-order index based on pdata.objects (which is already
1161 * in MIDX order; c.f., 'midx_pack_order_cmp()' for the definition of
1162 * this order).
1163 */
1164 ALLOC_ARRAY(index, pdata->nr_objects);
1165 for (i = 0; i < pdata->nr_objects; i++)
1166 index[i] = &pdata->objects[i].idx;
1167
1168 bitmap_writer_show_progress(flags & MIDX_PROGRESS);
1169 bitmap_writer_build_type_index(pdata, index, pdata->nr_objects);
1170
1171 /*
1172 * bitmap_writer_finish expects objects in lex order, but pack_order
1173 * gives us exactly that. use it directly instead of re-sorting the
1174 * array.
1175 *
1176 * This changes the order of objects in 'index' between
1177 * bitmap_writer_build_type_index and bitmap_writer_finish.
1178 *
1179 * The same re-ordering takes place in the single-pack bitmap code via
1180 * write_idx_file(), which is called by finish_tmp_packfile(), which
1181 * happens between bitmap_writer_build_type_index() and
1182 * bitmap_writer_finish().
1183 */
1184 for (i = 0; i < pdata->nr_objects; i++)
1185 index[pack_order[i]] = &pdata->objects[i].idx;
1186
1187 bitmap_writer_select_commits(commits, commits_nr, -1);
1188 ret = bitmap_writer_build(pdata);
1189 if (ret < 0)
1190 goto cleanup;
1191
1192 bitmap_writer_set_checksum(midx_hash);
1193 bitmap_writer_finish(index, pdata->nr_objects, bitmap_name, options);
1194
1195 cleanup:
1196 free(index);
1197 free(bitmap_name);
1198
1199 trace2_region_leave("midx", "write_midx_bitmap", the_repository);
1200
1201 return ret;
1202 }
1203
1204 static struct multi_pack_index *lookup_multi_pack_index(struct repository *r,
1205 const char *object_dir)
1206 {
1207 struct multi_pack_index *result = NULL;
1208 struct multi_pack_index *cur;
1209 char *obj_dir_real = real_pathdup(object_dir, 1);
1210 struct strbuf cur_path_real = STRBUF_INIT;
1211
1212 /* Ensure the given object_dir is local, or a known alternate. */
1213 find_odb(r, obj_dir_real);
1214
1215 for (cur = get_multi_pack_index(r); cur; cur = cur->next) {
1216 strbuf_realpath(&cur_path_real, cur->object_dir, 1);
1217 if (!strcmp(obj_dir_real, cur_path_real.buf)) {
1218 result = cur;
1219 goto cleanup;
1220 }
1221 }
1222
1223 cleanup:
1224 free(obj_dir_real);
1225 strbuf_release(&cur_path_real);
1226 return result;
1227 }
1228
1229 static int write_midx_internal(const char *object_dir,
1230 struct string_list *packs_to_include,
1231 struct string_list *packs_to_drop,
1232 const char *preferred_pack_name,
1233 const char *refs_snapshot,
1234 unsigned flags)
1235 {
1236 struct strbuf midx_name = STRBUF_INIT;
1237 unsigned char midx_hash[GIT_MAX_RAWSZ];
1238 uint32_t i;
1239 struct hashfile *f = NULL;
1240 struct lock_file lk;
1241 struct write_midx_context ctx = { 0 };
1242 int pack_name_concat_len = 0;
1243 int dropped_packs = 0;
1244 int result = 0;
1245 struct chunkfile *cf;
1246
1247 trace2_region_enter("midx", "write_midx_internal", the_repository);
1248
1249 get_midx_filename(&midx_name, object_dir);
1250 if (safe_create_leading_directories(midx_name.buf))
1251 die_errno(_("unable to create leading directories of %s"),
1252 midx_name.buf);
1253
1254 if (!packs_to_include) {
1255 /*
1256 * Only reference an existing MIDX when not filtering which
1257 * packs to include, since all packs and objects are copied
1258 * blindly from an existing MIDX if one is present.
1259 */
1260 ctx.m = lookup_multi_pack_index(the_repository, object_dir);
1261 }
1262
1263 if (ctx.m && !midx_checksum_valid(ctx.m)) {
1264 warning(_("ignoring existing multi-pack-index; checksum mismatch"));
1265 ctx.m = NULL;
1266 }
1267
1268 ctx.nr = 0;
1269 ctx.alloc = ctx.m ? ctx.m->num_packs : 16;
1270 ctx.info = NULL;
1271 ALLOC_ARRAY(ctx.info, ctx.alloc);
1272
1273 if (ctx.m) {
1274 for (i = 0; i < ctx.m->num_packs; i++) {
1275 ALLOC_GROW(ctx.info, ctx.nr + 1, ctx.alloc);
1276
1277 ctx.info[ctx.nr].orig_pack_int_id = i;
1278 ctx.info[ctx.nr].pack_name = xstrdup(ctx.m->pack_names[i]);
1279 ctx.info[ctx.nr].p = ctx.m->packs[i];
1280 ctx.info[ctx.nr].expired = 0;
1281
1282 if (flags & MIDX_WRITE_REV_INDEX) {
1283 /*
1284 * If generating a reverse index, need to have
1285 * packed_git's loaded to compare their
1286 * mtimes and object count.
1287 */
1288 if (prepare_midx_pack(the_repository, ctx.m, i)) {
1289 error(_("could not load pack"));
1290 result = 1;
1291 goto cleanup;
1292 }
1293
1294 if (open_pack_index(ctx.m->packs[i]))
1295 die(_("could not open index for %s"),
1296 ctx.m->packs[i]->pack_name);
1297 ctx.info[ctx.nr].p = ctx.m->packs[i];
1298 }
1299
1300 ctx.nr++;
1301 }
1302 }
1303
1304 ctx.pack_paths_checked = 0;
1305 if (flags & MIDX_PROGRESS)
1306 ctx.progress = start_delayed_progress(_("Adding packfiles to multi-pack-index"), 0);
1307 else
1308 ctx.progress = NULL;
1309
1310 ctx.to_include = packs_to_include;
1311
1312 for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &ctx);
1313 stop_progress(&ctx.progress);
1314
1315 if ((ctx.m && ctx.nr == ctx.m->num_packs) &&
1316 !(packs_to_include || packs_to_drop)) {
1317 struct bitmap_index *bitmap_git;
1318 int bitmap_exists;
1319 int want_bitmap = flags & MIDX_WRITE_BITMAP;
1320
1321 bitmap_git = prepare_midx_bitmap_git(ctx.m);
1322 bitmap_exists = bitmap_git && bitmap_is_midx(bitmap_git);
1323 free_bitmap_index(bitmap_git);
1324
1325 if (bitmap_exists || !want_bitmap) {
1326 /*
1327 * The correct MIDX already exists, and so does a
1328 * corresponding bitmap (or one wasn't requested).
1329 */
1330 if (!want_bitmap)
1331 clear_midx_files_ext(object_dir, ".bitmap",
1332 NULL);
1333 goto cleanup;
1334 }
1335 }
1336
1337 if (preferred_pack_name) {
1338 ctx.preferred_pack_idx = -1;
1339
1340 for (i = 0; i < ctx.nr; i++) {
1341 if (!cmp_idx_or_pack_name(preferred_pack_name,
1342 ctx.info[i].pack_name)) {
1343 ctx.preferred_pack_idx = i;
1344 break;
1345 }
1346 }
1347
1348 if (ctx.preferred_pack_idx == -1)
1349 warning(_("unknown preferred pack: '%s'"),
1350 preferred_pack_name);
1351 } else if (ctx.nr &&
1352 (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP))) {
1353 struct packed_git *oldest = ctx.info[ctx.preferred_pack_idx].p;
1354 ctx.preferred_pack_idx = 0;
1355
1356 if (packs_to_drop && packs_to_drop->nr)
1357 BUG("cannot write a MIDX bitmap during expiration");
1358
1359 /*
1360 * set a preferred pack when writing a bitmap to ensure that
1361 * the pack from which the first object is selected in pseudo
1362 * pack-order has all of its objects selected from that pack
1363 * (and not another pack containing a duplicate)
1364 */
1365 for (i = 1; i < ctx.nr; i++) {
1366 struct packed_git *p = ctx.info[i].p;
1367
1368 if (!oldest->num_objects || p->mtime < oldest->mtime) {
1369 oldest = p;
1370 ctx.preferred_pack_idx = i;
1371 }
1372 }
1373
1374 if (!oldest->num_objects) {
1375 /*
1376 * If all packs are empty; unset the preferred index.
1377 * This is acceptable since there will be no duplicate
1378 * objects to resolve, so the preferred value doesn't
1379 * matter.
1380 */
1381 ctx.preferred_pack_idx = -1;
1382 }
1383 } else {
1384 /*
1385 * otherwise don't mark any pack as preferred to avoid
1386 * interfering with expiration logic below
1387 */
1388 ctx.preferred_pack_idx = -1;
1389 }
1390
1391 if (ctx.preferred_pack_idx > -1) {
1392 struct packed_git *preferred = ctx.info[ctx.preferred_pack_idx].p;
1393 if (!preferred->num_objects) {
1394 error(_("cannot select preferred pack %s with no objects"),
1395 preferred->pack_name);
1396 result = 1;
1397 goto cleanup;
1398 }
1399 }
1400
1401 ctx.entries = get_sorted_entries(ctx.m, ctx.info, ctx.nr, &ctx.entries_nr,
1402 ctx.preferred_pack_idx);
1403
1404 ctx.large_offsets_needed = 0;
1405 for (i = 0; i < ctx.entries_nr; i++) {
1406 if (ctx.entries[i].offset > 0x7fffffff)
1407 ctx.num_large_offsets++;
1408 if (ctx.entries[i].offset > 0xffffffff)
1409 ctx.large_offsets_needed = 1;
1410 }
1411
1412 QSORT(ctx.info, ctx.nr, pack_info_compare);
1413
1414 if (packs_to_drop && packs_to_drop->nr) {
1415 int drop_index = 0;
1416 int missing_drops = 0;
1417
1418 for (i = 0; i < ctx.nr && drop_index < packs_to_drop->nr; i++) {
1419 int cmp = strcmp(ctx.info[i].pack_name,
1420 packs_to_drop->items[drop_index].string);
1421
1422 if (!cmp) {
1423 drop_index++;
1424 ctx.info[i].expired = 1;
1425 } else if (cmp > 0) {
1426 error(_("did not see pack-file %s to drop"),
1427 packs_to_drop->items[drop_index].string);
1428 drop_index++;
1429 missing_drops++;
1430 i--;
1431 } else {
1432 ctx.info[i].expired = 0;
1433 }
1434 }
1435
1436 if (missing_drops) {
1437 result = 1;
1438 goto cleanup;
1439 }
1440 }
1441
1442 /*
1443 * pack_perm stores a permutation between pack-int-ids from the
1444 * previous multi-pack-index to the new one we are writing:
1445 *
1446 * pack_perm[old_id] = new_id
1447 */
1448 ALLOC_ARRAY(ctx.pack_perm, ctx.nr);
1449 for (i = 0; i < ctx.nr; i++) {
1450 if (ctx.info[i].expired) {
1451 dropped_packs++;
1452 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = PACK_EXPIRED;
1453 } else {
1454 ctx.pack_perm[ctx.info[i].orig_pack_int_id] = i - dropped_packs;
1455 }
1456 }
1457
1458 for (i = 0; i < ctx.nr; i++) {
1459 if (!ctx.info[i].expired)
1460 pack_name_concat_len += strlen(ctx.info[i].pack_name) + 1;
1461 }
1462
1463 /* Check that the preferred pack wasn't expired (if given). */
1464 if (preferred_pack_name) {
1465 struct pack_info *preferred = bsearch(preferred_pack_name,
1466 ctx.info, ctx.nr,
1467 sizeof(*ctx.info),
1468 idx_or_pack_name_cmp);
1469 if (preferred) {
1470 uint32_t perm = ctx.pack_perm[preferred->orig_pack_int_id];
1471 if (perm == PACK_EXPIRED)
1472 warning(_("preferred pack '%s' is expired"),
1473 preferred_pack_name);
1474 }
1475 }
1476
1477 if (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
1478 pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
1479 (pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
1480
1481 hold_lock_file_for_update(&lk, midx_name.buf, LOCK_DIE_ON_ERROR);
1482 f = hashfd(get_lock_file_fd(&lk), get_lock_file_path(&lk));
1483
1484 if (ctx.nr - dropped_packs == 0) {
1485 error(_("no pack files to index."));
1486 result = 1;
1487 goto cleanup;
1488 }
1489
1490 if (!ctx.entries_nr) {
1491 if (flags & MIDX_WRITE_BITMAP)
1492 warning(_("refusing to write multi-pack .bitmap without any objects"));
1493 flags &= ~(MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP);
1494 }
1495
1496 cf = init_chunkfile(f);
1497
1498 add_chunk(cf, MIDX_CHUNKID_PACKNAMES, pack_name_concat_len,
1499 write_midx_pack_names);
1500 add_chunk(cf, MIDX_CHUNKID_OIDFANOUT, MIDX_CHUNK_FANOUT_SIZE,
1501 write_midx_oid_fanout);
1502 add_chunk(cf, MIDX_CHUNKID_OIDLOOKUP,
1503 st_mult(ctx.entries_nr, the_hash_algo->rawsz),
1504 write_midx_oid_lookup);
1505 add_chunk(cf, MIDX_CHUNKID_OBJECTOFFSETS,
1506 st_mult(ctx.entries_nr, MIDX_CHUNK_OFFSET_WIDTH),
1507 write_midx_object_offsets);
1508
1509 if (ctx.large_offsets_needed)
1510 add_chunk(cf, MIDX_CHUNKID_LARGEOFFSETS,
1511 st_mult(ctx.num_large_offsets,
1512 MIDX_CHUNK_LARGE_OFFSET_WIDTH),
1513 write_midx_large_offsets);
1514
1515 if (flags & (MIDX_WRITE_REV_INDEX | MIDX_WRITE_BITMAP)) {
1516 ctx.pack_order = midx_pack_order(&ctx);
1517 add_chunk(cf, MIDX_CHUNKID_REVINDEX,
1518 st_mult(ctx.entries_nr, sizeof(uint32_t)),
1519 write_midx_revindex);
1520 }
1521
1522 write_midx_header(f, get_num_chunks(cf), ctx.nr - dropped_packs);
1523 write_chunkfile(cf, &ctx);
1524
1525 finalize_hashfile(f, midx_hash, FSYNC_COMPONENT_PACK_METADATA,
1526 CSUM_FSYNC | CSUM_HASH_IN_STREAM);
1527 free_chunkfile(cf);
1528
1529 if (flags & MIDX_WRITE_REV_INDEX &&
1530 git_env_bool("GIT_TEST_MIDX_WRITE_REV", 0))
1531 write_midx_reverse_index(midx_name.buf, midx_hash, &ctx);
1532
1533 if (flags & MIDX_WRITE_BITMAP) {
1534 struct packing_data pdata;
1535 struct commit **commits;
1536 uint32_t commits_nr;
1537
1538 if (!ctx.entries_nr)
1539 BUG("cannot write a bitmap without any objects");
1540
1541 prepare_midx_packing_data(&pdata, &ctx);
1542
1543 commits = find_commits_for_midx_bitmap(&commits_nr, refs_snapshot, &ctx);
1544
1545 /*
1546 * The previous steps translated the information from
1547 * 'entries' into information suitable for constructing
1548 * bitmaps. We no longer need that array, so clear it to
1549 * reduce memory pressure.
1550 */
1551 FREE_AND_NULL(ctx.entries);
1552 ctx.entries_nr = 0;
1553
1554 if (write_midx_bitmap(midx_name.buf, midx_hash, &pdata,
1555 commits, commits_nr, ctx.pack_order,
1556 flags) < 0) {
1557 error(_("could not write multi-pack bitmap"));
1558 result = 1;
1559 goto cleanup;
1560 }
1561 }
1562 /*
1563 * NOTE: Do not use ctx.entries beyond this point, since it might
1564 * have been freed in the previous if block.
1565 */
1566
1567 if (ctx.m)
1568 close_object_store(the_repository->objects);
1569
1570 if (commit_lock_file(&lk) < 0)
1571 die_errno(_("could not write multi-pack-index"));
1572
1573 clear_midx_files_ext(object_dir, ".bitmap", midx_hash);
1574 clear_midx_files_ext(object_dir, ".rev", midx_hash);
1575
1576 cleanup:
1577 for (i = 0; i < ctx.nr; i++) {
1578 if (ctx.info[i].p) {
1579 close_pack(ctx.info[i].p);
1580 free(ctx.info[i].p);
1581 }
1582 free(ctx.info[i].pack_name);
1583 }
1584
1585 free(ctx.info);
1586 free(ctx.entries);
1587 free(ctx.pack_perm);
1588 free(ctx.pack_order);
1589 strbuf_release(&midx_name);
1590
1591 trace2_region_leave("midx", "write_midx_internal", the_repository);
1592
1593 return result;
1594 }
1595
1596 int write_midx_file(const char *object_dir,
1597 const char *preferred_pack_name,
1598 const char *refs_snapshot,
1599 unsigned flags)
1600 {
1601 return write_midx_internal(object_dir, NULL, NULL, preferred_pack_name,
1602 refs_snapshot, flags);
1603 }
1604
1605 int write_midx_file_only(const char *object_dir,
1606 struct string_list *packs_to_include,
1607 const char *preferred_pack_name,
1608 const char *refs_snapshot,
1609 unsigned flags)
1610 {
1611 return write_midx_internal(object_dir, packs_to_include, NULL,
1612 preferred_pack_name, refs_snapshot, flags);
1613 }
1614
1615 struct clear_midx_data {
1616 char *keep;
1617 const char *ext;
1618 };
1619
1620 static void clear_midx_file_ext(const char *full_path, size_t full_path_len UNUSED,
1621 const char *file_name, void *_data)
1622 {
1623 struct clear_midx_data *data = _data;
1624
1625 if (!(starts_with(file_name, "multi-pack-index-") &&
1626 ends_with(file_name, data->ext)))
1627 return;
1628 if (data->keep && !strcmp(data->keep, file_name))
1629 return;
1630
1631 if (unlink(full_path))
1632 die_errno(_("failed to remove %s"), full_path);
1633 }
1634
1635 static void clear_midx_files_ext(const char *object_dir, const char *ext,
1636 unsigned char *keep_hash)
1637 {
1638 struct clear_midx_data data;
1639 memset(&data, 0, sizeof(struct clear_midx_data));
1640
1641 if (keep_hash)
1642 data.keep = xstrfmt("multi-pack-index-%s%s",
1643 hash_to_hex(keep_hash), ext);
1644 data.ext = ext;
1645
1646 for_each_file_in_pack_dir(object_dir,
1647 clear_midx_file_ext,
1648 &data);
1649
1650 free(data.keep);
1651 }
1652
1653 void clear_midx_file(struct repository *r)
1654 {
1655 struct strbuf midx = STRBUF_INIT;
1656
1657 get_midx_filename(&midx, r->objects->odb->path);
1658
1659 if (r->objects && r->objects->multi_pack_index) {
1660 close_midx(r->objects->multi_pack_index);
1661 r->objects->multi_pack_index = NULL;
1662 }
1663
1664 if (remove_path(midx.buf))
1665 die(_("failed to clear multi-pack-index at %s"), midx.buf);
1666
1667 clear_midx_files_ext(r->objects->odb->path, ".bitmap", NULL);
1668 clear_midx_files_ext(r->objects->odb->path, ".rev", NULL);
1669
1670 strbuf_release(&midx);
1671 }
1672
1673 static int verify_midx_error;
1674
1675 __attribute__((format (printf, 1, 2)))
1676 static void midx_report(const char *fmt, ...)
1677 {
1678 va_list ap;
1679 verify_midx_error = 1;
1680 va_start(ap, fmt);
1681 vfprintf(stderr, fmt, ap);
1682 fprintf(stderr, "\n");
1683 va_end(ap);
1684 }
1685
1686 struct pair_pos_vs_id
1687 {
1688 uint32_t pos;
1689 uint32_t pack_int_id;
1690 };
1691
1692 static int compare_pair_pos_vs_id(const void *_a, const void *_b)
1693 {
1694 struct pair_pos_vs_id *a = (struct pair_pos_vs_id *)_a;
1695 struct pair_pos_vs_id *b = (struct pair_pos_vs_id *)_b;
1696
1697 return b->pack_int_id - a->pack_int_id;
1698 }
1699
1700 /*
1701 * Limit calls to display_progress() for performance reasons.
1702 * The interval here was arbitrarily chosen.
1703 */
1704 #define SPARSE_PROGRESS_INTERVAL (1 << 12)
1705 #define midx_display_sparse_progress(progress, n) \
1706 do { \
1707 uint64_t _n = (n); \
1708 if ((_n & (SPARSE_PROGRESS_INTERVAL - 1)) == 0) \
1709 display_progress(progress, _n); \
1710 } while (0)
1711
1712 int verify_midx_file(struct repository *r, const char *object_dir, unsigned flags)
1713 {
1714 struct pair_pos_vs_id *pairs = NULL;
1715 uint32_t i;
1716 struct progress *progress = NULL;
1717 struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
1718 verify_midx_error = 0;
1719
1720 if (!m) {
1721 int result = 0;
1722 struct stat sb;
1723 struct strbuf filename = STRBUF_INIT;
1724
1725 get_midx_filename(&filename, object_dir);
1726
1727 if (!stat(filename.buf, &sb)) {
1728 error(_("multi-pack-index file exists, but failed to parse"));
1729 result = 1;
1730 }
1731 strbuf_release(&filename);
1732 return result;
1733 }
1734
1735 if (!midx_checksum_valid(m))
1736 midx_report(_("incorrect checksum"));
1737
1738 if (flags & MIDX_PROGRESS)
1739 progress = start_delayed_progress(_("Looking for referenced packfiles"),
1740 m->num_packs);
1741 for (i = 0; i < m->num_packs; i++) {
1742 if (prepare_midx_pack(r, m, i))
1743 midx_report("failed to load pack in position %d", i);
1744
1745 display_progress(progress, i + 1);
1746 }
1747 stop_progress(&progress);
1748
1749 for (i = 0; i < 255; i++) {
1750 uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
1751 uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
1752
1753 if (oid_fanout1 > oid_fanout2)
1754 midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
1755 i, oid_fanout1, oid_fanout2, i + 1);
1756 }
1757
1758 if (m->num_objects == 0) {
1759 midx_report(_("the midx contains no oid"));
1760 /*
1761 * Remaining tests assume that we have objects, so we can
1762 * return here.
1763 */
1764 goto cleanup;
1765 }
1766
1767 if (flags & MIDX_PROGRESS)
1768 progress = start_sparse_progress(_("Verifying OID order in multi-pack-index"),
1769 m->num_objects - 1);
1770 for (i = 0; i < m->num_objects - 1; i++) {
1771 struct object_id oid1, oid2;
1772
1773 nth_midxed_object_oid(&oid1, m, i);
1774 nth_midxed_object_oid(&oid2, m, i + 1);
1775
1776 if (oidcmp(&oid1, &oid2) >= 0)
1777 midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
1778 i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
1779
1780 midx_display_sparse_progress(progress, i + 1);
1781 }
1782 stop_progress(&progress);
1783
1784 /*
1785 * Create an array mapping each object to its packfile id. Sort it
1786 * to group the objects by packfile. Use this permutation to visit
1787 * each of the objects and only require 1 packfile to be open at a
1788 * time.
1789 */
1790 ALLOC_ARRAY(pairs, m->num_objects);
1791 for (i = 0; i < m->num_objects; i++) {
1792 pairs[i].pos = i;
1793 pairs[i].pack_int_id = nth_midxed_pack_int_id(m, i);
1794 }
1795
1796 if (flags & MIDX_PROGRESS)
1797 progress = start_sparse_progress(_("Sorting objects by packfile"),
1798 m->num_objects);
1799 display_progress(progress, 0); /* TODO: Measure QSORT() progress */
1800 QSORT(pairs, m->num_objects, compare_pair_pos_vs_id);
1801 stop_progress(&progress);
1802
1803 if (flags & MIDX_PROGRESS)
1804 progress = start_sparse_progress(_("Verifying object offsets"), m->num_objects);
1805 for (i = 0; i < m->num_objects; i++) {
1806 struct object_id oid;
1807 struct pack_entry e;
1808 off_t m_offset, p_offset;
1809
1810 if (i > 0 && pairs[i-1].pack_int_id != pairs[i].pack_int_id &&
1811 m->packs[pairs[i-1].pack_int_id])
1812 {
1813 close_pack_fd(m->packs[pairs[i-1].pack_int_id]);
1814 close_pack_index(m->packs[pairs[i-1].pack_int_id]);
1815 }
1816
1817 nth_midxed_object_oid(&oid, m, pairs[i].pos);
1818
1819 if (!fill_midx_entry(r, &oid, &e, m)) {
1820 midx_report(_("failed to load pack entry for oid[%d] = %s"),
1821 pairs[i].pos, oid_to_hex(&oid));
1822 continue;
1823 }
1824
1825 if (open_pack_index(e.p)) {
1826 midx_report(_("failed to load pack-index for packfile %s"),
1827 e.p->pack_name);
1828 break;
1829 }
1830
1831 m_offset = e.offset;
1832 p_offset = find_pack_entry_one(oid.hash, e.p);
1833
1834 if (m_offset != p_offset)
1835 midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
1836 pairs[i].pos, oid_to_hex(&oid), m_offset, p_offset);
1837
1838 midx_display_sparse_progress(progress, i + 1);
1839 }
1840 stop_progress(&progress);
1841
1842 cleanup:
1843 free(pairs);
1844 close_midx(m);
1845
1846 return verify_midx_error;
1847 }
1848
1849 int expire_midx_packs(struct repository *r, const char *object_dir, unsigned flags)
1850 {
1851 uint32_t i, *count, result = 0;
1852 struct string_list packs_to_drop = STRING_LIST_INIT_DUP;
1853 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
1854 struct progress *progress = NULL;
1855
1856 if (!m)
1857 return 0;
1858
1859 CALLOC_ARRAY(count, m->num_packs);
1860
1861 if (flags & MIDX_PROGRESS)
1862 progress = start_delayed_progress(_("Counting referenced objects"),
1863 m->num_objects);
1864 for (i = 0; i < m->num_objects; i++) {
1865 int pack_int_id = nth_midxed_pack_int_id(m, i);
1866 count[pack_int_id]++;
1867 display_progress(progress, i + 1);
1868 }
1869 stop_progress(&progress);
1870
1871 if (flags & MIDX_PROGRESS)
1872 progress = start_delayed_progress(_("Finding and deleting unreferenced packfiles"),
1873 m->num_packs);
1874 for (i = 0; i < m->num_packs; i++) {
1875 char *pack_name;
1876 display_progress(progress, i + 1);
1877
1878 if (count[i])
1879 continue;
1880
1881 if (prepare_midx_pack(r, m, i))
1882 continue;
1883
1884 if (m->packs[i]->pack_keep || m->packs[i]->is_cruft)
1885 continue;
1886
1887 pack_name = xstrdup(m->packs[i]->pack_name);
1888 close_pack(m->packs[i]);
1889
1890 string_list_insert(&packs_to_drop, m->pack_names[i]);
1891 unlink_pack_path(pack_name, 0);
1892 free(pack_name);
1893 }
1894 stop_progress(&progress);
1895
1896 free(count);
1897
1898 if (packs_to_drop.nr)
1899 result = write_midx_internal(object_dir, NULL, &packs_to_drop, NULL, NULL, flags);
1900
1901 string_list_clear(&packs_to_drop, 0);
1902
1903 return result;
1904 }
1905
1906 struct repack_info {
1907 timestamp_t mtime;
1908 uint32_t referenced_objects;
1909 uint32_t pack_int_id;
1910 };
1911
1912 static int compare_by_mtime(const void *a_, const void *b_)
1913 {
1914 const struct repack_info *a, *b;
1915
1916 a = (const struct repack_info *)a_;
1917 b = (const struct repack_info *)b_;
1918
1919 if (a->mtime < b->mtime)
1920 return -1;
1921 if (a->mtime > b->mtime)
1922 return 1;
1923 return 0;
1924 }
1925
1926 static int fill_included_packs_all(struct repository *r,
1927 struct multi_pack_index *m,
1928 unsigned char *include_pack)
1929 {
1930 uint32_t i, count = 0;
1931 int pack_kept_objects = 0;
1932
1933 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1934
1935 for (i = 0; i < m->num_packs; i++) {
1936 if (prepare_midx_pack(r, m, i))
1937 continue;
1938 if (!pack_kept_objects && m->packs[i]->pack_keep)
1939 continue;
1940 if (m->packs[i]->is_cruft)
1941 continue;
1942
1943 include_pack[i] = 1;
1944 count++;
1945 }
1946
1947 return count < 2;
1948 }
1949
1950 static int fill_included_packs_batch(struct repository *r,
1951 struct multi_pack_index *m,
1952 unsigned char *include_pack,
1953 size_t batch_size)
1954 {
1955 uint32_t i, packs_to_repack;
1956 size_t total_size;
1957 struct repack_info *pack_info;
1958 int pack_kept_objects = 0;
1959
1960 CALLOC_ARRAY(pack_info, m->num_packs);
1961
1962 repo_config_get_bool(r, "repack.packkeptobjects", &pack_kept_objects);
1963
1964 for (i = 0; i < m->num_packs; i++) {
1965 pack_info[i].pack_int_id = i;
1966
1967 if (prepare_midx_pack(r, m, i))
1968 continue;
1969
1970 pack_info[i].mtime = m->packs[i]->mtime;
1971 }
1972
1973 for (i = 0; i < m->num_objects; i++) {
1974 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
1975 pack_info[pack_int_id].referenced_objects++;
1976 }
1977
1978 QSORT(pack_info, m->num_packs, compare_by_mtime);
1979
1980 total_size = 0;
1981 packs_to_repack = 0;
1982 for (i = 0; total_size < batch_size && i < m->num_packs; i++) {
1983 int pack_int_id = pack_info[i].pack_int_id;
1984 struct packed_git *p = m->packs[pack_int_id];
1985 size_t expected_size;
1986
1987 if (!p)
1988 continue;
1989 if (!pack_kept_objects && p->pack_keep)
1990 continue;
1991 if (p->is_cruft)
1992 continue;
1993 if (open_pack_index(p) || !p->num_objects)
1994 continue;
1995
1996 expected_size = st_mult(p->pack_size,
1997 pack_info[i].referenced_objects);
1998 expected_size /= p->num_objects;
1999
2000 if (expected_size >= batch_size)
2001 continue;
2002
2003 packs_to_repack++;
2004 total_size += expected_size;
2005 include_pack[pack_int_id] = 1;
2006 }
2007
2008 free(pack_info);
2009
2010 if (packs_to_repack < 2)
2011 return 1;
2012
2013 return 0;
2014 }
2015
2016 int midx_repack(struct repository *r, const char *object_dir, size_t batch_size, unsigned flags)
2017 {
2018 int result = 0;
2019 uint32_t i;
2020 unsigned char *include_pack;
2021 struct child_process cmd = CHILD_PROCESS_INIT;
2022 FILE *cmd_in;
2023 struct strbuf base_name = STRBUF_INIT;
2024 struct multi_pack_index *m = lookup_multi_pack_index(r, object_dir);
2025
2026 /*
2027 * When updating the default for these configuration
2028 * variables in builtin/repack.c, these must be adjusted
2029 * to match.
2030 */
2031 int delta_base_offset = 1;
2032 int use_delta_islands = 0;
2033
2034 if (!m)
2035 return 0;
2036
2037 CALLOC_ARRAY(include_pack, m->num_packs);
2038
2039 if (batch_size) {
2040 if (fill_included_packs_batch(r, m, include_pack, batch_size))
2041 goto cleanup;
2042 } else if (fill_included_packs_all(r, m, include_pack))
2043 goto cleanup;
2044
2045 repo_config_get_bool(r, "repack.usedeltabaseoffset", &delta_base_offset);
2046 repo_config_get_bool(r, "repack.usedeltaislands", &use_delta_islands);
2047
2048 strvec_push(&cmd.args, "pack-objects");
2049
2050 strbuf_addstr(&base_name, object_dir);
2051 strbuf_addstr(&base_name, "/pack/pack");
2052 strvec_push(&cmd.args, base_name.buf);
2053
2054 if (delta_base_offset)
2055 strvec_push(&cmd.args, "--delta-base-offset");
2056 if (use_delta_islands)
2057 strvec_push(&cmd.args, "--delta-islands");
2058
2059 if (flags & MIDX_PROGRESS)
2060 strvec_push(&cmd.args, "--progress");
2061 else
2062 strvec_push(&cmd.args, "-q");
2063
2064 strbuf_release(&base_name);
2065
2066 cmd.git_cmd = 1;
2067 cmd.in = cmd.out = -1;
2068
2069 if (start_command(&cmd)) {
2070 error(_("could not start pack-objects"));
2071 result = 1;
2072 goto cleanup;
2073 }
2074
2075 cmd_in = xfdopen(cmd.in, "w");
2076
2077 for (i = 0; i < m->num_objects; i++) {
2078 struct object_id oid;
2079 uint32_t pack_int_id = nth_midxed_pack_int_id(m, i);
2080
2081 if (!include_pack[pack_int_id])
2082 continue;
2083
2084 nth_midxed_object_oid(&oid, m, i);
2085 fprintf(cmd_in, "%s\n", oid_to_hex(&oid));
2086 }
2087 fclose(cmd_in);
2088
2089 if (finish_command(&cmd)) {
2090 error(_("could not finish pack-objects"));
2091 result = 1;
2092 goto cleanup;
2093 }
2094
2095 result = write_midx_internal(object_dir, NULL, NULL, NULL, NULL, flags);
2096
2097 cleanup:
2098 free(include_pack);
2099 return result;
2100 }