object-store: move packed_git and packed_git_mru to object store
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "attr.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "commit.h"
9 #include "tag.h"
10 #include "tree.h"
11 #include "delta.h"
12 #include "pack.h"
13 #include "pack-revindex.h"
14 #include "csum-file.h"
15 #include "tree-walk.h"
16 #include "diff.h"
17 #include "revision.h"
18 #include "list-objects.h"
19 #include "list-objects-filter.h"
20 #include "list-objects-filter-options.h"
21 #include "pack-objects.h"
22 #include "progress.h"
23 #include "refs.h"
24 #include "streaming.h"
25 #include "thread-utils.h"
26 #include "pack-bitmap.h"
27 #include "reachable.h"
28 #include "sha1-array.h"
29 #include "argv-array.h"
30 #include "list.h"
31 #include "packfile.h"
32 #include "object-store.h"
33
34 static const char *pack_usage[] = {
35 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
36 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
37 NULL
38 };
39
40 /*
41 * Objects we are going to pack are collected in the `to_pack` structure.
42 * It contains an array (dynamically expanded) of the object data, and a map
43 * that can resolve SHA1s to their position in the array.
44 */
45 static struct packing_data to_pack;
46
47 static struct pack_idx_entry **written_list;
48 static uint32_t nr_result, nr_written;
49
50 static int non_empty;
51 static int reuse_delta = 1, reuse_object = 1;
52 static int keep_unreachable, unpack_unreachable, include_tag;
53 static timestamp_t unpack_unreachable_expiration;
54 static int pack_loose_unreachable;
55 static int local;
56 static int have_non_local_packs;
57 static int incremental;
58 static int ignore_packed_keep;
59 static int allow_ofs_delta;
60 static struct pack_idx_option pack_idx_opts;
61 static const char *base_name;
62 static int progress = 1;
63 static int window = 10;
64 static unsigned long pack_size_limit;
65 static int depth = 50;
66 static int delta_search_threads;
67 static int pack_to_stdout;
68 static int num_preferred_base;
69 static struct progress *progress_state;
70
71 static struct packed_git *reuse_packfile;
72 static uint32_t reuse_packfile_objects;
73 static off_t reuse_packfile_offset;
74
75 static int use_bitmap_index_default = 1;
76 static int use_bitmap_index = -1;
77 static int write_bitmap_index;
78 static uint16_t write_bitmap_options;
79
80 static int exclude_promisor_objects;
81
82 static unsigned long delta_cache_size = 0;
83 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
84 static unsigned long cache_max_small_delta_size = 1000;
85
86 static unsigned long window_memory_limit = 0;
87
88 static struct list_objects_filter_options filter_options;
89
90 enum missing_action {
91 MA_ERROR = 0, /* fail if any missing objects are encountered */
92 MA_ALLOW_ANY, /* silently allow ALL missing objects */
93 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
94 };
95 static enum missing_action arg_missing_action;
96 static show_object_fn fn_show_object;
97
98 /*
99 * stats
100 */
101 static uint32_t written, written_delta;
102 static uint32_t reused, reused_delta;
103
104 /*
105 * Indexed commits
106 */
107 static struct commit **indexed_commits;
108 static unsigned int indexed_commits_nr;
109 static unsigned int indexed_commits_alloc;
110
111 static void index_commit_for_bitmap(struct commit *commit)
112 {
113 if (indexed_commits_nr >= indexed_commits_alloc) {
114 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
115 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
116 }
117
118 indexed_commits[indexed_commits_nr++] = commit;
119 }
120
121 static void *get_delta(struct object_entry *entry)
122 {
123 unsigned long size, base_size, delta_size;
124 void *buf, *base_buf, *delta_buf;
125 enum object_type type;
126
127 buf = read_sha1_file(entry->idx.oid.hash, &type, &size);
128 if (!buf)
129 die("unable to read %s", oid_to_hex(&entry->idx.oid));
130 base_buf = read_sha1_file(entry->delta->idx.oid.hash, &type,
131 &base_size);
132 if (!base_buf)
133 die("unable to read %s",
134 oid_to_hex(&entry->delta->idx.oid));
135 delta_buf = diff_delta(base_buf, base_size,
136 buf, size, &delta_size, 0);
137 if (!delta_buf || delta_size != entry->delta_size)
138 die("delta size changed");
139 free(buf);
140 free(base_buf);
141 return delta_buf;
142 }
143
144 static unsigned long do_compress(void **pptr, unsigned long size)
145 {
146 git_zstream stream;
147 void *in, *out;
148 unsigned long maxsize;
149
150 git_deflate_init(&stream, pack_compression_level);
151 maxsize = git_deflate_bound(&stream, size);
152
153 in = *pptr;
154 out = xmalloc(maxsize);
155 *pptr = out;
156
157 stream.next_in = in;
158 stream.avail_in = size;
159 stream.next_out = out;
160 stream.avail_out = maxsize;
161 while (git_deflate(&stream, Z_FINISH) == Z_OK)
162 ; /* nothing */
163 git_deflate_end(&stream);
164
165 free(in);
166 return stream.total_out;
167 }
168
169 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
170 const struct object_id *oid)
171 {
172 git_zstream stream;
173 unsigned char ibuf[1024 * 16];
174 unsigned char obuf[1024 * 16];
175 unsigned long olen = 0;
176
177 git_deflate_init(&stream, pack_compression_level);
178
179 for (;;) {
180 ssize_t readlen;
181 int zret = Z_OK;
182 readlen = read_istream(st, ibuf, sizeof(ibuf));
183 if (readlen == -1)
184 die(_("unable to read %s"), oid_to_hex(oid));
185
186 stream.next_in = ibuf;
187 stream.avail_in = readlen;
188 while ((stream.avail_in || readlen == 0) &&
189 (zret == Z_OK || zret == Z_BUF_ERROR)) {
190 stream.next_out = obuf;
191 stream.avail_out = sizeof(obuf);
192 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
193 sha1write(f, obuf, stream.next_out - obuf);
194 olen += stream.next_out - obuf;
195 }
196 if (stream.avail_in)
197 die(_("deflate error (%d)"), zret);
198 if (readlen == 0) {
199 if (zret != Z_STREAM_END)
200 die(_("deflate error (%d)"), zret);
201 break;
202 }
203 }
204 git_deflate_end(&stream);
205 return olen;
206 }
207
208 /*
209 * we are going to reuse the existing object data as is. make
210 * sure it is not corrupt.
211 */
212 static int check_pack_inflate(struct packed_git *p,
213 struct pack_window **w_curs,
214 off_t offset,
215 off_t len,
216 unsigned long expect)
217 {
218 git_zstream stream;
219 unsigned char fakebuf[4096], *in;
220 int st;
221
222 memset(&stream, 0, sizeof(stream));
223 git_inflate_init(&stream);
224 do {
225 in = use_pack(p, w_curs, offset, &stream.avail_in);
226 stream.next_in = in;
227 stream.next_out = fakebuf;
228 stream.avail_out = sizeof(fakebuf);
229 st = git_inflate(&stream, Z_FINISH);
230 offset += stream.next_in - in;
231 } while (st == Z_OK || st == Z_BUF_ERROR);
232 git_inflate_end(&stream);
233 return (st == Z_STREAM_END &&
234 stream.total_out == expect &&
235 stream.total_in == len) ? 0 : -1;
236 }
237
238 static void copy_pack_data(struct sha1file *f,
239 struct packed_git *p,
240 struct pack_window **w_curs,
241 off_t offset,
242 off_t len)
243 {
244 unsigned char *in;
245 unsigned long avail;
246
247 while (len) {
248 in = use_pack(p, w_curs, offset, &avail);
249 if (avail > len)
250 avail = (unsigned long)len;
251 sha1write(f, in, avail);
252 offset += avail;
253 len -= avail;
254 }
255 }
256
257 /* Return 0 if we will bust the pack-size limit */
258 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
259 unsigned long limit, int usable_delta)
260 {
261 unsigned long size, datalen;
262 unsigned char header[MAX_PACK_OBJECT_HEADER],
263 dheader[MAX_PACK_OBJECT_HEADER];
264 unsigned hdrlen;
265 enum object_type type;
266 void *buf;
267 struct git_istream *st = NULL;
268
269 if (!usable_delta) {
270 if (entry->type == OBJ_BLOB &&
271 entry->size > big_file_threshold &&
272 (st = open_istream(entry->idx.oid.hash, &type, &size, NULL)) != NULL)
273 buf = NULL;
274 else {
275 buf = read_sha1_file(entry->idx.oid.hash, &type,
276 &size);
277 if (!buf)
278 die(_("unable to read %s"),
279 oid_to_hex(&entry->idx.oid));
280 }
281 /*
282 * make sure no cached delta data remains from a
283 * previous attempt before a pack split occurred.
284 */
285 FREE_AND_NULL(entry->delta_data);
286 entry->z_delta_size = 0;
287 } else if (entry->delta_data) {
288 size = entry->delta_size;
289 buf = entry->delta_data;
290 entry->delta_data = NULL;
291 type = (allow_ofs_delta && entry->delta->idx.offset) ?
292 OBJ_OFS_DELTA : OBJ_REF_DELTA;
293 } else {
294 buf = get_delta(entry);
295 size = entry->delta_size;
296 type = (allow_ofs_delta && entry->delta->idx.offset) ?
297 OBJ_OFS_DELTA : OBJ_REF_DELTA;
298 }
299
300 if (st) /* large blob case, just assume we don't compress well */
301 datalen = size;
302 else if (entry->z_delta_size)
303 datalen = entry->z_delta_size;
304 else
305 datalen = do_compress(&buf, size);
306
307 /*
308 * The object header is a byte of 'type' followed by zero or
309 * more bytes of length.
310 */
311 hdrlen = encode_in_pack_object_header(header, sizeof(header),
312 type, size);
313
314 if (type == OBJ_OFS_DELTA) {
315 /*
316 * Deltas with relative base contain an additional
317 * encoding of the relative offset for the delta
318 * base from this object's position in the pack.
319 */
320 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
321 unsigned pos = sizeof(dheader) - 1;
322 dheader[pos] = ofs & 127;
323 while (ofs >>= 7)
324 dheader[--pos] = 128 | (--ofs & 127);
325 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
326 if (st)
327 close_istream(st);
328 free(buf);
329 return 0;
330 }
331 sha1write(f, header, hdrlen);
332 sha1write(f, dheader + pos, sizeof(dheader) - pos);
333 hdrlen += sizeof(dheader) - pos;
334 } else if (type == OBJ_REF_DELTA) {
335 /*
336 * Deltas with a base reference contain
337 * an additional 20 bytes for the base sha1.
338 */
339 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
340 if (st)
341 close_istream(st);
342 free(buf);
343 return 0;
344 }
345 sha1write(f, header, hdrlen);
346 sha1write(f, entry->delta->idx.oid.hash, 20);
347 hdrlen += 20;
348 } else {
349 if (limit && hdrlen + datalen + 20 >= limit) {
350 if (st)
351 close_istream(st);
352 free(buf);
353 return 0;
354 }
355 sha1write(f, header, hdrlen);
356 }
357 if (st) {
358 datalen = write_large_blob_data(st, f, &entry->idx.oid);
359 close_istream(st);
360 } else {
361 sha1write(f, buf, datalen);
362 free(buf);
363 }
364
365 return hdrlen + datalen;
366 }
367
368 /* Return 0 if we will bust the pack-size limit */
369 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
370 unsigned long limit, int usable_delta)
371 {
372 struct packed_git *p = entry->in_pack;
373 struct pack_window *w_curs = NULL;
374 struct revindex_entry *revidx;
375 off_t offset;
376 enum object_type type = entry->type;
377 off_t datalen;
378 unsigned char header[MAX_PACK_OBJECT_HEADER],
379 dheader[MAX_PACK_OBJECT_HEADER];
380 unsigned hdrlen;
381
382 if (entry->delta)
383 type = (allow_ofs_delta && entry->delta->idx.offset) ?
384 OBJ_OFS_DELTA : OBJ_REF_DELTA;
385 hdrlen = encode_in_pack_object_header(header, sizeof(header),
386 type, entry->size);
387
388 offset = entry->in_pack_offset;
389 revidx = find_pack_revindex(p, offset);
390 datalen = revidx[1].offset - offset;
391 if (!pack_to_stdout && p->index_version > 1 &&
392 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
393 error("bad packed object CRC for %s",
394 oid_to_hex(&entry->idx.oid));
395 unuse_pack(&w_curs);
396 return write_no_reuse_object(f, entry, limit, usable_delta);
397 }
398
399 offset += entry->in_pack_header_size;
400 datalen -= entry->in_pack_header_size;
401
402 if (!pack_to_stdout && p->index_version == 1 &&
403 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
404 error("corrupt packed object for %s",
405 oid_to_hex(&entry->idx.oid));
406 unuse_pack(&w_curs);
407 return write_no_reuse_object(f, entry, limit, usable_delta);
408 }
409
410 if (type == OBJ_OFS_DELTA) {
411 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
412 unsigned pos = sizeof(dheader) - 1;
413 dheader[pos] = ofs & 127;
414 while (ofs >>= 7)
415 dheader[--pos] = 128 | (--ofs & 127);
416 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
417 unuse_pack(&w_curs);
418 return 0;
419 }
420 sha1write(f, header, hdrlen);
421 sha1write(f, dheader + pos, sizeof(dheader) - pos);
422 hdrlen += sizeof(dheader) - pos;
423 reused_delta++;
424 } else if (type == OBJ_REF_DELTA) {
425 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
426 unuse_pack(&w_curs);
427 return 0;
428 }
429 sha1write(f, header, hdrlen);
430 sha1write(f, entry->delta->idx.oid.hash, 20);
431 hdrlen += 20;
432 reused_delta++;
433 } else {
434 if (limit && hdrlen + datalen + 20 >= limit) {
435 unuse_pack(&w_curs);
436 return 0;
437 }
438 sha1write(f, header, hdrlen);
439 }
440 copy_pack_data(f, p, &w_curs, offset, datalen);
441 unuse_pack(&w_curs);
442 reused++;
443 return hdrlen + datalen;
444 }
445
446 /* Return 0 if we will bust the pack-size limit */
447 static off_t write_object(struct sha1file *f,
448 struct object_entry *entry,
449 off_t write_offset)
450 {
451 unsigned long limit;
452 off_t len;
453 int usable_delta, to_reuse;
454
455 if (!pack_to_stdout)
456 crc32_begin(f);
457
458 /* apply size limit if limited packsize and not first object */
459 if (!pack_size_limit || !nr_written)
460 limit = 0;
461 else if (pack_size_limit <= write_offset)
462 /*
463 * the earlier object did not fit the limit; avoid
464 * mistaking this with unlimited (i.e. limit = 0).
465 */
466 limit = 1;
467 else
468 limit = pack_size_limit - write_offset;
469
470 if (!entry->delta)
471 usable_delta = 0; /* no delta */
472 else if (!pack_size_limit)
473 usable_delta = 1; /* unlimited packfile */
474 else if (entry->delta->idx.offset == (off_t)-1)
475 usable_delta = 0; /* base was written to another pack */
476 else if (entry->delta->idx.offset)
477 usable_delta = 1; /* base already exists in this pack */
478 else
479 usable_delta = 0; /* base could end up in another pack */
480
481 if (!reuse_object)
482 to_reuse = 0; /* explicit */
483 else if (!entry->in_pack)
484 to_reuse = 0; /* can't reuse what we don't have */
485 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
486 /* check_object() decided it for us ... */
487 to_reuse = usable_delta;
488 /* ... but pack split may override that */
489 else if (entry->type != entry->in_pack_type)
490 to_reuse = 0; /* pack has delta which is unusable */
491 else if (entry->delta)
492 to_reuse = 0; /* we want to pack afresh */
493 else
494 to_reuse = 1; /* we have it in-pack undeltified,
495 * and we do not need to deltify it.
496 */
497
498 if (!to_reuse)
499 len = write_no_reuse_object(f, entry, limit, usable_delta);
500 else
501 len = write_reuse_object(f, entry, limit, usable_delta);
502 if (!len)
503 return 0;
504
505 if (usable_delta)
506 written_delta++;
507 written++;
508 if (!pack_to_stdout)
509 entry->idx.crc32 = crc32_end(f);
510 return len;
511 }
512
513 enum write_one_status {
514 WRITE_ONE_SKIP = -1, /* already written */
515 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
516 WRITE_ONE_WRITTEN = 1, /* normal */
517 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
518 };
519
520 static enum write_one_status write_one(struct sha1file *f,
521 struct object_entry *e,
522 off_t *offset)
523 {
524 off_t size;
525 int recursing;
526
527 /*
528 * we set offset to 1 (which is an impossible value) to mark
529 * the fact that this object is involved in "write its base
530 * first before writing a deltified object" recursion.
531 */
532 recursing = (e->idx.offset == 1);
533 if (recursing) {
534 warning("recursive delta detected for object %s",
535 oid_to_hex(&e->idx.oid));
536 return WRITE_ONE_RECURSIVE;
537 } else if (e->idx.offset || e->preferred_base) {
538 /* offset is non zero if object is written already. */
539 return WRITE_ONE_SKIP;
540 }
541
542 /* if we are deltified, write out base object first. */
543 if (e->delta) {
544 e->idx.offset = 1; /* now recurse */
545 switch (write_one(f, e->delta, offset)) {
546 case WRITE_ONE_RECURSIVE:
547 /* we cannot depend on this one */
548 e->delta = NULL;
549 break;
550 default:
551 break;
552 case WRITE_ONE_BREAK:
553 e->idx.offset = recursing;
554 return WRITE_ONE_BREAK;
555 }
556 }
557
558 e->idx.offset = *offset;
559 size = write_object(f, e, *offset);
560 if (!size) {
561 e->idx.offset = recursing;
562 return WRITE_ONE_BREAK;
563 }
564 written_list[nr_written++] = &e->idx;
565
566 /* make sure off_t is sufficiently large not to wrap */
567 if (signed_add_overflows(*offset, size))
568 die("pack too large for current definition of off_t");
569 *offset += size;
570 return WRITE_ONE_WRITTEN;
571 }
572
573 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
574 void *cb_data)
575 {
576 struct object_id peeled;
577 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
578
579 if (entry)
580 entry->tagged = 1;
581 if (!peel_ref(path, &peeled)) {
582 entry = packlist_find(&to_pack, peeled.hash, NULL);
583 if (entry)
584 entry->tagged = 1;
585 }
586 return 0;
587 }
588
589 static inline void add_to_write_order(struct object_entry **wo,
590 unsigned int *endp,
591 struct object_entry *e)
592 {
593 if (e->filled)
594 return;
595 wo[(*endp)++] = e;
596 e->filled = 1;
597 }
598
599 static void add_descendants_to_write_order(struct object_entry **wo,
600 unsigned int *endp,
601 struct object_entry *e)
602 {
603 int add_to_order = 1;
604 while (e) {
605 if (add_to_order) {
606 struct object_entry *s;
607 /* add this node... */
608 add_to_write_order(wo, endp, e);
609 /* all its siblings... */
610 for (s = e->delta_sibling; s; s = s->delta_sibling) {
611 add_to_write_order(wo, endp, s);
612 }
613 }
614 /* drop down a level to add left subtree nodes if possible */
615 if (e->delta_child) {
616 add_to_order = 1;
617 e = e->delta_child;
618 } else {
619 add_to_order = 0;
620 /* our sibling might have some children, it is next */
621 if (e->delta_sibling) {
622 e = e->delta_sibling;
623 continue;
624 }
625 /* go back to our parent node */
626 e = e->delta;
627 while (e && !e->delta_sibling) {
628 /* we're on the right side of a subtree, keep
629 * going up until we can go right again */
630 e = e->delta;
631 }
632 if (!e) {
633 /* done- we hit our original root node */
634 return;
635 }
636 /* pass it off to sibling at this level */
637 e = e->delta_sibling;
638 }
639 };
640 }
641
642 static void add_family_to_write_order(struct object_entry **wo,
643 unsigned int *endp,
644 struct object_entry *e)
645 {
646 struct object_entry *root;
647
648 for (root = e; root->delta; root = root->delta)
649 ; /* nothing */
650 add_descendants_to_write_order(wo, endp, root);
651 }
652
653 static struct object_entry **compute_write_order(void)
654 {
655 unsigned int i, wo_end, last_untagged;
656
657 struct object_entry **wo;
658 struct object_entry *objects = to_pack.objects;
659
660 for (i = 0; i < to_pack.nr_objects; i++) {
661 objects[i].tagged = 0;
662 objects[i].filled = 0;
663 objects[i].delta_child = NULL;
664 objects[i].delta_sibling = NULL;
665 }
666
667 /*
668 * Fully connect delta_child/delta_sibling network.
669 * Make sure delta_sibling is sorted in the original
670 * recency order.
671 */
672 for (i = to_pack.nr_objects; i > 0;) {
673 struct object_entry *e = &objects[--i];
674 if (!e->delta)
675 continue;
676 /* Mark me as the first child */
677 e->delta_sibling = e->delta->delta_child;
678 e->delta->delta_child = e;
679 }
680
681 /*
682 * Mark objects that are at the tip of tags.
683 */
684 for_each_tag_ref(mark_tagged, NULL);
685
686 /*
687 * Give the objects in the original recency order until
688 * we see a tagged tip.
689 */
690 ALLOC_ARRAY(wo, to_pack.nr_objects);
691 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
692 if (objects[i].tagged)
693 break;
694 add_to_write_order(wo, &wo_end, &objects[i]);
695 }
696 last_untagged = i;
697
698 /*
699 * Then fill all the tagged tips.
700 */
701 for (; i < to_pack.nr_objects; i++) {
702 if (objects[i].tagged)
703 add_to_write_order(wo, &wo_end, &objects[i]);
704 }
705
706 /*
707 * And then all remaining commits and tags.
708 */
709 for (i = last_untagged; i < to_pack.nr_objects; i++) {
710 if (objects[i].type != OBJ_COMMIT &&
711 objects[i].type != OBJ_TAG)
712 continue;
713 add_to_write_order(wo, &wo_end, &objects[i]);
714 }
715
716 /*
717 * And then all the trees.
718 */
719 for (i = last_untagged; i < to_pack.nr_objects; i++) {
720 if (objects[i].type != OBJ_TREE)
721 continue;
722 add_to_write_order(wo, &wo_end, &objects[i]);
723 }
724
725 /*
726 * Finally all the rest in really tight order
727 */
728 for (i = last_untagged; i < to_pack.nr_objects; i++) {
729 if (!objects[i].filled)
730 add_family_to_write_order(wo, &wo_end, &objects[i]);
731 }
732
733 if (wo_end != to_pack.nr_objects)
734 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
735
736 return wo;
737 }
738
739 static off_t write_reused_pack(struct sha1file *f)
740 {
741 unsigned char buffer[8192];
742 off_t to_write, total;
743 int fd;
744
745 if (!is_pack_valid(reuse_packfile))
746 die("packfile is invalid: %s", reuse_packfile->pack_name);
747
748 fd = git_open(reuse_packfile->pack_name);
749 if (fd < 0)
750 die_errno("unable to open packfile for reuse: %s",
751 reuse_packfile->pack_name);
752
753 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
754 die_errno("unable to seek in reused packfile");
755
756 if (reuse_packfile_offset < 0)
757 reuse_packfile_offset = reuse_packfile->pack_size - 20;
758
759 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
760
761 while (to_write) {
762 int read_pack = xread(fd, buffer, sizeof(buffer));
763
764 if (read_pack <= 0)
765 die_errno("unable to read from reused packfile");
766
767 if (read_pack > to_write)
768 read_pack = to_write;
769
770 sha1write(f, buffer, read_pack);
771 to_write -= read_pack;
772
773 /*
774 * We don't know the actual number of objects written,
775 * only how many bytes written, how many bytes total, and
776 * how many objects total. So we can fake it by pretending all
777 * objects we are writing are the same size. This gives us a
778 * smooth progress meter, and at the end it matches the true
779 * answer.
780 */
781 written = reuse_packfile_objects *
782 (((double)(total - to_write)) / total);
783 display_progress(progress_state, written);
784 }
785
786 close(fd);
787 written = reuse_packfile_objects;
788 display_progress(progress_state, written);
789 return reuse_packfile_offset - sizeof(struct pack_header);
790 }
791
792 static const char no_split_warning[] = N_(
793 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
794 );
795
796 static void write_pack_file(void)
797 {
798 uint32_t i = 0, j;
799 struct sha1file *f;
800 off_t offset;
801 uint32_t nr_remaining = nr_result;
802 time_t last_mtime = 0;
803 struct object_entry **write_order;
804
805 if (progress > pack_to_stdout)
806 progress_state = start_progress(_("Writing objects"), nr_result);
807 ALLOC_ARRAY(written_list, to_pack.nr_objects);
808 write_order = compute_write_order();
809
810 do {
811 struct object_id oid;
812 char *pack_tmp_name = NULL;
813
814 if (pack_to_stdout)
815 f = sha1fd_throughput(1, "<stdout>", progress_state);
816 else
817 f = create_tmp_packfile(&pack_tmp_name);
818
819 offset = write_pack_header(f, nr_remaining);
820
821 if (reuse_packfile) {
822 off_t packfile_size;
823 assert(pack_to_stdout);
824
825 packfile_size = write_reused_pack(f);
826 offset += packfile_size;
827 }
828
829 nr_written = 0;
830 for (; i < to_pack.nr_objects; i++) {
831 struct object_entry *e = write_order[i];
832 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
833 break;
834 display_progress(progress_state, written);
835 }
836
837 /*
838 * Did we write the wrong # entries in the header?
839 * If so, rewrite it like in fast-import
840 */
841 if (pack_to_stdout) {
842 sha1close(f, oid.hash, CSUM_CLOSE);
843 } else if (nr_written == nr_remaining) {
844 sha1close(f, oid.hash, CSUM_FSYNC);
845 } else {
846 int fd = sha1close(f, oid.hash, 0);
847 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
848 nr_written, oid.hash, offset);
849 close(fd);
850 if (write_bitmap_index) {
851 warning(_(no_split_warning));
852 write_bitmap_index = 0;
853 }
854 }
855
856 if (!pack_to_stdout) {
857 struct stat st;
858 struct strbuf tmpname = STRBUF_INIT;
859
860 /*
861 * Packs are runtime accessed in their mtime
862 * order since newer packs are more likely to contain
863 * younger objects. So if we are creating multiple
864 * packs then we should modify the mtime of later ones
865 * to preserve this property.
866 */
867 if (stat(pack_tmp_name, &st) < 0) {
868 warning_errno("failed to stat %s", pack_tmp_name);
869 } else if (!last_mtime) {
870 last_mtime = st.st_mtime;
871 } else {
872 struct utimbuf utb;
873 utb.actime = st.st_atime;
874 utb.modtime = --last_mtime;
875 if (utime(pack_tmp_name, &utb) < 0)
876 warning_errno("failed utime() on %s", pack_tmp_name);
877 }
878
879 strbuf_addf(&tmpname, "%s-", base_name);
880
881 if (write_bitmap_index) {
882 bitmap_writer_set_checksum(oid.hash);
883 bitmap_writer_build_type_index(written_list, nr_written);
884 }
885
886 finish_tmp_packfile(&tmpname, pack_tmp_name,
887 written_list, nr_written,
888 &pack_idx_opts, oid.hash);
889
890 if (write_bitmap_index) {
891 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
892
893 stop_progress(&progress_state);
894
895 bitmap_writer_show_progress(progress);
896 bitmap_writer_reuse_bitmaps(&to_pack);
897 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
898 bitmap_writer_build(&to_pack);
899 bitmap_writer_finish(written_list, nr_written,
900 tmpname.buf, write_bitmap_options);
901 write_bitmap_index = 0;
902 }
903
904 strbuf_release(&tmpname);
905 free(pack_tmp_name);
906 puts(oid_to_hex(&oid));
907 }
908
909 /* mark written objects as written to previous pack */
910 for (j = 0; j < nr_written; j++) {
911 written_list[j]->offset = (off_t)-1;
912 }
913 nr_remaining -= nr_written;
914 } while (nr_remaining && i < to_pack.nr_objects);
915
916 free(written_list);
917 free(write_order);
918 stop_progress(&progress_state);
919 if (written != nr_result)
920 die("wrote %"PRIu32" objects while expecting %"PRIu32,
921 written, nr_result);
922 }
923
924 static int no_try_delta(const char *path)
925 {
926 static struct attr_check *check;
927
928 if (!check)
929 check = attr_check_initl("delta", NULL);
930 if (git_check_attr(path, check))
931 return 0;
932 if (ATTR_FALSE(check->items[0].value))
933 return 1;
934 return 0;
935 }
936
937 /*
938 * When adding an object, check whether we have already added it
939 * to our packing list. If so, we can skip. However, if we are
940 * being asked to excludei t, but the previous mention was to include
941 * it, make sure to adjust its flags and tweak our numbers accordingly.
942 *
943 * As an optimization, we pass out the index position where we would have
944 * found the item, since that saves us from having to look it up again a
945 * few lines later when we want to add the new entry.
946 */
947 static int have_duplicate_entry(const struct object_id *oid,
948 int exclude,
949 uint32_t *index_pos)
950 {
951 struct object_entry *entry;
952
953 entry = packlist_find(&to_pack, oid->hash, index_pos);
954 if (!entry)
955 return 0;
956
957 if (exclude) {
958 if (!entry->preferred_base)
959 nr_result--;
960 entry->preferred_base = 1;
961 }
962
963 return 1;
964 }
965
966 static int want_found_object(int exclude, struct packed_git *p)
967 {
968 if (exclude)
969 return 1;
970 if (incremental)
971 return 0;
972
973 /*
974 * When asked to do --local (do not include an object that appears in a
975 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
976 * an object that appears in a pack marked with .keep), finding a pack
977 * that matches the criteria is sufficient for us to decide to omit it.
978 * However, even if this pack does not satisfy the criteria, we need to
979 * make sure no copy of this object appears in _any_ pack that makes us
980 * to omit the object, so we need to check all the packs.
981 *
982 * We can however first check whether these options can possible matter;
983 * if they do not matter we know we want the object in generated pack.
984 * Otherwise, we signal "-1" at the end to tell the caller that we do
985 * not know either way, and it needs to check more packs.
986 */
987 if (!ignore_packed_keep &&
988 (!local || !have_non_local_packs))
989 return 1;
990
991 if (local && !p->pack_local)
992 return 0;
993 if (ignore_packed_keep && p->pack_local && p->pack_keep)
994 return 0;
995
996 /* we don't know yet; keep looking for more packs */
997 return -1;
998 }
999
1000 /*
1001 * Check whether we want the object in the pack (e.g., we do not want
1002 * objects found in non-local stores if the "--local" option was used).
1003 *
1004 * If the caller already knows an existing pack it wants to take the object
1005 * from, that is passed in *found_pack and *found_offset; otherwise this
1006 * function finds if there is any pack that has the object and returns the pack
1007 * and its offset in these variables.
1008 */
1009 static int want_object_in_pack(const struct object_id *oid,
1010 int exclude,
1011 struct packed_git **found_pack,
1012 off_t *found_offset)
1013 {
1014 int want;
1015 struct list_head *pos;
1016
1017 if (!exclude && local && has_loose_object_nonlocal(oid->hash))
1018 return 0;
1019
1020 /*
1021 * If we already know the pack object lives in, start checks from that
1022 * pack - in the usual case when neither --local was given nor .keep files
1023 * are present we will determine the answer right now.
1024 */
1025 if (*found_pack) {
1026 want = want_found_object(exclude, *found_pack);
1027 if (want != -1)
1028 return want;
1029 }
1030 list_for_each(pos, get_packed_git_mru(the_repository)) {
1031 struct packed_git *p = list_entry(pos, struct packed_git, mru);
1032 off_t offset;
1033
1034 if (p == *found_pack)
1035 offset = *found_offset;
1036 else
1037 offset = find_pack_entry_one(oid->hash, p);
1038
1039 if (offset) {
1040 if (!*found_pack) {
1041 if (!is_pack_valid(p))
1042 continue;
1043 *found_offset = offset;
1044 *found_pack = p;
1045 }
1046 want = want_found_object(exclude, p);
1047 if (!exclude && want > 0)
1048 list_move(&p->mru,
1049 get_packed_git_mru(the_repository));
1050 if (want != -1)
1051 return want;
1052 }
1053 }
1054
1055 return 1;
1056 }
1057
1058 static void create_object_entry(const struct object_id *oid,
1059 enum object_type type,
1060 uint32_t hash,
1061 int exclude,
1062 int no_try_delta,
1063 uint32_t index_pos,
1064 struct packed_git *found_pack,
1065 off_t found_offset)
1066 {
1067 struct object_entry *entry;
1068
1069 entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1070 entry->hash = hash;
1071 if (type)
1072 entry->type = type;
1073 if (exclude)
1074 entry->preferred_base = 1;
1075 else
1076 nr_result++;
1077 if (found_pack) {
1078 entry->in_pack = found_pack;
1079 entry->in_pack_offset = found_offset;
1080 }
1081
1082 entry->no_try_delta = no_try_delta;
1083 }
1084
1085 static const char no_closure_warning[] = N_(
1086 "disabling bitmap writing, as some objects are not being packed"
1087 );
1088
1089 static int add_object_entry(const struct object_id *oid, enum object_type type,
1090 const char *name, int exclude)
1091 {
1092 struct packed_git *found_pack = NULL;
1093 off_t found_offset = 0;
1094 uint32_t index_pos;
1095
1096 if (have_duplicate_entry(oid, exclude, &index_pos))
1097 return 0;
1098
1099 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1100 /* The pack is missing an object, so it will not have closure */
1101 if (write_bitmap_index) {
1102 warning(_(no_closure_warning));
1103 write_bitmap_index = 0;
1104 }
1105 return 0;
1106 }
1107
1108 create_object_entry(oid, type, pack_name_hash(name),
1109 exclude, name && no_try_delta(name),
1110 index_pos, found_pack, found_offset);
1111
1112 display_progress(progress_state, nr_result);
1113 return 1;
1114 }
1115
1116 static int add_object_entry_from_bitmap(const struct object_id *oid,
1117 enum object_type type,
1118 int flags, uint32_t name_hash,
1119 struct packed_git *pack, off_t offset)
1120 {
1121 uint32_t index_pos;
1122
1123 if (have_duplicate_entry(oid, 0, &index_pos))
1124 return 0;
1125
1126 if (!want_object_in_pack(oid, 0, &pack, &offset))
1127 return 0;
1128
1129 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1130
1131 display_progress(progress_state, nr_result);
1132 return 1;
1133 }
1134
1135 struct pbase_tree_cache {
1136 struct object_id oid;
1137 int ref;
1138 int temporary;
1139 void *tree_data;
1140 unsigned long tree_size;
1141 };
1142
1143 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1144 static int pbase_tree_cache_ix(const struct object_id *oid)
1145 {
1146 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1147 }
1148 static int pbase_tree_cache_ix_incr(int ix)
1149 {
1150 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1151 }
1152
1153 static struct pbase_tree {
1154 struct pbase_tree *next;
1155 /* This is a phony "cache" entry; we are not
1156 * going to evict it or find it through _get()
1157 * mechanism -- this is for the toplevel node that
1158 * would almost always change with any commit.
1159 */
1160 struct pbase_tree_cache pcache;
1161 } *pbase_tree;
1162
1163 static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1164 {
1165 struct pbase_tree_cache *ent, *nent;
1166 void *data;
1167 unsigned long size;
1168 enum object_type type;
1169 int neigh;
1170 int my_ix = pbase_tree_cache_ix(oid);
1171 int available_ix = -1;
1172
1173 /* pbase-tree-cache acts as a limited hashtable.
1174 * your object will be found at your index or within a few
1175 * slots after that slot if it is cached.
1176 */
1177 for (neigh = 0; neigh < 8; neigh++) {
1178 ent = pbase_tree_cache[my_ix];
1179 if (ent && !oidcmp(&ent->oid, oid)) {
1180 ent->ref++;
1181 return ent;
1182 }
1183 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1184 ((0 <= available_ix) &&
1185 (!ent && pbase_tree_cache[available_ix])))
1186 available_ix = my_ix;
1187 if (!ent)
1188 break;
1189 my_ix = pbase_tree_cache_ix_incr(my_ix);
1190 }
1191
1192 /* Did not find one. Either we got a bogus request or
1193 * we need to read and perhaps cache.
1194 */
1195 data = read_sha1_file(oid->hash, &type, &size);
1196 if (!data)
1197 return NULL;
1198 if (type != OBJ_TREE) {
1199 free(data);
1200 return NULL;
1201 }
1202
1203 /* We need to either cache or return a throwaway copy */
1204
1205 if (available_ix < 0)
1206 ent = NULL;
1207 else {
1208 ent = pbase_tree_cache[available_ix];
1209 my_ix = available_ix;
1210 }
1211
1212 if (!ent) {
1213 nent = xmalloc(sizeof(*nent));
1214 nent->temporary = (available_ix < 0);
1215 }
1216 else {
1217 /* evict and reuse */
1218 free(ent->tree_data);
1219 nent = ent;
1220 }
1221 oidcpy(&nent->oid, oid);
1222 nent->tree_data = data;
1223 nent->tree_size = size;
1224 nent->ref = 1;
1225 if (!nent->temporary)
1226 pbase_tree_cache[my_ix] = nent;
1227 return nent;
1228 }
1229
1230 static void pbase_tree_put(struct pbase_tree_cache *cache)
1231 {
1232 if (!cache->temporary) {
1233 cache->ref--;
1234 return;
1235 }
1236 free(cache->tree_data);
1237 free(cache);
1238 }
1239
1240 static int name_cmp_len(const char *name)
1241 {
1242 int i;
1243 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1244 ;
1245 return i;
1246 }
1247
1248 static void add_pbase_object(struct tree_desc *tree,
1249 const char *name,
1250 int cmplen,
1251 const char *fullname)
1252 {
1253 struct name_entry entry;
1254 int cmp;
1255
1256 while (tree_entry(tree,&entry)) {
1257 if (S_ISGITLINK(entry.mode))
1258 continue;
1259 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1260 memcmp(name, entry.path, cmplen);
1261 if (cmp > 0)
1262 continue;
1263 if (cmp < 0)
1264 return;
1265 if (name[cmplen] != '/') {
1266 add_object_entry(entry.oid,
1267 object_type(entry.mode),
1268 fullname, 1);
1269 return;
1270 }
1271 if (S_ISDIR(entry.mode)) {
1272 struct tree_desc sub;
1273 struct pbase_tree_cache *tree;
1274 const char *down = name+cmplen+1;
1275 int downlen = name_cmp_len(down);
1276
1277 tree = pbase_tree_get(entry.oid);
1278 if (!tree)
1279 return;
1280 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1281
1282 add_pbase_object(&sub, down, downlen, fullname);
1283 pbase_tree_put(tree);
1284 }
1285 }
1286 }
1287
1288 static unsigned *done_pbase_paths;
1289 static int done_pbase_paths_num;
1290 static int done_pbase_paths_alloc;
1291 static int done_pbase_path_pos(unsigned hash)
1292 {
1293 int lo = 0;
1294 int hi = done_pbase_paths_num;
1295 while (lo < hi) {
1296 int mi = lo + (hi - lo) / 2;
1297 if (done_pbase_paths[mi] == hash)
1298 return mi;
1299 if (done_pbase_paths[mi] < hash)
1300 hi = mi;
1301 else
1302 lo = mi + 1;
1303 }
1304 return -lo-1;
1305 }
1306
1307 static int check_pbase_path(unsigned hash)
1308 {
1309 int pos = done_pbase_path_pos(hash);
1310 if (0 <= pos)
1311 return 1;
1312 pos = -pos - 1;
1313 ALLOC_GROW(done_pbase_paths,
1314 done_pbase_paths_num + 1,
1315 done_pbase_paths_alloc);
1316 done_pbase_paths_num++;
1317 if (pos < done_pbase_paths_num)
1318 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1319 done_pbase_paths_num - pos - 1);
1320 done_pbase_paths[pos] = hash;
1321 return 0;
1322 }
1323
1324 static void add_preferred_base_object(const char *name)
1325 {
1326 struct pbase_tree *it;
1327 int cmplen;
1328 unsigned hash = pack_name_hash(name);
1329
1330 if (!num_preferred_base || check_pbase_path(hash))
1331 return;
1332
1333 cmplen = name_cmp_len(name);
1334 for (it = pbase_tree; it; it = it->next) {
1335 if (cmplen == 0) {
1336 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1337 }
1338 else {
1339 struct tree_desc tree;
1340 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1341 add_pbase_object(&tree, name, cmplen, name);
1342 }
1343 }
1344 }
1345
1346 static void add_preferred_base(struct object_id *oid)
1347 {
1348 struct pbase_tree *it;
1349 void *data;
1350 unsigned long size;
1351 struct object_id tree_oid;
1352
1353 if (window <= num_preferred_base++)
1354 return;
1355
1356 data = read_object_with_reference(oid->hash, tree_type, &size, tree_oid.hash);
1357 if (!data)
1358 return;
1359
1360 for (it = pbase_tree; it; it = it->next) {
1361 if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1362 free(data);
1363 return;
1364 }
1365 }
1366
1367 it = xcalloc(1, sizeof(*it));
1368 it->next = pbase_tree;
1369 pbase_tree = it;
1370
1371 oidcpy(&it->pcache.oid, &tree_oid);
1372 it->pcache.tree_data = data;
1373 it->pcache.tree_size = size;
1374 }
1375
1376 static void cleanup_preferred_base(void)
1377 {
1378 struct pbase_tree *it;
1379 unsigned i;
1380
1381 it = pbase_tree;
1382 pbase_tree = NULL;
1383 while (it) {
1384 struct pbase_tree *this = it;
1385 it = this->next;
1386 free(this->pcache.tree_data);
1387 free(this);
1388 }
1389
1390 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1391 if (!pbase_tree_cache[i])
1392 continue;
1393 free(pbase_tree_cache[i]->tree_data);
1394 FREE_AND_NULL(pbase_tree_cache[i]);
1395 }
1396
1397 FREE_AND_NULL(done_pbase_paths);
1398 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1399 }
1400
1401 static void check_object(struct object_entry *entry)
1402 {
1403 if (entry->in_pack) {
1404 struct packed_git *p = entry->in_pack;
1405 struct pack_window *w_curs = NULL;
1406 const unsigned char *base_ref = NULL;
1407 struct object_entry *base_entry;
1408 unsigned long used, used_0;
1409 unsigned long avail;
1410 off_t ofs;
1411 unsigned char *buf, c;
1412
1413 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1414
1415 /*
1416 * We want in_pack_type even if we do not reuse delta
1417 * since non-delta representations could still be reused.
1418 */
1419 used = unpack_object_header_buffer(buf, avail,
1420 &entry->in_pack_type,
1421 &entry->size);
1422 if (used == 0)
1423 goto give_up;
1424
1425 /*
1426 * Determine if this is a delta and if so whether we can
1427 * reuse it or not. Otherwise let's find out as cheaply as
1428 * possible what the actual type and size for this object is.
1429 */
1430 switch (entry->in_pack_type) {
1431 default:
1432 /* Not a delta hence we've already got all we need. */
1433 entry->type = entry->in_pack_type;
1434 entry->in_pack_header_size = used;
1435 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1436 goto give_up;
1437 unuse_pack(&w_curs);
1438 return;
1439 case OBJ_REF_DELTA:
1440 if (reuse_delta && !entry->preferred_base)
1441 base_ref = use_pack(p, &w_curs,
1442 entry->in_pack_offset + used, NULL);
1443 entry->in_pack_header_size = used + 20;
1444 break;
1445 case OBJ_OFS_DELTA:
1446 buf = use_pack(p, &w_curs,
1447 entry->in_pack_offset + used, NULL);
1448 used_0 = 0;
1449 c = buf[used_0++];
1450 ofs = c & 127;
1451 while (c & 128) {
1452 ofs += 1;
1453 if (!ofs || MSB(ofs, 7)) {
1454 error("delta base offset overflow in pack for %s",
1455 oid_to_hex(&entry->idx.oid));
1456 goto give_up;
1457 }
1458 c = buf[used_0++];
1459 ofs = (ofs << 7) + (c & 127);
1460 }
1461 ofs = entry->in_pack_offset - ofs;
1462 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1463 error("delta base offset out of bound for %s",
1464 oid_to_hex(&entry->idx.oid));
1465 goto give_up;
1466 }
1467 if (reuse_delta && !entry->preferred_base) {
1468 struct revindex_entry *revidx;
1469 revidx = find_pack_revindex(p, ofs);
1470 if (!revidx)
1471 goto give_up;
1472 base_ref = nth_packed_object_sha1(p, revidx->nr);
1473 }
1474 entry->in_pack_header_size = used + used_0;
1475 break;
1476 }
1477
1478 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1479 /*
1480 * If base_ref was set above that means we wish to
1481 * reuse delta data, and we even found that base
1482 * in the list of objects we want to pack. Goodie!
1483 *
1484 * Depth value does not matter - find_deltas() will
1485 * never consider reused delta as the base object to
1486 * deltify other objects against, in order to avoid
1487 * circular deltas.
1488 */
1489 entry->type = entry->in_pack_type;
1490 entry->delta = base_entry;
1491 entry->delta_size = entry->size;
1492 entry->delta_sibling = base_entry->delta_child;
1493 base_entry->delta_child = entry;
1494 unuse_pack(&w_curs);
1495 return;
1496 }
1497
1498 if (entry->type) {
1499 /*
1500 * This must be a delta and we already know what the
1501 * final object type is. Let's extract the actual
1502 * object size from the delta header.
1503 */
1504 entry->size = get_size_from_delta(p, &w_curs,
1505 entry->in_pack_offset + entry->in_pack_header_size);
1506 if (entry->size == 0)
1507 goto give_up;
1508 unuse_pack(&w_curs);
1509 return;
1510 }
1511
1512 /*
1513 * No choice but to fall back to the recursive delta walk
1514 * with sha1_object_info() to find about the object type
1515 * at this point...
1516 */
1517 give_up:
1518 unuse_pack(&w_curs);
1519 }
1520
1521 entry->type = sha1_object_info(entry->idx.oid.hash, &entry->size);
1522 /*
1523 * The error condition is checked in prepare_pack(). This is
1524 * to permit a missing preferred base object to be ignored
1525 * as a preferred base. Doing so can result in a larger
1526 * pack file, but the transfer will still take place.
1527 */
1528 }
1529
1530 static int pack_offset_sort(const void *_a, const void *_b)
1531 {
1532 const struct object_entry *a = *(struct object_entry **)_a;
1533 const struct object_entry *b = *(struct object_entry **)_b;
1534
1535 /* avoid filesystem trashing with loose objects */
1536 if (!a->in_pack && !b->in_pack)
1537 return oidcmp(&a->idx.oid, &b->idx.oid);
1538
1539 if (a->in_pack < b->in_pack)
1540 return -1;
1541 if (a->in_pack > b->in_pack)
1542 return 1;
1543 return a->in_pack_offset < b->in_pack_offset ? -1 :
1544 (a->in_pack_offset > b->in_pack_offset);
1545 }
1546
1547 /*
1548 * Drop an on-disk delta we were planning to reuse. Naively, this would
1549 * just involve blanking out the "delta" field, but we have to deal
1550 * with some extra book-keeping:
1551 *
1552 * 1. Removing ourselves from the delta_sibling linked list.
1553 *
1554 * 2. Updating our size/type to the non-delta representation. These were
1555 * either not recorded initially (size) or overwritten with the delta type
1556 * (type) when check_object() decided to reuse the delta.
1557 *
1558 * 3. Resetting our delta depth, as we are now a base object.
1559 */
1560 static void drop_reused_delta(struct object_entry *entry)
1561 {
1562 struct object_entry **p = &entry->delta->delta_child;
1563 struct object_info oi = OBJECT_INFO_INIT;
1564
1565 while (*p) {
1566 if (*p == entry)
1567 *p = (*p)->delta_sibling;
1568 else
1569 p = &(*p)->delta_sibling;
1570 }
1571 entry->delta = NULL;
1572 entry->depth = 0;
1573
1574 oi.sizep = &entry->size;
1575 oi.typep = &entry->type;
1576 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1577 /*
1578 * We failed to get the info from this pack for some reason;
1579 * fall back to sha1_object_info, which may find another copy.
1580 * And if that fails, the error will be recorded in entry->type
1581 * and dealt with in prepare_pack().
1582 */
1583 entry->type = sha1_object_info(entry->idx.oid.hash,
1584 &entry->size);
1585 }
1586 }
1587
1588 /*
1589 * Follow the chain of deltas from this entry onward, throwing away any links
1590 * that cause us to hit a cycle (as determined by the DFS state flags in
1591 * the entries).
1592 *
1593 * We also detect too-long reused chains that would violate our --depth
1594 * limit.
1595 */
1596 static void break_delta_chains(struct object_entry *entry)
1597 {
1598 /*
1599 * The actual depth of each object we will write is stored as an int,
1600 * as it cannot exceed our int "depth" limit. But before we break
1601 * changes based no that limit, we may potentially go as deep as the
1602 * number of objects, which is elsewhere bounded to a uint32_t.
1603 */
1604 uint32_t total_depth;
1605 struct object_entry *cur, *next;
1606
1607 for (cur = entry, total_depth = 0;
1608 cur;
1609 cur = cur->delta, total_depth++) {
1610 if (cur->dfs_state == DFS_DONE) {
1611 /*
1612 * We've already seen this object and know it isn't
1613 * part of a cycle. We do need to append its depth
1614 * to our count.
1615 */
1616 total_depth += cur->depth;
1617 break;
1618 }
1619
1620 /*
1621 * We break cycles before looping, so an ACTIVE state (or any
1622 * other cruft which made its way into the state variable)
1623 * is a bug.
1624 */
1625 if (cur->dfs_state != DFS_NONE)
1626 die("BUG: confusing delta dfs state in first pass: %d",
1627 cur->dfs_state);
1628
1629 /*
1630 * Now we know this is the first time we've seen the object. If
1631 * it's not a delta, we're done traversing, but we'll mark it
1632 * done to save time on future traversals.
1633 */
1634 if (!cur->delta) {
1635 cur->dfs_state = DFS_DONE;
1636 break;
1637 }
1638
1639 /*
1640 * Mark ourselves as active and see if the next step causes
1641 * us to cycle to another active object. It's important to do
1642 * this _before_ we loop, because it impacts where we make the
1643 * cut, and thus how our total_depth counter works.
1644 * E.g., We may see a partial loop like:
1645 *
1646 * A -> B -> C -> D -> B
1647 *
1648 * Cutting B->C breaks the cycle. But now the depth of A is
1649 * only 1, and our total_depth counter is at 3. The size of the
1650 * error is always one less than the size of the cycle we
1651 * broke. Commits C and D were "lost" from A's chain.
1652 *
1653 * If we instead cut D->B, then the depth of A is correct at 3.
1654 * We keep all commits in the chain that we examined.
1655 */
1656 cur->dfs_state = DFS_ACTIVE;
1657 if (cur->delta->dfs_state == DFS_ACTIVE) {
1658 drop_reused_delta(cur);
1659 cur->dfs_state = DFS_DONE;
1660 break;
1661 }
1662 }
1663
1664 /*
1665 * And now that we've gone all the way to the bottom of the chain, we
1666 * need to clear the active flags and set the depth fields as
1667 * appropriate. Unlike the loop above, which can quit when it drops a
1668 * delta, we need to keep going to look for more depth cuts. So we need
1669 * an extra "next" pointer to keep going after we reset cur->delta.
1670 */
1671 for (cur = entry; cur; cur = next) {
1672 next = cur->delta;
1673
1674 /*
1675 * We should have a chain of zero or more ACTIVE states down to
1676 * a final DONE. We can quit after the DONE, because either it
1677 * has no bases, or we've already handled them in a previous
1678 * call.
1679 */
1680 if (cur->dfs_state == DFS_DONE)
1681 break;
1682 else if (cur->dfs_state != DFS_ACTIVE)
1683 die("BUG: confusing delta dfs state in second pass: %d",
1684 cur->dfs_state);
1685
1686 /*
1687 * If the total_depth is more than depth, then we need to snip
1688 * the chain into two or more smaller chains that don't exceed
1689 * the maximum depth. Most of the resulting chains will contain
1690 * (depth + 1) entries (i.e., depth deltas plus one base), and
1691 * the last chain (i.e., the one containing entry) will contain
1692 * whatever entries are left over, namely
1693 * (total_depth % (depth + 1)) of them.
1694 *
1695 * Since we are iterating towards decreasing depth, we need to
1696 * decrement total_depth as we go, and we need to write to the
1697 * entry what its final depth will be after all of the
1698 * snipping. Since we're snipping into chains of length (depth
1699 * + 1) entries, the final depth of an entry will be its
1700 * original depth modulo (depth + 1). Any time we encounter an
1701 * entry whose final depth is supposed to be zero, we snip it
1702 * from its delta base, thereby making it so.
1703 */
1704 cur->depth = (total_depth--) % (depth + 1);
1705 if (!cur->depth)
1706 drop_reused_delta(cur);
1707
1708 cur->dfs_state = DFS_DONE;
1709 }
1710 }
1711
1712 static void get_object_details(void)
1713 {
1714 uint32_t i;
1715 struct object_entry **sorted_by_offset;
1716
1717 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1718 for (i = 0; i < to_pack.nr_objects; i++)
1719 sorted_by_offset[i] = to_pack.objects + i;
1720 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1721
1722 for (i = 0; i < to_pack.nr_objects; i++) {
1723 struct object_entry *entry = sorted_by_offset[i];
1724 check_object(entry);
1725 if (big_file_threshold < entry->size)
1726 entry->no_try_delta = 1;
1727 }
1728
1729 /*
1730 * This must happen in a second pass, since we rely on the delta
1731 * information for the whole list being completed.
1732 */
1733 for (i = 0; i < to_pack.nr_objects; i++)
1734 break_delta_chains(&to_pack.objects[i]);
1735
1736 free(sorted_by_offset);
1737 }
1738
1739 /*
1740 * We search for deltas in a list sorted by type, by filename hash, and then
1741 * by size, so that we see progressively smaller and smaller files.
1742 * That's because we prefer deltas to be from the bigger file
1743 * to the smaller -- deletes are potentially cheaper, but perhaps
1744 * more importantly, the bigger file is likely the more recent
1745 * one. The deepest deltas are therefore the oldest objects which are
1746 * less susceptible to be accessed often.
1747 */
1748 static int type_size_sort(const void *_a, const void *_b)
1749 {
1750 const struct object_entry *a = *(struct object_entry **)_a;
1751 const struct object_entry *b = *(struct object_entry **)_b;
1752
1753 if (a->type > b->type)
1754 return -1;
1755 if (a->type < b->type)
1756 return 1;
1757 if (a->hash > b->hash)
1758 return -1;
1759 if (a->hash < b->hash)
1760 return 1;
1761 if (a->preferred_base > b->preferred_base)
1762 return -1;
1763 if (a->preferred_base < b->preferred_base)
1764 return 1;
1765 if (a->size > b->size)
1766 return -1;
1767 if (a->size < b->size)
1768 return 1;
1769 return a < b ? -1 : (a > b); /* newest first */
1770 }
1771
1772 struct unpacked {
1773 struct object_entry *entry;
1774 void *data;
1775 struct delta_index *index;
1776 unsigned depth;
1777 };
1778
1779 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1780 unsigned long delta_size)
1781 {
1782 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1783 return 0;
1784
1785 if (delta_size < cache_max_small_delta_size)
1786 return 1;
1787
1788 /* cache delta, if objects are large enough compared to delta size */
1789 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1790 return 1;
1791
1792 return 0;
1793 }
1794
1795 #ifndef NO_PTHREADS
1796
1797 static pthread_mutex_t read_mutex;
1798 #define read_lock() pthread_mutex_lock(&read_mutex)
1799 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1800
1801 static pthread_mutex_t cache_mutex;
1802 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1803 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1804
1805 static pthread_mutex_t progress_mutex;
1806 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1807 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1808
1809 #else
1810
1811 #define read_lock() (void)0
1812 #define read_unlock() (void)0
1813 #define cache_lock() (void)0
1814 #define cache_unlock() (void)0
1815 #define progress_lock() (void)0
1816 #define progress_unlock() (void)0
1817
1818 #endif
1819
1820 static int try_delta(struct unpacked *trg, struct unpacked *src,
1821 unsigned max_depth, unsigned long *mem_usage)
1822 {
1823 struct object_entry *trg_entry = trg->entry;
1824 struct object_entry *src_entry = src->entry;
1825 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1826 unsigned ref_depth;
1827 enum object_type type;
1828 void *delta_buf;
1829
1830 /* Don't bother doing diffs between different types */
1831 if (trg_entry->type != src_entry->type)
1832 return -1;
1833
1834 /*
1835 * We do not bother to try a delta that we discarded on an
1836 * earlier try, but only when reusing delta data. Note that
1837 * src_entry that is marked as the preferred_base should always
1838 * be considered, as even if we produce a suboptimal delta against
1839 * it, we will still save the transfer cost, as we already know
1840 * the other side has it and we won't send src_entry at all.
1841 */
1842 if (reuse_delta && trg_entry->in_pack &&
1843 trg_entry->in_pack == src_entry->in_pack &&
1844 !src_entry->preferred_base &&
1845 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1846 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1847 return 0;
1848
1849 /* Let's not bust the allowed depth. */
1850 if (src->depth >= max_depth)
1851 return 0;
1852
1853 /* Now some size filtering heuristics. */
1854 trg_size = trg_entry->size;
1855 if (!trg_entry->delta) {
1856 max_size = trg_size/2 - 20;
1857 ref_depth = 1;
1858 } else {
1859 max_size = trg_entry->delta_size;
1860 ref_depth = trg->depth;
1861 }
1862 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1863 (max_depth - ref_depth + 1);
1864 if (max_size == 0)
1865 return 0;
1866 src_size = src_entry->size;
1867 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1868 if (sizediff >= max_size)
1869 return 0;
1870 if (trg_size < src_size / 32)
1871 return 0;
1872
1873 /* Load data if not already done */
1874 if (!trg->data) {
1875 read_lock();
1876 trg->data = read_sha1_file(trg_entry->idx.oid.hash, &type,
1877 &sz);
1878 read_unlock();
1879 if (!trg->data)
1880 die("object %s cannot be read",
1881 oid_to_hex(&trg_entry->idx.oid));
1882 if (sz != trg_size)
1883 die("object %s inconsistent object length (%lu vs %lu)",
1884 oid_to_hex(&trg_entry->idx.oid), sz,
1885 trg_size);
1886 *mem_usage += sz;
1887 }
1888 if (!src->data) {
1889 read_lock();
1890 src->data = read_sha1_file(src_entry->idx.oid.hash, &type,
1891 &sz);
1892 read_unlock();
1893 if (!src->data) {
1894 if (src_entry->preferred_base) {
1895 static int warned = 0;
1896 if (!warned++)
1897 warning("object %s cannot be read",
1898 oid_to_hex(&src_entry->idx.oid));
1899 /*
1900 * Those objects are not included in the
1901 * resulting pack. Be resilient and ignore
1902 * them if they can't be read, in case the
1903 * pack could be created nevertheless.
1904 */
1905 return 0;
1906 }
1907 die("object %s cannot be read",
1908 oid_to_hex(&src_entry->idx.oid));
1909 }
1910 if (sz != src_size)
1911 die("object %s inconsistent object length (%lu vs %lu)",
1912 oid_to_hex(&src_entry->idx.oid), sz,
1913 src_size);
1914 *mem_usage += sz;
1915 }
1916 if (!src->index) {
1917 src->index = create_delta_index(src->data, src_size);
1918 if (!src->index) {
1919 static int warned = 0;
1920 if (!warned++)
1921 warning("suboptimal pack - out of memory");
1922 return 0;
1923 }
1924 *mem_usage += sizeof_delta_index(src->index);
1925 }
1926
1927 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1928 if (!delta_buf)
1929 return 0;
1930
1931 if (trg_entry->delta) {
1932 /* Prefer only shallower same-sized deltas. */
1933 if (delta_size == trg_entry->delta_size &&
1934 src->depth + 1 >= trg->depth) {
1935 free(delta_buf);
1936 return 0;
1937 }
1938 }
1939
1940 /*
1941 * Handle memory allocation outside of the cache
1942 * accounting lock. Compiler will optimize the strangeness
1943 * away when NO_PTHREADS is defined.
1944 */
1945 free(trg_entry->delta_data);
1946 cache_lock();
1947 if (trg_entry->delta_data) {
1948 delta_cache_size -= trg_entry->delta_size;
1949 trg_entry->delta_data = NULL;
1950 }
1951 if (delta_cacheable(src_size, trg_size, delta_size)) {
1952 delta_cache_size += delta_size;
1953 cache_unlock();
1954 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1955 } else {
1956 cache_unlock();
1957 free(delta_buf);
1958 }
1959
1960 trg_entry->delta = src_entry;
1961 trg_entry->delta_size = delta_size;
1962 trg->depth = src->depth + 1;
1963
1964 return 1;
1965 }
1966
1967 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1968 {
1969 struct object_entry *child = me->delta_child;
1970 unsigned int m = n;
1971 while (child) {
1972 unsigned int c = check_delta_limit(child, n + 1);
1973 if (m < c)
1974 m = c;
1975 child = child->delta_sibling;
1976 }
1977 return m;
1978 }
1979
1980 static unsigned long free_unpacked(struct unpacked *n)
1981 {
1982 unsigned long freed_mem = sizeof_delta_index(n->index);
1983 free_delta_index(n->index);
1984 n->index = NULL;
1985 if (n->data) {
1986 freed_mem += n->entry->size;
1987 FREE_AND_NULL(n->data);
1988 }
1989 n->entry = NULL;
1990 n->depth = 0;
1991 return freed_mem;
1992 }
1993
1994 static void find_deltas(struct object_entry **list, unsigned *list_size,
1995 int window, int depth, unsigned *processed)
1996 {
1997 uint32_t i, idx = 0, count = 0;
1998 struct unpacked *array;
1999 unsigned long mem_usage = 0;
2000
2001 array = xcalloc(window, sizeof(struct unpacked));
2002
2003 for (;;) {
2004 struct object_entry *entry;
2005 struct unpacked *n = array + idx;
2006 int j, max_depth, best_base = -1;
2007
2008 progress_lock();
2009 if (!*list_size) {
2010 progress_unlock();
2011 break;
2012 }
2013 entry = *list++;
2014 (*list_size)--;
2015 if (!entry->preferred_base) {
2016 (*processed)++;
2017 display_progress(progress_state, *processed);
2018 }
2019 progress_unlock();
2020
2021 mem_usage -= free_unpacked(n);
2022 n->entry = entry;
2023
2024 while (window_memory_limit &&
2025 mem_usage > window_memory_limit &&
2026 count > 1) {
2027 uint32_t tail = (idx + window - count) % window;
2028 mem_usage -= free_unpacked(array + tail);
2029 count--;
2030 }
2031
2032 /* We do not compute delta to *create* objects we are not
2033 * going to pack.
2034 */
2035 if (entry->preferred_base)
2036 goto next;
2037
2038 /*
2039 * If the current object is at pack edge, take the depth the
2040 * objects that depend on the current object into account
2041 * otherwise they would become too deep.
2042 */
2043 max_depth = depth;
2044 if (entry->delta_child) {
2045 max_depth -= check_delta_limit(entry, 0);
2046 if (max_depth <= 0)
2047 goto next;
2048 }
2049
2050 j = window;
2051 while (--j > 0) {
2052 int ret;
2053 uint32_t other_idx = idx + j;
2054 struct unpacked *m;
2055 if (other_idx >= window)
2056 other_idx -= window;
2057 m = array + other_idx;
2058 if (!m->entry)
2059 break;
2060 ret = try_delta(n, m, max_depth, &mem_usage);
2061 if (ret < 0)
2062 break;
2063 else if (ret > 0)
2064 best_base = other_idx;
2065 }
2066
2067 /*
2068 * If we decided to cache the delta data, then it is best
2069 * to compress it right away. First because we have to do
2070 * it anyway, and doing it here while we're threaded will
2071 * save a lot of time in the non threaded write phase,
2072 * as well as allow for caching more deltas within
2073 * the same cache size limit.
2074 * ...
2075 * But only if not writing to stdout, since in that case
2076 * the network is most likely throttling writes anyway,
2077 * and therefore it is best to go to the write phase ASAP
2078 * instead, as we can afford spending more time compressing
2079 * between writes at that moment.
2080 */
2081 if (entry->delta_data && !pack_to_stdout) {
2082 entry->z_delta_size = do_compress(&entry->delta_data,
2083 entry->delta_size);
2084 cache_lock();
2085 delta_cache_size -= entry->delta_size;
2086 delta_cache_size += entry->z_delta_size;
2087 cache_unlock();
2088 }
2089
2090 /* if we made n a delta, and if n is already at max
2091 * depth, leaving it in the window is pointless. we
2092 * should evict it first.
2093 */
2094 if (entry->delta && max_depth <= n->depth)
2095 continue;
2096
2097 /*
2098 * Move the best delta base up in the window, after the
2099 * currently deltified object, to keep it longer. It will
2100 * be the first base object to be attempted next.
2101 */
2102 if (entry->delta) {
2103 struct unpacked swap = array[best_base];
2104 int dist = (window + idx - best_base) % window;
2105 int dst = best_base;
2106 while (dist--) {
2107 int src = (dst + 1) % window;
2108 array[dst] = array[src];
2109 dst = src;
2110 }
2111 array[dst] = swap;
2112 }
2113
2114 next:
2115 idx++;
2116 if (count + 1 < window)
2117 count++;
2118 if (idx >= window)
2119 idx = 0;
2120 }
2121
2122 for (i = 0; i < window; ++i) {
2123 free_delta_index(array[i].index);
2124 free(array[i].data);
2125 }
2126 free(array);
2127 }
2128
2129 #ifndef NO_PTHREADS
2130
2131 static void try_to_free_from_threads(size_t size)
2132 {
2133 read_lock();
2134 release_pack_memory(size);
2135 read_unlock();
2136 }
2137
2138 static try_to_free_t old_try_to_free_routine;
2139
2140 /*
2141 * The main thread waits on the condition that (at least) one of the workers
2142 * has stopped working (which is indicated in the .working member of
2143 * struct thread_params).
2144 * When a work thread has completed its work, it sets .working to 0 and
2145 * signals the main thread and waits on the condition that .data_ready
2146 * becomes 1.
2147 */
2148
2149 struct thread_params {
2150 pthread_t thread;
2151 struct object_entry **list;
2152 unsigned list_size;
2153 unsigned remaining;
2154 int window;
2155 int depth;
2156 int working;
2157 int data_ready;
2158 pthread_mutex_t mutex;
2159 pthread_cond_t cond;
2160 unsigned *processed;
2161 };
2162
2163 static pthread_cond_t progress_cond;
2164
2165 /*
2166 * Mutex and conditional variable can't be statically-initialized on Windows.
2167 */
2168 static void init_threaded_search(void)
2169 {
2170 init_recursive_mutex(&read_mutex);
2171 pthread_mutex_init(&cache_mutex, NULL);
2172 pthread_mutex_init(&progress_mutex, NULL);
2173 pthread_cond_init(&progress_cond, NULL);
2174 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2175 }
2176
2177 static void cleanup_threaded_search(void)
2178 {
2179 set_try_to_free_routine(old_try_to_free_routine);
2180 pthread_cond_destroy(&progress_cond);
2181 pthread_mutex_destroy(&read_mutex);
2182 pthread_mutex_destroy(&cache_mutex);
2183 pthread_mutex_destroy(&progress_mutex);
2184 }
2185
2186 static void *threaded_find_deltas(void *arg)
2187 {
2188 struct thread_params *me = arg;
2189
2190 progress_lock();
2191 while (me->remaining) {
2192 progress_unlock();
2193
2194 find_deltas(me->list, &me->remaining,
2195 me->window, me->depth, me->processed);
2196
2197 progress_lock();
2198 me->working = 0;
2199 pthread_cond_signal(&progress_cond);
2200 progress_unlock();
2201
2202 /*
2203 * We must not set ->data_ready before we wait on the
2204 * condition because the main thread may have set it to 1
2205 * before we get here. In order to be sure that new
2206 * work is available if we see 1 in ->data_ready, it
2207 * was initialized to 0 before this thread was spawned
2208 * and we reset it to 0 right away.
2209 */
2210 pthread_mutex_lock(&me->mutex);
2211 while (!me->data_ready)
2212 pthread_cond_wait(&me->cond, &me->mutex);
2213 me->data_ready = 0;
2214 pthread_mutex_unlock(&me->mutex);
2215
2216 progress_lock();
2217 }
2218 progress_unlock();
2219 /* leave ->working 1 so that this doesn't get more work assigned */
2220 return NULL;
2221 }
2222
2223 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2224 int window, int depth, unsigned *processed)
2225 {
2226 struct thread_params *p;
2227 int i, ret, active_threads = 0;
2228
2229 init_threaded_search();
2230
2231 if (delta_search_threads <= 1) {
2232 find_deltas(list, &list_size, window, depth, processed);
2233 cleanup_threaded_search();
2234 return;
2235 }
2236 if (progress > pack_to_stdout)
2237 fprintf(stderr, "Delta compression using up to %d threads.\n",
2238 delta_search_threads);
2239 p = xcalloc(delta_search_threads, sizeof(*p));
2240
2241 /* Partition the work amongst work threads. */
2242 for (i = 0; i < delta_search_threads; i++) {
2243 unsigned sub_size = list_size / (delta_search_threads - i);
2244
2245 /* don't use too small segments or no deltas will be found */
2246 if (sub_size < 2*window && i+1 < delta_search_threads)
2247 sub_size = 0;
2248
2249 p[i].window = window;
2250 p[i].depth = depth;
2251 p[i].processed = processed;
2252 p[i].working = 1;
2253 p[i].data_ready = 0;
2254
2255 /* try to split chunks on "path" boundaries */
2256 while (sub_size && sub_size < list_size &&
2257 list[sub_size]->hash &&
2258 list[sub_size]->hash == list[sub_size-1]->hash)
2259 sub_size++;
2260
2261 p[i].list = list;
2262 p[i].list_size = sub_size;
2263 p[i].remaining = sub_size;
2264
2265 list += sub_size;
2266 list_size -= sub_size;
2267 }
2268
2269 /* Start work threads. */
2270 for (i = 0; i < delta_search_threads; i++) {
2271 if (!p[i].list_size)
2272 continue;
2273 pthread_mutex_init(&p[i].mutex, NULL);
2274 pthread_cond_init(&p[i].cond, NULL);
2275 ret = pthread_create(&p[i].thread, NULL,
2276 threaded_find_deltas, &p[i]);
2277 if (ret)
2278 die("unable to create thread: %s", strerror(ret));
2279 active_threads++;
2280 }
2281
2282 /*
2283 * Now let's wait for work completion. Each time a thread is done
2284 * with its work, we steal half of the remaining work from the
2285 * thread with the largest number of unprocessed objects and give
2286 * it to that newly idle thread. This ensure good load balancing
2287 * until the remaining object list segments are simply too short
2288 * to be worth splitting anymore.
2289 */
2290 while (active_threads) {
2291 struct thread_params *target = NULL;
2292 struct thread_params *victim = NULL;
2293 unsigned sub_size = 0;
2294
2295 progress_lock();
2296 for (;;) {
2297 for (i = 0; !target && i < delta_search_threads; i++)
2298 if (!p[i].working)
2299 target = &p[i];
2300 if (target)
2301 break;
2302 pthread_cond_wait(&progress_cond, &progress_mutex);
2303 }
2304
2305 for (i = 0; i < delta_search_threads; i++)
2306 if (p[i].remaining > 2*window &&
2307 (!victim || victim->remaining < p[i].remaining))
2308 victim = &p[i];
2309 if (victim) {
2310 sub_size = victim->remaining / 2;
2311 list = victim->list + victim->list_size - sub_size;
2312 while (sub_size && list[0]->hash &&
2313 list[0]->hash == list[-1]->hash) {
2314 list++;
2315 sub_size--;
2316 }
2317 if (!sub_size) {
2318 /*
2319 * It is possible for some "paths" to have
2320 * so many objects that no hash boundary
2321 * might be found. Let's just steal the
2322 * exact half in that case.
2323 */
2324 sub_size = victim->remaining / 2;
2325 list -= sub_size;
2326 }
2327 target->list = list;
2328 victim->list_size -= sub_size;
2329 victim->remaining -= sub_size;
2330 }
2331 target->list_size = sub_size;
2332 target->remaining = sub_size;
2333 target->working = 1;
2334 progress_unlock();
2335
2336 pthread_mutex_lock(&target->mutex);
2337 target->data_ready = 1;
2338 pthread_cond_signal(&target->cond);
2339 pthread_mutex_unlock(&target->mutex);
2340
2341 if (!sub_size) {
2342 pthread_join(target->thread, NULL);
2343 pthread_cond_destroy(&target->cond);
2344 pthread_mutex_destroy(&target->mutex);
2345 active_threads--;
2346 }
2347 }
2348 cleanup_threaded_search();
2349 free(p);
2350 }
2351
2352 #else
2353 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2354 #endif
2355
2356 static void add_tag_chain(const struct object_id *oid)
2357 {
2358 struct tag *tag;
2359
2360 /*
2361 * We catch duplicates already in add_object_entry(), but we'd
2362 * prefer to do this extra check to avoid having to parse the
2363 * tag at all if we already know that it's being packed (e.g., if
2364 * it was included via bitmaps, we would not have parsed it
2365 * previously).
2366 */
2367 if (packlist_find(&to_pack, oid->hash, NULL))
2368 return;
2369
2370 tag = lookup_tag(oid);
2371 while (1) {
2372 if (!tag || parse_tag(tag) || !tag->tagged)
2373 die("unable to pack objects reachable from tag %s",
2374 oid_to_hex(oid));
2375
2376 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2377
2378 if (tag->tagged->type != OBJ_TAG)
2379 return;
2380
2381 tag = (struct tag *)tag->tagged;
2382 }
2383 }
2384
2385 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2386 {
2387 struct object_id peeled;
2388
2389 if (starts_with(path, "refs/tags/") && /* is a tag? */
2390 !peel_ref(path, &peeled) && /* peelable? */
2391 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2392 add_tag_chain(oid);
2393 return 0;
2394 }
2395
2396 static void prepare_pack(int window, int depth)
2397 {
2398 struct object_entry **delta_list;
2399 uint32_t i, nr_deltas;
2400 unsigned n;
2401
2402 get_object_details();
2403
2404 /*
2405 * If we're locally repacking then we need to be doubly careful
2406 * from now on in order to make sure no stealth corruption gets
2407 * propagated to the new pack. Clients receiving streamed packs
2408 * should validate everything they get anyway so no need to incur
2409 * the additional cost here in that case.
2410 */
2411 if (!pack_to_stdout)
2412 do_check_packed_object_crc = 1;
2413
2414 if (!to_pack.nr_objects || !window || !depth)
2415 return;
2416
2417 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2418 nr_deltas = n = 0;
2419
2420 for (i = 0; i < to_pack.nr_objects; i++) {
2421 struct object_entry *entry = to_pack.objects + i;
2422
2423 if (entry->delta)
2424 /* This happens if we decided to reuse existing
2425 * delta from a pack. "reuse_delta &&" is implied.
2426 */
2427 continue;
2428
2429 if (entry->size < 50)
2430 continue;
2431
2432 if (entry->no_try_delta)
2433 continue;
2434
2435 if (!entry->preferred_base) {
2436 nr_deltas++;
2437 if (entry->type < 0)
2438 die("unable to get type of object %s",
2439 oid_to_hex(&entry->idx.oid));
2440 } else {
2441 if (entry->type < 0) {
2442 /*
2443 * This object is not found, but we
2444 * don't have to include it anyway.
2445 */
2446 continue;
2447 }
2448 }
2449
2450 delta_list[n++] = entry;
2451 }
2452
2453 if (nr_deltas && n > 1) {
2454 unsigned nr_done = 0;
2455 if (progress)
2456 progress_state = start_progress(_("Compressing objects"),
2457 nr_deltas);
2458 QSORT(delta_list, n, type_size_sort);
2459 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2460 stop_progress(&progress_state);
2461 if (nr_done != nr_deltas)
2462 die("inconsistency with delta count");
2463 }
2464 free(delta_list);
2465 }
2466
2467 static int git_pack_config(const char *k, const char *v, void *cb)
2468 {
2469 if (!strcmp(k, "pack.window")) {
2470 window = git_config_int(k, v);
2471 return 0;
2472 }
2473 if (!strcmp(k, "pack.windowmemory")) {
2474 window_memory_limit = git_config_ulong(k, v);
2475 return 0;
2476 }
2477 if (!strcmp(k, "pack.depth")) {
2478 depth = git_config_int(k, v);
2479 return 0;
2480 }
2481 if (!strcmp(k, "pack.deltacachesize")) {
2482 max_delta_cache_size = git_config_int(k, v);
2483 return 0;
2484 }
2485 if (!strcmp(k, "pack.deltacachelimit")) {
2486 cache_max_small_delta_size = git_config_int(k, v);
2487 return 0;
2488 }
2489 if (!strcmp(k, "pack.writebitmaphashcache")) {
2490 if (git_config_bool(k, v))
2491 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2492 else
2493 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2494 }
2495 if (!strcmp(k, "pack.usebitmaps")) {
2496 use_bitmap_index_default = git_config_bool(k, v);
2497 return 0;
2498 }
2499 if (!strcmp(k, "pack.threads")) {
2500 delta_search_threads = git_config_int(k, v);
2501 if (delta_search_threads < 0)
2502 die("invalid number of threads specified (%d)",
2503 delta_search_threads);
2504 #ifdef NO_PTHREADS
2505 if (delta_search_threads != 1) {
2506 warning("no threads support, ignoring %s", k);
2507 delta_search_threads = 0;
2508 }
2509 #endif
2510 return 0;
2511 }
2512 if (!strcmp(k, "pack.indexversion")) {
2513 pack_idx_opts.version = git_config_int(k, v);
2514 if (pack_idx_opts.version > 2)
2515 die("bad pack.indexversion=%"PRIu32,
2516 pack_idx_opts.version);
2517 return 0;
2518 }
2519 return git_default_config(k, v, cb);
2520 }
2521
2522 static void read_object_list_from_stdin(void)
2523 {
2524 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2525 struct object_id oid;
2526 const char *p;
2527
2528 for (;;) {
2529 if (!fgets(line, sizeof(line), stdin)) {
2530 if (feof(stdin))
2531 break;
2532 if (!ferror(stdin))
2533 die("fgets returned NULL, not EOF, not error!");
2534 if (errno != EINTR)
2535 die_errno("fgets");
2536 clearerr(stdin);
2537 continue;
2538 }
2539 if (line[0] == '-') {
2540 if (get_oid_hex(line+1, &oid))
2541 die("expected edge object ID, got garbage:\n %s",
2542 line);
2543 add_preferred_base(&oid);
2544 continue;
2545 }
2546 if (parse_oid_hex(line, &oid, &p))
2547 die("expected object ID, got garbage:\n %s", line);
2548
2549 add_preferred_base_object(p + 1);
2550 add_object_entry(&oid, 0, p + 1, 0);
2551 }
2552 }
2553
2554 #define OBJECT_ADDED (1u<<20)
2555
2556 static void show_commit(struct commit *commit, void *data)
2557 {
2558 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2559 commit->object.flags |= OBJECT_ADDED;
2560
2561 if (write_bitmap_index)
2562 index_commit_for_bitmap(commit);
2563 }
2564
2565 static void show_object(struct object *obj, const char *name, void *data)
2566 {
2567 add_preferred_base_object(name);
2568 add_object_entry(&obj->oid, obj->type, name, 0);
2569 obj->flags |= OBJECT_ADDED;
2570 }
2571
2572 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2573 {
2574 assert(arg_missing_action == MA_ALLOW_ANY);
2575
2576 /*
2577 * Quietly ignore ALL missing objects. This avoids problems with
2578 * staging them now and getting an odd error later.
2579 */
2580 if (!has_object_file(&obj->oid))
2581 return;
2582
2583 show_object(obj, name, data);
2584 }
2585
2586 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2587 {
2588 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2589
2590 /*
2591 * Quietly ignore EXPECTED missing objects. This avoids problems with
2592 * staging them now and getting an odd error later.
2593 */
2594 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2595 return;
2596
2597 show_object(obj, name, data);
2598 }
2599
2600 static int option_parse_missing_action(const struct option *opt,
2601 const char *arg, int unset)
2602 {
2603 assert(arg);
2604 assert(!unset);
2605
2606 if (!strcmp(arg, "error")) {
2607 arg_missing_action = MA_ERROR;
2608 fn_show_object = show_object;
2609 return 0;
2610 }
2611
2612 if (!strcmp(arg, "allow-any")) {
2613 arg_missing_action = MA_ALLOW_ANY;
2614 fetch_if_missing = 0;
2615 fn_show_object = show_object__ma_allow_any;
2616 return 0;
2617 }
2618
2619 if (!strcmp(arg, "allow-promisor")) {
2620 arg_missing_action = MA_ALLOW_PROMISOR;
2621 fetch_if_missing = 0;
2622 fn_show_object = show_object__ma_allow_promisor;
2623 return 0;
2624 }
2625
2626 die(_("invalid value for --missing"));
2627 return 0;
2628 }
2629
2630 static void show_edge(struct commit *commit)
2631 {
2632 add_preferred_base(&commit->object.oid);
2633 }
2634
2635 struct in_pack_object {
2636 off_t offset;
2637 struct object *object;
2638 };
2639
2640 struct in_pack {
2641 unsigned int alloc;
2642 unsigned int nr;
2643 struct in_pack_object *array;
2644 };
2645
2646 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2647 {
2648 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2649 in_pack->array[in_pack->nr].object = object;
2650 in_pack->nr++;
2651 }
2652
2653 /*
2654 * Compare the objects in the offset order, in order to emulate the
2655 * "git rev-list --objects" output that produced the pack originally.
2656 */
2657 static int ofscmp(const void *a_, const void *b_)
2658 {
2659 struct in_pack_object *a = (struct in_pack_object *)a_;
2660 struct in_pack_object *b = (struct in_pack_object *)b_;
2661
2662 if (a->offset < b->offset)
2663 return -1;
2664 else if (a->offset > b->offset)
2665 return 1;
2666 else
2667 return oidcmp(&a->object->oid, &b->object->oid);
2668 }
2669
2670 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2671 {
2672 struct packed_git *p;
2673 struct in_pack in_pack;
2674 uint32_t i;
2675
2676 memset(&in_pack, 0, sizeof(in_pack));
2677
2678 for (p = get_packed_git(the_repository); p; p = p->next) {
2679 struct object_id oid;
2680 struct object *o;
2681
2682 if (!p->pack_local || p->pack_keep)
2683 continue;
2684 if (open_pack_index(p))
2685 die("cannot open pack index");
2686
2687 ALLOC_GROW(in_pack.array,
2688 in_pack.nr + p->num_objects,
2689 in_pack.alloc);
2690
2691 for (i = 0; i < p->num_objects; i++) {
2692 nth_packed_object_oid(&oid, p, i);
2693 o = lookup_unknown_object(oid.hash);
2694 if (!(o->flags & OBJECT_ADDED))
2695 mark_in_pack_object(o, p, &in_pack);
2696 o->flags |= OBJECT_ADDED;
2697 }
2698 }
2699
2700 if (in_pack.nr) {
2701 QSORT(in_pack.array, in_pack.nr, ofscmp);
2702 for (i = 0; i < in_pack.nr; i++) {
2703 struct object *o = in_pack.array[i].object;
2704 add_object_entry(&o->oid, o->type, "", 0);
2705 }
2706 }
2707 free(in_pack.array);
2708 }
2709
2710 static int add_loose_object(const struct object_id *oid, const char *path,
2711 void *data)
2712 {
2713 enum object_type type = sha1_object_info(oid->hash, NULL);
2714
2715 if (type < 0) {
2716 warning("loose object at %s could not be examined", path);
2717 return 0;
2718 }
2719
2720 add_object_entry(oid, type, "", 0);
2721 return 0;
2722 }
2723
2724 /*
2725 * We actually don't even have to worry about reachability here.
2726 * add_object_entry will weed out duplicates, so we just add every
2727 * loose object we find.
2728 */
2729 static void add_unreachable_loose_objects(void)
2730 {
2731 for_each_loose_file_in_objdir(get_object_directory(),
2732 add_loose_object,
2733 NULL, NULL, NULL);
2734 }
2735
2736 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2737 {
2738 static struct packed_git *last_found = (void *)1;
2739 struct packed_git *p;
2740
2741 p = (last_found != (void *)1) ? last_found :
2742 get_packed_git(the_repository);
2743
2744 while (p) {
2745 if ((!p->pack_local || p->pack_keep) &&
2746 find_pack_entry_one(oid->hash, p)) {
2747 last_found = p;
2748 return 1;
2749 }
2750 if (p == last_found)
2751 p = get_packed_git(the_repository);
2752 else
2753 p = p->next;
2754 if (p == last_found)
2755 p = p->next;
2756 }
2757 return 0;
2758 }
2759
2760 /*
2761 * Store a list of sha1s that are should not be discarded
2762 * because they are either written too recently, or are
2763 * reachable from another object that was.
2764 *
2765 * This is filled by get_object_list.
2766 */
2767 static struct oid_array recent_objects;
2768
2769 static int loosened_object_can_be_discarded(const struct object_id *oid,
2770 timestamp_t mtime)
2771 {
2772 if (!unpack_unreachable_expiration)
2773 return 0;
2774 if (mtime > unpack_unreachable_expiration)
2775 return 0;
2776 if (oid_array_lookup(&recent_objects, oid) >= 0)
2777 return 0;
2778 return 1;
2779 }
2780
2781 static void loosen_unused_packed_objects(struct rev_info *revs)
2782 {
2783 struct packed_git *p;
2784 uint32_t i;
2785 struct object_id oid;
2786
2787 for (p = get_packed_git(the_repository); p; p = p->next) {
2788 if (!p->pack_local || p->pack_keep)
2789 continue;
2790
2791 if (open_pack_index(p))
2792 die("cannot open pack index");
2793
2794 for (i = 0; i < p->num_objects; i++) {
2795 nth_packed_object_oid(&oid, p, i);
2796 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2797 !has_sha1_pack_kept_or_nonlocal(&oid) &&
2798 !loosened_object_can_be_discarded(&oid, p->mtime))
2799 if (force_object_loose(oid.hash, p->mtime))
2800 die("unable to force loose object");
2801 }
2802 }
2803 }
2804
2805 /*
2806 * This tracks any options which pack-reuse code expects to be on, or which a
2807 * reader of the pack might not understand, and which would therefore prevent
2808 * blind reuse of what we have on disk.
2809 */
2810 static int pack_options_allow_reuse(void)
2811 {
2812 return pack_to_stdout &&
2813 allow_ofs_delta &&
2814 !ignore_packed_keep &&
2815 (!local || !have_non_local_packs) &&
2816 !incremental;
2817 }
2818
2819 static int get_object_list_from_bitmap(struct rev_info *revs)
2820 {
2821 if (prepare_bitmap_walk(revs) < 0)
2822 return -1;
2823
2824 if (pack_options_allow_reuse() &&
2825 !reuse_partial_packfile_from_bitmap(
2826 &reuse_packfile,
2827 &reuse_packfile_objects,
2828 &reuse_packfile_offset)) {
2829 assert(reuse_packfile_objects);
2830 nr_result += reuse_packfile_objects;
2831 display_progress(progress_state, nr_result);
2832 }
2833
2834 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2835 return 0;
2836 }
2837
2838 static void record_recent_object(struct object *obj,
2839 const char *name,
2840 void *data)
2841 {
2842 oid_array_append(&recent_objects, &obj->oid);
2843 }
2844
2845 static void record_recent_commit(struct commit *commit, void *data)
2846 {
2847 oid_array_append(&recent_objects, &commit->object.oid);
2848 }
2849
2850 static void get_object_list(int ac, const char **av)
2851 {
2852 struct rev_info revs;
2853 char line[1000];
2854 int flags = 0;
2855
2856 init_revisions(&revs, NULL);
2857 save_commit_buffer = 0;
2858 setup_revisions(ac, av, &revs, NULL);
2859
2860 /* make sure shallows are read */
2861 is_repository_shallow();
2862
2863 while (fgets(line, sizeof(line), stdin) != NULL) {
2864 int len = strlen(line);
2865 if (len && line[len - 1] == '\n')
2866 line[--len] = 0;
2867 if (!len)
2868 break;
2869 if (*line == '-') {
2870 if (!strcmp(line, "--not")) {
2871 flags ^= UNINTERESTING;
2872 write_bitmap_index = 0;
2873 continue;
2874 }
2875 if (starts_with(line, "--shallow ")) {
2876 struct object_id oid;
2877 if (get_oid_hex(line + 10, &oid))
2878 die("not an SHA-1 '%s'", line + 10);
2879 register_shallow(&oid);
2880 use_bitmap_index = 0;
2881 continue;
2882 }
2883 die("not a rev '%s'", line);
2884 }
2885 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2886 die("bad revision '%s'", line);
2887 }
2888
2889 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2890 return;
2891
2892 if (prepare_revision_walk(&revs))
2893 die("revision walk setup failed");
2894 mark_edges_uninteresting(&revs, show_edge);
2895
2896 if (!fn_show_object)
2897 fn_show_object = show_object;
2898 traverse_commit_list_filtered(&filter_options, &revs,
2899 show_commit, fn_show_object, NULL,
2900 NULL);
2901
2902 if (unpack_unreachable_expiration) {
2903 revs.ignore_missing_links = 1;
2904 if (add_unseen_recent_objects_to_traversal(&revs,
2905 unpack_unreachable_expiration))
2906 die("unable to add recent objects");
2907 if (prepare_revision_walk(&revs))
2908 die("revision walk setup failed");
2909 traverse_commit_list(&revs, record_recent_commit,
2910 record_recent_object, NULL);
2911 }
2912
2913 if (keep_unreachable)
2914 add_objects_in_unpacked_packs(&revs);
2915 if (pack_loose_unreachable)
2916 add_unreachable_loose_objects();
2917 if (unpack_unreachable)
2918 loosen_unused_packed_objects(&revs);
2919
2920 oid_array_clear(&recent_objects);
2921 }
2922
2923 static int option_parse_index_version(const struct option *opt,
2924 const char *arg, int unset)
2925 {
2926 char *c;
2927 const char *val = arg;
2928 pack_idx_opts.version = strtoul(val, &c, 10);
2929 if (pack_idx_opts.version > 2)
2930 die(_("unsupported index version %s"), val);
2931 if (*c == ',' && c[1])
2932 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2933 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2934 die(_("bad index version '%s'"), val);
2935 return 0;
2936 }
2937
2938 static int option_parse_unpack_unreachable(const struct option *opt,
2939 const char *arg, int unset)
2940 {
2941 if (unset) {
2942 unpack_unreachable = 0;
2943 unpack_unreachable_expiration = 0;
2944 }
2945 else {
2946 unpack_unreachable = 1;
2947 if (arg)
2948 unpack_unreachable_expiration = approxidate(arg);
2949 }
2950 return 0;
2951 }
2952
2953 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2954 {
2955 int use_internal_rev_list = 0;
2956 int thin = 0;
2957 int shallow = 0;
2958 int all_progress_implied = 0;
2959 struct argv_array rp = ARGV_ARRAY_INIT;
2960 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2961 int rev_list_index = 0;
2962 struct option pack_objects_options[] = {
2963 OPT_SET_INT('q', "quiet", &progress,
2964 N_("do not show progress meter"), 0),
2965 OPT_SET_INT(0, "progress", &progress,
2966 N_("show progress meter"), 1),
2967 OPT_SET_INT(0, "all-progress", &progress,
2968 N_("show progress meter during object writing phase"), 2),
2969 OPT_BOOL(0, "all-progress-implied",
2970 &all_progress_implied,
2971 N_("similar to --all-progress when progress meter is shown")),
2972 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2973 N_("write the pack index file in the specified idx format version"),
2974 0, option_parse_index_version },
2975 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2976 N_("maximum size of each output pack file")),
2977 OPT_BOOL(0, "local", &local,
2978 N_("ignore borrowed objects from alternate object store")),
2979 OPT_BOOL(0, "incremental", &incremental,
2980 N_("ignore packed objects")),
2981 OPT_INTEGER(0, "window", &window,
2982 N_("limit pack window by objects")),
2983 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2984 N_("limit pack window by memory in addition to object limit")),
2985 OPT_INTEGER(0, "depth", &depth,
2986 N_("maximum length of delta chain allowed in the resulting pack")),
2987 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2988 N_("reuse existing deltas")),
2989 OPT_BOOL(0, "reuse-object", &reuse_object,
2990 N_("reuse existing objects")),
2991 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2992 N_("use OFS_DELTA objects")),
2993 OPT_INTEGER(0, "threads", &delta_search_threads,
2994 N_("use threads when searching for best delta matches")),
2995 OPT_BOOL(0, "non-empty", &non_empty,
2996 N_("do not create an empty pack output")),
2997 OPT_BOOL(0, "revs", &use_internal_rev_list,
2998 N_("read revision arguments from standard input")),
2999 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
3000 N_("limit the objects to those that are not yet packed"),
3001 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3002 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
3003 N_("include objects reachable from any reference"),
3004 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3005 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
3006 N_("include objects referred by reflog entries"),
3007 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3008 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
3009 N_("include objects referred to by the index"),
3010 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3011 OPT_BOOL(0, "stdout", &pack_to_stdout,
3012 N_("output pack to stdout")),
3013 OPT_BOOL(0, "include-tag", &include_tag,
3014 N_("include tag objects that refer to objects to be packed")),
3015 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3016 N_("keep unreachable objects")),
3017 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3018 N_("pack loose unreachable objects")),
3019 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3020 N_("unpack unreachable objects newer than <time>"),
3021 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3022 OPT_BOOL(0, "thin", &thin,
3023 N_("create thin packs")),
3024 OPT_BOOL(0, "shallow", &shallow,
3025 N_("create packs suitable for shallow fetches")),
3026 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
3027 N_("ignore packs that have companion .keep file")),
3028 OPT_INTEGER(0, "compression", &pack_compression_level,
3029 N_("pack compression level")),
3030 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3031 N_("do not hide commits by grafts"), 0),
3032 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3033 N_("use a bitmap index if available to speed up counting objects")),
3034 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3035 N_("write a bitmap index together with the pack index")),
3036 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3037 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3038 N_("handling for missing objects"), PARSE_OPT_NONEG,
3039 option_parse_missing_action },
3040 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3041 N_("do not pack objects in promisor packfiles")),
3042 OPT_END(),
3043 };
3044
3045 check_replace_refs = 0;
3046
3047 reset_pack_idx_option(&pack_idx_opts);
3048 git_config(git_pack_config, NULL);
3049
3050 progress = isatty(2);
3051 argc = parse_options(argc, argv, prefix, pack_objects_options,
3052 pack_usage, 0);
3053
3054 if (argc) {
3055 base_name = argv[0];
3056 argc--;
3057 }
3058 if (pack_to_stdout != !base_name || argc)
3059 usage_with_options(pack_usage, pack_objects_options);
3060
3061 argv_array_push(&rp, "pack-objects");
3062 if (thin) {
3063 use_internal_rev_list = 1;
3064 argv_array_push(&rp, shallow
3065 ? "--objects-edge-aggressive"
3066 : "--objects-edge");
3067 } else
3068 argv_array_push(&rp, "--objects");
3069
3070 if (rev_list_all) {
3071 use_internal_rev_list = 1;
3072 argv_array_push(&rp, "--all");
3073 }
3074 if (rev_list_reflog) {
3075 use_internal_rev_list = 1;
3076 argv_array_push(&rp, "--reflog");
3077 }
3078 if (rev_list_index) {
3079 use_internal_rev_list = 1;
3080 argv_array_push(&rp, "--indexed-objects");
3081 }
3082 if (rev_list_unpacked) {
3083 use_internal_rev_list = 1;
3084 argv_array_push(&rp, "--unpacked");
3085 }
3086
3087 if (exclude_promisor_objects) {
3088 use_internal_rev_list = 1;
3089 fetch_if_missing = 0;
3090 argv_array_push(&rp, "--exclude-promisor-objects");
3091 }
3092
3093 if (!reuse_object)
3094 reuse_delta = 0;
3095 if (pack_compression_level == -1)
3096 pack_compression_level = Z_DEFAULT_COMPRESSION;
3097 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3098 die("bad pack compression level %d", pack_compression_level);
3099
3100 if (!delta_search_threads) /* --threads=0 means autodetect */
3101 delta_search_threads = online_cpus();
3102
3103 #ifdef NO_PTHREADS
3104 if (delta_search_threads != 1)
3105 warning("no threads support, ignoring --threads");
3106 #endif
3107 if (!pack_to_stdout && !pack_size_limit)
3108 pack_size_limit = pack_size_limit_cfg;
3109 if (pack_to_stdout && pack_size_limit)
3110 die("--max-pack-size cannot be used to build a pack for transfer.");
3111 if (pack_size_limit && pack_size_limit < 1024*1024) {
3112 warning("minimum pack size limit is 1 MiB");
3113 pack_size_limit = 1024*1024;
3114 }
3115
3116 if (!pack_to_stdout && thin)
3117 die("--thin cannot be used to build an indexable pack.");
3118
3119 if (keep_unreachable && unpack_unreachable)
3120 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3121 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3122 unpack_unreachable_expiration = 0;
3123
3124 if (filter_options.choice) {
3125 if (!pack_to_stdout)
3126 die("cannot use --filter without --stdout.");
3127 use_bitmap_index = 0;
3128 }
3129
3130 /*
3131 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3132 *
3133 * - to produce good pack (with bitmap index not-yet-packed objects are
3134 * packed in suboptimal order).
3135 *
3136 * - to use more robust pack-generation codepath (avoiding possible
3137 * bugs in bitmap code and possible bitmap index corruption).
3138 */
3139 if (!pack_to_stdout)
3140 use_bitmap_index_default = 0;
3141
3142 if (use_bitmap_index < 0)
3143 use_bitmap_index = use_bitmap_index_default;
3144
3145 /* "hard" reasons not to use bitmaps; these just won't work at all */
3146 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3147 use_bitmap_index = 0;
3148
3149 if (pack_to_stdout || !rev_list_all)
3150 write_bitmap_index = 0;
3151
3152 if (progress && all_progress_implied)
3153 progress = 2;
3154
3155 prepare_packed_git();
3156 if (ignore_packed_keep) {
3157 struct packed_git *p;
3158 for (p = get_packed_git(the_repository); p; p = p->next)
3159 if (p->pack_local && p->pack_keep)
3160 break;
3161 if (!p) /* no keep-able packs found */
3162 ignore_packed_keep = 0;
3163 }
3164 if (local) {
3165 /*
3166 * unlike ignore_packed_keep above, we do not want to
3167 * unset "local" based on looking at packs, as it
3168 * also covers non-local objects
3169 */
3170 struct packed_git *p;
3171 for (p = get_packed_git(the_repository); p; p = p->next) {
3172 if (!p->pack_local) {
3173 have_non_local_packs = 1;
3174 break;
3175 }
3176 }
3177 }
3178
3179 if (progress)
3180 progress_state = start_progress(_("Counting objects"), 0);
3181 if (!use_internal_rev_list)
3182 read_object_list_from_stdin();
3183 else {
3184 get_object_list(rp.argc, rp.argv);
3185 argv_array_clear(&rp);
3186 }
3187 cleanup_preferred_base();
3188 if (include_tag && nr_result)
3189 for_each_ref(add_ref_tag, NULL);
3190 stop_progress(&progress_state);
3191
3192 if (non_empty && !nr_result)
3193 return 0;
3194 if (nr_result)
3195 prepare_pack(window, depth);
3196 write_pack_file();
3197 if (progress)
3198 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3199 " reused %"PRIu32" (delta %"PRIu32")\n",
3200 written, written_delta, reused, reused_delta);
3201 return 0;
3202 }