t1400: use consistent style for test_expect_success calls
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "pack-objects.h"
18 #include "progress.h"
19 #include "refs.h"
20 #include "streaming.h"
21 #include "thread-utils.h"
22 #include "pack-bitmap.h"
23 #include "reachable.h"
24 #include "sha1-array.h"
25 #include "argv-array.h"
26 #include "mru.h"
27
28 static const char *pack_usage[] = {
29 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
30 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
31 NULL
32 };
33
34 /*
35 * Objects we are going to pack are collected in the `to_pack` structure.
36 * It contains an array (dynamically expanded) of the object data, and a map
37 * that can resolve SHA1s to their position in the array.
38 */
39 static struct packing_data to_pack;
40
41 static struct pack_idx_entry **written_list;
42 static uint32_t nr_result, nr_written;
43
44 static int non_empty;
45 static int reuse_delta = 1, reuse_object = 1;
46 static int keep_unreachable, unpack_unreachable, include_tag;
47 static unsigned long unpack_unreachable_expiration;
48 static int pack_loose_unreachable;
49 static int local;
50 static int have_non_local_packs;
51 static int incremental;
52 static int ignore_packed_keep;
53 static int allow_ofs_delta;
54 static struct pack_idx_option pack_idx_opts;
55 static const char *base_name;
56 static int progress = 1;
57 static int window = 10;
58 static unsigned long pack_size_limit;
59 static int depth = 50;
60 static int delta_search_threads;
61 static int pack_to_stdout;
62 static int num_preferred_base;
63 static struct progress *progress_state;
64
65 static struct packed_git *reuse_packfile;
66 static uint32_t reuse_packfile_objects;
67 static off_t reuse_packfile_offset;
68
69 static int use_bitmap_index_default = 1;
70 static int use_bitmap_index = -1;
71 static int write_bitmap_index;
72 static uint16_t write_bitmap_options;
73
74 static unsigned long delta_cache_size = 0;
75 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
76 static unsigned long cache_max_small_delta_size = 1000;
77
78 static unsigned long window_memory_limit = 0;
79
80 /*
81 * stats
82 */
83 static uint32_t written, written_delta;
84 static uint32_t reused, reused_delta;
85
86 /*
87 * Indexed commits
88 */
89 static struct commit **indexed_commits;
90 static unsigned int indexed_commits_nr;
91 static unsigned int indexed_commits_alloc;
92
93 static void index_commit_for_bitmap(struct commit *commit)
94 {
95 if (indexed_commits_nr >= indexed_commits_alloc) {
96 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
97 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
98 }
99
100 indexed_commits[indexed_commits_nr++] = commit;
101 }
102
103 static void *get_delta(struct object_entry *entry)
104 {
105 unsigned long size, base_size, delta_size;
106 void *buf, *base_buf, *delta_buf;
107 enum object_type type;
108
109 buf = read_sha1_file(entry->idx.sha1, &type, &size);
110 if (!buf)
111 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
112 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
113 if (!base_buf)
114 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
115 delta_buf = diff_delta(base_buf, base_size,
116 buf, size, &delta_size, 0);
117 if (!delta_buf || delta_size != entry->delta_size)
118 die("delta size changed");
119 free(buf);
120 free(base_buf);
121 return delta_buf;
122 }
123
124 static unsigned long do_compress(void **pptr, unsigned long size)
125 {
126 git_zstream stream;
127 void *in, *out;
128 unsigned long maxsize;
129
130 git_deflate_init(&stream, pack_compression_level);
131 maxsize = git_deflate_bound(&stream, size);
132
133 in = *pptr;
134 out = xmalloc(maxsize);
135 *pptr = out;
136
137 stream.next_in = in;
138 stream.avail_in = size;
139 stream.next_out = out;
140 stream.avail_out = maxsize;
141 while (git_deflate(&stream, Z_FINISH) == Z_OK)
142 ; /* nothing */
143 git_deflate_end(&stream);
144
145 free(in);
146 return stream.total_out;
147 }
148
149 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
150 const unsigned char *sha1)
151 {
152 git_zstream stream;
153 unsigned char ibuf[1024 * 16];
154 unsigned char obuf[1024 * 16];
155 unsigned long olen = 0;
156
157 git_deflate_init(&stream, pack_compression_level);
158
159 for (;;) {
160 ssize_t readlen;
161 int zret = Z_OK;
162 readlen = read_istream(st, ibuf, sizeof(ibuf));
163 if (readlen == -1)
164 die(_("unable to read %s"), sha1_to_hex(sha1));
165
166 stream.next_in = ibuf;
167 stream.avail_in = readlen;
168 while ((stream.avail_in || readlen == 0) &&
169 (zret == Z_OK || zret == Z_BUF_ERROR)) {
170 stream.next_out = obuf;
171 stream.avail_out = sizeof(obuf);
172 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
173 sha1write(f, obuf, stream.next_out - obuf);
174 olen += stream.next_out - obuf;
175 }
176 if (stream.avail_in)
177 die(_("deflate error (%d)"), zret);
178 if (readlen == 0) {
179 if (zret != Z_STREAM_END)
180 die(_("deflate error (%d)"), zret);
181 break;
182 }
183 }
184 git_deflate_end(&stream);
185 return olen;
186 }
187
188 /*
189 * we are going to reuse the existing object data as is. make
190 * sure it is not corrupt.
191 */
192 static int check_pack_inflate(struct packed_git *p,
193 struct pack_window **w_curs,
194 off_t offset,
195 off_t len,
196 unsigned long expect)
197 {
198 git_zstream stream;
199 unsigned char fakebuf[4096], *in;
200 int st;
201
202 memset(&stream, 0, sizeof(stream));
203 git_inflate_init(&stream);
204 do {
205 in = use_pack(p, w_curs, offset, &stream.avail_in);
206 stream.next_in = in;
207 stream.next_out = fakebuf;
208 stream.avail_out = sizeof(fakebuf);
209 st = git_inflate(&stream, Z_FINISH);
210 offset += stream.next_in - in;
211 } while (st == Z_OK || st == Z_BUF_ERROR);
212 git_inflate_end(&stream);
213 return (st == Z_STREAM_END &&
214 stream.total_out == expect &&
215 stream.total_in == len) ? 0 : -1;
216 }
217
218 static void copy_pack_data(struct sha1file *f,
219 struct packed_git *p,
220 struct pack_window **w_curs,
221 off_t offset,
222 off_t len)
223 {
224 unsigned char *in;
225 unsigned long avail;
226
227 while (len) {
228 in = use_pack(p, w_curs, offset, &avail);
229 if (avail > len)
230 avail = (unsigned long)len;
231 sha1write(f, in, avail);
232 offset += avail;
233 len -= avail;
234 }
235 }
236
237 /* Return 0 if we will bust the pack-size limit */
238 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
239 unsigned long limit, int usable_delta)
240 {
241 unsigned long size, datalen;
242 unsigned char header[10], dheader[10];
243 unsigned hdrlen;
244 enum object_type type;
245 void *buf;
246 struct git_istream *st = NULL;
247
248 if (!usable_delta) {
249 if (entry->type == OBJ_BLOB &&
250 entry->size > big_file_threshold &&
251 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
252 buf = NULL;
253 else {
254 buf = read_sha1_file(entry->idx.sha1, &type, &size);
255 if (!buf)
256 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
257 }
258 /*
259 * make sure no cached delta data remains from a
260 * previous attempt before a pack split occurred.
261 */
262 free(entry->delta_data);
263 entry->delta_data = NULL;
264 entry->z_delta_size = 0;
265 } else if (entry->delta_data) {
266 size = entry->delta_size;
267 buf = entry->delta_data;
268 entry->delta_data = NULL;
269 type = (allow_ofs_delta && entry->delta->idx.offset) ?
270 OBJ_OFS_DELTA : OBJ_REF_DELTA;
271 } else {
272 buf = get_delta(entry);
273 size = entry->delta_size;
274 type = (allow_ofs_delta && entry->delta->idx.offset) ?
275 OBJ_OFS_DELTA : OBJ_REF_DELTA;
276 }
277
278 if (st) /* large blob case, just assume we don't compress well */
279 datalen = size;
280 else if (entry->z_delta_size)
281 datalen = entry->z_delta_size;
282 else
283 datalen = do_compress(&buf, size);
284
285 /*
286 * The object header is a byte of 'type' followed by zero or
287 * more bytes of length.
288 */
289 hdrlen = encode_in_pack_object_header(type, size, header);
290
291 if (type == OBJ_OFS_DELTA) {
292 /*
293 * Deltas with relative base contain an additional
294 * encoding of the relative offset for the delta
295 * base from this object's position in the pack.
296 */
297 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
298 unsigned pos = sizeof(dheader) - 1;
299 dheader[pos] = ofs & 127;
300 while (ofs >>= 7)
301 dheader[--pos] = 128 | (--ofs & 127);
302 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
303 if (st)
304 close_istream(st);
305 free(buf);
306 return 0;
307 }
308 sha1write(f, header, hdrlen);
309 sha1write(f, dheader + pos, sizeof(dheader) - pos);
310 hdrlen += sizeof(dheader) - pos;
311 } else if (type == OBJ_REF_DELTA) {
312 /*
313 * Deltas with a base reference contain
314 * an additional 20 bytes for the base sha1.
315 */
316 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
317 if (st)
318 close_istream(st);
319 free(buf);
320 return 0;
321 }
322 sha1write(f, header, hdrlen);
323 sha1write(f, entry->delta->idx.sha1, 20);
324 hdrlen += 20;
325 } else {
326 if (limit && hdrlen + datalen + 20 >= limit) {
327 if (st)
328 close_istream(st);
329 free(buf);
330 return 0;
331 }
332 sha1write(f, header, hdrlen);
333 }
334 if (st) {
335 datalen = write_large_blob_data(st, f, entry->idx.sha1);
336 close_istream(st);
337 } else {
338 sha1write(f, buf, datalen);
339 free(buf);
340 }
341
342 return hdrlen + datalen;
343 }
344
345 /* Return 0 if we will bust the pack-size limit */
346 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
347 unsigned long limit, int usable_delta)
348 {
349 struct packed_git *p = entry->in_pack;
350 struct pack_window *w_curs = NULL;
351 struct revindex_entry *revidx;
352 off_t offset;
353 enum object_type type = entry->type;
354 off_t datalen;
355 unsigned char header[10], dheader[10];
356 unsigned hdrlen;
357
358 if (entry->delta)
359 type = (allow_ofs_delta && entry->delta->idx.offset) ?
360 OBJ_OFS_DELTA : OBJ_REF_DELTA;
361 hdrlen = encode_in_pack_object_header(type, entry->size, header);
362
363 offset = entry->in_pack_offset;
364 revidx = find_pack_revindex(p, offset);
365 datalen = revidx[1].offset - offset;
366 if (!pack_to_stdout && p->index_version > 1 &&
367 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
368 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
369 unuse_pack(&w_curs);
370 return write_no_reuse_object(f, entry, limit, usable_delta);
371 }
372
373 offset += entry->in_pack_header_size;
374 datalen -= entry->in_pack_header_size;
375
376 if (!pack_to_stdout && p->index_version == 1 &&
377 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
378 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
379 unuse_pack(&w_curs);
380 return write_no_reuse_object(f, entry, limit, usable_delta);
381 }
382
383 if (type == OBJ_OFS_DELTA) {
384 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
385 unsigned pos = sizeof(dheader) - 1;
386 dheader[pos] = ofs & 127;
387 while (ofs >>= 7)
388 dheader[--pos] = 128 | (--ofs & 127);
389 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
390 unuse_pack(&w_curs);
391 return 0;
392 }
393 sha1write(f, header, hdrlen);
394 sha1write(f, dheader + pos, sizeof(dheader) - pos);
395 hdrlen += sizeof(dheader) - pos;
396 reused_delta++;
397 } else if (type == OBJ_REF_DELTA) {
398 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
399 unuse_pack(&w_curs);
400 return 0;
401 }
402 sha1write(f, header, hdrlen);
403 sha1write(f, entry->delta->idx.sha1, 20);
404 hdrlen += 20;
405 reused_delta++;
406 } else {
407 if (limit && hdrlen + datalen + 20 >= limit) {
408 unuse_pack(&w_curs);
409 return 0;
410 }
411 sha1write(f, header, hdrlen);
412 }
413 copy_pack_data(f, p, &w_curs, offset, datalen);
414 unuse_pack(&w_curs);
415 reused++;
416 return hdrlen + datalen;
417 }
418
419 /* Return 0 if we will bust the pack-size limit */
420 static off_t write_object(struct sha1file *f,
421 struct object_entry *entry,
422 off_t write_offset)
423 {
424 unsigned long limit;
425 off_t len;
426 int usable_delta, to_reuse;
427
428 if (!pack_to_stdout)
429 crc32_begin(f);
430
431 /* apply size limit if limited packsize and not first object */
432 if (!pack_size_limit || !nr_written)
433 limit = 0;
434 else if (pack_size_limit <= write_offset)
435 /*
436 * the earlier object did not fit the limit; avoid
437 * mistaking this with unlimited (i.e. limit = 0).
438 */
439 limit = 1;
440 else
441 limit = pack_size_limit - write_offset;
442
443 if (!entry->delta)
444 usable_delta = 0; /* no delta */
445 else if (!pack_size_limit)
446 usable_delta = 1; /* unlimited packfile */
447 else if (entry->delta->idx.offset == (off_t)-1)
448 usable_delta = 0; /* base was written to another pack */
449 else if (entry->delta->idx.offset)
450 usable_delta = 1; /* base already exists in this pack */
451 else
452 usable_delta = 0; /* base could end up in another pack */
453
454 if (!reuse_object)
455 to_reuse = 0; /* explicit */
456 else if (!entry->in_pack)
457 to_reuse = 0; /* can't reuse what we don't have */
458 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
459 /* check_object() decided it for us ... */
460 to_reuse = usable_delta;
461 /* ... but pack split may override that */
462 else if (entry->type != entry->in_pack_type)
463 to_reuse = 0; /* pack has delta which is unusable */
464 else if (entry->delta)
465 to_reuse = 0; /* we want to pack afresh */
466 else
467 to_reuse = 1; /* we have it in-pack undeltified,
468 * and we do not need to deltify it.
469 */
470
471 if (!to_reuse)
472 len = write_no_reuse_object(f, entry, limit, usable_delta);
473 else
474 len = write_reuse_object(f, entry, limit, usable_delta);
475 if (!len)
476 return 0;
477
478 if (usable_delta)
479 written_delta++;
480 written++;
481 if (!pack_to_stdout)
482 entry->idx.crc32 = crc32_end(f);
483 return len;
484 }
485
486 enum write_one_status {
487 WRITE_ONE_SKIP = -1, /* already written */
488 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
489 WRITE_ONE_WRITTEN = 1, /* normal */
490 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
491 };
492
493 static enum write_one_status write_one(struct sha1file *f,
494 struct object_entry *e,
495 off_t *offset)
496 {
497 off_t size;
498 int recursing;
499
500 /*
501 * we set offset to 1 (which is an impossible value) to mark
502 * the fact that this object is involved in "write its base
503 * first before writing a deltified object" recursion.
504 */
505 recursing = (e->idx.offset == 1);
506 if (recursing) {
507 warning("recursive delta detected for object %s",
508 sha1_to_hex(e->idx.sha1));
509 return WRITE_ONE_RECURSIVE;
510 } else if (e->idx.offset || e->preferred_base) {
511 /* offset is non zero if object is written already. */
512 return WRITE_ONE_SKIP;
513 }
514
515 /* if we are deltified, write out base object first. */
516 if (e->delta) {
517 e->idx.offset = 1; /* now recurse */
518 switch (write_one(f, e->delta, offset)) {
519 case WRITE_ONE_RECURSIVE:
520 /* we cannot depend on this one */
521 e->delta = NULL;
522 break;
523 default:
524 break;
525 case WRITE_ONE_BREAK:
526 e->idx.offset = recursing;
527 return WRITE_ONE_BREAK;
528 }
529 }
530
531 e->idx.offset = *offset;
532 size = write_object(f, e, *offset);
533 if (!size) {
534 e->idx.offset = recursing;
535 return WRITE_ONE_BREAK;
536 }
537 written_list[nr_written++] = &e->idx;
538
539 /* make sure off_t is sufficiently large not to wrap */
540 if (signed_add_overflows(*offset, size))
541 die("pack too large for current definition of off_t");
542 *offset += size;
543 return WRITE_ONE_WRITTEN;
544 }
545
546 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
547 void *cb_data)
548 {
549 unsigned char peeled[20];
550 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
551
552 if (entry)
553 entry->tagged = 1;
554 if (!peel_ref(path, peeled)) {
555 entry = packlist_find(&to_pack, peeled, NULL);
556 if (entry)
557 entry->tagged = 1;
558 }
559 return 0;
560 }
561
562 static inline void add_to_write_order(struct object_entry **wo,
563 unsigned int *endp,
564 struct object_entry *e)
565 {
566 if (e->filled)
567 return;
568 wo[(*endp)++] = e;
569 e->filled = 1;
570 }
571
572 static void add_descendants_to_write_order(struct object_entry **wo,
573 unsigned int *endp,
574 struct object_entry *e)
575 {
576 int add_to_order = 1;
577 while (e) {
578 if (add_to_order) {
579 struct object_entry *s;
580 /* add this node... */
581 add_to_write_order(wo, endp, e);
582 /* all its siblings... */
583 for (s = e->delta_sibling; s; s = s->delta_sibling) {
584 add_to_write_order(wo, endp, s);
585 }
586 }
587 /* drop down a level to add left subtree nodes if possible */
588 if (e->delta_child) {
589 add_to_order = 1;
590 e = e->delta_child;
591 } else {
592 add_to_order = 0;
593 /* our sibling might have some children, it is next */
594 if (e->delta_sibling) {
595 e = e->delta_sibling;
596 continue;
597 }
598 /* go back to our parent node */
599 e = e->delta;
600 while (e && !e->delta_sibling) {
601 /* we're on the right side of a subtree, keep
602 * going up until we can go right again */
603 e = e->delta;
604 }
605 if (!e) {
606 /* done- we hit our original root node */
607 return;
608 }
609 /* pass it off to sibling at this level */
610 e = e->delta_sibling;
611 }
612 };
613 }
614
615 static void add_family_to_write_order(struct object_entry **wo,
616 unsigned int *endp,
617 struct object_entry *e)
618 {
619 struct object_entry *root;
620
621 for (root = e; root->delta; root = root->delta)
622 ; /* nothing */
623 add_descendants_to_write_order(wo, endp, root);
624 }
625
626 static struct object_entry **compute_write_order(void)
627 {
628 unsigned int i, wo_end, last_untagged;
629
630 struct object_entry **wo;
631 struct object_entry *objects = to_pack.objects;
632
633 for (i = 0; i < to_pack.nr_objects; i++) {
634 objects[i].tagged = 0;
635 objects[i].filled = 0;
636 objects[i].delta_child = NULL;
637 objects[i].delta_sibling = NULL;
638 }
639
640 /*
641 * Fully connect delta_child/delta_sibling network.
642 * Make sure delta_sibling is sorted in the original
643 * recency order.
644 */
645 for (i = to_pack.nr_objects; i > 0;) {
646 struct object_entry *e = &objects[--i];
647 if (!e->delta)
648 continue;
649 /* Mark me as the first child */
650 e->delta_sibling = e->delta->delta_child;
651 e->delta->delta_child = e;
652 }
653
654 /*
655 * Mark objects that are at the tip of tags.
656 */
657 for_each_tag_ref(mark_tagged, NULL);
658
659 /*
660 * Give the objects in the original recency order until
661 * we see a tagged tip.
662 */
663 ALLOC_ARRAY(wo, to_pack.nr_objects);
664 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
665 if (objects[i].tagged)
666 break;
667 add_to_write_order(wo, &wo_end, &objects[i]);
668 }
669 last_untagged = i;
670
671 /*
672 * Then fill all the tagged tips.
673 */
674 for (; i < to_pack.nr_objects; i++) {
675 if (objects[i].tagged)
676 add_to_write_order(wo, &wo_end, &objects[i]);
677 }
678
679 /*
680 * And then all remaining commits and tags.
681 */
682 for (i = last_untagged; i < to_pack.nr_objects; i++) {
683 if (objects[i].type != OBJ_COMMIT &&
684 objects[i].type != OBJ_TAG)
685 continue;
686 add_to_write_order(wo, &wo_end, &objects[i]);
687 }
688
689 /*
690 * And then all the trees.
691 */
692 for (i = last_untagged; i < to_pack.nr_objects; i++) {
693 if (objects[i].type != OBJ_TREE)
694 continue;
695 add_to_write_order(wo, &wo_end, &objects[i]);
696 }
697
698 /*
699 * Finally all the rest in really tight order
700 */
701 for (i = last_untagged; i < to_pack.nr_objects; i++) {
702 if (!objects[i].filled)
703 add_family_to_write_order(wo, &wo_end, &objects[i]);
704 }
705
706 if (wo_end != to_pack.nr_objects)
707 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
708
709 return wo;
710 }
711
712 static off_t write_reused_pack(struct sha1file *f)
713 {
714 unsigned char buffer[8192];
715 off_t to_write, total;
716 int fd;
717
718 if (!is_pack_valid(reuse_packfile))
719 die("packfile is invalid: %s", reuse_packfile->pack_name);
720
721 fd = git_open(reuse_packfile->pack_name);
722 if (fd < 0)
723 die_errno("unable to open packfile for reuse: %s",
724 reuse_packfile->pack_name);
725
726 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
727 die_errno("unable to seek in reused packfile");
728
729 if (reuse_packfile_offset < 0)
730 reuse_packfile_offset = reuse_packfile->pack_size - 20;
731
732 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
733
734 while (to_write) {
735 int read_pack = xread(fd, buffer, sizeof(buffer));
736
737 if (read_pack <= 0)
738 die_errno("unable to read from reused packfile");
739
740 if (read_pack > to_write)
741 read_pack = to_write;
742
743 sha1write(f, buffer, read_pack);
744 to_write -= read_pack;
745
746 /*
747 * We don't know the actual number of objects written,
748 * only how many bytes written, how many bytes total, and
749 * how many objects total. So we can fake it by pretending all
750 * objects we are writing are the same size. This gives us a
751 * smooth progress meter, and at the end it matches the true
752 * answer.
753 */
754 written = reuse_packfile_objects *
755 (((double)(total - to_write)) / total);
756 display_progress(progress_state, written);
757 }
758
759 close(fd);
760 written = reuse_packfile_objects;
761 display_progress(progress_state, written);
762 return reuse_packfile_offset - sizeof(struct pack_header);
763 }
764
765 static const char no_split_warning[] = N_(
766 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
767 );
768
769 static void write_pack_file(void)
770 {
771 uint32_t i = 0, j;
772 struct sha1file *f;
773 off_t offset;
774 uint32_t nr_remaining = nr_result;
775 time_t last_mtime = 0;
776 struct object_entry **write_order;
777
778 if (progress > pack_to_stdout)
779 progress_state = start_progress(_("Writing objects"), nr_result);
780 ALLOC_ARRAY(written_list, to_pack.nr_objects);
781 write_order = compute_write_order();
782
783 do {
784 unsigned char sha1[20];
785 char *pack_tmp_name = NULL;
786
787 if (pack_to_stdout)
788 f = sha1fd_throughput(1, "<stdout>", progress_state);
789 else
790 f = create_tmp_packfile(&pack_tmp_name);
791
792 offset = write_pack_header(f, nr_remaining);
793
794 if (reuse_packfile) {
795 off_t packfile_size;
796 assert(pack_to_stdout);
797
798 packfile_size = write_reused_pack(f);
799 offset += packfile_size;
800 }
801
802 nr_written = 0;
803 for (; i < to_pack.nr_objects; i++) {
804 struct object_entry *e = write_order[i];
805 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
806 break;
807 display_progress(progress_state, written);
808 }
809
810 /*
811 * Did we write the wrong # entries in the header?
812 * If so, rewrite it like in fast-import
813 */
814 if (pack_to_stdout) {
815 sha1close(f, sha1, CSUM_CLOSE);
816 } else if (nr_written == nr_remaining) {
817 sha1close(f, sha1, CSUM_FSYNC);
818 } else {
819 int fd = sha1close(f, sha1, 0);
820 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
821 nr_written, sha1, offset);
822 close(fd);
823 if (write_bitmap_index) {
824 warning(_(no_split_warning));
825 write_bitmap_index = 0;
826 }
827 }
828
829 if (!pack_to_stdout) {
830 struct stat st;
831 struct strbuf tmpname = STRBUF_INIT;
832
833 /*
834 * Packs are runtime accessed in their mtime
835 * order since newer packs are more likely to contain
836 * younger objects. So if we are creating multiple
837 * packs then we should modify the mtime of later ones
838 * to preserve this property.
839 */
840 if (stat(pack_tmp_name, &st) < 0) {
841 warning_errno("failed to stat %s", pack_tmp_name);
842 } else if (!last_mtime) {
843 last_mtime = st.st_mtime;
844 } else {
845 struct utimbuf utb;
846 utb.actime = st.st_atime;
847 utb.modtime = --last_mtime;
848 if (utime(pack_tmp_name, &utb) < 0)
849 warning_errno("failed utime() on %s", pack_tmp_name);
850 }
851
852 strbuf_addf(&tmpname, "%s-", base_name);
853
854 if (write_bitmap_index) {
855 bitmap_writer_set_checksum(sha1);
856 bitmap_writer_build_type_index(written_list, nr_written);
857 }
858
859 finish_tmp_packfile(&tmpname, pack_tmp_name,
860 written_list, nr_written,
861 &pack_idx_opts, sha1);
862
863 if (write_bitmap_index) {
864 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
865
866 stop_progress(&progress_state);
867
868 bitmap_writer_show_progress(progress);
869 bitmap_writer_reuse_bitmaps(&to_pack);
870 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
871 bitmap_writer_build(&to_pack);
872 bitmap_writer_finish(written_list, nr_written,
873 tmpname.buf, write_bitmap_options);
874 write_bitmap_index = 0;
875 }
876
877 strbuf_release(&tmpname);
878 free(pack_tmp_name);
879 puts(sha1_to_hex(sha1));
880 }
881
882 /* mark written objects as written to previous pack */
883 for (j = 0; j < nr_written; j++) {
884 written_list[j]->offset = (off_t)-1;
885 }
886 nr_remaining -= nr_written;
887 } while (nr_remaining && i < to_pack.nr_objects);
888
889 free(written_list);
890 free(write_order);
891 stop_progress(&progress_state);
892 if (written != nr_result)
893 die("wrote %"PRIu32" objects while expecting %"PRIu32,
894 written, nr_result);
895 }
896
897 static int no_try_delta(const char *path)
898 {
899 static struct attr_check *check;
900
901 if (!check)
902 check = attr_check_initl("delta", NULL);
903 if (git_check_attr(path, check))
904 return 0;
905 if (ATTR_FALSE(check->items[0].value))
906 return 1;
907 return 0;
908 }
909
910 /*
911 * When adding an object, check whether we have already added it
912 * to our packing list. If so, we can skip. However, if we are
913 * being asked to excludei t, but the previous mention was to include
914 * it, make sure to adjust its flags and tweak our numbers accordingly.
915 *
916 * As an optimization, we pass out the index position where we would have
917 * found the item, since that saves us from having to look it up again a
918 * few lines later when we want to add the new entry.
919 */
920 static int have_duplicate_entry(const unsigned char *sha1,
921 int exclude,
922 uint32_t *index_pos)
923 {
924 struct object_entry *entry;
925
926 entry = packlist_find(&to_pack, sha1, index_pos);
927 if (!entry)
928 return 0;
929
930 if (exclude) {
931 if (!entry->preferred_base)
932 nr_result--;
933 entry->preferred_base = 1;
934 }
935
936 return 1;
937 }
938
939 static int want_found_object(int exclude, struct packed_git *p)
940 {
941 if (exclude)
942 return 1;
943 if (incremental)
944 return 0;
945
946 /*
947 * When asked to do --local (do not include an object that appears in a
948 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
949 * an object that appears in a pack marked with .keep), finding a pack
950 * that matches the criteria is sufficient for us to decide to omit it.
951 * However, even if this pack does not satisfy the criteria, we need to
952 * make sure no copy of this object appears in _any_ pack that makes us
953 * to omit the object, so we need to check all the packs.
954 *
955 * We can however first check whether these options can possible matter;
956 * if they do not matter we know we want the object in generated pack.
957 * Otherwise, we signal "-1" at the end to tell the caller that we do
958 * not know either way, and it needs to check more packs.
959 */
960 if (!ignore_packed_keep &&
961 (!local || !have_non_local_packs))
962 return 1;
963
964 if (local && !p->pack_local)
965 return 0;
966 if (ignore_packed_keep && p->pack_local && p->pack_keep)
967 return 0;
968
969 /* we don't know yet; keep looking for more packs */
970 return -1;
971 }
972
973 /*
974 * Check whether we want the object in the pack (e.g., we do not want
975 * objects found in non-local stores if the "--local" option was used).
976 *
977 * If the caller already knows an existing pack it wants to take the object
978 * from, that is passed in *found_pack and *found_offset; otherwise this
979 * function finds if there is any pack that has the object and returns the pack
980 * and its offset in these variables.
981 */
982 static int want_object_in_pack(const unsigned char *sha1,
983 int exclude,
984 struct packed_git **found_pack,
985 off_t *found_offset)
986 {
987 struct mru_entry *entry;
988 int want;
989
990 if (!exclude && local && has_loose_object_nonlocal(sha1))
991 return 0;
992
993 /*
994 * If we already know the pack object lives in, start checks from that
995 * pack - in the usual case when neither --local was given nor .keep files
996 * are present we will determine the answer right now.
997 */
998 if (*found_pack) {
999 want = want_found_object(exclude, *found_pack);
1000 if (want != -1)
1001 return want;
1002 }
1003
1004 for (entry = packed_git_mru->head; entry; entry = entry->next) {
1005 struct packed_git *p = entry->item;
1006 off_t offset;
1007
1008 if (p == *found_pack)
1009 offset = *found_offset;
1010 else
1011 offset = find_pack_entry_one(sha1, p);
1012
1013 if (offset) {
1014 if (!*found_pack) {
1015 if (!is_pack_valid(p))
1016 continue;
1017 *found_offset = offset;
1018 *found_pack = p;
1019 }
1020 want = want_found_object(exclude, p);
1021 if (!exclude && want > 0)
1022 mru_mark(packed_git_mru, entry);
1023 if (want != -1)
1024 return want;
1025 }
1026 }
1027
1028 return 1;
1029 }
1030
1031 static void create_object_entry(const unsigned char *sha1,
1032 enum object_type type,
1033 uint32_t hash,
1034 int exclude,
1035 int no_try_delta,
1036 uint32_t index_pos,
1037 struct packed_git *found_pack,
1038 off_t found_offset)
1039 {
1040 struct object_entry *entry;
1041
1042 entry = packlist_alloc(&to_pack, sha1, index_pos);
1043 entry->hash = hash;
1044 if (type)
1045 entry->type = type;
1046 if (exclude)
1047 entry->preferred_base = 1;
1048 else
1049 nr_result++;
1050 if (found_pack) {
1051 entry->in_pack = found_pack;
1052 entry->in_pack_offset = found_offset;
1053 }
1054
1055 entry->no_try_delta = no_try_delta;
1056 }
1057
1058 static const char no_closure_warning[] = N_(
1059 "disabling bitmap writing, as some objects are not being packed"
1060 );
1061
1062 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1063 const char *name, int exclude)
1064 {
1065 struct packed_git *found_pack = NULL;
1066 off_t found_offset = 0;
1067 uint32_t index_pos;
1068
1069 if (have_duplicate_entry(sha1, exclude, &index_pos))
1070 return 0;
1071
1072 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1073 /* The pack is missing an object, so it will not have closure */
1074 if (write_bitmap_index) {
1075 warning(_(no_closure_warning));
1076 write_bitmap_index = 0;
1077 }
1078 return 0;
1079 }
1080
1081 create_object_entry(sha1, type, pack_name_hash(name),
1082 exclude, name && no_try_delta(name),
1083 index_pos, found_pack, found_offset);
1084
1085 display_progress(progress_state, nr_result);
1086 return 1;
1087 }
1088
1089 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1090 enum object_type type,
1091 int flags, uint32_t name_hash,
1092 struct packed_git *pack, off_t offset)
1093 {
1094 uint32_t index_pos;
1095
1096 if (have_duplicate_entry(sha1, 0, &index_pos))
1097 return 0;
1098
1099 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1100 return 0;
1101
1102 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1103
1104 display_progress(progress_state, nr_result);
1105 return 1;
1106 }
1107
1108 struct pbase_tree_cache {
1109 unsigned char sha1[20];
1110 int ref;
1111 int temporary;
1112 void *tree_data;
1113 unsigned long tree_size;
1114 };
1115
1116 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1117 static int pbase_tree_cache_ix(const unsigned char *sha1)
1118 {
1119 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1120 }
1121 static int pbase_tree_cache_ix_incr(int ix)
1122 {
1123 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1124 }
1125
1126 static struct pbase_tree {
1127 struct pbase_tree *next;
1128 /* This is a phony "cache" entry; we are not
1129 * going to evict it or find it through _get()
1130 * mechanism -- this is for the toplevel node that
1131 * would almost always change with any commit.
1132 */
1133 struct pbase_tree_cache pcache;
1134 } *pbase_tree;
1135
1136 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1137 {
1138 struct pbase_tree_cache *ent, *nent;
1139 void *data;
1140 unsigned long size;
1141 enum object_type type;
1142 int neigh;
1143 int my_ix = pbase_tree_cache_ix(sha1);
1144 int available_ix = -1;
1145
1146 /* pbase-tree-cache acts as a limited hashtable.
1147 * your object will be found at your index or within a few
1148 * slots after that slot if it is cached.
1149 */
1150 for (neigh = 0; neigh < 8; neigh++) {
1151 ent = pbase_tree_cache[my_ix];
1152 if (ent && !hashcmp(ent->sha1, sha1)) {
1153 ent->ref++;
1154 return ent;
1155 }
1156 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1157 ((0 <= available_ix) &&
1158 (!ent && pbase_tree_cache[available_ix])))
1159 available_ix = my_ix;
1160 if (!ent)
1161 break;
1162 my_ix = pbase_tree_cache_ix_incr(my_ix);
1163 }
1164
1165 /* Did not find one. Either we got a bogus request or
1166 * we need to read and perhaps cache.
1167 */
1168 data = read_sha1_file(sha1, &type, &size);
1169 if (!data)
1170 return NULL;
1171 if (type != OBJ_TREE) {
1172 free(data);
1173 return NULL;
1174 }
1175
1176 /* We need to either cache or return a throwaway copy */
1177
1178 if (available_ix < 0)
1179 ent = NULL;
1180 else {
1181 ent = pbase_tree_cache[available_ix];
1182 my_ix = available_ix;
1183 }
1184
1185 if (!ent) {
1186 nent = xmalloc(sizeof(*nent));
1187 nent->temporary = (available_ix < 0);
1188 }
1189 else {
1190 /* evict and reuse */
1191 free(ent->tree_data);
1192 nent = ent;
1193 }
1194 hashcpy(nent->sha1, sha1);
1195 nent->tree_data = data;
1196 nent->tree_size = size;
1197 nent->ref = 1;
1198 if (!nent->temporary)
1199 pbase_tree_cache[my_ix] = nent;
1200 return nent;
1201 }
1202
1203 static void pbase_tree_put(struct pbase_tree_cache *cache)
1204 {
1205 if (!cache->temporary) {
1206 cache->ref--;
1207 return;
1208 }
1209 free(cache->tree_data);
1210 free(cache);
1211 }
1212
1213 static int name_cmp_len(const char *name)
1214 {
1215 int i;
1216 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1217 ;
1218 return i;
1219 }
1220
1221 static void add_pbase_object(struct tree_desc *tree,
1222 const char *name,
1223 int cmplen,
1224 const char *fullname)
1225 {
1226 struct name_entry entry;
1227 int cmp;
1228
1229 while (tree_entry(tree,&entry)) {
1230 if (S_ISGITLINK(entry.mode))
1231 continue;
1232 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1233 memcmp(name, entry.path, cmplen);
1234 if (cmp > 0)
1235 continue;
1236 if (cmp < 0)
1237 return;
1238 if (name[cmplen] != '/') {
1239 add_object_entry(entry.oid->hash,
1240 object_type(entry.mode),
1241 fullname, 1);
1242 return;
1243 }
1244 if (S_ISDIR(entry.mode)) {
1245 struct tree_desc sub;
1246 struct pbase_tree_cache *tree;
1247 const char *down = name+cmplen+1;
1248 int downlen = name_cmp_len(down);
1249
1250 tree = pbase_tree_get(entry.oid->hash);
1251 if (!tree)
1252 return;
1253 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1254
1255 add_pbase_object(&sub, down, downlen, fullname);
1256 pbase_tree_put(tree);
1257 }
1258 }
1259 }
1260
1261 static unsigned *done_pbase_paths;
1262 static int done_pbase_paths_num;
1263 static int done_pbase_paths_alloc;
1264 static int done_pbase_path_pos(unsigned hash)
1265 {
1266 int lo = 0;
1267 int hi = done_pbase_paths_num;
1268 while (lo < hi) {
1269 int mi = (hi + lo) / 2;
1270 if (done_pbase_paths[mi] == hash)
1271 return mi;
1272 if (done_pbase_paths[mi] < hash)
1273 hi = mi;
1274 else
1275 lo = mi + 1;
1276 }
1277 return -lo-1;
1278 }
1279
1280 static int check_pbase_path(unsigned hash)
1281 {
1282 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1283 if (0 <= pos)
1284 return 1;
1285 pos = -pos - 1;
1286 ALLOC_GROW(done_pbase_paths,
1287 done_pbase_paths_num + 1,
1288 done_pbase_paths_alloc);
1289 done_pbase_paths_num++;
1290 if (pos < done_pbase_paths_num)
1291 memmove(done_pbase_paths + pos + 1,
1292 done_pbase_paths + pos,
1293 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1294 done_pbase_paths[pos] = hash;
1295 return 0;
1296 }
1297
1298 static void add_preferred_base_object(const char *name)
1299 {
1300 struct pbase_tree *it;
1301 int cmplen;
1302 unsigned hash = pack_name_hash(name);
1303
1304 if (!num_preferred_base || check_pbase_path(hash))
1305 return;
1306
1307 cmplen = name_cmp_len(name);
1308 for (it = pbase_tree; it; it = it->next) {
1309 if (cmplen == 0) {
1310 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1311 }
1312 else {
1313 struct tree_desc tree;
1314 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1315 add_pbase_object(&tree, name, cmplen, name);
1316 }
1317 }
1318 }
1319
1320 static void add_preferred_base(unsigned char *sha1)
1321 {
1322 struct pbase_tree *it;
1323 void *data;
1324 unsigned long size;
1325 unsigned char tree_sha1[20];
1326
1327 if (window <= num_preferred_base++)
1328 return;
1329
1330 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1331 if (!data)
1332 return;
1333
1334 for (it = pbase_tree; it; it = it->next) {
1335 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1336 free(data);
1337 return;
1338 }
1339 }
1340
1341 it = xcalloc(1, sizeof(*it));
1342 it->next = pbase_tree;
1343 pbase_tree = it;
1344
1345 hashcpy(it->pcache.sha1, tree_sha1);
1346 it->pcache.tree_data = data;
1347 it->pcache.tree_size = size;
1348 }
1349
1350 static void cleanup_preferred_base(void)
1351 {
1352 struct pbase_tree *it;
1353 unsigned i;
1354
1355 it = pbase_tree;
1356 pbase_tree = NULL;
1357 while (it) {
1358 struct pbase_tree *this = it;
1359 it = this->next;
1360 free(this->pcache.tree_data);
1361 free(this);
1362 }
1363
1364 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1365 if (!pbase_tree_cache[i])
1366 continue;
1367 free(pbase_tree_cache[i]->tree_data);
1368 free(pbase_tree_cache[i]);
1369 pbase_tree_cache[i] = NULL;
1370 }
1371
1372 free(done_pbase_paths);
1373 done_pbase_paths = NULL;
1374 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1375 }
1376
1377 static void check_object(struct object_entry *entry)
1378 {
1379 if (entry->in_pack) {
1380 struct packed_git *p = entry->in_pack;
1381 struct pack_window *w_curs = NULL;
1382 const unsigned char *base_ref = NULL;
1383 struct object_entry *base_entry;
1384 unsigned long used, used_0;
1385 unsigned long avail;
1386 off_t ofs;
1387 unsigned char *buf, c;
1388
1389 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1390
1391 /*
1392 * We want in_pack_type even if we do not reuse delta
1393 * since non-delta representations could still be reused.
1394 */
1395 used = unpack_object_header_buffer(buf, avail,
1396 &entry->in_pack_type,
1397 &entry->size);
1398 if (used == 0)
1399 goto give_up;
1400
1401 /*
1402 * Determine if this is a delta and if so whether we can
1403 * reuse it or not. Otherwise let's find out as cheaply as
1404 * possible what the actual type and size for this object is.
1405 */
1406 switch (entry->in_pack_type) {
1407 default:
1408 /* Not a delta hence we've already got all we need. */
1409 entry->type = entry->in_pack_type;
1410 entry->in_pack_header_size = used;
1411 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1412 goto give_up;
1413 unuse_pack(&w_curs);
1414 return;
1415 case OBJ_REF_DELTA:
1416 if (reuse_delta && !entry->preferred_base)
1417 base_ref = use_pack(p, &w_curs,
1418 entry->in_pack_offset + used, NULL);
1419 entry->in_pack_header_size = used + 20;
1420 break;
1421 case OBJ_OFS_DELTA:
1422 buf = use_pack(p, &w_curs,
1423 entry->in_pack_offset + used, NULL);
1424 used_0 = 0;
1425 c = buf[used_0++];
1426 ofs = c & 127;
1427 while (c & 128) {
1428 ofs += 1;
1429 if (!ofs || MSB(ofs, 7)) {
1430 error("delta base offset overflow in pack for %s",
1431 sha1_to_hex(entry->idx.sha1));
1432 goto give_up;
1433 }
1434 c = buf[used_0++];
1435 ofs = (ofs << 7) + (c & 127);
1436 }
1437 ofs = entry->in_pack_offset - ofs;
1438 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1439 error("delta base offset out of bound for %s",
1440 sha1_to_hex(entry->idx.sha1));
1441 goto give_up;
1442 }
1443 if (reuse_delta && !entry->preferred_base) {
1444 struct revindex_entry *revidx;
1445 revidx = find_pack_revindex(p, ofs);
1446 if (!revidx)
1447 goto give_up;
1448 base_ref = nth_packed_object_sha1(p, revidx->nr);
1449 }
1450 entry->in_pack_header_size = used + used_0;
1451 break;
1452 }
1453
1454 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1455 /*
1456 * If base_ref was set above that means we wish to
1457 * reuse delta data, and we even found that base
1458 * in the list of objects we want to pack. Goodie!
1459 *
1460 * Depth value does not matter - find_deltas() will
1461 * never consider reused delta as the base object to
1462 * deltify other objects against, in order to avoid
1463 * circular deltas.
1464 */
1465 entry->type = entry->in_pack_type;
1466 entry->delta = base_entry;
1467 entry->delta_size = entry->size;
1468 entry->delta_sibling = base_entry->delta_child;
1469 base_entry->delta_child = entry;
1470 unuse_pack(&w_curs);
1471 return;
1472 }
1473
1474 if (entry->type) {
1475 /*
1476 * This must be a delta and we already know what the
1477 * final object type is. Let's extract the actual
1478 * object size from the delta header.
1479 */
1480 entry->size = get_size_from_delta(p, &w_curs,
1481 entry->in_pack_offset + entry->in_pack_header_size);
1482 if (entry->size == 0)
1483 goto give_up;
1484 unuse_pack(&w_curs);
1485 return;
1486 }
1487
1488 /*
1489 * No choice but to fall back to the recursive delta walk
1490 * with sha1_object_info() to find about the object type
1491 * at this point...
1492 */
1493 give_up:
1494 unuse_pack(&w_curs);
1495 }
1496
1497 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1498 /*
1499 * The error condition is checked in prepare_pack(). This is
1500 * to permit a missing preferred base object to be ignored
1501 * as a preferred base. Doing so can result in a larger
1502 * pack file, but the transfer will still take place.
1503 */
1504 }
1505
1506 static int pack_offset_sort(const void *_a, const void *_b)
1507 {
1508 const struct object_entry *a = *(struct object_entry **)_a;
1509 const struct object_entry *b = *(struct object_entry **)_b;
1510
1511 /* avoid filesystem trashing with loose objects */
1512 if (!a->in_pack && !b->in_pack)
1513 return hashcmp(a->idx.sha1, b->idx.sha1);
1514
1515 if (a->in_pack < b->in_pack)
1516 return -1;
1517 if (a->in_pack > b->in_pack)
1518 return 1;
1519 return a->in_pack_offset < b->in_pack_offset ? -1 :
1520 (a->in_pack_offset > b->in_pack_offset);
1521 }
1522
1523 /*
1524 * Drop an on-disk delta we were planning to reuse. Naively, this would
1525 * just involve blanking out the "delta" field, but we have to deal
1526 * with some extra book-keeping:
1527 *
1528 * 1. Removing ourselves from the delta_sibling linked list.
1529 *
1530 * 2. Updating our size/type to the non-delta representation. These were
1531 * either not recorded initially (size) or overwritten with the delta type
1532 * (type) when check_object() decided to reuse the delta.
1533 *
1534 * 3. Resetting our delta depth, as we are now a base object.
1535 */
1536 static void drop_reused_delta(struct object_entry *entry)
1537 {
1538 struct object_entry **p = &entry->delta->delta_child;
1539 struct object_info oi = OBJECT_INFO_INIT;
1540
1541 while (*p) {
1542 if (*p == entry)
1543 *p = (*p)->delta_sibling;
1544 else
1545 p = &(*p)->delta_sibling;
1546 }
1547 entry->delta = NULL;
1548 entry->depth = 0;
1549
1550 oi.sizep = &entry->size;
1551 oi.typep = &entry->type;
1552 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1553 /*
1554 * We failed to get the info from this pack for some reason;
1555 * fall back to sha1_object_info, which may find another copy.
1556 * And if that fails, the error will be recorded in entry->type
1557 * and dealt with in prepare_pack().
1558 */
1559 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1560 }
1561 }
1562
1563 /*
1564 * Follow the chain of deltas from this entry onward, throwing away any links
1565 * that cause us to hit a cycle (as determined by the DFS state flags in
1566 * the entries).
1567 *
1568 * We also detect too-long reused chains that would violate our --depth
1569 * limit.
1570 */
1571 static void break_delta_chains(struct object_entry *entry)
1572 {
1573 /*
1574 * The actual depth of each object we will write is stored as an int,
1575 * as it cannot exceed our int "depth" limit. But before we break
1576 * changes based no that limit, we may potentially go as deep as the
1577 * number of objects, which is elsewhere bounded to a uint32_t.
1578 */
1579 uint32_t total_depth;
1580 struct object_entry *cur, *next;
1581
1582 for (cur = entry, total_depth = 0;
1583 cur;
1584 cur = cur->delta, total_depth++) {
1585 if (cur->dfs_state == DFS_DONE) {
1586 /*
1587 * We've already seen this object and know it isn't
1588 * part of a cycle. We do need to append its depth
1589 * to our count.
1590 */
1591 total_depth += cur->depth;
1592 break;
1593 }
1594
1595 /*
1596 * We break cycles before looping, so an ACTIVE state (or any
1597 * other cruft which made its way into the state variable)
1598 * is a bug.
1599 */
1600 if (cur->dfs_state != DFS_NONE)
1601 die("BUG: confusing delta dfs state in first pass: %d",
1602 cur->dfs_state);
1603
1604 /*
1605 * Now we know this is the first time we've seen the object. If
1606 * it's not a delta, we're done traversing, but we'll mark it
1607 * done to save time on future traversals.
1608 */
1609 if (!cur->delta) {
1610 cur->dfs_state = DFS_DONE;
1611 break;
1612 }
1613
1614 /*
1615 * Mark ourselves as active and see if the next step causes
1616 * us to cycle to another active object. It's important to do
1617 * this _before_ we loop, because it impacts where we make the
1618 * cut, and thus how our total_depth counter works.
1619 * E.g., We may see a partial loop like:
1620 *
1621 * A -> B -> C -> D -> B
1622 *
1623 * Cutting B->C breaks the cycle. But now the depth of A is
1624 * only 1, and our total_depth counter is at 3. The size of the
1625 * error is always one less than the size of the cycle we
1626 * broke. Commits C and D were "lost" from A's chain.
1627 *
1628 * If we instead cut D->B, then the depth of A is correct at 3.
1629 * We keep all commits in the chain that we examined.
1630 */
1631 cur->dfs_state = DFS_ACTIVE;
1632 if (cur->delta->dfs_state == DFS_ACTIVE) {
1633 drop_reused_delta(cur);
1634 cur->dfs_state = DFS_DONE;
1635 break;
1636 }
1637 }
1638
1639 /*
1640 * And now that we've gone all the way to the bottom of the chain, we
1641 * need to clear the active flags and set the depth fields as
1642 * appropriate. Unlike the loop above, which can quit when it drops a
1643 * delta, we need to keep going to look for more depth cuts. So we need
1644 * an extra "next" pointer to keep going after we reset cur->delta.
1645 */
1646 for (cur = entry; cur; cur = next) {
1647 next = cur->delta;
1648
1649 /*
1650 * We should have a chain of zero or more ACTIVE states down to
1651 * a final DONE. We can quit after the DONE, because either it
1652 * has no bases, or we've already handled them in a previous
1653 * call.
1654 */
1655 if (cur->dfs_state == DFS_DONE)
1656 break;
1657 else if (cur->dfs_state != DFS_ACTIVE)
1658 die("BUG: confusing delta dfs state in second pass: %d",
1659 cur->dfs_state);
1660
1661 /*
1662 * If the total_depth is more than depth, then we need to snip
1663 * the chain into two or more smaller chains that don't exceed
1664 * the maximum depth. Most of the resulting chains will contain
1665 * (depth + 1) entries (i.e., depth deltas plus one base), and
1666 * the last chain (i.e., the one containing entry) will contain
1667 * whatever entries are left over, namely
1668 * (total_depth % (depth + 1)) of them.
1669 *
1670 * Since we are iterating towards decreasing depth, we need to
1671 * decrement total_depth as we go, and we need to write to the
1672 * entry what its final depth will be after all of the
1673 * snipping. Since we're snipping into chains of length (depth
1674 * + 1) entries, the final depth of an entry will be its
1675 * original depth modulo (depth + 1). Any time we encounter an
1676 * entry whose final depth is supposed to be zero, we snip it
1677 * from its delta base, thereby making it so.
1678 */
1679 cur->depth = (total_depth--) % (depth + 1);
1680 if (!cur->depth)
1681 drop_reused_delta(cur);
1682
1683 cur->dfs_state = DFS_DONE;
1684 }
1685 }
1686
1687 static void get_object_details(void)
1688 {
1689 uint32_t i;
1690 struct object_entry **sorted_by_offset;
1691
1692 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1693 for (i = 0; i < to_pack.nr_objects; i++)
1694 sorted_by_offset[i] = to_pack.objects + i;
1695 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1696
1697 for (i = 0; i < to_pack.nr_objects; i++) {
1698 struct object_entry *entry = sorted_by_offset[i];
1699 check_object(entry);
1700 if (big_file_threshold < entry->size)
1701 entry->no_try_delta = 1;
1702 }
1703
1704 /*
1705 * This must happen in a second pass, since we rely on the delta
1706 * information for the whole list being completed.
1707 */
1708 for (i = 0; i < to_pack.nr_objects; i++)
1709 break_delta_chains(&to_pack.objects[i]);
1710
1711 free(sorted_by_offset);
1712 }
1713
1714 /*
1715 * We search for deltas in a list sorted by type, by filename hash, and then
1716 * by size, so that we see progressively smaller and smaller files.
1717 * That's because we prefer deltas to be from the bigger file
1718 * to the smaller -- deletes are potentially cheaper, but perhaps
1719 * more importantly, the bigger file is likely the more recent
1720 * one. The deepest deltas are therefore the oldest objects which are
1721 * less susceptible to be accessed often.
1722 */
1723 static int type_size_sort(const void *_a, const void *_b)
1724 {
1725 const struct object_entry *a = *(struct object_entry **)_a;
1726 const struct object_entry *b = *(struct object_entry **)_b;
1727
1728 if (a->type > b->type)
1729 return -1;
1730 if (a->type < b->type)
1731 return 1;
1732 if (a->hash > b->hash)
1733 return -1;
1734 if (a->hash < b->hash)
1735 return 1;
1736 if (a->preferred_base > b->preferred_base)
1737 return -1;
1738 if (a->preferred_base < b->preferred_base)
1739 return 1;
1740 if (a->size > b->size)
1741 return -1;
1742 if (a->size < b->size)
1743 return 1;
1744 return a < b ? -1 : (a > b); /* newest first */
1745 }
1746
1747 struct unpacked {
1748 struct object_entry *entry;
1749 void *data;
1750 struct delta_index *index;
1751 unsigned depth;
1752 };
1753
1754 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1755 unsigned long delta_size)
1756 {
1757 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1758 return 0;
1759
1760 if (delta_size < cache_max_small_delta_size)
1761 return 1;
1762
1763 /* cache delta, if objects are large enough compared to delta size */
1764 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1765 return 1;
1766
1767 return 0;
1768 }
1769
1770 #ifndef NO_PTHREADS
1771
1772 static pthread_mutex_t read_mutex;
1773 #define read_lock() pthread_mutex_lock(&read_mutex)
1774 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1775
1776 static pthread_mutex_t cache_mutex;
1777 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1778 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1779
1780 static pthread_mutex_t progress_mutex;
1781 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1782 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1783
1784 #else
1785
1786 #define read_lock() (void)0
1787 #define read_unlock() (void)0
1788 #define cache_lock() (void)0
1789 #define cache_unlock() (void)0
1790 #define progress_lock() (void)0
1791 #define progress_unlock() (void)0
1792
1793 #endif
1794
1795 static int try_delta(struct unpacked *trg, struct unpacked *src,
1796 unsigned max_depth, unsigned long *mem_usage)
1797 {
1798 struct object_entry *trg_entry = trg->entry;
1799 struct object_entry *src_entry = src->entry;
1800 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1801 unsigned ref_depth;
1802 enum object_type type;
1803 void *delta_buf;
1804
1805 /* Don't bother doing diffs between different types */
1806 if (trg_entry->type != src_entry->type)
1807 return -1;
1808
1809 /*
1810 * We do not bother to try a delta that we discarded on an
1811 * earlier try, but only when reusing delta data. Note that
1812 * src_entry that is marked as the preferred_base should always
1813 * be considered, as even if we produce a suboptimal delta against
1814 * it, we will still save the transfer cost, as we already know
1815 * the other side has it and we won't send src_entry at all.
1816 */
1817 if (reuse_delta && trg_entry->in_pack &&
1818 trg_entry->in_pack == src_entry->in_pack &&
1819 !src_entry->preferred_base &&
1820 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1821 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1822 return 0;
1823
1824 /* Let's not bust the allowed depth. */
1825 if (src->depth >= max_depth)
1826 return 0;
1827
1828 /* Now some size filtering heuristics. */
1829 trg_size = trg_entry->size;
1830 if (!trg_entry->delta) {
1831 max_size = trg_size/2 - 20;
1832 ref_depth = 1;
1833 } else {
1834 max_size = trg_entry->delta_size;
1835 ref_depth = trg->depth;
1836 }
1837 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1838 (max_depth - ref_depth + 1);
1839 if (max_size == 0)
1840 return 0;
1841 src_size = src_entry->size;
1842 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1843 if (sizediff >= max_size)
1844 return 0;
1845 if (trg_size < src_size / 32)
1846 return 0;
1847
1848 /* Load data if not already done */
1849 if (!trg->data) {
1850 read_lock();
1851 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1852 read_unlock();
1853 if (!trg->data)
1854 die("object %s cannot be read",
1855 sha1_to_hex(trg_entry->idx.sha1));
1856 if (sz != trg_size)
1857 die("object %s inconsistent object length (%lu vs %lu)",
1858 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1859 *mem_usage += sz;
1860 }
1861 if (!src->data) {
1862 read_lock();
1863 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1864 read_unlock();
1865 if (!src->data) {
1866 if (src_entry->preferred_base) {
1867 static int warned = 0;
1868 if (!warned++)
1869 warning("object %s cannot be read",
1870 sha1_to_hex(src_entry->idx.sha1));
1871 /*
1872 * Those objects are not included in the
1873 * resulting pack. Be resilient and ignore
1874 * them if they can't be read, in case the
1875 * pack could be created nevertheless.
1876 */
1877 return 0;
1878 }
1879 die("object %s cannot be read",
1880 sha1_to_hex(src_entry->idx.sha1));
1881 }
1882 if (sz != src_size)
1883 die("object %s inconsistent object length (%lu vs %lu)",
1884 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1885 *mem_usage += sz;
1886 }
1887 if (!src->index) {
1888 src->index = create_delta_index(src->data, src_size);
1889 if (!src->index) {
1890 static int warned = 0;
1891 if (!warned++)
1892 warning("suboptimal pack - out of memory");
1893 return 0;
1894 }
1895 *mem_usage += sizeof_delta_index(src->index);
1896 }
1897
1898 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1899 if (!delta_buf)
1900 return 0;
1901
1902 if (trg_entry->delta) {
1903 /* Prefer only shallower same-sized deltas. */
1904 if (delta_size == trg_entry->delta_size &&
1905 src->depth + 1 >= trg->depth) {
1906 free(delta_buf);
1907 return 0;
1908 }
1909 }
1910
1911 /*
1912 * Handle memory allocation outside of the cache
1913 * accounting lock. Compiler will optimize the strangeness
1914 * away when NO_PTHREADS is defined.
1915 */
1916 free(trg_entry->delta_data);
1917 cache_lock();
1918 if (trg_entry->delta_data) {
1919 delta_cache_size -= trg_entry->delta_size;
1920 trg_entry->delta_data = NULL;
1921 }
1922 if (delta_cacheable(src_size, trg_size, delta_size)) {
1923 delta_cache_size += delta_size;
1924 cache_unlock();
1925 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1926 } else {
1927 cache_unlock();
1928 free(delta_buf);
1929 }
1930
1931 trg_entry->delta = src_entry;
1932 trg_entry->delta_size = delta_size;
1933 trg->depth = src->depth + 1;
1934
1935 return 1;
1936 }
1937
1938 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1939 {
1940 struct object_entry *child = me->delta_child;
1941 unsigned int m = n;
1942 while (child) {
1943 unsigned int c = check_delta_limit(child, n + 1);
1944 if (m < c)
1945 m = c;
1946 child = child->delta_sibling;
1947 }
1948 return m;
1949 }
1950
1951 static unsigned long free_unpacked(struct unpacked *n)
1952 {
1953 unsigned long freed_mem = sizeof_delta_index(n->index);
1954 free_delta_index(n->index);
1955 n->index = NULL;
1956 if (n->data) {
1957 freed_mem += n->entry->size;
1958 free(n->data);
1959 n->data = NULL;
1960 }
1961 n->entry = NULL;
1962 n->depth = 0;
1963 return freed_mem;
1964 }
1965
1966 static void find_deltas(struct object_entry **list, unsigned *list_size,
1967 int window, int depth, unsigned *processed)
1968 {
1969 uint32_t i, idx = 0, count = 0;
1970 struct unpacked *array;
1971 unsigned long mem_usage = 0;
1972
1973 array = xcalloc(window, sizeof(struct unpacked));
1974
1975 for (;;) {
1976 struct object_entry *entry;
1977 struct unpacked *n = array + idx;
1978 int j, max_depth, best_base = -1;
1979
1980 progress_lock();
1981 if (!*list_size) {
1982 progress_unlock();
1983 break;
1984 }
1985 entry = *list++;
1986 (*list_size)--;
1987 if (!entry->preferred_base) {
1988 (*processed)++;
1989 display_progress(progress_state, *processed);
1990 }
1991 progress_unlock();
1992
1993 mem_usage -= free_unpacked(n);
1994 n->entry = entry;
1995
1996 while (window_memory_limit &&
1997 mem_usage > window_memory_limit &&
1998 count > 1) {
1999 uint32_t tail = (idx + window - count) % window;
2000 mem_usage -= free_unpacked(array + tail);
2001 count--;
2002 }
2003
2004 /* We do not compute delta to *create* objects we are not
2005 * going to pack.
2006 */
2007 if (entry->preferred_base)
2008 goto next;
2009
2010 /*
2011 * If the current object is at pack edge, take the depth the
2012 * objects that depend on the current object into account
2013 * otherwise they would become too deep.
2014 */
2015 max_depth = depth;
2016 if (entry->delta_child) {
2017 max_depth -= check_delta_limit(entry, 0);
2018 if (max_depth <= 0)
2019 goto next;
2020 }
2021
2022 j = window;
2023 while (--j > 0) {
2024 int ret;
2025 uint32_t other_idx = idx + j;
2026 struct unpacked *m;
2027 if (other_idx >= window)
2028 other_idx -= window;
2029 m = array + other_idx;
2030 if (!m->entry)
2031 break;
2032 ret = try_delta(n, m, max_depth, &mem_usage);
2033 if (ret < 0)
2034 break;
2035 else if (ret > 0)
2036 best_base = other_idx;
2037 }
2038
2039 /*
2040 * If we decided to cache the delta data, then it is best
2041 * to compress it right away. First because we have to do
2042 * it anyway, and doing it here while we're threaded will
2043 * save a lot of time in the non threaded write phase,
2044 * as well as allow for caching more deltas within
2045 * the same cache size limit.
2046 * ...
2047 * But only if not writing to stdout, since in that case
2048 * the network is most likely throttling writes anyway,
2049 * and therefore it is best to go to the write phase ASAP
2050 * instead, as we can afford spending more time compressing
2051 * between writes at that moment.
2052 */
2053 if (entry->delta_data && !pack_to_stdout) {
2054 entry->z_delta_size = do_compress(&entry->delta_data,
2055 entry->delta_size);
2056 cache_lock();
2057 delta_cache_size -= entry->delta_size;
2058 delta_cache_size += entry->z_delta_size;
2059 cache_unlock();
2060 }
2061
2062 /* if we made n a delta, and if n is already at max
2063 * depth, leaving it in the window is pointless. we
2064 * should evict it first.
2065 */
2066 if (entry->delta && max_depth <= n->depth)
2067 continue;
2068
2069 /*
2070 * Move the best delta base up in the window, after the
2071 * currently deltified object, to keep it longer. It will
2072 * be the first base object to be attempted next.
2073 */
2074 if (entry->delta) {
2075 struct unpacked swap = array[best_base];
2076 int dist = (window + idx - best_base) % window;
2077 int dst = best_base;
2078 while (dist--) {
2079 int src = (dst + 1) % window;
2080 array[dst] = array[src];
2081 dst = src;
2082 }
2083 array[dst] = swap;
2084 }
2085
2086 next:
2087 idx++;
2088 if (count + 1 < window)
2089 count++;
2090 if (idx >= window)
2091 idx = 0;
2092 }
2093
2094 for (i = 0; i < window; ++i) {
2095 free_delta_index(array[i].index);
2096 free(array[i].data);
2097 }
2098 free(array);
2099 }
2100
2101 #ifndef NO_PTHREADS
2102
2103 static void try_to_free_from_threads(size_t size)
2104 {
2105 read_lock();
2106 release_pack_memory(size);
2107 read_unlock();
2108 }
2109
2110 static try_to_free_t old_try_to_free_routine;
2111
2112 /*
2113 * The main thread waits on the condition that (at least) one of the workers
2114 * has stopped working (which is indicated in the .working member of
2115 * struct thread_params).
2116 * When a work thread has completed its work, it sets .working to 0 and
2117 * signals the main thread and waits on the condition that .data_ready
2118 * becomes 1.
2119 */
2120
2121 struct thread_params {
2122 pthread_t thread;
2123 struct object_entry **list;
2124 unsigned list_size;
2125 unsigned remaining;
2126 int window;
2127 int depth;
2128 int working;
2129 int data_ready;
2130 pthread_mutex_t mutex;
2131 pthread_cond_t cond;
2132 unsigned *processed;
2133 };
2134
2135 static pthread_cond_t progress_cond;
2136
2137 /*
2138 * Mutex and conditional variable can't be statically-initialized on Windows.
2139 */
2140 static void init_threaded_search(void)
2141 {
2142 init_recursive_mutex(&read_mutex);
2143 pthread_mutex_init(&cache_mutex, NULL);
2144 pthread_mutex_init(&progress_mutex, NULL);
2145 pthread_cond_init(&progress_cond, NULL);
2146 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2147 }
2148
2149 static void cleanup_threaded_search(void)
2150 {
2151 set_try_to_free_routine(old_try_to_free_routine);
2152 pthread_cond_destroy(&progress_cond);
2153 pthread_mutex_destroy(&read_mutex);
2154 pthread_mutex_destroy(&cache_mutex);
2155 pthread_mutex_destroy(&progress_mutex);
2156 }
2157
2158 static void *threaded_find_deltas(void *arg)
2159 {
2160 struct thread_params *me = arg;
2161
2162 while (me->remaining) {
2163 find_deltas(me->list, &me->remaining,
2164 me->window, me->depth, me->processed);
2165
2166 progress_lock();
2167 me->working = 0;
2168 pthread_cond_signal(&progress_cond);
2169 progress_unlock();
2170
2171 /*
2172 * We must not set ->data_ready before we wait on the
2173 * condition because the main thread may have set it to 1
2174 * before we get here. In order to be sure that new
2175 * work is available if we see 1 in ->data_ready, it
2176 * was initialized to 0 before this thread was spawned
2177 * and we reset it to 0 right away.
2178 */
2179 pthread_mutex_lock(&me->mutex);
2180 while (!me->data_ready)
2181 pthread_cond_wait(&me->cond, &me->mutex);
2182 me->data_ready = 0;
2183 pthread_mutex_unlock(&me->mutex);
2184 }
2185 /* leave ->working 1 so that this doesn't get more work assigned */
2186 return NULL;
2187 }
2188
2189 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2190 int window, int depth, unsigned *processed)
2191 {
2192 struct thread_params *p;
2193 int i, ret, active_threads = 0;
2194
2195 init_threaded_search();
2196
2197 if (delta_search_threads <= 1) {
2198 find_deltas(list, &list_size, window, depth, processed);
2199 cleanup_threaded_search();
2200 return;
2201 }
2202 if (progress > pack_to_stdout)
2203 fprintf(stderr, "Delta compression using up to %d threads.\n",
2204 delta_search_threads);
2205 p = xcalloc(delta_search_threads, sizeof(*p));
2206
2207 /* Partition the work amongst work threads. */
2208 for (i = 0; i < delta_search_threads; i++) {
2209 unsigned sub_size = list_size / (delta_search_threads - i);
2210
2211 /* don't use too small segments or no deltas will be found */
2212 if (sub_size < 2*window && i+1 < delta_search_threads)
2213 sub_size = 0;
2214
2215 p[i].window = window;
2216 p[i].depth = depth;
2217 p[i].processed = processed;
2218 p[i].working = 1;
2219 p[i].data_ready = 0;
2220
2221 /* try to split chunks on "path" boundaries */
2222 while (sub_size && sub_size < list_size &&
2223 list[sub_size]->hash &&
2224 list[sub_size]->hash == list[sub_size-1]->hash)
2225 sub_size++;
2226
2227 p[i].list = list;
2228 p[i].list_size = sub_size;
2229 p[i].remaining = sub_size;
2230
2231 list += sub_size;
2232 list_size -= sub_size;
2233 }
2234
2235 /* Start work threads. */
2236 for (i = 0; i < delta_search_threads; i++) {
2237 if (!p[i].list_size)
2238 continue;
2239 pthread_mutex_init(&p[i].mutex, NULL);
2240 pthread_cond_init(&p[i].cond, NULL);
2241 ret = pthread_create(&p[i].thread, NULL,
2242 threaded_find_deltas, &p[i]);
2243 if (ret)
2244 die("unable to create thread: %s", strerror(ret));
2245 active_threads++;
2246 }
2247
2248 /*
2249 * Now let's wait for work completion. Each time a thread is done
2250 * with its work, we steal half of the remaining work from the
2251 * thread with the largest number of unprocessed objects and give
2252 * it to that newly idle thread. This ensure good load balancing
2253 * until the remaining object list segments are simply too short
2254 * to be worth splitting anymore.
2255 */
2256 while (active_threads) {
2257 struct thread_params *target = NULL;
2258 struct thread_params *victim = NULL;
2259 unsigned sub_size = 0;
2260
2261 progress_lock();
2262 for (;;) {
2263 for (i = 0; !target && i < delta_search_threads; i++)
2264 if (!p[i].working)
2265 target = &p[i];
2266 if (target)
2267 break;
2268 pthread_cond_wait(&progress_cond, &progress_mutex);
2269 }
2270
2271 for (i = 0; i < delta_search_threads; i++)
2272 if (p[i].remaining > 2*window &&
2273 (!victim || victim->remaining < p[i].remaining))
2274 victim = &p[i];
2275 if (victim) {
2276 sub_size = victim->remaining / 2;
2277 list = victim->list + victim->list_size - sub_size;
2278 while (sub_size && list[0]->hash &&
2279 list[0]->hash == list[-1]->hash) {
2280 list++;
2281 sub_size--;
2282 }
2283 if (!sub_size) {
2284 /*
2285 * It is possible for some "paths" to have
2286 * so many objects that no hash boundary
2287 * might be found. Let's just steal the
2288 * exact half in that case.
2289 */
2290 sub_size = victim->remaining / 2;
2291 list -= sub_size;
2292 }
2293 target->list = list;
2294 victim->list_size -= sub_size;
2295 victim->remaining -= sub_size;
2296 }
2297 target->list_size = sub_size;
2298 target->remaining = sub_size;
2299 target->working = 1;
2300 progress_unlock();
2301
2302 pthread_mutex_lock(&target->mutex);
2303 target->data_ready = 1;
2304 pthread_cond_signal(&target->cond);
2305 pthread_mutex_unlock(&target->mutex);
2306
2307 if (!sub_size) {
2308 pthread_join(target->thread, NULL);
2309 pthread_cond_destroy(&target->cond);
2310 pthread_mutex_destroy(&target->mutex);
2311 active_threads--;
2312 }
2313 }
2314 cleanup_threaded_search();
2315 free(p);
2316 }
2317
2318 #else
2319 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2320 #endif
2321
2322 static void add_tag_chain(const struct object_id *oid)
2323 {
2324 struct tag *tag;
2325
2326 /*
2327 * We catch duplicates already in add_object_entry(), but we'd
2328 * prefer to do this extra check to avoid having to parse the
2329 * tag at all if we already know that it's being packed (e.g., if
2330 * it was included via bitmaps, we would not have parsed it
2331 * previously).
2332 */
2333 if (packlist_find(&to_pack, oid->hash, NULL))
2334 return;
2335
2336 tag = lookup_tag(oid->hash);
2337 while (1) {
2338 if (!tag || parse_tag(tag) || !tag->tagged)
2339 die("unable to pack objects reachable from tag %s",
2340 oid_to_hex(oid));
2341
2342 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2343
2344 if (tag->tagged->type != OBJ_TAG)
2345 return;
2346
2347 tag = (struct tag *)tag->tagged;
2348 }
2349 }
2350
2351 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2352 {
2353 struct object_id peeled;
2354
2355 if (starts_with(path, "refs/tags/") && /* is a tag? */
2356 !peel_ref(path, peeled.hash) && /* peelable? */
2357 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2358 add_tag_chain(oid);
2359 return 0;
2360 }
2361
2362 static void prepare_pack(int window, int depth)
2363 {
2364 struct object_entry **delta_list;
2365 uint32_t i, nr_deltas;
2366 unsigned n;
2367
2368 get_object_details();
2369
2370 /*
2371 * If we're locally repacking then we need to be doubly careful
2372 * from now on in order to make sure no stealth corruption gets
2373 * propagated to the new pack. Clients receiving streamed packs
2374 * should validate everything they get anyway so no need to incur
2375 * the additional cost here in that case.
2376 */
2377 if (!pack_to_stdout)
2378 do_check_packed_object_crc = 1;
2379
2380 if (!to_pack.nr_objects || !window || !depth)
2381 return;
2382
2383 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2384 nr_deltas = n = 0;
2385
2386 for (i = 0; i < to_pack.nr_objects; i++) {
2387 struct object_entry *entry = to_pack.objects + i;
2388
2389 if (entry->delta)
2390 /* This happens if we decided to reuse existing
2391 * delta from a pack. "reuse_delta &&" is implied.
2392 */
2393 continue;
2394
2395 if (entry->size < 50)
2396 continue;
2397
2398 if (entry->no_try_delta)
2399 continue;
2400
2401 if (!entry->preferred_base) {
2402 nr_deltas++;
2403 if (entry->type < 0)
2404 die("unable to get type of object %s",
2405 sha1_to_hex(entry->idx.sha1));
2406 } else {
2407 if (entry->type < 0) {
2408 /*
2409 * This object is not found, but we
2410 * don't have to include it anyway.
2411 */
2412 continue;
2413 }
2414 }
2415
2416 delta_list[n++] = entry;
2417 }
2418
2419 if (nr_deltas && n > 1) {
2420 unsigned nr_done = 0;
2421 if (progress)
2422 progress_state = start_progress(_("Compressing objects"),
2423 nr_deltas);
2424 QSORT(delta_list, n, type_size_sort);
2425 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2426 stop_progress(&progress_state);
2427 if (nr_done != nr_deltas)
2428 die("inconsistency with delta count");
2429 }
2430 free(delta_list);
2431 }
2432
2433 static int git_pack_config(const char *k, const char *v, void *cb)
2434 {
2435 if (!strcmp(k, "pack.window")) {
2436 window = git_config_int(k, v);
2437 return 0;
2438 }
2439 if (!strcmp(k, "pack.windowmemory")) {
2440 window_memory_limit = git_config_ulong(k, v);
2441 return 0;
2442 }
2443 if (!strcmp(k, "pack.depth")) {
2444 depth = git_config_int(k, v);
2445 return 0;
2446 }
2447 if (!strcmp(k, "pack.deltacachesize")) {
2448 max_delta_cache_size = git_config_int(k, v);
2449 return 0;
2450 }
2451 if (!strcmp(k, "pack.deltacachelimit")) {
2452 cache_max_small_delta_size = git_config_int(k, v);
2453 return 0;
2454 }
2455 if (!strcmp(k, "pack.writebitmaphashcache")) {
2456 if (git_config_bool(k, v))
2457 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2458 else
2459 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2460 }
2461 if (!strcmp(k, "pack.usebitmaps")) {
2462 use_bitmap_index_default = git_config_bool(k, v);
2463 return 0;
2464 }
2465 if (!strcmp(k, "pack.threads")) {
2466 delta_search_threads = git_config_int(k, v);
2467 if (delta_search_threads < 0)
2468 die("invalid number of threads specified (%d)",
2469 delta_search_threads);
2470 #ifdef NO_PTHREADS
2471 if (delta_search_threads != 1)
2472 warning("no threads support, ignoring %s", k);
2473 #endif
2474 return 0;
2475 }
2476 if (!strcmp(k, "pack.indexversion")) {
2477 pack_idx_opts.version = git_config_int(k, v);
2478 if (pack_idx_opts.version > 2)
2479 die("bad pack.indexversion=%"PRIu32,
2480 pack_idx_opts.version);
2481 return 0;
2482 }
2483 return git_default_config(k, v, cb);
2484 }
2485
2486 static void read_object_list_from_stdin(void)
2487 {
2488 char line[40 + 1 + PATH_MAX + 2];
2489 unsigned char sha1[20];
2490
2491 for (;;) {
2492 if (!fgets(line, sizeof(line), stdin)) {
2493 if (feof(stdin))
2494 break;
2495 if (!ferror(stdin))
2496 die("fgets returned NULL, not EOF, not error!");
2497 if (errno != EINTR)
2498 die_errno("fgets");
2499 clearerr(stdin);
2500 continue;
2501 }
2502 if (line[0] == '-') {
2503 if (get_sha1_hex(line+1, sha1))
2504 die("expected edge sha1, got garbage:\n %s",
2505 line);
2506 add_preferred_base(sha1);
2507 continue;
2508 }
2509 if (get_sha1_hex(line, sha1))
2510 die("expected sha1, got garbage:\n %s", line);
2511
2512 add_preferred_base_object(line+41);
2513 add_object_entry(sha1, 0, line+41, 0);
2514 }
2515 }
2516
2517 #define OBJECT_ADDED (1u<<20)
2518
2519 static void show_commit(struct commit *commit, void *data)
2520 {
2521 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2522 commit->object.flags |= OBJECT_ADDED;
2523
2524 if (write_bitmap_index)
2525 index_commit_for_bitmap(commit);
2526 }
2527
2528 static void show_object(struct object *obj, const char *name, void *data)
2529 {
2530 add_preferred_base_object(name);
2531 add_object_entry(obj->oid.hash, obj->type, name, 0);
2532 obj->flags |= OBJECT_ADDED;
2533 }
2534
2535 static void show_edge(struct commit *commit)
2536 {
2537 add_preferred_base(commit->object.oid.hash);
2538 }
2539
2540 struct in_pack_object {
2541 off_t offset;
2542 struct object *object;
2543 };
2544
2545 struct in_pack {
2546 int alloc;
2547 int nr;
2548 struct in_pack_object *array;
2549 };
2550
2551 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2552 {
2553 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2554 in_pack->array[in_pack->nr].object = object;
2555 in_pack->nr++;
2556 }
2557
2558 /*
2559 * Compare the objects in the offset order, in order to emulate the
2560 * "git rev-list --objects" output that produced the pack originally.
2561 */
2562 static int ofscmp(const void *a_, const void *b_)
2563 {
2564 struct in_pack_object *a = (struct in_pack_object *)a_;
2565 struct in_pack_object *b = (struct in_pack_object *)b_;
2566
2567 if (a->offset < b->offset)
2568 return -1;
2569 else if (a->offset > b->offset)
2570 return 1;
2571 else
2572 return oidcmp(&a->object->oid, &b->object->oid);
2573 }
2574
2575 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2576 {
2577 struct packed_git *p;
2578 struct in_pack in_pack;
2579 uint32_t i;
2580
2581 memset(&in_pack, 0, sizeof(in_pack));
2582
2583 for (p = packed_git; p; p = p->next) {
2584 const unsigned char *sha1;
2585 struct object *o;
2586
2587 if (!p->pack_local || p->pack_keep)
2588 continue;
2589 if (open_pack_index(p))
2590 die("cannot open pack index");
2591
2592 ALLOC_GROW(in_pack.array,
2593 in_pack.nr + p->num_objects,
2594 in_pack.alloc);
2595
2596 for (i = 0; i < p->num_objects; i++) {
2597 sha1 = nth_packed_object_sha1(p, i);
2598 o = lookup_unknown_object(sha1);
2599 if (!(o->flags & OBJECT_ADDED))
2600 mark_in_pack_object(o, p, &in_pack);
2601 o->flags |= OBJECT_ADDED;
2602 }
2603 }
2604
2605 if (in_pack.nr) {
2606 QSORT(in_pack.array, in_pack.nr, ofscmp);
2607 for (i = 0; i < in_pack.nr; i++) {
2608 struct object *o = in_pack.array[i].object;
2609 add_object_entry(o->oid.hash, o->type, "", 0);
2610 }
2611 }
2612 free(in_pack.array);
2613 }
2614
2615 static int add_loose_object(const struct object_id *oid, const char *path,
2616 void *data)
2617 {
2618 enum object_type type = sha1_object_info(oid->hash, NULL);
2619
2620 if (type < 0) {
2621 warning("loose object at %s could not be examined", path);
2622 return 0;
2623 }
2624
2625 add_object_entry(oid->hash, type, "", 0);
2626 return 0;
2627 }
2628
2629 /*
2630 * We actually don't even have to worry about reachability here.
2631 * add_object_entry will weed out duplicates, so we just add every
2632 * loose object we find.
2633 */
2634 static void add_unreachable_loose_objects(void)
2635 {
2636 for_each_loose_file_in_objdir(get_object_directory(),
2637 add_loose_object,
2638 NULL, NULL, NULL);
2639 }
2640
2641 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2642 {
2643 static struct packed_git *last_found = (void *)1;
2644 struct packed_git *p;
2645
2646 p = (last_found != (void *)1) ? last_found : packed_git;
2647
2648 while (p) {
2649 if ((!p->pack_local || p->pack_keep) &&
2650 find_pack_entry_one(sha1, p)) {
2651 last_found = p;
2652 return 1;
2653 }
2654 if (p == last_found)
2655 p = packed_git;
2656 else
2657 p = p->next;
2658 if (p == last_found)
2659 p = p->next;
2660 }
2661 return 0;
2662 }
2663
2664 /*
2665 * Store a list of sha1s that are should not be discarded
2666 * because they are either written too recently, or are
2667 * reachable from another object that was.
2668 *
2669 * This is filled by get_object_list.
2670 */
2671 static struct sha1_array recent_objects;
2672
2673 static int loosened_object_can_be_discarded(const unsigned char *sha1,
2674 unsigned long mtime)
2675 {
2676 if (!unpack_unreachable_expiration)
2677 return 0;
2678 if (mtime > unpack_unreachable_expiration)
2679 return 0;
2680 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2681 return 0;
2682 return 1;
2683 }
2684
2685 static void loosen_unused_packed_objects(struct rev_info *revs)
2686 {
2687 struct packed_git *p;
2688 uint32_t i;
2689 const unsigned char *sha1;
2690
2691 for (p = packed_git; p; p = p->next) {
2692 if (!p->pack_local || p->pack_keep)
2693 continue;
2694
2695 if (open_pack_index(p))
2696 die("cannot open pack index");
2697
2698 for (i = 0; i < p->num_objects; i++) {
2699 sha1 = nth_packed_object_sha1(p, i);
2700 if (!packlist_find(&to_pack, sha1, NULL) &&
2701 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2702 !loosened_object_can_be_discarded(sha1, p->mtime))
2703 if (force_object_loose(sha1, p->mtime))
2704 die("unable to force loose object");
2705 }
2706 }
2707 }
2708
2709 /*
2710 * This tracks any options which pack-reuse code expects to be on, or which a
2711 * reader of the pack might not understand, and which would therefore prevent
2712 * blind reuse of what we have on disk.
2713 */
2714 static int pack_options_allow_reuse(void)
2715 {
2716 return pack_to_stdout && allow_ofs_delta;
2717 }
2718
2719 static int get_object_list_from_bitmap(struct rev_info *revs)
2720 {
2721 if (prepare_bitmap_walk(revs) < 0)
2722 return -1;
2723
2724 if (pack_options_allow_reuse() &&
2725 !reuse_partial_packfile_from_bitmap(
2726 &reuse_packfile,
2727 &reuse_packfile_objects,
2728 &reuse_packfile_offset)) {
2729 assert(reuse_packfile_objects);
2730 nr_result += reuse_packfile_objects;
2731 display_progress(progress_state, nr_result);
2732 }
2733
2734 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2735 return 0;
2736 }
2737
2738 static void record_recent_object(struct object *obj,
2739 const char *name,
2740 void *data)
2741 {
2742 sha1_array_append(&recent_objects, obj->oid.hash);
2743 }
2744
2745 static void record_recent_commit(struct commit *commit, void *data)
2746 {
2747 sha1_array_append(&recent_objects, commit->object.oid.hash);
2748 }
2749
2750 static void get_object_list(int ac, const char **av)
2751 {
2752 struct rev_info revs;
2753 char line[1000];
2754 int flags = 0;
2755
2756 init_revisions(&revs, NULL);
2757 save_commit_buffer = 0;
2758 setup_revisions(ac, av, &revs, NULL);
2759
2760 /* make sure shallows are read */
2761 is_repository_shallow();
2762
2763 while (fgets(line, sizeof(line), stdin) != NULL) {
2764 int len = strlen(line);
2765 if (len && line[len - 1] == '\n')
2766 line[--len] = 0;
2767 if (!len)
2768 break;
2769 if (*line == '-') {
2770 if (!strcmp(line, "--not")) {
2771 flags ^= UNINTERESTING;
2772 write_bitmap_index = 0;
2773 continue;
2774 }
2775 if (starts_with(line, "--shallow ")) {
2776 unsigned char sha1[20];
2777 if (get_sha1_hex(line + 10, sha1))
2778 die("not an SHA-1 '%s'", line + 10);
2779 register_shallow(sha1);
2780 use_bitmap_index = 0;
2781 continue;
2782 }
2783 die("not a rev '%s'", line);
2784 }
2785 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2786 die("bad revision '%s'", line);
2787 }
2788
2789 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2790 return;
2791
2792 if (prepare_revision_walk(&revs))
2793 die("revision walk setup failed");
2794 mark_edges_uninteresting(&revs, show_edge);
2795 traverse_commit_list(&revs, show_commit, show_object, NULL);
2796
2797 if (unpack_unreachable_expiration) {
2798 revs.ignore_missing_links = 1;
2799 if (add_unseen_recent_objects_to_traversal(&revs,
2800 unpack_unreachable_expiration))
2801 die("unable to add recent objects");
2802 if (prepare_revision_walk(&revs))
2803 die("revision walk setup failed");
2804 traverse_commit_list(&revs, record_recent_commit,
2805 record_recent_object, NULL);
2806 }
2807
2808 if (keep_unreachable)
2809 add_objects_in_unpacked_packs(&revs);
2810 if (pack_loose_unreachable)
2811 add_unreachable_loose_objects();
2812 if (unpack_unreachable)
2813 loosen_unused_packed_objects(&revs);
2814
2815 sha1_array_clear(&recent_objects);
2816 }
2817
2818 static int option_parse_index_version(const struct option *opt,
2819 const char *arg, int unset)
2820 {
2821 char *c;
2822 const char *val = arg;
2823 pack_idx_opts.version = strtoul(val, &c, 10);
2824 if (pack_idx_opts.version > 2)
2825 die(_("unsupported index version %s"), val);
2826 if (*c == ',' && c[1])
2827 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2828 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2829 die(_("bad index version '%s'"), val);
2830 return 0;
2831 }
2832
2833 static int option_parse_unpack_unreachable(const struct option *opt,
2834 const char *arg, int unset)
2835 {
2836 if (unset) {
2837 unpack_unreachable = 0;
2838 unpack_unreachable_expiration = 0;
2839 }
2840 else {
2841 unpack_unreachable = 1;
2842 if (arg)
2843 unpack_unreachable_expiration = approxidate(arg);
2844 }
2845 return 0;
2846 }
2847
2848 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2849 {
2850 int use_internal_rev_list = 0;
2851 int thin = 0;
2852 int shallow = 0;
2853 int all_progress_implied = 0;
2854 struct argv_array rp = ARGV_ARRAY_INIT;
2855 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2856 int rev_list_index = 0;
2857 struct option pack_objects_options[] = {
2858 OPT_SET_INT('q', "quiet", &progress,
2859 N_("do not show progress meter"), 0),
2860 OPT_SET_INT(0, "progress", &progress,
2861 N_("show progress meter"), 1),
2862 OPT_SET_INT(0, "all-progress", &progress,
2863 N_("show progress meter during object writing phase"), 2),
2864 OPT_BOOL(0, "all-progress-implied",
2865 &all_progress_implied,
2866 N_("similar to --all-progress when progress meter is shown")),
2867 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2868 N_("write the pack index file in the specified idx format version"),
2869 0, option_parse_index_version },
2870 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2871 N_("maximum size of each output pack file")),
2872 OPT_BOOL(0, "local", &local,
2873 N_("ignore borrowed objects from alternate object store")),
2874 OPT_BOOL(0, "incremental", &incremental,
2875 N_("ignore packed objects")),
2876 OPT_INTEGER(0, "window", &window,
2877 N_("limit pack window by objects")),
2878 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2879 N_("limit pack window by memory in addition to object limit")),
2880 OPT_INTEGER(0, "depth", &depth,
2881 N_("maximum length of delta chain allowed in the resulting pack")),
2882 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2883 N_("reuse existing deltas")),
2884 OPT_BOOL(0, "reuse-object", &reuse_object,
2885 N_("reuse existing objects")),
2886 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2887 N_("use OFS_DELTA objects")),
2888 OPT_INTEGER(0, "threads", &delta_search_threads,
2889 N_("use threads when searching for best delta matches")),
2890 OPT_BOOL(0, "non-empty", &non_empty,
2891 N_("do not create an empty pack output")),
2892 OPT_BOOL(0, "revs", &use_internal_rev_list,
2893 N_("read revision arguments from standard input")),
2894 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2895 N_("limit the objects to those that are not yet packed"),
2896 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2897 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2898 N_("include objects reachable from any reference"),
2899 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2900 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2901 N_("include objects referred by reflog entries"),
2902 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2903 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2904 N_("include objects referred to by the index"),
2905 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2906 OPT_BOOL(0, "stdout", &pack_to_stdout,
2907 N_("output pack to stdout")),
2908 OPT_BOOL(0, "include-tag", &include_tag,
2909 N_("include tag objects that refer to objects to be packed")),
2910 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2911 N_("keep unreachable objects")),
2912 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2913 N_("pack loose unreachable objects")),
2914 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2915 N_("unpack unreachable objects newer than <time>"),
2916 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2917 OPT_BOOL(0, "thin", &thin,
2918 N_("create thin packs")),
2919 OPT_BOOL(0, "shallow", &shallow,
2920 N_("create packs suitable for shallow fetches")),
2921 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2922 N_("ignore packs that have companion .keep file")),
2923 OPT_INTEGER(0, "compression", &pack_compression_level,
2924 N_("pack compression level")),
2925 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2926 N_("do not hide commits by grafts"), 0),
2927 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2928 N_("use a bitmap index if available to speed up counting objects")),
2929 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2930 N_("write a bitmap index together with the pack index")),
2931 OPT_END(),
2932 };
2933
2934 check_replace_refs = 0;
2935
2936 reset_pack_idx_option(&pack_idx_opts);
2937 git_config(git_pack_config, NULL);
2938
2939 progress = isatty(2);
2940 argc = parse_options(argc, argv, prefix, pack_objects_options,
2941 pack_usage, 0);
2942
2943 if (argc) {
2944 base_name = argv[0];
2945 argc--;
2946 }
2947 if (pack_to_stdout != !base_name || argc)
2948 usage_with_options(pack_usage, pack_objects_options);
2949
2950 argv_array_push(&rp, "pack-objects");
2951 if (thin) {
2952 use_internal_rev_list = 1;
2953 argv_array_push(&rp, shallow
2954 ? "--objects-edge-aggressive"
2955 : "--objects-edge");
2956 } else
2957 argv_array_push(&rp, "--objects");
2958
2959 if (rev_list_all) {
2960 use_internal_rev_list = 1;
2961 argv_array_push(&rp, "--all");
2962 }
2963 if (rev_list_reflog) {
2964 use_internal_rev_list = 1;
2965 argv_array_push(&rp, "--reflog");
2966 }
2967 if (rev_list_index) {
2968 use_internal_rev_list = 1;
2969 argv_array_push(&rp, "--indexed-objects");
2970 }
2971 if (rev_list_unpacked) {
2972 use_internal_rev_list = 1;
2973 argv_array_push(&rp, "--unpacked");
2974 }
2975
2976 if (!reuse_object)
2977 reuse_delta = 0;
2978 if (pack_compression_level == -1)
2979 pack_compression_level = Z_DEFAULT_COMPRESSION;
2980 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2981 die("bad pack compression level %d", pack_compression_level);
2982
2983 if (!delta_search_threads) /* --threads=0 means autodetect */
2984 delta_search_threads = online_cpus();
2985
2986 #ifdef NO_PTHREADS
2987 if (delta_search_threads != 1)
2988 warning("no threads support, ignoring --threads");
2989 #endif
2990 if (!pack_to_stdout && !pack_size_limit)
2991 pack_size_limit = pack_size_limit_cfg;
2992 if (pack_to_stdout && pack_size_limit)
2993 die("--max-pack-size cannot be used to build a pack for transfer.");
2994 if (pack_size_limit && pack_size_limit < 1024*1024) {
2995 warning("minimum pack size limit is 1 MiB");
2996 pack_size_limit = 1024*1024;
2997 }
2998
2999 if (!pack_to_stdout && thin)
3000 die("--thin cannot be used to build an indexable pack.");
3001
3002 if (keep_unreachable && unpack_unreachable)
3003 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3004 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3005 unpack_unreachable_expiration = 0;
3006
3007 /*
3008 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3009 *
3010 * - to produce good pack (with bitmap index not-yet-packed objects are
3011 * packed in suboptimal order).
3012 *
3013 * - to use more robust pack-generation codepath (avoiding possible
3014 * bugs in bitmap code and possible bitmap index corruption).
3015 */
3016 if (!pack_to_stdout)
3017 use_bitmap_index_default = 0;
3018
3019 if (use_bitmap_index < 0)
3020 use_bitmap_index = use_bitmap_index_default;
3021
3022 /* "hard" reasons not to use bitmaps; these just won't work at all */
3023 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3024 use_bitmap_index = 0;
3025
3026 if (pack_to_stdout || !rev_list_all)
3027 write_bitmap_index = 0;
3028
3029 if (progress && all_progress_implied)
3030 progress = 2;
3031
3032 prepare_packed_git();
3033 if (ignore_packed_keep) {
3034 struct packed_git *p;
3035 for (p = packed_git; p; p = p->next)
3036 if (p->pack_local && p->pack_keep)
3037 break;
3038 if (!p) /* no keep-able packs found */
3039 ignore_packed_keep = 0;
3040 }
3041 if (local) {
3042 /*
3043 * unlike ignore_packed_keep above, we do not want to
3044 * unset "local" based on looking at packs, as it
3045 * also covers non-local objects
3046 */
3047 struct packed_git *p;
3048 for (p = packed_git; p; p = p->next) {
3049 if (!p->pack_local) {
3050 have_non_local_packs = 1;
3051 break;
3052 }
3053 }
3054 }
3055
3056 if (progress)
3057 progress_state = start_progress(_("Counting objects"), 0);
3058 if (!use_internal_rev_list)
3059 read_object_list_from_stdin();
3060 else {
3061 get_object_list(rp.argc, rp.argv);
3062 argv_array_clear(&rp);
3063 }
3064 cleanup_preferred_base();
3065 if (include_tag && nr_result)
3066 for_each_ref(add_ref_tag, NULL);
3067 stop_progress(&progress_state);
3068
3069 if (non_empty && !nr_result)
3070 return 0;
3071 if (nr_result)
3072 prepare_pack(window, depth);
3073 write_pack_file();
3074 if (progress)
3075 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3076 " reused %"PRIu32" (delta %"PRIu32")\n",
3077 written, written_delta, reused, reused_delta);
3078 return 0;
3079 }