Merge branch 'rs/move-array' into maint
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "config.h"
4 #include "attr.h"
5 #include "object.h"
6 #include "blob.h"
7 #include "commit.h"
8 #include "tag.h"
9 #include "tree.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "pack-revindex.h"
13 #include "csum-file.h"
14 #include "tree-walk.h"
15 #include "diff.h"
16 #include "revision.h"
17 #include "list-objects.h"
18 #include "pack-objects.h"
19 #include "progress.h"
20 #include "refs.h"
21 #include "streaming.h"
22 #include "thread-utils.h"
23 #include "pack-bitmap.h"
24 #include "reachable.h"
25 #include "sha1-array.h"
26 #include "argv-array.h"
27 #include "mru.h"
28
29 static const char *pack_usage[] = {
30 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
31 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
32 NULL
33 };
34
35 /*
36 * Objects we are going to pack are collected in the `to_pack` structure.
37 * It contains an array (dynamically expanded) of the object data, and a map
38 * that can resolve SHA1s to their position in the array.
39 */
40 static struct packing_data to_pack;
41
42 static struct pack_idx_entry **written_list;
43 static uint32_t nr_result, nr_written;
44
45 static int non_empty;
46 static int reuse_delta = 1, reuse_object = 1;
47 static int keep_unreachable, unpack_unreachable, include_tag;
48 static timestamp_t unpack_unreachable_expiration;
49 static int pack_loose_unreachable;
50 static int local;
51 static int have_non_local_packs;
52 static int incremental;
53 static int ignore_packed_keep;
54 static int allow_ofs_delta;
55 static struct pack_idx_option pack_idx_opts;
56 static const char *base_name;
57 static int progress = 1;
58 static int window = 10;
59 static unsigned long pack_size_limit;
60 static int depth = 50;
61 static int delta_search_threads;
62 static int pack_to_stdout;
63 static int num_preferred_base;
64 static struct progress *progress_state;
65
66 static struct packed_git *reuse_packfile;
67 static uint32_t reuse_packfile_objects;
68 static off_t reuse_packfile_offset;
69
70 static int use_bitmap_index_default = 1;
71 static int use_bitmap_index = -1;
72 static int write_bitmap_index;
73 static uint16_t write_bitmap_options;
74
75 static unsigned long delta_cache_size = 0;
76 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
77 static unsigned long cache_max_small_delta_size = 1000;
78
79 static unsigned long window_memory_limit = 0;
80
81 /*
82 * stats
83 */
84 static uint32_t written, written_delta;
85 static uint32_t reused, reused_delta;
86
87 /*
88 * Indexed commits
89 */
90 static struct commit **indexed_commits;
91 static unsigned int indexed_commits_nr;
92 static unsigned int indexed_commits_alloc;
93
94 static void index_commit_for_bitmap(struct commit *commit)
95 {
96 if (indexed_commits_nr >= indexed_commits_alloc) {
97 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
98 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
99 }
100
101 indexed_commits[indexed_commits_nr++] = commit;
102 }
103
104 static void *get_delta(struct object_entry *entry)
105 {
106 unsigned long size, base_size, delta_size;
107 void *buf, *base_buf, *delta_buf;
108 enum object_type type;
109
110 buf = read_sha1_file(entry->idx.oid.hash, &type, &size);
111 if (!buf)
112 die("unable to read %s", oid_to_hex(&entry->idx.oid));
113 base_buf = read_sha1_file(entry->delta->idx.oid.hash, &type,
114 &base_size);
115 if (!base_buf)
116 die("unable to read %s",
117 oid_to_hex(&entry->delta->idx.oid));
118 delta_buf = diff_delta(base_buf, base_size,
119 buf, size, &delta_size, 0);
120 if (!delta_buf || delta_size != entry->delta_size)
121 die("delta size changed");
122 free(buf);
123 free(base_buf);
124 return delta_buf;
125 }
126
127 static unsigned long do_compress(void **pptr, unsigned long size)
128 {
129 git_zstream stream;
130 void *in, *out;
131 unsigned long maxsize;
132
133 git_deflate_init(&stream, pack_compression_level);
134 maxsize = git_deflate_bound(&stream, size);
135
136 in = *pptr;
137 out = xmalloc(maxsize);
138 *pptr = out;
139
140 stream.next_in = in;
141 stream.avail_in = size;
142 stream.next_out = out;
143 stream.avail_out = maxsize;
144 while (git_deflate(&stream, Z_FINISH) == Z_OK)
145 ; /* nothing */
146 git_deflate_end(&stream);
147
148 free(in);
149 return stream.total_out;
150 }
151
152 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
153 const unsigned char *sha1)
154 {
155 git_zstream stream;
156 unsigned char ibuf[1024 * 16];
157 unsigned char obuf[1024 * 16];
158 unsigned long olen = 0;
159
160 git_deflate_init(&stream, pack_compression_level);
161
162 for (;;) {
163 ssize_t readlen;
164 int zret = Z_OK;
165 readlen = read_istream(st, ibuf, sizeof(ibuf));
166 if (readlen == -1)
167 die(_("unable to read %s"), sha1_to_hex(sha1));
168
169 stream.next_in = ibuf;
170 stream.avail_in = readlen;
171 while ((stream.avail_in || readlen == 0) &&
172 (zret == Z_OK || zret == Z_BUF_ERROR)) {
173 stream.next_out = obuf;
174 stream.avail_out = sizeof(obuf);
175 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
176 sha1write(f, obuf, stream.next_out - obuf);
177 olen += stream.next_out - obuf;
178 }
179 if (stream.avail_in)
180 die(_("deflate error (%d)"), zret);
181 if (readlen == 0) {
182 if (zret != Z_STREAM_END)
183 die(_("deflate error (%d)"), zret);
184 break;
185 }
186 }
187 git_deflate_end(&stream);
188 return olen;
189 }
190
191 /*
192 * we are going to reuse the existing object data as is. make
193 * sure it is not corrupt.
194 */
195 static int check_pack_inflate(struct packed_git *p,
196 struct pack_window **w_curs,
197 off_t offset,
198 off_t len,
199 unsigned long expect)
200 {
201 git_zstream stream;
202 unsigned char fakebuf[4096], *in;
203 int st;
204
205 memset(&stream, 0, sizeof(stream));
206 git_inflate_init(&stream);
207 do {
208 in = use_pack(p, w_curs, offset, &stream.avail_in);
209 stream.next_in = in;
210 stream.next_out = fakebuf;
211 stream.avail_out = sizeof(fakebuf);
212 st = git_inflate(&stream, Z_FINISH);
213 offset += stream.next_in - in;
214 } while (st == Z_OK || st == Z_BUF_ERROR);
215 git_inflate_end(&stream);
216 return (st == Z_STREAM_END &&
217 stream.total_out == expect &&
218 stream.total_in == len) ? 0 : -1;
219 }
220
221 static void copy_pack_data(struct sha1file *f,
222 struct packed_git *p,
223 struct pack_window **w_curs,
224 off_t offset,
225 off_t len)
226 {
227 unsigned char *in;
228 unsigned long avail;
229
230 while (len) {
231 in = use_pack(p, w_curs, offset, &avail);
232 if (avail > len)
233 avail = (unsigned long)len;
234 sha1write(f, in, avail);
235 offset += avail;
236 len -= avail;
237 }
238 }
239
240 /* Return 0 if we will bust the pack-size limit */
241 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
242 unsigned long limit, int usable_delta)
243 {
244 unsigned long size, datalen;
245 unsigned char header[MAX_PACK_OBJECT_HEADER],
246 dheader[MAX_PACK_OBJECT_HEADER];
247 unsigned hdrlen;
248 enum object_type type;
249 void *buf;
250 struct git_istream *st = NULL;
251
252 if (!usable_delta) {
253 if (entry->type == OBJ_BLOB &&
254 entry->size > big_file_threshold &&
255 (st = open_istream(entry->idx.oid.hash, &type, &size, NULL)) != NULL)
256 buf = NULL;
257 else {
258 buf = read_sha1_file(entry->idx.oid.hash, &type,
259 &size);
260 if (!buf)
261 die(_("unable to read %s"),
262 oid_to_hex(&entry->idx.oid));
263 }
264 /*
265 * make sure no cached delta data remains from a
266 * previous attempt before a pack split occurred.
267 */
268 FREE_AND_NULL(entry->delta_data);
269 entry->z_delta_size = 0;
270 } else if (entry->delta_data) {
271 size = entry->delta_size;
272 buf = entry->delta_data;
273 entry->delta_data = NULL;
274 type = (allow_ofs_delta && entry->delta->idx.offset) ?
275 OBJ_OFS_DELTA : OBJ_REF_DELTA;
276 } else {
277 buf = get_delta(entry);
278 size = entry->delta_size;
279 type = (allow_ofs_delta && entry->delta->idx.offset) ?
280 OBJ_OFS_DELTA : OBJ_REF_DELTA;
281 }
282
283 if (st) /* large blob case, just assume we don't compress well */
284 datalen = size;
285 else if (entry->z_delta_size)
286 datalen = entry->z_delta_size;
287 else
288 datalen = do_compress(&buf, size);
289
290 /*
291 * The object header is a byte of 'type' followed by zero or
292 * more bytes of length.
293 */
294 hdrlen = encode_in_pack_object_header(header, sizeof(header),
295 type, size);
296
297 if (type == OBJ_OFS_DELTA) {
298 /*
299 * Deltas with relative base contain an additional
300 * encoding of the relative offset for the delta
301 * base from this object's position in the pack.
302 */
303 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
304 unsigned pos = sizeof(dheader) - 1;
305 dheader[pos] = ofs & 127;
306 while (ofs >>= 7)
307 dheader[--pos] = 128 | (--ofs & 127);
308 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
309 if (st)
310 close_istream(st);
311 free(buf);
312 return 0;
313 }
314 sha1write(f, header, hdrlen);
315 sha1write(f, dheader + pos, sizeof(dheader) - pos);
316 hdrlen += sizeof(dheader) - pos;
317 } else if (type == OBJ_REF_DELTA) {
318 /*
319 * Deltas with a base reference contain
320 * an additional 20 bytes for the base sha1.
321 */
322 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
323 if (st)
324 close_istream(st);
325 free(buf);
326 return 0;
327 }
328 sha1write(f, header, hdrlen);
329 sha1write(f, entry->delta->idx.oid.hash, 20);
330 hdrlen += 20;
331 } else {
332 if (limit && hdrlen + datalen + 20 >= limit) {
333 if (st)
334 close_istream(st);
335 free(buf);
336 return 0;
337 }
338 sha1write(f, header, hdrlen);
339 }
340 if (st) {
341 datalen = write_large_blob_data(st, f, entry->idx.oid.hash);
342 close_istream(st);
343 } else {
344 sha1write(f, buf, datalen);
345 free(buf);
346 }
347
348 return hdrlen + datalen;
349 }
350
351 /* Return 0 if we will bust the pack-size limit */
352 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
353 unsigned long limit, int usable_delta)
354 {
355 struct packed_git *p = entry->in_pack;
356 struct pack_window *w_curs = NULL;
357 struct revindex_entry *revidx;
358 off_t offset;
359 enum object_type type = entry->type;
360 off_t datalen;
361 unsigned char header[MAX_PACK_OBJECT_HEADER],
362 dheader[MAX_PACK_OBJECT_HEADER];
363 unsigned hdrlen;
364
365 if (entry->delta)
366 type = (allow_ofs_delta && entry->delta->idx.offset) ?
367 OBJ_OFS_DELTA : OBJ_REF_DELTA;
368 hdrlen = encode_in_pack_object_header(header, sizeof(header),
369 type, entry->size);
370
371 offset = entry->in_pack_offset;
372 revidx = find_pack_revindex(p, offset);
373 datalen = revidx[1].offset - offset;
374 if (!pack_to_stdout && p->index_version > 1 &&
375 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
376 error("bad packed object CRC for %s",
377 oid_to_hex(&entry->idx.oid));
378 unuse_pack(&w_curs);
379 return write_no_reuse_object(f, entry, limit, usable_delta);
380 }
381
382 offset += entry->in_pack_header_size;
383 datalen -= entry->in_pack_header_size;
384
385 if (!pack_to_stdout && p->index_version == 1 &&
386 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
387 error("corrupt packed object for %s",
388 oid_to_hex(&entry->idx.oid));
389 unuse_pack(&w_curs);
390 return write_no_reuse_object(f, entry, limit, usable_delta);
391 }
392
393 if (type == OBJ_OFS_DELTA) {
394 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
395 unsigned pos = sizeof(dheader) - 1;
396 dheader[pos] = ofs & 127;
397 while (ofs >>= 7)
398 dheader[--pos] = 128 | (--ofs & 127);
399 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
400 unuse_pack(&w_curs);
401 return 0;
402 }
403 sha1write(f, header, hdrlen);
404 sha1write(f, dheader + pos, sizeof(dheader) - pos);
405 hdrlen += sizeof(dheader) - pos;
406 reused_delta++;
407 } else if (type == OBJ_REF_DELTA) {
408 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
409 unuse_pack(&w_curs);
410 return 0;
411 }
412 sha1write(f, header, hdrlen);
413 sha1write(f, entry->delta->idx.oid.hash, 20);
414 hdrlen += 20;
415 reused_delta++;
416 } else {
417 if (limit && hdrlen + datalen + 20 >= limit) {
418 unuse_pack(&w_curs);
419 return 0;
420 }
421 sha1write(f, header, hdrlen);
422 }
423 copy_pack_data(f, p, &w_curs, offset, datalen);
424 unuse_pack(&w_curs);
425 reused++;
426 return hdrlen + datalen;
427 }
428
429 /* Return 0 if we will bust the pack-size limit */
430 static off_t write_object(struct sha1file *f,
431 struct object_entry *entry,
432 off_t write_offset)
433 {
434 unsigned long limit;
435 off_t len;
436 int usable_delta, to_reuse;
437
438 if (!pack_to_stdout)
439 crc32_begin(f);
440
441 /* apply size limit if limited packsize and not first object */
442 if (!pack_size_limit || !nr_written)
443 limit = 0;
444 else if (pack_size_limit <= write_offset)
445 /*
446 * the earlier object did not fit the limit; avoid
447 * mistaking this with unlimited (i.e. limit = 0).
448 */
449 limit = 1;
450 else
451 limit = pack_size_limit - write_offset;
452
453 if (!entry->delta)
454 usable_delta = 0; /* no delta */
455 else if (!pack_size_limit)
456 usable_delta = 1; /* unlimited packfile */
457 else if (entry->delta->idx.offset == (off_t)-1)
458 usable_delta = 0; /* base was written to another pack */
459 else if (entry->delta->idx.offset)
460 usable_delta = 1; /* base already exists in this pack */
461 else
462 usable_delta = 0; /* base could end up in another pack */
463
464 if (!reuse_object)
465 to_reuse = 0; /* explicit */
466 else if (!entry->in_pack)
467 to_reuse = 0; /* can't reuse what we don't have */
468 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
469 /* check_object() decided it for us ... */
470 to_reuse = usable_delta;
471 /* ... but pack split may override that */
472 else if (entry->type != entry->in_pack_type)
473 to_reuse = 0; /* pack has delta which is unusable */
474 else if (entry->delta)
475 to_reuse = 0; /* we want to pack afresh */
476 else
477 to_reuse = 1; /* we have it in-pack undeltified,
478 * and we do not need to deltify it.
479 */
480
481 if (!to_reuse)
482 len = write_no_reuse_object(f, entry, limit, usable_delta);
483 else
484 len = write_reuse_object(f, entry, limit, usable_delta);
485 if (!len)
486 return 0;
487
488 if (usable_delta)
489 written_delta++;
490 written++;
491 if (!pack_to_stdout)
492 entry->idx.crc32 = crc32_end(f);
493 return len;
494 }
495
496 enum write_one_status {
497 WRITE_ONE_SKIP = -1, /* already written */
498 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
499 WRITE_ONE_WRITTEN = 1, /* normal */
500 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
501 };
502
503 static enum write_one_status write_one(struct sha1file *f,
504 struct object_entry *e,
505 off_t *offset)
506 {
507 off_t size;
508 int recursing;
509
510 /*
511 * we set offset to 1 (which is an impossible value) to mark
512 * the fact that this object is involved in "write its base
513 * first before writing a deltified object" recursion.
514 */
515 recursing = (e->idx.offset == 1);
516 if (recursing) {
517 warning("recursive delta detected for object %s",
518 oid_to_hex(&e->idx.oid));
519 return WRITE_ONE_RECURSIVE;
520 } else if (e->idx.offset || e->preferred_base) {
521 /* offset is non zero if object is written already. */
522 return WRITE_ONE_SKIP;
523 }
524
525 /* if we are deltified, write out base object first. */
526 if (e->delta) {
527 e->idx.offset = 1; /* now recurse */
528 switch (write_one(f, e->delta, offset)) {
529 case WRITE_ONE_RECURSIVE:
530 /* we cannot depend on this one */
531 e->delta = NULL;
532 break;
533 default:
534 break;
535 case WRITE_ONE_BREAK:
536 e->idx.offset = recursing;
537 return WRITE_ONE_BREAK;
538 }
539 }
540
541 e->idx.offset = *offset;
542 size = write_object(f, e, *offset);
543 if (!size) {
544 e->idx.offset = recursing;
545 return WRITE_ONE_BREAK;
546 }
547 written_list[nr_written++] = &e->idx;
548
549 /* make sure off_t is sufficiently large not to wrap */
550 if (signed_add_overflows(*offset, size))
551 die("pack too large for current definition of off_t");
552 *offset += size;
553 return WRITE_ONE_WRITTEN;
554 }
555
556 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
557 void *cb_data)
558 {
559 unsigned char peeled[20];
560 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
561
562 if (entry)
563 entry->tagged = 1;
564 if (!peel_ref(path, peeled)) {
565 entry = packlist_find(&to_pack, peeled, NULL);
566 if (entry)
567 entry->tagged = 1;
568 }
569 return 0;
570 }
571
572 static inline void add_to_write_order(struct object_entry **wo,
573 unsigned int *endp,
574 struct object_entry *e)
575 {
576 if (e->filled)
577 return;
578 wo[(*endp)++] = e;
579 e->filled = 1;
580 }
581
582 static void add_descendants_to_write_order(struct object_entry **wo,
583 unsigned int *endp,
584 struct object_entry *e)
585 {
586 int add_to_order = 1;
587 while (e) {
588 if (add_to_order) {
589 struct object_entry *s;
590 /* add this node... */
591 add_to_write_order(wo, endp, e);
592 /* all its siblings... */
593 for (s = e->delta_sibling; s; s = s->delta_sibling) {
594 add_to_write_order(wo, endp, s);
595 }
596 }
597 /* drop down a level to add left subtree nodes if possible */
598 if (e->delta_child) {
599 add_to_order = 1;
600 e = e->delta_child;
601 } else {
602 add_to_order = 0;
603 /* our sibling might have some children, it is next */
604 if (e->delta_sibling) {
605 e = e->delta_sibling;
606 continue;
607 }
608 /* go back to our parent node */
609 e = e->delta;
610 while (e && !e->delta_sibling) {
611 /* we're on the right side of a subtree, keep
612 * going up until we can go right again */
613 e = e->delta;
614 }
615 if (!e) {
616 /* done- we hit our original root node */
617 return;
618 }
619 /* pass it off to sibling at this level */
620 e = e->delta_sibling;
621 }
622 };
623 }
624
625 static void add_family_to_write_order(struct object_entry **wo,
626 unsigned int *endp,
627 struct object_entry *e)
628 {
629 struct object_entry *root;
630
631 for (root = e; root->delta; root = root->delta)
632 ; /* nothing */
633 add_descendants_to_write_order(wo, endp, root);
634 }
635
636 static struct object_entry **compute_write_order(void)
637 {
638 unsigned int i, wo_end, last_untagged;
639
640 struct object_entry **wo;
641 struct object_entry *objects = to_pack.objects;
642
643 for (i = 0; i < to_pack.nr_objects; i++) {
644 objects[i].tagged = 0;
645 objects[i].filled = 0;
646 objects[i].delta_child = NULL;
647 objects[i].delta_sibling = NULL;
648 }
649
650 /*
651 * Fully connect delta_child/delta_sibling network.
652 * Make sure delta_sibling is sorted in the original
653 * recency order.
654 */
655 for (i = to_pack.nr_objects; i > 0;) {
656 struct object_entry *e = &objects[--i];
657 if (!e->delta)
658 continue;
659 /* Mark me as the first child */
660 e->delta_sibling = e->delta->delta_child;
661 e->delta->delta_child = e;
662 }
663
664 /*
665 * Mark objects that are at the tip of tags.
666 */
667 for_each_tag_ref(mark_tagged, NULL);
668
669 /*
670 * Give the objects in the original recency order until
671 * we see a tagged tip.
672 */
673 ALLOC_ARRAY(wo, to_pack.nr_objects);
674 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
675 if (objects[i].tagged)
676 break;
677 add_to_write_order(wo, &wo_end, &objects[i]);
678 }
679 last_untagged = i;
680
681 /*
682 * Then fill all the tagged tips.
683 */
684 for (; i < to_pack.nr_objects; i++) {
685 if (objects[i].tagged)
686 add_to_write_order(wo, &wo_end, &objects[i]);
687 }
688
689 /*
690 * And then all remaining commits and tags.
691 */
692 for (i = last_untagged; i < to_pack.nr_objects; i++) {
693 if (objects[i].type != OBJ_COMMIT &&
694 objects[i].type != OBJ_TAG)
695 continue;
696 add_to_write_order(wo, &wo_end, &objects[i]);
697 }
698
699 /*
700 * And then all the trees.
701 */
702 for (i = last_untagged; i < to_pack.nr_objects; i++) {
703 if (objects[i].type != OBJ_TREE)
704 continue;
705 add_to_write_order(wo, &wo_end, &objects[i]);
706 }
707
708 /*
709 * Finally all the rest in really tight order
710 */
711 for (i = last_untagged; i < to_pack.nr_objects; i++) {
712 if (!objects[i].filled)
713 add_family_to_write_order(wo, &wo_end, &objects[i]);
714 }
715
716 if (wo_end != to_pack.nr_objects)
717 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
718
719 return wo;
720 }
721
722 static off_t write_reused_pack(struct sha1file *f)
723 {
724 unsigned char buffer[8192];
725 off_t to_write, total;
726 int fd;
727
728 if (!is_pack_valid(reuse_packfile))
729 die("packfile is invalid: %s", reuse_packfile->pack_name);
730
731 fd = git_open(reuse_packfile->pack_name);
732 if (fd < 0)
733 die_errno("unable to open packfile for reuse: %s",
734 reuse_packfile->pack_name);
735
736 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
737 die_errno("unable to seek in reused packfile");
738
739 if (reuse_packfile_offset < 0)
740 reuse_packfile_offset = reuse_packfile->pack_size - 20;
741
742 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
743
744 while (to_write) {
745 int read_pack = xread(fd, buffer, sizeof(buffer));
746
747 if (read_pack <= 0)
748 die_errno("unable to read from reused packfile");
749
750 if (read_pack > to_write)
751 read_pack = to_write;
752
753 sha1write(f, buffer, read_pack);
754 to_write -= read_pack;
755
756 /*
757 * We don't know the actual number of objects written,
758 * only how many bytes written, how many bytes total, and
759 * how many objects total. So we can fake it by pretending all
760 * objects we are writing are the same size. This gives us a
761 * smooth progress meter, and at the end it matches the true
762 * answer.
763 */
764 written = reuse_packfile_objects *
765 (((double)(total - to_write)) / total);
766 display_progress(progress_state, written);
767 }
768
769 close(fd);
770 written = reuse_packfile_objects;
771 display_progress(progress_state, written);
772 return reuse_packfile_offset - sizeof(struct pack_header);
773 }
774
775 static const char no_split_warning[] = N_(
776 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
777 );
778
779 static void write_pack_file(void)
780 {
781 uint32_t i = 0, j;
782 struct sha1file *f;
783 off_t offset;
784 uint32_t nr_remaining = nr_result;
785 time_t last_mtime = 0;
786 struct object_entry **write_order;
787
788 if (progress > pack_to_stdout)
789 progress_state = start_progress(_("Writing objects"), nr_result);
790 ALLOC_ARRAY(written_list, to_pack.nr_objects);
791 write_order = compute_write_order();
792
793 do {
794 unsigned char sha1[20];
795 char *pack_tmp_name = NULL;
796
797 if (pack_to_stdout)
798 f = sha1fd_throughput(1, "<stdout>", progress_state);
799 else
800 f = create_tmp_packfile(&pack_tmp_name);
801
802 offset = write_pack_header(f, nr_remaining);
803
804 if (reuse_packfile) {
805 off_t packfile_size;
806 assert(pack_to_stdout);
807
808 packfile_size = write_reused_pack(f);
809 offset += packfile_size;
810 }
811
812 nr_written = 0;
813 for (; i < to_pack.nr_objects; i++) {
814 struct object_entry *e = write_order[i];
815 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
816 break;
817 display_progress(progress_state, written);
818 }
819
820 /*
821 * Did we write the wrong # entries in the header?
822 * If so, rewrite it like in fast-import
823 */
824 if (pack_to_stdout) {
825 sha1close(f, sha1, CSUM_CLOSE);
826 } else if (nr_written == nr_remaining) {
827 sha1close(f, sha1, CSUM_FSYNC);
828 } else {
829 int fd = sha1close(f, sha1, 0);
830 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
831 nr_written, sha1, offset);
832 close(fd);
833 if (write_bitmap_index) {
834 warning(_(no_split_warning));
835 write_bitmap_index = 0;
836 }
837 }
838
839 if (!pack_to_stdout) {
840 struct stat st;
841 struct strbuf tmpname = STRBUF_INIT;
842
843 /*
844 * Packs are runtime accessed in their mtime
845 * order since newer packs are more likely to contain
846 * younger objects. So if we are creating multiple
847 * packs then we should modify the mtime of later ones
848 * to preserve this property.
849 */
850 if (stat(pack_tmp_name, &st) < 0) {
851 warning_errno("failed to stat %s", pack_tmp_name);
852 } else if (!last_mtime) {
853 last_mtime = st.st_mtime;
854 } else {
855 struct utimbuf utb;
856 utb.actime = st.st_atime;
857 utb.modtime = --last_mtime;
858 if (utime(pack_tmp_name, &utb) < 0)
859 warning_errno("failed utime() on %s", pack_tmp_name);
860 }
861
862 strbuf_addf(&tmpname, "%s-", base_name);
863
864 if (write_bitmap_index) {
865 bitmap_writer_set_checksum(sha1);
866 bitmap_writer_build_type_index(written_list, nr_written);
867 }
868
869 finish_tmp_packfile(&tmpname, pack_tmp_name,
870 written_list, nr_written,
871 &pack_idx_opts, sha1);
872
873 if (write_bitmap_index) {
874 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
875
876 stop_progress(&progress_state);
877
878 bitmap_writer_show_progress(progress);
879 bitmap_writer_reuse_bitmaps(&to_pack);
880 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
881 bitmap_writer_build(&to_pack);
882 bitmap_writer_finish(written_list, nr_written,
883 tmpname.buf, write_bitmap_options);
884 write_bitmap_index = 0;
885 }
886
887 strbuf_release(&tmpname);
888 free(pack_tmp_name);
889 puts(sha1_to_hex(sha1));
890 }
891
892 /* mark written objects as written to previous pack */
893 for (j = 0; j < nr_written; j++) {
894 written_list[j]->offset = (off_t)-1;
895 }
896 nr_remaining -= nr_written;
897 } while (nr_remaining && i < to_pack.nr_objects);
898
899 free(written_list);
900 free(write_order);
901 stop_progress(&progress_state);
902 if (written != nr_result)
903 die("wrote %"PRIu32" objects while expecting %"PRIu32,
904 written, nr_result);
905 }
906
907 static int no_try_delta(const char *path)
908 {
909 static struct attr_check *check;
910
911 if (!check)
912 check = attr_check_initl("delta", NULL);
913 if (git_check_attr(path, check))
914 return 0;
915 if (ATTR_FALSE(check->items[0].value))
916 return 1;
917 return 0;
918 }
919
920 /*
921 * When adding an object, check whether we have already added it
922 * to our packing list. If so, we can skip. However, if we are
923 * being asked to excludei t, but the previous mention was to include
924 * it, make sure to adjust its flags and tweak our numbers accordingly.
925 *
926 * As an optimization, we pass out the index position where we would have
927 * found the item, since that saves us from having to look it up again a
928 * few lines later when we want to add the new entry.
929 */
930 static int have_duplicate_entry(const unsigned char *sha1,
931 int exclude,
932 uint32_t *index_pos)
933 {
934 struct object_entry *entry;
935
936 entry = packlist_find(&to_pack, sha1, index_pos);
937 if (!entry)
938 return 0;
939
940 if (exclude) {
941 if (!entry->preferred_base)
942 nr_result--;
943 entry->preferred_base = 1;
944 }
945
946 return 1;
947 }
948
949 static int want_found_object(int exclude, struct packed_git *p)
950 {
951 if (exclude)
952 return 1;
953 if (incremental)
954 return 0;
955
956 /*
957 * When asked to do --local (do not include an object that appears in a
958 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
959 * an object that appears in a pack marked with .keep), finding a pack
960 * that matches the criteria is sufficient for us to decide to omit it.
961 * However, even if this pack does not satisfy the criteria, we need to
962 * make sure no copy of this object appears in _any_ pack that makes us
963 * to omit the object, so we need to check all the packs.
964 *
965 * We can however first check whether these options can possible matter;
966 * if they do not matter we know we want the object in generated pack.
967 * Otherwise, we signal "-1" at the end to tell the caller that we do
968 * not know either way, and it needs to check more packs.
969 */
970 if (!ignore_packed_keep &&
971 (!local || !have_non_local_packs))
972 return 1;
973
974 if (local && !p->pack_local)
975 return 0;
976 if (ignore_packed_keep && p->pack_local && p->pack_keep)
977 return 0;
978
979 /* we don't know yet; keep looking for more packs */
980 return -1;
981 }
982
983 /*
984 * Check whether we want the object in the pack (e.g., we do not want
985 * objects found in non-local stores if the "--local" option was used).
986 *
987 * If the caller already knows an existing pack it wants to take the object
988 * from, that is passed in *found_pack and *found_offset; otherwise this
989 * function finds if there is any pack that has the object and returns the pack
990 * and its offset in these variables.
991 */
992 static int want_object_in_pack(const unsigned char *sha1,
993 int exclude,
994 struct packed_git **found_pack,
995 off_t *found_offset)
996 {
997 struct mru_entry *entry;
998 int want;
999
1000 if (!exclude && local && has_loose_object_nonlocal(sha1))
1001 return 0;
1002
1003 /*
1004 * If we already know the pack object lives in, start checks from that
1005 * pack - in the usual case when neither --local was given nor .keep files
1006 * are present we will determine the answer right now.
1007 */
1008 if (*found_pack) {
1009 want = want_found_object(exclude, *found_pack);
1010 if (want != -1)
1011 return want;
1012 }
1013
1014 for (entry = packed_git_mru->head; entry; entry = entry->next) {
1015 struct packed_git *p = entry->item;
1016 off_t offset;
1017
1018 if (p == *found_pack)
1019 offset = *found_offset;
1020 else
1021 offset = find_pack_entry_one(sha1, p);
1022
1023 if (offset) {
1024 if (!*found_pack) {
1025 if (!is_pack_valid(p))
1026 continue;
1027 *found_offset = offset;
1028 *found_pack = p;
1029 }
1030 want = want_found_object(exclude, p);
1031 if (!exclude && want > 0)
1032 mru_mark(packed_git_mru, entry);
1033 if (want != -1)
1034 return want;
1035 }
1036 }
1037
1038 return 1;
1039 }
1040
1041 static void create_object_entry(const unsigned char *sha1,
1042 enum object_type type,
1043 uint32_t hash,
1044 int exclude,
1045 int no_try_delta,
1046 uint32_t index_pos,
1047 struct packed_git *found_pack,
1048 off_t found_offset)
1049 {
1050 struct object_entry *entry;
1051
1052 entry = packlist_alloc(&to_pack, sha1, index_pos);
1053 entry->hash = hash;
1054 if (type)
1055 entry->type = type;
1056 if (exclude)
1057 entry->preferred_base = 1;
1058 else
1059 nr_result++;
1060 if (found_pack) {
1061 entry->in_pack = found_pack;
1062 entry->in_pack_offset = found_offset;
1063 }
1064
1065 entry->no_try_delta = no_try_delta;
1066 }
1067
1068 static const char no_closure_warning[] = N_(
1069 "disabling bitmap writing, as some objects are not being packed"
1070 );
1071
1072 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1073 const char *name, int exclude)
1074 {
1075 struct packed_git *found_pack = NULL;
1076 off_t found_offset = 0;
1077 uint32_t index_pos;
1078
1079 if (have_duplicate_entry(sha1, exclude, &index_pos))
1080 return 0;
1081
1082 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1083 /* The pack is missing an object, so it will not have closure */
1084 if (write_bitmap_index) {
1085 warning(_(no_closure_warning));
1086 write_bitmap_index = 0;
1087 }
1088 return 0;
1089 }
1090
1091 create_object_entry(sha1, type, pack_name_hash(name),
1092 exclude, name && no_try_delta(name),
1093 index_pos, found_pack, found_offset);
1094
1095 display_progress(progress_state, nr_result);
1096 return 1;
1097 }
1098
1099 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1100 enum object_type type,
1101 int flags, uint32_t name_hash,
1102 struct packed_git *pack, off_t offset)
1103 {
1104 uint32_t index_pos;
1105
1106 if (have_duplicate_entry(sha1, 0, &index_pos))
1107 return 0;
1108
1109 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1110 return 0;
1111
1112 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1113
1114 display_progress(progress_state, nr_result);
1115 return 1;
1116 }
1117
1118 struct pbase_tree_cache {
1119 unsigned char sha1[20];
1120 int ref;
1121 int temporary;
1122 void *tree_data;
1123 unsigned long tree_size;
1124 };
1125
1126 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1127 static int pbase_tree_cache_ix(const unsigned char *sha1)
1128 {
1129 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1130 }
1131 static int pbase_tree_cache_ix_incr(int ix)
1132 {
1133 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1134 }
1135
1136 static struct pbase_tree {
1137 struct pbase_tree *next;
1138 /* This is a phony "cache" entry; we are not
1139 * going to evict it or find it through _get()
1140 * mechanism -- this is for the toplevel node that
1141 * would almost always change with any commit.
1142 */
1143 struct pbase_tree_cache pcache;
1144 } *pbase_tree;
1145
1146 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1147 {
1148 struct pbase_tree_cache *ent, *nent;
1149 void *data;
1150 unsigned long size;
1151 enum object_type type;
1152 int neigh;
1153 int my_ix = pbase_tree_cache_ix(sha1);
1154 int available_ix = -1;
1155
1156 /* pbase-tree-cache acts as a limited hashtable.
1157 * your object will be found at your index or within a few
1158 * slots after that slot if it is cached.
1159 */
1160 for (neigh = 0; neigh < 8; neigh++) {
1161 ent = pbase_tree_cache[my_ix];
1162 if (ent && !hashcmp(ent->sha1, sha1)) {
1163 ent->ref++;
1164 return ent;
1165 }
1166 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1167 ((0 <= available_ix) &&
1168 (!ent && pbase_tree_cache[available_ix])))
1169 available_ix = my_ix;
1170 if (!ent)
1171 break;
1172 my_ix = pbase_tree_cache_ix_incr(my_ix);
1173 }
1174
1175 /* Did not find one. Either we got a bogus request or
1176 * we need to read and perhaps cache.
1177 */
1178 data = read_sha1_file(sha1, &type, &size);
1179 if (!data)
1180 return NULL;
1181 if (type != OBJ_TREE) {
1182 free(data);
1183 return NULL;
1184 }
1185
1186 /* We need to either cache or return a throwaway copy */
1187
1188 if (available_ix < 0)
1189 ent = NULL;
1190 else {
1191 ent = pbase_tree_cache[available_ix];
1192 my_ix = available_ix;
1193 }
1194
1195 if (!ent) {
1196 nent = xmalloc(sizeof(*nent));
1197 nent->temporary = (available_ix < 0);
1198 }
1199 else {
1200 /* evict and reuse */
1201 free(ent->tree_data);
1202 nent = ent;
1203 }
1204 hashcpy(nent->sha1, sha1);
1205 nent->tree_data = data;
1206 nent->tree_size = size;
1207 nent->ref = 1;
1208 if (!nent->temporary)
1209 pbase_tree_cache[my_ix] = nent;
1210 return nent;
1211 }
1212
1213 static void pbase_tree_put(struct pbase_tree_cache *cache)
1214 {
1215 if (!cache->temporary) {
1216 cache->ref--;
1217 return;
1218 }
1219 free(cache->tree_data);
1220 free(cache);
1221 }
1222
1223 static int name_cmp_len(const char *name)
1224 {
1225 int i;
1226 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1227 ;
1228 return i;
1229 }
1230
1231 static void add_pbase_object(struct tree_desc *tree,
1232 const char *name,
1233 int cmplen,
1234 const char *fullname)
1235 {
1236 struct name_entry entry;
1237 int cmp;
1238
1239 while (tree_entry(tree,&entry)) {
1240 if (S_ISGITLINK(entry.mode))
1241 continue;
1242 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1243 memcmp(name, entry.path, cmplen);
1244 if (cmp > 0)
1245 continue;
1246 if (cmp < 0)
1247 return;
1248 if (name[cmplen] != '/') {
1249 add_object_entry(entry.oid->hash,
1250 object_type(entry.mode),
1251 fullname, 1);
1252 return;
1253 }
1254 if (S_ISDIR(entry.mode)) {
1255 struct tree_desc sub;
1256 struct pbase_tree_cache *tree;
1257 const char *down = name+cmplen+1;
1258 int downlen = name_cmp_len(down);
1259
1260 tree = pbase_tree_get(entry.oid->hash);
1261 if (!tree)
1262 return;
1263 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1264
1265 add_pbase_object(&sub, down, downlen, fullname);
1266 pbase_tree_put(tree);
1267 }
1268 }
1269 }
1270
1271 static unsigned *done_pbase_paths;
1272 static int done_pbase_paths_num;
1273 static int done_pbase_paths_alloc;
1274 static int done_pbase_path_pos(unsigned hash)
1275 {
1276 int lo = 0;
1277 int hi = done_pbase_paths_num;
1278 while (lo < hi) {
1279 int mi = (hi + lo) / 2;
1280 if (done_pbase_paths[mi] == hash)
1281 return mi;
1282 if (done_pbase_paths[mi] < hash)
1283 hi = mi;
1284 else
1285 lo = mi + 1;
1286 }
1287 return -lo-1;
1288 }
1289
1290 static int check_pbase_path(unsigned hash)
1291 {
1292 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1293 if (0 <= pos)
1294 return 1;
1295 pos = -pos - 1;
1296 ALLOC_GROW(done_pbase_paths,
1297 done_pbase_paths_num + 1,
1298 done_pbase_paths_alloc);
1299 done_pbase_paths_num++;
1300 if (pos < done_pbase_paths_num)
1301 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1302 done_pbase_paths_num - pos - 1);
1303 done_pbase_paths[pos] = hash;
1304 return 0;
1305 }
1306
1307 static void add_preferred_base_object(const char *name)
1308 {
1309 struct pbase_tree *it;
1310 int cmplen;
1311 unsigned hash = pack_name_hash(name);
1312
1313 if (!num_preferred_base || check_pbase_path(hash))
1314 return;
1315
1316 cmplen = name_cmp_len(name);
1317 for (it = pbase_tree; it; it = it->next) {
1318 if (cmplen == 0) {
1319 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1320 }
1321 else {
1322 struct tree_desc tree;
1323 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1324 add_pbase_object(&tree, name, cmplen, name);
1325 }
1326 }
1327 }
1328
1329 static void add_preferred_base(unsigned char *sha1)
1330 {
1331 struct pbase_tree *it;
1332 void *data;
1333 unsigned long size;
1334 unsigned char tree_sha1[20];
1335
1336 if (window <= num_preferred_base++)
1337 return;
1338
1339 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1340 if (!data)
1341 return;
1342
1343 for (it = pbase_tree; it; it = it->next) {
1344 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1345 free(data);
1346 return;
1347 }
1348 }
1349
1350 it = xcalloc(1, sizeof(*it));
1351 it->next = pbase_tree;
1352 pbase_tree = it;
1353
1354 hashcpy(it->pcache.sha1, tree_sha1);
1355 it->pcache.tree_data = data;
1356 it->pcache.tree_size = size;
1357 }
1358
1359 static void cleanup_preferred_base(void)
1360 {
1361 struct pbase_tree *it;
1362 unsigned i;
1363
1364 it = pbase_tree;
1365 pbase_tree = NULL;
1366 while (it) {
1367 struct pbase_tree *this = it;
1368 it = this->next;
1369 free(this->pcache.tree_data);
1370 free(this);
1371 }
1372
1373 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1374 if (!pbase_tree_cache[i])
1375 continue;
1376 free(pbase_tree_cache[i]->tree_data);
1377 FREE_AND_NULL(pbase_tree_cache[i]);
1378 }
1379
1380 FREE_AND_NULL(done_pbase_paths);
1381 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1382 }
1383
1384 static void check_object(struct object_entry *entry)
1385 {
1386 if (entry->in_pack) {
1387 struct packed_git *p = entry->in_pack;
1388 struct pack_window *w_curs = NULL;
1389 const unsigned char *base_ref = NULL;
1390 struct object_entry *base_entry;
1391 unsigned long used, used_0;
1392 unsigned long avail;
1393 off_t ofs;
1394 unsigned char *buf, c;
1395
1396 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1397
1398 /*
1399 * We want in_pack_type even if we do not reuse delta
1400 * since non-delta representations could still be reused.
1401 */
1402 used = unpack_object_header_buffer(buf, avail,
1403 &entry->in_pack_type,
1404 &entry->size);
1405 if (used == 0)
1406 goto give_up;
1407
1408 /*
1409 * Determine if this is a delta and if so whether we can
1410 * reuse it or not. Otherwise let's find out as cheaply as
1411 * possible what the actual type and size for this object is.
1412 */
1413 switch (entry->in_pack_type) {
1414 default:
1415 /* Not a delta hence we've already got all we need. */
1416 entry->type = entry->in_pack_type;
1417 entry->in_pack_header_size = used;
1418 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1419 goto give_up;
1420 unuse_pack(&w_curs);
1421 return;
1422 case OBJ_REF_DELTA:
1423 if (reuse_delta && !entry->preferred_base)
1424 base_ref = use_pack(p, &w_curs,
1425 entry->in_pack_offset + used, NULL);
1426 entry->in_pack_header_size = used + 20;
1427 break;
1428 case OBJ_OFS_DELTA:
1429 buf = use_pack(p, &w_curs,
1430 entry->in_pack_offset + used, NULL);
1431 used_0 = 0;
1432 c = buf[used_0++];
1433 ofs = c & 127;
1434 while (c & 128) {
1435 ofs += 1;
1436 if (!ofs || MSB(ofs, 7)) {
1437 error("delta base offset overflow in pack for %s",
1438 oid_to_hex(&entry->idx.oid));
1439 goto give_up;
1440 }
1441 c = buf[used_0++];
1442 ofs = (ofs << 7) + (c & 127);
1443 }
1444 ofs = entry->in_pack_offset - ofs;
1445 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1446 error("delta base offset out of bound for %s",
1447 oid_to_hex(&entry->idx.oid));
1448 goto give_up;
1449 }
1450 if (reuse_delta && !entry->preferred_base) {
1451 struct revindex_entry *revidx;
1452 revidx = find_pack_revindex(p, ofs);
1453 if (!revidx)
1454 goto give_up;
1455 base_ref = nth_packed_object_sha1(p, revidx->nr);
1456 }
1457 entry->in_pack_header_size = used + used_0;
1458 break;
1459 }
1460
1461 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1462 /*
1463 * If base_ref was set above that means we wish to
1464 * reuse delta data, and we even found that base
1465 * in the list of objects we want to pack. Goodie!
1466 *
1467 * Depth value does not matter - find_deltas() will
1468 * never consider reused delta as the base object to
1469 * deltify other objects against, in order to avoid
1470 * circular deltas.
1471 */
1472 entry->type = entry->in_pack_type;
1473 entry->delta = base_entry;
1474 entry->delta_size = entry->size;
1475 entry->delta_sibling = base_entry->delta_child;
1476 base_entry->delta_child = entry;
1477 unuse_pack(&w_curs);
1478 return;
1479 }
1480
1481 if (entry->type) {
1482 /*
1483 * This must be a delta and we already know what the
1484 * final object type is. Let's extract the actual
1485 * object size from the delta header.
1486 */
1487 entry->size = get_size_from_delta(p, &w_curs,
1488 entry->in_pack_offset + entry->in_pack_header_size);
1489 if (entry->size == 0)
1490 goto give_up;
1491 unuse_pack(&w_curs);
1492 return;
1493 }
1494
1495 /*
1496 * No choice but to fall back to the recursive delta walk
1497 * with sha1_object_info() to find about the object type
1498 * at this point...
1499 */
1500 give_up:
1501 unuse_pack(&w_curs);
1502 }
1503
1504 entry->type = sha1_object_info(entry->idx.oid.hash, &entry->size);
1505 /*
1506 * The error condition is checked in prepare_pack(). This is
1507 * to permit a missing preferred base object to be ignored
1508 * as a preferred base. Doing so can result in a larger
1509 * pack file, but the transfer will still take place.
1510 */
1511 }
1512
1513 static int pack_offset_sort(const void *_a, const void *_b)
1514 {
1515 const struct object_entry *a = *(struct object_entry **)_a;
1516 const struct object_entry *b = *(struct object_entry **)_b;
1517
1518 /* avoid filesystem trashing with loose objects */
1519 if (!a->in_pack && !b->in_pack)
1520 return oidcmp(&a->idx.oid, &b->idx.oid);
1521
1522 if (a->in_pack < b->in_pack)
1523 return -1;
1524 if (a->in_pack > b->in_pack)
1525 return 1;
1526 return a->in_pack_offset < b->in_pack_offset ? -1 :
1527 (a->in_pack_offset > b->in_pack_offset);
1528 }
1529
1530 /*
1531 * Drop an on-disk delta we were planning to reuse. Naively, this would
1532 * just involve blanking out the "delta" field, but we have to deal
1533 * with some extra book-keeping:
1534 *
1535 * 1. Removing ourselves from the delta_sibling linked list.
1536 *
1537 * 2. Updating our size/type to the non-delta representation. These were
1538 * either not recorded initially (size) or overwritten with the delta type
1539 * (type) when check_object() decided to reuse the delta.
1540 *
1541 * 3. Resetting our delta depth, as we are now a base object.
1542 */
1543 static void drop_reused_delta(struct object_entry *entry)
1544 {
1545 struct object_entry **p = &entry->delta->delta_child;
1546 struct object_info oi = OBJECT_INFO_INIT;
1547
1548 while (*p) {
1549 if (*p == entry)
1550 *p = (*p)->delta_sibling;
1551 else
1552 p = &(*p)->delta_sibling;
1553 }
1554 entry->delta = NULL;
1555 entry->depth = 0;
1556
1557 oi.sizep = &entry->size;
1558 oi.typep = &entry->type;
1559 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1560 /*
1561 * We failed to get the info from this pack for some reason;
1562 * fall back to sha1_object_info, which may find another copy.
1563 * And if that fails, the error will be recorded in entry->type
1564 * and dealt with in prepare_pack().
1565 */
1566 entry->type = sha1_object_info(entry->idx.oid.hash,
1567 &entry->size);
1568 }
1569 }
1570
1571 /*
1572 * Follow the chain of deltas from this entry onward, throwing away any links
1573 * that cause us to hit a cycle (as determined by the DFS state flags in
1574 * the entries).
1575 *
1576 * We also detect too-long reused chains that would violate our --depth
1577 * limit.
1578 */
1579 static void break_delta_chains(struct object_entry *entry)
1580 {
1581 /*
1582 * The actual depth of each object we will write is stored as an int,
1583 * as it cannot exceed our int "depth" limit. But before we break
1584 * changes based no that limit, we may potentially go as deep as the
1585 * number of objects, which is elsewhere bounded to a uint32_t.
1586 */
1587 uint32_t total_depth;
1588 struct object_entry *cur, *next;
1589
1590 for (cur = entry, total_depth = 0;
1591 cur;
1592 cur = cur->delta, total_depth++) {
1593 if (cur->dfs_state == DFS_DONE) {
1594 /*
1595 * We've already seen this object and know it isn't
1596 * part of a cycle. We do need to append its depth
1597 * to our count.
1598 */
1599 total_depth += cur->depth;
1600 break;
1601 }
1602
1603 /*
1604 * We break cycles before looping, so an ACTIVE state (or any
1605 * other cruft which made its way into the state variable)
1606 * is a bug.
1607 */
1608 if (cur->dfs_state != DFS_NONE)
1609 die("BUG: confusing delta dfs state in first pass: %d",
1610 cur->dfs_state);
1611
1612 /*
1613 * Now we know this is the first time we've seen the object. If
1614 * it's not a delta, we're done traversing, but we'll mark it
1615 * done to save time on future traversals.
1616 */
1617 if (!cur->delta) {
1618 cur->dfs_state = DFS_DONE;
1619 break;
1620 }
1621
1622 /*
1623 * Mark ourselves as active and see if the next step causes
1624 * us to cycle to another active object. It's important to do
1625 * this _before_ we loop, because it impacts where we make the
1626 * cut, and thus how our total_depth counter works.
1627 * E.g., We may see a partial loop like:
1628 *
1629 * A -> B -> C -> D -> B
1630 *
1631 * Cutting B->C breaks the cycle. But now the depth of A is
1632 * only 1, and our total_depth counter is at 3. The size of the
1633 * error is always one less than the size of the cycle we
1634 * broke. Commits C and D were "lost" from A's chain.
1635 *
1636 * If we instead cut D->B, then the depth of A is correct at 3.
1637 * We keep all commits in the chain that we examined.
1638 */
1639 cur->dfs_state = DFS_ACTIVE;
1640 if (cur->delta->dfs_state == DFS_ACTIVE) {
1641 drop_reused_delta(cur);
1642 cur->dfs_state = DFS_DONE;
1643 break;
1644 }
1645 }
1646
1647 /*
1648 * And now that we've gone all the way to the bottom of the chain, we
1649 * need to clear the active flags and set the depth fields as
1650 * appropriate. Unlike the loop above, which can quit when it drops a
1651 * delta, we need to keep going to look for more depth cuts. So we need
1652 * an extra "next" pointer to keep going after we reset cur->delta.
1653 */
1654 for (cur = entry; cur; cur = next) {
1655 next = cur->delta;
1656
1657 /*
1658 * We should have a chain of zero or more ACTIVE states down to
1659 * a final DONE. We can quit after the DONE, because either it
1660 * has no bases, or we've already handled them in a previous
1661 * call.
1662 */
1663 if (cur->dfs_state == DFS_DONE)
1664 break;
1665 else if (cur->dfs_state != DFS_ACTIVE)
1666 die("BUG: confusing delta dfs state in second pass: %d",
1667 cur->dfs_state);
1668
1669 /*
1670 * If the total_depth is more than depth, then we need to snip
1671 * the chain into two or more smaller chains that don't exceed
1672 * the maximum depth. Most of the resulting chains will contain
1673 * (depth + 1) entries (i.e., depth deltas plus one base), and
1674 * the last chain (i.e., the one containing entry) will contain
1675 * whatever entries are left over, namely
1676 * (total_depth % (depth + 1)) of them.
1677 *
1678 * Since we are iterating towards decreasing depth, we need to
1679 * decrement total_depth as we go, and we need to write to the
1680 * entry what its final depth will be after all of the
1681 * snipping. Since we're snipping into chains of length (depth
1682 * + 1) entries, the final depth of an entry will be its
1683 * original depth modulo (depth + 1). Any time we encounter an
1684 * entry whose final depth is supposed to be zero, we snip it
1685 * from its delta base, thereby making it so.
1686 */
1687 cur->depth = (total_depth--) % (depth + 1);
1688 if (!cur->depth)
1689 drop_reused_delta(cur);
1690
1691 cur->dfs_state = DFS_DONE;
1692 }
1693 }
1694
1695 static void get_object_details(void)
1696 {
1697 uint32_t i;
1698 struct object_entry **sorted_by_offset;
1699
1700 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1701 for (i = 0; i < to_pack.nr_objects; i++)
1702 sorted_by_offset[i] = to_pack.objects + i;
1703 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1704
1705 for (i = 0; i < to_pack.nr_objects; i++) {
1706 struct object_entry *entry = sorted_by_offset[i];
1707 check_object(entry);
1708 if (big_file_threshold < entry->size)
1709 entry->no_try_delta = 1;
1710 }
1711
1712 /*
1713 * This must happen in a second pass, since we rely on the delta
1714 * information for the whole list being completed.
1715 */
1716 for (i = 0; i < to_pack.nr_objects; i++)
1717 break_delta_chains(&to_pack.objects[i]);
1718
1719 free(sorted_by_offset);
1720 }
1721
1722 /*
1723 * We search for deltas in a list sorted by type, by filename hash, and then
1724 * by size, so that we see progressively smaller and smaller files.
1725 * That's because we prefer deltas to be from the bigger file
1726 * to the smaller -- deletes are potentially cheaper, but perhaps
1727 * more importantly, the bigger file is likely the more recent
1728 * one. The deepest deltas are therefore the oldest objects which are
1729 * less susceptible to be accessed often.
1730 */
1731 static int type_size_sort(const void *_a, const void *_b)
1732 {
1733 const struct object_entry *a = *(struct object_entry **)_a;
1734 const struct object_entry *b = *(struct object_entry **)_b;
1735
1736 if (a->type > b->type)
1737 return -1;
1738 if (a->type < b->type)
1739 return 1;
1740 if (a->hash > b->hash)
1741 return -1;
1742 if (a->hash < b->hash)
1743 return 1;
1744 if (a->preferred_base > b->preferred_base)
1745 return -1;
1746 if (a->preferred_base < b->preferred_base)
1747 return 1;
1748 if (a->size > b->size)
1749 return -1;
1750 if (a->size < b->size)
1751 return 1;
1752 return a < b ? -1 : (a > b); /* newest first */
1753 }
1754
1755 struct unpacked {
1756 struct object_entry *entry;
1757 void *data;
1758 struct delta_index *index;
1759 unsigned depth;
1760 };
1761
1762 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1763 unsigned long delta_size)
1764 {
1765 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1766 return 0;
1767
1768 if (delta_size < cache_max_small_delta_size)
1769 return 1;
1770
1771 /* cache delta, if objects are large enough compared to delta size */
1772 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1773 return 1;
1774
1775 return 0;
1776 }
1777
1778 #ifndef NO_PTHREADS
1779
1780 static pthread_mutex_t read_mutex;
1781 #define read_lock() pthread_mutex_lock(&read_mutex)
1782 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1783
1784 static pthread_mutex_t cache_mutex;
1785 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1786 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1787
1788 static pthread_mutex_t progress_mutex;
1789 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1790 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1791
1792 #else
1793
1794 #define read_lock() (void)0
1795 #define read_unlock() (void)0
1796 #define cache_lock() (void)0
1797 #define cache_unlock() (void)0
1798 #define progress_lock() (void)0
1799 #define progress_unlock() (void)0
1800
1801 #endif
1802
1803 static int try_delta(struct unpacked *trg, struct unpacked *src,
1804 unsigned max_depth, unsigned long *mem_usage)
1805 {
1806 struct object_entry *trg_entry = trg->entry;
1807 struct object_entry *src_entry = src->entry;
1808 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1809 unsigned ref_depth;
1810 enum object_type type;
1811 void *delta_buf;
1812
1813 /* Don't bother doing diffs between different types */
1814 if (trg_entry->type != src_entry->type)
1815 return -1;
1816
1817 /*
1818 * We do not bother to try a delta that we discarded on an
1819 * earlier try, but only when reusing delta data. Note that
1820 * src_entry that is marked as the preferred_base should always
1821 * be considered, as even if we produce a suboptimal delta against
1822 * it, we will still save the transfer cost, as we already know
1823 * the other side has it and we won't send src_entry at all.
1824 */
1825 if (reuse_delta && trg_entry->in_pack &&
1826 trg_entry->in_pack == src_entry->in_pack &&
1827 !src_entry->preferred_base &&
1828 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1829 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1830 return 0;
1831
1832 /* Let's not bust the allowed depth. */
1833 if (src->depth >= max_depth)
1834 return 0;
1835
1836 /* Now some size filtering heuristics. */
1837 trg_size = trg_entry->size;
1838 if (!trg_entry->delta) {
1839 max_size = trg_size/2 - 20;
1840 ref_depth = 1;
1841 } else {
1842 max_size = trg_entry->delta_size;
1843 ref_depth = trg->depth;
1844 }
1845 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1846 (max_depth - ref_depth + 1);
1847 if (max_size == 0)
1848 return 0;
1849 src_size = src_entry->size;
1850 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1851 if (sizediff >= max_size)
1852 return 0;
1853 if (trg_size < src_size / 32)
1854 return 0;
1855
1856 /* Load data if not already done */
1857 if (!trg->data) {
1858 read_lock();
1859 trg->data = read_sha1_file(trg_entry->idx.oid.hash, &type,
1860 &sz);
1861 read_unlock();
1862 if (!trg->data)
1863 die("object %s cannot be read",
1864 oid_to_hex(&trg_entry->idx.oid));
1865 if (sz != trg_size)
1866 die("object %s inconsistent object length (%lu vs %lu)",
1867 oid_to_hex(&trg_entry->idx.oid), sz,
1868 trg_size);
1869 *mem_usage += sz;
1870 }
1871 if (!src->data) {
1872 read_lock();
1873 src->data = read_sha1_file(src_entry->idx.oid.hash, &type,
1874 &sz);
1875 read_unlock();
1876 if (!src->data) {
1877 if (src_entry->preferred_base) {
1878 static int warned = 0;
1879 if (!warned++)
1880 warning("object %s cannot be read",
1881 oid_to_hex(&src_entry->idx.oid));
1882 /*
1883 * Those objects are not included in the
1884 * resulting pack. Be resilient and ignore
1885 * them if they can't be read, in case the
1886 * pack could be created nevertheless.
1887 */
1888 return 0;
1889 }
1890 die("object %s cannot be read",
1891 oid_to_hex(&src_entry->idx.oid));
1892 }
1893 if (sz != src_size)
1894 die("object %s inconsistent object length (%lu vs %lu)",
1895 oid_to_hex(&src_entry->idx.oid), sz,
1896 src_size);
1897 *mem_usage += sz;
1898 }
1899 if (!src->index) {
1900 src->index = create_delta_index(src->data, src_size);
1901 if (!src->index) {
1902 static int warned = 0;
1903 if (!warned++)
1904 warning("suboptimal pack - out of memory");
1905 return 0;
1906 }
1907 *mem_usage += sizeof_delta_index(src->index);
1908 }
1909
1910 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1911 if (!delta_buf)
1912 return 0;
1913
1914 if (trg_entry->delta) {
1915 /* Prefer only shallower same-sized deltas. */
1916 if (delta_size == trg_entry->delta_size &&
1917 src->depth + 1 >= trg->depth) {
1918 free(delta_buf);
1919 return 0;
1920 }
1921 }
1922
1923 /*
1924 * Handle memory allocation outside of the cache
1925 * accounting lock. Compiler will optimize the strangeness
1926 * away when NO_PTHREADS is defined.
1927 */
1928 free(trg_entry->delta_data);
1929 cache_lock();
1930 if (trg_entry->delta_data) {
1931 delta_cache_size -= trg_entry->delta_size;
1932 trg_entry->delta_data = NULL;
1933 }
1934 if (delta_cacheable(src_size, trg_size, delta_size)) {
1935 delta_cache_size += delta_size;
1936 cache_unlock();
1937 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1938 } else {
1939 cache_unlock();
1940 free(delta_buf);
1941 }
1942
1943 trg_entry->delta = src_entry;
1944 trg_entry->delta_size = delta_size;
1945 trg->depth = src->depth + 1;
1946
1947 return 1;
1948 }
1949
1950 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1951 {
1952 struct object_entry *child = me->delta_child;
1953 unsigned int m = n;
1954 while (child) {
1955 unsigned int c = check_delta_limit(child, n + 1);
1956 if (m < c)
1957 m = c;
1958 child = child->delta_sibling;
1959 }
1960 return m;
1961 }
1962
1963 static unsigned long free_unpacked(struct unpacked *n)
1964 {
1965 unsigned long freed_mem = sizeof_delta_index(n->index);
1966 free_delta_index(n->index);
1967 n->index = NULL;
1968 if (n->data) {
1969 freed_mem += n->entry->size;
1970 FREE_AND_NULL(n->data);
1971 }
1972 n->entry = NULL;
1973 n->depth = 0;
1974 return freed_mem;
1975 }
1976
1977 static void find_deltas(struct object_entry **list, unsigned *list_size,
1978 int window, int depth, unsigned *processed)
1979 {
1980 uint32_t i, idx = 0, count = 0;
1981 struct unpacked *array;
1982 unsigned long mem_usage = 0;
1983
1984 array = xcalloc(window, sizeof(struct unpacked));
1985
1986 for (;;) {
1987 struct object_entry *entry;
1988 struct unpacked *n = array + idx;
1989 int j, max_depth, best_base = -1;
1990
1991 progress_lock();
1992 if (!*list_size) {
1993 progress_unlock();
1994 break;
1995 }
1996 entry = *list++;
1997 (*list_size)--;
1998 if (!entry->preferred_base) {
1999 (*processed)++;
2000 display_progress(progress_state, *processed);
2001 }
2002 progress_unlock();
2003
2004 mem_usage -= free_unpacked(n);
2005 n->entry = entry;
2006
2007 while (window_memory_limit &&
2008 mem_usage > window_memory_limit &&
2009 count > 1) {
2010 uint32_t tail = (idx + window - count) % window;
2011 mem_usage -= free_unpacked(array + tail);
2012 count--;
2013 }
2014
2015 /* We do not compute delta to *create* objects we are not
2016 * going to pack.
2017 */
2018 if (entry->preferred_base)
2019 goto next;
2020
2021 /*
2022 * If the current object is at pack edge, take the depth the
2023 * objects that depend on the current object into account
2024 * otherwise they would become too deep.
2025 */
2026 max_depth = depth;
2027 if (entry->delta_child) {
2028 max_depth -= check_delta_limit(entry, 0);
2029 if (max_depth <= 0)
2030 goto next;
2031 }
2032
2033 j = window;
2034 while (--j > 0) {
2035 int ret;
2036 uint32_t other_idx = idx + j;
2037 struct unpacked *m;
2038 if (other_idx >= window)
2039 other_idx -= window;
2040 m = array + other_idx;
2041 if (!m->entry)
2042 break;
2043 ret = try_delta(n, m, max_depth, &mem_usage);
2044 if (ret < 0)
2045 break;
2046 else if (ret > 0)
2047 best_base = other_idx;
2048 }
2049
2050 /*
2051 * If we decided to cache the delta data, then it is best
2052 * to compress it right away. First because we have to do
2053 * it anyway, and doing it here while we're threaded will
2054 * save a lot of time in the non threaded write phase,
2055 * as well as allow for caching more deltas within
2056 * the same cache size limit.
2057 * ...
2058 * But only if not writing to stdout, since in that case
2059 * the network is most likely throttling writes anyway,
2060 * and therefore it is best to go to the write phase ASAP
2061 * instead, as we can afford spending more time compressing
2062 * between writes at that moment.
2063 */
2064 if (entry->delta_data && !pack_to_stdout) {
2065 entry->z_delta_size = do_compress(&entry->delta_data,
2066 entry->delta_size);
2067 cache_lock();
2068 delta_cache_size -= entry->delta_size;
2069 delta_cache_size += entry->z_delta_size;
2070 cache_unlock();
2071 }
2072
2073 /* if we made n a delta, and if n is already at max
2074 * depth, leaving it in the window is pointless. we
2075 * should evict it first.
2076 */
2077 if (entry->delta && max_depth <= n->depth)
2078 continue;
2079
2080 /*
2081 * Move the best delta base up in the window, after the
2082 * currently deltified object, to keep it longer. It will
2083 * be the first base object to be attempted next.
2084 */
2085 if (entry->delta) {
2086 struct unpacked swap = array[best_base];
2087 int dist = (window + idx - best_base) % window;
2088 int dst = best_base;
2089 while (dist--) {
2090 int src = (dst + 1) % window;
2091 array[dst] = array[src];
2092 dst = src;
2093 }
2094 array[dst] = swap;
2095 }
2096
2097 next:
2098 idx++;
2099 if (count + 1 < window)
2100 count++;
2101 if (idx >= window)
2102 idx = 0;
2103 }
2104
2105 for (i = 0; i < window; ++i) {
2106 free_delta_index(array[i].index);
2107 free(array[i].data);
2108 }
2109 free(array);
2110 }
2111
2112 #ifndef NO_PTHREADS
2113
2114 static void try_to_free_from_threads(size_t size)
2115 {
2116 read_lock();
2117 release_pack_memory(size);
2118 read_unlock();
2119 }
2120
2121 static try_to_free_t old_try_to_free_routine;
2122
2123 /*
2124 * The main thread waits on the condition that (at least) one of the workers
2125 * has stopped working (which is indicated in the .working member of
2126 * struct thread_params).
2127 * When a work thread has completed its work, it sets .working to 0 and
2128 * signals the main thread and waits on the condition that .data_ready
2129 * becomes 1.
2130 */
2131
2132 struct thread_params {
2133 pthread_t thread;
2134 struct object_entry **list;
2135 unsigned list_size;
2136 unsigned remaining;
2137 int window;
2138 int depth;
2139 int working;
2140 int data_ready;
2141 pthread_mutex_t mutex;
2142 pthread_cond_t cond;
2143 unsigned *processed;
2144 };
2145
2146 static pthread_cond_t progress_cond;
2147
2148 /*
2149 * Mutex and conditional variable can't be statically-initialized on Windows.
2150 */
2151 static void init_threaded_search(void)
2152 {
2153 init_recursive_mutex(&read_mutex);
2154 pthread_mutex_init(&cache_mutex, NULL);
2155 pthread_mutex_init(&progress_mutex, NULL);
2156 pthread_cond_init(&progress_cond, NULL);
2157 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2158 }
2159
2160 static void cleanup_threaded_search(void)
2161 {
2162 set_try_to_free_routine(old_try_to_free_routine);
2163 pthread_cond_destroy(&progress_cond);
2164 pthread_mutex_destroy(&read_mutex);
2165 pthread_mutex_destroy(&cache_mutex);
2166 pthread_mutex_destroy(&progress_mutex);
2167 }
2168
2169 static void *threaded_find_deltas(void *arg)
2170 {
2171 struct thread_params *me = arg;
2172
2173 while (me->remaining) {
2174 find_deltas(me->list, &me->remaining,
2175 me->window, me->depth, me->processed);
2176
2177 progress_lock();
2178 me->working = 0;
2179 pthread_cond_signal(&progress_cond);
2180 progress_unlock();
2181
2182 /*
2183 * We must not set ->data_ready before we wait on the
2184 * condition because the main thread may have set it to 1
2185 * before we get here. In order to be sure that new
2186 * work is available if we see 1 in ->data_ready, it
2187 * was initialized to 0 before this thread was spawned
2188 * and we reset it to 0 right away.
2189 */
2190 pthread_mutex_lock(&me->mutex);
2191 while (!me->data_ready)
2192 pthread_cond_wait(&me->cond, &me->mutex);
2193 me->data_ready = 0;
2194 pthread_mutex_unlock(&me->mutex);
2195 }
2196 /* leave ->working 1 so that this doesn't get more work assigned */
2197 return NULL;
2198 }
2199
2200 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2201 int window, int depth, unsigned *processed)
2202 {
2203 struct thread_params *p;
2204 int i, ret, active_threads = 0;
2205
2206 init_threaded_search();
2207
2208 if (delta_search_threads <= 1) {
2209 find_deltas(list, &list_size, window, depth, processed);
2210 cleanup_threaded_search();
2211 return;
2212 }
2213 if (progress > pack_to_stdout)
2214 fprintf(stderr, "Delta compression using up to %d threads.\n",
2215 delta_search_threads);
2216 p = xcalloc(delta_search_threads, sizeof(*p));
2217
2218 /* Partition the work amongst work threads. */
2219 for (i = 0; i < delta_search_threads; i++) {
2220 unsigned sub_size = list_size / (delta_search_threads - i);
2221
2222 /* don't use too small segments or no deltas will be found */
2223 if (sub_size < 2*window && i+1 < delta_search_threads)
2224 sub_size = 0;
2225
2226 p[i].window = window;
2227 p[i].depth = depth;
2228 p[i].processed = processed;
2229 p[i].working = 1;
2230 p[i].data_ready = 0;
2231
2232 /* try to split chunks on "path" boundaries */
2233 while (sub_size && sub_size < list_size &&
2234 list[sub_size]->hash &&
2235 list[sub_size]->hash == list[sub_size-1]->hash)
2236 sub_size++;
2237
2238 p[i].list = list;
2239 p[i].list_size = sub_size;
2240 p[i].remaining = sub_size;
2241
2242 list += sub_size;
2243 list_size -= sub_size;
2244 }
2245
2246 /* Start work threads. */
2247 for (i = 0; i < delta_search_threads; i++) {
2248 if (!p[i].list_size)
2249 continue;
2250 pthread_mutex_init(&p[i].mutex, NULL);
2251 pthread_cond_init(&p[i].cond, NULL);
2252 ret = pthread_create(&p[i].thread, NULL,
2253 threaded_find_deltas, &p[i]);
2254 if (ret)
2255 die("unable to create thread: %s", strerror(ret));
2256 active_threads++;
2257 }
2258
2259 /*
2260 * Now let's wait for work completion. Each time a thread is done
2261 * with its work, we steal half of the remaining work from the
2262 * thread with the largest number of unprocessed objects and give
2263 * it to that newly idle thread. This ensure good load balancing
2264 * until the remaining object list segments are simply too short
2265 * to be worth splitting anymore.
2266 */
2267 while (active_threads) {
2268 struct thread_params *target = NULL;
2269 struct thread_params *victim = NULL;
2270 unsigned sub_size = 0;
2271
2272 progress_lock();
2273 for (;;) {
2274 for (i = 0; !target && i < delta_search_threads; i++)
2275 if (!p[i].working)
2276 target = &p[i];
2277 if (target)
2278 break;
2279 pthread_cond_wait(&progress_cond, &progress_mutex);
2280 }
2281
2282 for (i = 0; i < delta_search_threads; i++)
2283 if (p[i].remaining > 2*window &&
2284 (!victim || victim->remaining < p[i].remaining))
2285 victim = &p[i];
2286 if (victim) {
2287 sub_size = victim->remaining / 2;
2288 list = victim->list + victim->list_size - sub_size;
2289 while (sub_size && list[0]->hash &&
2290 list[0]->hash == list[-1]->hash) {
2291 list++;
2292 sub_size--;
2293 }
2294 if (!sub_size) {
2295 /*
2296 * It is possible for some "paths" to have
2297 * so many objects that no hash boundary
2298 * might be found. Let's just steal the
2299 * exact half in that case.
2300 */
2301 sub_size = victim->remaining / 2;
2302 list -= sub_size;
2303 }
2304 target->list = list;
2305 victim->list_size -= sub_size;
2306 victim->remaining -= sub_size;
2307 }
2308 target->list_size = sub_size;
2309 target->remaining = sub_size;
2310 target->working = 1;
2311 progress_unlock();
2312
2313 pthread_mutex_lock(&target->mutex);
2314 target->data_ready = 1;
2315 pthread_cond_signal(&target->cond);
2316 pthread_mutex_unlock(&target->mutex);
2317
2318 if (!sub_size) {
2319 pthread_join(target->thread, NULL);
2320 pthread_cond_destroy(&target->cond);
2321 pthread_mutex_destroy(&target->mutex);
2322 active_threads--;
2323 }
2324 }
2325 cleanup_threaded_search();
2326 free(p);
2327 }
2328
2329 #else
2330 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2331 #endif
2332
2333 static void add_tag_chain(const struct object_id *oid)
2334 {
2335 struct tag *tag;
2336
2337 /*
2338 * We catch duplicates already in add_object_entry(), but we'd
2339 * prefer to do this extra check to avoid having to parse the
2340 * tag at all if we already know that it's being packed (e.g., if
2341 * it was included via bitmaps, we would not have parsed it
2342 * previously).
2343 */
2344 if (packlist_find(&to_pack, oid->hash, NULL))
2345 return;
2346
2347 tag = lookup_tag(oid);
2348 while (1) {
2349 if (!tag || parse_tag(tag) || !tag->tagged)
2350 die("unable to pack objects reachable from tag %s",
2351 oid_to_hex(oid));
2352
2353 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2354
2355 if (tag->tagged->type != OBJ_TAG)
2356 return;
2357
2358 tag = (struct tag *)tag->tagged;
2359 }
2360 }
2361
2362 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2363 {
2364 struct object_id peeled;
2365
2366 if (starts_with(path, "refs/tags/") && /* is a tag? */
2367 !peel_ref(path, peeled.hash) && /* peelable? */
2368 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2369 add_tag_chain(oid);
2370 return 0;
2371 }
2372
2373 static void prepare_pack(int window, int depth)
2374 {
2375 struct object_entry **delta_list;
2376 uint32_t i, nr_deltas;
2377 unsigned n;
2378
2379 get_object_details();
2380
2381 /*
2382 * If we're locally repacking then we need to be doubly careful
2383 * from now on in order to make sure no stealth corruption gets
2384 * propagated to the new pack. Clients receiving streamed packs
2385 * should validate everything they get anyway so no need to incur
2386 * the additional cost here in that case.
2387 */
2388 if (!pack_to_stdout)
2389 do_check_packed_object_crc = 1;
2390
2391 if (!to_pack.nr_objects || !window || !depth)
2392 return;
2393
2394 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2395 nr_deltas = n = 0;
2396
2397 for (i = 0; i < to_pack.nr_objects; i++) {
2398 struct object_entry *entry = to_pack.objects + i;
2399
2400 if (entry->delta)
2401 /* This happens if we decided to reuse existing
2402 * delta from a pack. "reuse_delta &&" is implied.
2403 */
2404 continue;
2405
2406 if (entry->size < 50)
2407 continue;
2408
2409 if (entry->no_try_delta)
2410 continue;
2411
2412 if (!entry->preferred_base) {
2413 nr_deltas++;
2414 if (entry->type < 0)
2415 die("unable to get type of object %s",
2416 oid_to_hex(&entry->idx.oid));
2417 } else {
2418 if (entry->type < 0) {
2419 /*
2420 * This object is not found, but we
2421 * don't have to include it anyway.
2422 */
2423 continue;
2424 }
2425 }
2426
2427 delta_list[n++] = entry;
2428 }
2429
2430 if (nr_deltas && n > 1) {
2431 unsigned nr_done = 0;
2432 if (progress)
2433 progress_state = start_progress(_("Compressing objects"),
2434 nr_deltas);
2435 QSORT(delta_list, n, type_size_sort);
2436 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2437 stop_progress(&progress_state);
2438 if (nr_done != nr_deltas)
2439 die("inconsistency with delta count");
2440 }
2441 free(delta_list);
2442 }
2443
2444 static int git_pack_config(const char *k, const char *v, void *cb)
2445 {
2446 if (!strcmp(k, "pack.window")) {
2447 window = git_config_int(k, v);
2448 return 0;
2449 }
2450 if (!strcmp(k, "pack.windowmemory")) {
2451 window_memory_limit = git_config_ulong(k, v);
2452 return 0;
2453 }
2454 if (!strcmp(k, "pack.depth")) {
2455 depth = git_config_int(k, v);
2456 return 0;
2457 }
2458 if (!strcmp(k, "pack.deltacachesize")) {
2459 max_delta_cache_size = git_config_int(k, v);
2460 return 0;
2461 }
2462 if (!strcmp(k, "pack.deltacachelimit")) {
2463 cache_max_small_delta_size = git_config_int(k, v);
2464 return 0;
2465 }
2466 if (!strcmp(k, "pack.writebitmaphashcache")) {
2467 if (git_config_bool(k, v))
2468 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2469 else
2470 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2471 }
2472 if (!strcmp(k, "pack.usebitmaps")) {
2473 use_bitmap_index_default = git_config_bool(k, v);
2474 return 0;
2475 }
2476 if (!strcmp(k, "pack.threads")) {
2477 delta_search_threads = git_config_int(k, v);
2478 if (delta_search_threads < 0)
2479 die("invalid number of threads specified (%d)",
2480 delta_search_threads);
2481 #ifdef NO_PTHREADS
2482 if (delta_search_threads != 1) {
2483 warning("no threads support, ignoring %s", k);
2484 delta_search_threads = 0;
2485 }
2486 #endif
2487 return 0;
2488 }
2489 if (!strcmp(k, "pack.indexversion")) {
2490 pack_idx_opts.version = git_config_int(k, v);
2491 if (pack_idx_opts.version > 2)
2492 die("bad pack.indexversion=%"PRIu32,
2493 pack_idx_opts.version);
2494 return 0;
2495 }
2496 return git_default_config(k, v, cb);
2497 }
2498
2499 static void read_object_list_from_stdin(void)
2500 {
2501 char line[40 + 1 + PATH_MAX + 2];
2502 unsigned char sha1[20];
2503
2504 for (;;) {
2505 if (!fgets(line, sizeof(line), stdin)) {
2506 if (feof(stdin))
2507 break;
2508 if (!ferror(stdin))
2509 die("fgets returned NULL, not EOF, not error!");
2510 if (errno != EINTR)
2511 die_errno("fgets");
2512 clearerr(stdin);
2513 continue;
2514 }
2515 if (line[0] == '-') {
2516 if (get_sha1_hex(line+1, sha1))
2517 die("expected edge sha1, got garbage:\n %s",
2518 line);
2519 add_preferred_base(sha1);
2520 continue;
2521 }
2522 if (get_sha1_hex(line, sha1))
2523 die("expected sha1, got garbage:\n %s", line);
2524
2525 add_preferred_base_object(line+41);
2526 add_object_entry(sha1, 0, line+41, 0);
2527 }
2528 }
2529
2530 #define OBJECT_ADDED (1u<<20)
2531
2532 static void show_commit(struct commit *commit, void *data)
2533 {
2534 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2535 commit->object.flags |= OBJECT_ADDED;
2536
2537 if (write_bitmap_index)
2538 index_commit_for_bitmap(commit);
2539 }
2540
2541 static void show_object(struct object *obj, const char *name, void *data)
2542 {
2543 add_preferred_base_object(name);
2544 add_object_entry(obj->oid.hash, obj->type, name, 0);
2545 obj->flags |= OBJECT_ADDED;
2546 }
2547
2548 static void show_edge(struct commit *commit)
2549 {
2550 add_preferred_base(commit->object.oid.hash);
2551 }
2552
2553 struct in_pack_object {
2554 off_t offset;
2555 struct object *object;
2556 };
2557
2558 struct in_pack {
2559 int alloc;
2560 int nr;
2561 struct in_pack_object *array;
2562 };
2563
2564 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2565 {
2566 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2567 in_pack->array[in_pack->nr].object = object;
2568 in_pack->nr++;
2569 }
2570
2571 /*
2572 * Compare the objects in the offset order, in order to emulate the
2573 * "git rev-list --objects" output that produced the pack originally.
2574 */
2575 static int ofscmp(const void *a_, const void *b_)
2576 {
2577 struct in_pack_object *a = (struct in_pack_object *)a_;
2578 struct in_pack_object *b = (struct in_pack_object *)b_;
2579
2580 if (a->offset < b->offset)
2581 return -1;
2582 else if (a->offset > b->offset)
2583 return 1;
2584 else
2585 return oidcmp(&a->object->oid, &b->object->oid);
2586 }
2587
2588 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2589 {
2590 struct packed_git *p;
2591 struct in_pack in_pack;
2592 uint32_t i;
2593
2594 memset(&in_pack, 0, sizeof(in_pack));
2595
2596 for (p = packed_git; p; p = p->next) {
2597 const unsigned char *sha1;
2598 struct object *o;
2599
2600 if (!p->pack_local || p->pack_keep)
2601 continue;
2602 if (open_pack_index(p))
2603 die("cannot open pack index");
2604
2605 ALLOC_GROW(in_pack.array,
2606 in_pack.nr + p->num_objects,
2607 in_pack.alloc);
2608
2609 for (i = 0; i < p->num_objects; i++) {
2610 sha1 = nth_packed_object_sha1(p, i);
2611 o = lookup_unknown_object(sha1);
2612 if (!(o->flags & OBJECT_ADDED))
2613 mark_in_pack_object(o, p, &in_pack);
2614 o->flags |= OBJECT_ADDED;
2615 }
2616 }
2617
2618 if (in_pack.nr) {
2619 QSORT(in_pack.array, in_pack.nr, ofscmp);
2620 for (i = 0; i < in_pack.nr; i++) {
2621 struct object *o = in_pack.array[i].object;
2622 add_object_entry(o->oid.hash, o->type, "", 0);
2623 }
2624 }
2625 free(in_pack.array);
2626 }
2627
2628 static int add_loose_object(const struct object_id *oid, const char *path,
2629 void *data)
2630 {
2631 enum object_type type = sha1_object_info(oid->hash, NULL);
2632
2633 if (type < 0) {
2634 warning("loose object at %s could not be examined", path);
2635 return 0;
2636 }
2637
2638 add_object_entry(oid->hash, type, "", 0);
2639 return 0;
2640 }
2641
2642 /*
2643 * We actually don't even have to worry about reachability here.
2644 * add_object_entry will weed out duplicates, so we just add every
2645 * loose object we find.
2646 */
2647 static void add_unreachable_loose_objects(void)
2648 {
2649 for_each_loose_file_in_objdir(get_object_directory(),
2650 add_loose_object,
2651 NULL, NULL, NULL);
2652 }
2653
2654 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2655 {
2656 static struct packed_git *last_found = (void *)1;
2657 struct packed_git *p;
2658
2659 p = (last_found != (void *)1) ? last_found : packed_git;
2660
2661 while (p) {
2662 if ((!p->pack_local || p->pack_keep) &&
2663 find_pack_entry_one(sha1, p)) {
2664 last_found = p;
2665 return 1;
2666 }
2667 if (p == last_found)
2668 p = packed_git;
2669 else
2670 p = p->next;
2671 if (p == last_found)
2672 p = p->next;
2673 }
2674 return 0;
2675 }
2676
2677 /*
2678 * Store a list of sha1s that are should not be discarded
2679 * because they are either written too recently, or are
2680 * reachable from another object that was.
2681 *
2682 * This is filled by get_object_list.
2683 */
2684 static struct oid_array recent_objects;
2685
2686 static int loosened_object_can_be_discarded(const struct object_id *oid,
2687 timestamp_t mtime)
2688 {
2689 if (!unpack_unreachable_expiration)
2690 return 0;
2691 if (mtime > unpack_unreachable_expiration)
2692 return 0;
2693 if (oid_array_lookup(&recent_objects, oid) >= 0)
2694 return 0;
2695 return 1;
2696 }
2697
2698 static void loosen_unused_packed_objects(struct rev_info *revs)
2699 {
2700 struct packed_git *p;
2701 uint32_t i;
2702 struct object_id oid;
2703
2704 for (p = packed_git; p; p = p->next) {
2705 if (!p->pack_local || p->pack_keep)
2706 continue;
2707
2708 if (open_pack_index(p))
2709 die("cannot open pack index");
2710
2711 for (i = 0; i < p->num_objects; i++) {
2712 nth_packed_object_oid(&oid, p, i);
2713 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2714 !has_sha1_pack_kept_or_nonlocal(oid.hash) &&
2715 !loosened_object_can_be_discarded(&oid, p->mtime))
2716 if (force_object_loose(oid.hash, p->mtime))
2717 die("unable to force loose object");
2718 }
2719 }
2720 }
2721
2722 /*
2723 * This tracks any options which pack-reuse code expects to be on, or which a
2724 * reader of the pack might not understand, and which would therefore prevent
2725 * blind reuse of what we have on disk.
2726 */
2727 static int pack_options_allow_reuse(void)
2728 {
2729 return pack_to_stdout &&
2730 allow_ofs_delta &&
2731 !ignore_packed_keep &&
2732 (!local || !have_non_local_packs) &&
2733 !incremental;
2734 }
2735
2736 static int get_object_list_from_bitmap(struct rev_info *revs)
2737 {
2738 if (prepare_bitmap_walk(revs) < 0)
2739 return -1;
2740
2741 if (pack_options_allow_reuse() &&
2742 !reuse_partial_packfile_from_bitmap(
2743 &reuse_packfile,
2744 &reuse_packfile_objects,
2745 &reuse_packfile_offset)) {
2746 assert(reuse_packfile_objects);
2747 nr_result += reuse_packfile_objects;
2748 display_progress(progress_state, nr_result);
2749 }
2750
2751 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2752 return 0;
2753 }
2754
2755 static void record_recent_object(struct object *obj,
2756 const char *name,
2757 void *data)
2758 {
2759 oid_array_append(&recent_objects, &obj->oid);
2760 }
2761
2762 static void record_recent_commit(struct commit *commit, void *data)
2763 {
2764 oid_array_append(&recent_objects, &commit->object.oid);
2765 }
2766
2767 static void get_object_list(int ac, const char **av)
2768 {
2769 struct rev_info revs;
2770 char line[1000];
2771 int flags = 0;
2772
2773 init_revisions(&revs, NULL);
2774 save_commit_buffer = 0;
2775 setup_revisions(ac, av, &revs, NULL);
2776
2777 /* make sure shallows are read */
2778 is_repository_shallow();
2779
2780 while (fgets(line, sizeof(line), stdin) != NULL) {
2781 int len = strlen(line);
2782 if (len && line[len - 1] == '\n')
2783 line[--len] = 0;
2784 if (!len)
2785 break;
2786 if (*line == '-') {
2787 if (!strcmp(line, "--not")) {
2788 flags ^= UNINTERESTING;
2789 write_bitmap_index = 0;
2790 continue;
2791 }
2792 if (starts_with(line, "--shallow ")) {
2793 struct object_id oid;
2794 if (get_oid_hex(line + 10, &oid))
2795 die("not an SHA-1 '%s'", line + 10);
2796 register_shallow(&oid);
2797 use_bitmap_index = 0;
2798 continue;
2799 }
2800 die("not a rev '%s'", line);
2801 }
2802 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2803 die("bad revision '%s'", line);
2804 }
2805
2806 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2807 return;
2808
2809 if (prepare_revision_walk(&revs))
2810 die("revision walk setup failed");
2811 mark_edges_uninteresting(&revs, show_edge);
2812 traverse_commit_list(&revs, show_commit, show_object, NULL);
2813
2814 if (unpack_unreachable_expiration) {
2815 revs.ignore_missing_links = 1;
2816 if (add_unseen_recent_objects_to_traversal(&revs,
2817 unpack_unreachable_expiration))
2818 die("unable to add recent objects");
2819 if (prepare_revision_walk(&revs))
2820 die("revision walk setup failed");
2821 traverse_commit_list(&revs, record_recent_commit,
2822 record_recent_object, NULL);
2823 }
2824
2825 if (keep_unreachable)
2826 add_objects_in_unpacked_packs(&revs);
2827 if (pack_loose_unreachable)
2828 add_unreachable_loose_objects();
2829 if (unpack_unreachable)
2830 loosen_unused_packed_objects(&revs);
2831
2832 oid_array_clear(&recent_objects);
2833 }
2834
2835 static int option_parse_index_version(const struct option *opt,
2836 const char *arg, int unset)
2837 {
2838 char *c;
2839 const char *val = arg;
2840 pack_idx_opts.version = strtoul(val, &c, 10);
2841 if (pack_idx_opts.version > 2)
2842 die(_("unsupported index version %s"), val);
2843 if (*c == ',' && c[1])
2844 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2845 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2846 die(_("bad index version '%s'"), val);
2847 return 0;
2848 }
2849
2850 static int option_parse_unpack_unreachable(const struct option *opt,
2851 const char *arg, int unset)
2852 {
2853 if (unset) {
2854 unpack_unreachable = 0;
2855 unpack_unreachable_expiration = 0;
2856 }
2857 else {
2858 unpack_unreachable = 1;
2859 if (arg)
2860 unpack_unreachable_expiration = approxidate(arg);
2861 }
2862 return 0;
2863 }
2864
2865 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2866 {
2867 int use_internal_rev_list = 0;
2868 int thin = 0;
2869 int shallow = 0;
2870 int all_progress_implied = 0;
2871 struct argv_array rp = ARGV_ARRAY_INIT;
2872 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2873 int rev_list_index = 0;
2874 struct option pack_objects_options[] = {
2875 OPT_SET_INT('q', "quiet", &progress,
2876 N_("do not show progress meter"), 0),
2877 OPT_SET_INT(0, "progress", &progress,
2878 N_("show progress meter"), 1),
2879 OPT_SET_INT(0, "all-progress", &progress,
2880 N_("show progress meter during object writing phase"), 2),
2881 OPT_BOOL(0, "all-progress-implied",
2882 &all_progress_implied,
2883 N_("similar to --all-progress when progress meter is shown")),
2884 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2885 N_("write the pack index file in the specified idx format version"),
2886 0, option_parse_index_version },
2887 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2888 N_("maximum size of each output pack file")),
2889 OPT_BOOL(0, "local", &local,
2890 N_("ignore borrowed objects from alternate object store")),
2891 OPT_BOOL(0, "incremental", &incremental,
2892 N_("ignore packed objects")),
2893 OPT_INTEGER(0, "window", &window,
2894 N_("limit pack window by objects")),
2895 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2896 N_("limit pack window by memory in addition to object limit")),
2897 OPT_INTEGER(0, "depth", &depth,
2898 N_("maximum length of delta chain allowed in the resulting pack")),
2899 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2900 N_("reuse existing deltas")),
2901 OPT_BOOL(0, "reuse-object", &reuse_object,
2902 N_("reuse existing objects")),
2903 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2904 N_("use OFS_DELTA objects")),
2905 OPT_INTEGER(0, "threads", &delta_search_threads,
2906 N_("use threads when searching for best delta matches")),
2907 OPT_BOOL(0, "non-empty", &non_empty,
2908 N_("do not create an empty pack output")),
2909 OPT_BOOL(0, "revs", &use_internal_rev_list,
2910 N_("read revision arguments from standard input")),
2911 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2912 N_("limit the objects to those that are not yet packed"),
2913 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2914 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2915 N_("include objects reachable from any reference"),
2916 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2917 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2918 N_("include objects referred by reflog entries"),
2919 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2920 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2921 N_("include objects referred to by the index"),
2922 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2923 OPT_BOOL(0, "stdout", &pack_to_stdout,
2924 N_("output pack to stdout")),
2925 OPT_BOOL(0, "include-tag", &include_tag,
2926 N_("include tag objects that refer to objects to be packed")),
2927 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2928 N_("keep unreachable objects")),
2929 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2930 N_("pack loose unreachable objects")),
2931 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2932 N_("unpack unreachable objects newer than <time>"),
2933 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2934 OPT_BOOL(0, "thin", &thin,
2935 N_("create thin packs")),
2936 OPT_BOOL(0, "shallow", &shallow,
2937 N_("create packs suitable for shallow fetches")),
2938 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2939 N_("ignore packs that have companion .keep file")),
2940 OPT_INTEGER(0, "compression", &pack_compression_level,
2941 N_("pack compression level")),
2942 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2943 N_("do not hide commits by grafts"), 0),
2944 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2945 N_("use a bitmap index if available to speed up counting objects")),
2946 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2947 N_("write a bitmap index together with the pack index")),
2948 OPT_END(),
2949 };
2950
2951 check_replace_refs = 0;
2952
2953 reset_pack_idx_option(&pack_idx_opts);
2954 git_config(git_pack_config, NULL);
2955
2956 progress = isatty(2);
2957 argc = parse_options(argc, argv, prefix, pack_objects_options,
2958 pack_usage, 0);
2959
2960 if (argc) {
2961 base_name = argv[0];
2962 argc--;
2963 }
2964 if (pack_to_stdout != !base_name || argc)
2965 usage_with_options(pack_usage, pack_objects_options);
2966
2967 argv_array_push(&rp, "pack-objects");
2968 if (thin) {
2969 use_internal_rev_list = 1;
2970 argv_array_push(&rp, shallow
2971 ? "--objects-edge-aggressive"
2972 : "--objects-edge");
2973 } else
2974 argv_array_push(&rp, "--objects");
2975
2976 if (rev_list_all) {
2977 use_internal_rev_list = 1;
2978 argv_array_push(&rp, "--all");
2979 }
2980 if (rev_list_reflog) {
2981 use_internal_rev_list = 1;
2982 argv_array_push(&rp, "--reflog");
2983 }
2984 if (rev_list_index) {
2985 use_internal_rev_list = 1;
2986 argv_array_push(&rp, "--indexed-objects");
2987 }
2988 if (rev_list_unpacked) {
2989 use_internal_rev_list = 1;
2990 argv_array_push(&rp, "--unpacked");
2991 }
2992
2993 if (!reuse_object)
2994 reuse_delta = 0;
2995 if (pack_compression_level == -1)
2996 pack_compression_level = Z_DEFAULT_COMPRESSION;
2997 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2998 die("bad pack compression level %d", pack_compression_level);
2999
3000 if (!delta_search_threads) /* --threads=0 means autodetect */
3001 delta_search_threads = online_cpus();
3002
3003 #ifdef NO_PTHREADS
3004 if (delta_search_threads != 1)
3005 warning("no threads support, ignoring --threads");
3006 #endif
3007 if (!pack_to_stdout && !pack_size_limit)
3008 pack_size_limit = pack_size_limit_cfg;
3009 if (pack_to_stdout && pack_size_limit)
3010 die("--max-pack-size cannot be used to build a pack for transfer.");
3011 if (pack_size_limit && pack_size_limit < 1024*1024) {
3012 warning("minimum pack size limit is 1 MiB");
3013 pack_size_limit = 1024*1024;
3014 }
3015
3016 if (!pack_to_stdout && thin)
3017 die("--thin cannot be used to build an indexable pack.");
3018
3019 if (keep_unreachable && unpack_unreachable)
3020 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3021 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3022 unpack_unreachable_expiration = 0;
3023
3024 /*
3025 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3026 *
3027 * - to produce good pack (with bitmap index not-yet-packed objects are
3028 * packed in suboptimal order).
3029 *
3030 * - to use more robust pack-generation codepath (avoiding possible
3031 * bugs in bitmap code and possible bitmap index corruption).
3032 */
3033 if (!pack_to_stdout)
3034 use_bitmap_index_default = 0;
3035
3036 if (use_bitmap_index < 0)
3037 use_bitmap_index = use_bitmap_index_default;
3038
3039 /* "hard" reasons not to use bitmaps; these just won't work at all */
3040 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3041 use_bitmap_index = 0;
3042
3043 if (pack_to_stdout || !rev_list_all)
3044 write_bitmap_index = 0;
3045
3046 if (progress && all_progress_implied)
3047 progress = 2;
3048
3049 prepare_packed_git();
3050 if (ignore_packed_keep) {
3051 struct packed_git *p;
3052 for (p = packed_git; p; p = p->next)
3053 if (p->pack_local && p->pack_keep)
3054 break;
3055 if (!p) /* no keep-able packs found */
3056 ignore_packed_keep = 0;
3057 }
3058 if (local) {
3059 /*
3060 * unlike ignore_packed_keep above, we do not want to
3061 * unset "local" based on looking at packs, as it
3062 * also covers non-local objects
3063 */
3064 struct packed_git *p;
3065 for (p = packed_git; p; p = p->next) {
3066 if (!p->pack_local) {
3067 have_non_local_packs = 1;
3068 break;
3069 }
3070 }
3071 }
3072
3073 if (progress)
3074 progress_state = start_progress(_("Counting objects"), 0);
3075 if (!use_internal_rev_list)
3076 read_object_list_from_stdin();
3077 else {
3078 get_object_list(rp.argc, rp.argv);
3079 argv_array_clear(&rp);
3080 }
3081 cleanup_preferred_base();
3082 if (include_tag && nr_result)
3083 for_each_ref(add_ref_tag, NULL);
3084 stop_progress(&progress_state);
3085
3086 if (non_empty && !nr_result)
3087 return 0;
3088 if (nr_result)
3089 prepare_pack(window, depth);
3090 write_pack_file();
3091 if (progress)
3092 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3093 " reused %"PRIu32" (delta %"PRIu32")\n",
3094 written, written_delta, reused, reused_delta);
3095 return 0;
3096 }