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