t5300: abstract away SHA-1-specific constants
[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 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
841 } else if (nr_written == nr_remaining) {
842 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
843 } else {
844 int fd = finalize_hashfile(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(&entry->idx.oid, &entry->size);
1520 /*
1521 * The error condition is checked in prepare_pack(). This is
1522 * to permit a missing preferred base object to be ignored
1523 * as a preferred base. Doing so can result in a larger
1524 * pack file, but the transfer will still take place.
1525 */
1526 }
1527
1528 static int pack_offset_sort(const void *_a, const void *_b)
1529 {
1530 const struct object_entry *a = *(struct object_entry **)_a;
1531 const struct object_entry *b = *(struct object_entry **)_b;
1532
1533 /* avoid filesystem trashing with loose objects */
1534 if (!a->in_pack && !b->in_pack)
1535 return oidcmp(&a->idx.oid, &b->idx.oid);
1536
1537 if (a->in_pack < b->in_pack)
1538 return -1;
1539 if (a->in_pack > b->in_pack)
1540 return 1;
1541 return a->in_pack_offset < b->in_pack_offset ? -1 :
1542 (a->in_pack_offset > b->in_pack_offset);
1543 }
1544
1545 /*
1546 * Drop an on-disk delta we were planning to reuse. Naively, this would
1547 * just involve blanking out the "delta" field, but we have to deal
1548 * with some extra book-keeping:
1549 *
1550 * 1. Removing ourselves from the delta_sibling linked list.
1551 *
1552 * 2. Updating our size/type to the non-delta representation. These were
1553 * either not recorded initially (size) or overwritten with the delta type
1554 * (type) when check_object() decided to reuse the delta.
1555 *
1556 * 3. Resetting our delta depth, as we are now a base object.
1557 */
1558 static void drop_reused_delta(struct object_entry *entry)
1559 {
1560 struct object_entry **p = &entry->delta->delta_child;
1561 struct object_info oi = OBJECT_INFO_INIT;
1562
1563 while (*p) {
1564 if (*p == entry)
1565 *p = (*p)->delta_sibling;
1566 else
1567 p = &(*p)->delta_sibling;
1568 }
1569 entry->delta = NULL;
1570 entry->depth = 0;
1571
1572 oi.sizep = &entry->size;
1573 oi.typep = &entry->type;
1574 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1575 /*
1576 * We failed to get the info from this pack for some reason;
1577 * fall back to sha1_object_info, which may find another copy.
1578 * And if that fails, the error will be recorded in entry->type
1579 * and dealt with in prepare_pack().
1580 */
1581 entry->type = oid_object_info(&entry->idx.oid, &entry->size);
1582 }
1583 }
1584
1585 /*
1586 * Follow the chain of deltas from this entry onward, throwing away any links
1587 * that cause us to hit a cycle (as determined by the DFS state flags in
1588 * the entries).
1589 *
1590 * We also detect too-long reused chains that would violate our --depth
1591 * limit.
1592 */
1593 static void break_delta_chains(struct object_entry *entry)
1594 {
1595 /*
1596 * The actual depth of each object we will write is stored as an int,
1597 * as it cannot exceed our int "depth" limit. But before we break
1598 * changes based no that limit, we may potentially go as deep as the
1599 * number of objects, which is elsewhere bounded to a uint32_t.
1600 */
1601 uint32_t total_depth;
1602 struct object_entry *cur, *next;
1603
1604 for (cur = entry, total_depth = 0;
1605 cur;
1606 cur = cur->delta, total_depth++) {
1607 if (cur->dfs_state == DFS_DONE) {
1608 /*
1609 * We've already seen this object and know it isn't
1610 * part of a cycle. We do need to append its depth
1611 * to our count.
1612 */
1613 total_depth += cur->depth;
1614 break;
1615 }
1616
1617 /*
1618 * We break cycles before looping, so an ACTIVE state (or any
1619 * other cruft which made its way into the state variable)
1620 * is a bug.
1621 */
1622 if (cur->dfs_state != DFS_NONE)
1623 die("BUG: confusing delta dfs state in first pass: %d",
1624 cur->dfs_state);
1625
1626 /*
1627 * Now we know this is the first time we've seen the object. If
1628 * it's not a delta, we're done traversing, but we'll mark it
1629 * done to save time on future traversals.
1630 */
1631 if (!cur->delta) {
1632 cur->dfs_state = DFS_DONE;
1633 break;
1634 }
1635
1636 /*
1637 * Mark ourselves as active and see if the next step causes
1638 * us to cycle to another active object. It's important to do
1639 * this _before_ we loop, because it impacts where we make the
1640 * cut, and thus how our total_depth counter works.
1641 * E.g., We may see a partial loop like:
1642 *
1643 * A -> B -> C -> D -> B
1644 *
1645 * Cutting B->C breaks the cycle. But now the depth of A is
1646 * only 1, and our total_depth counter is at 3. The size of the
1647 * error is always one less than the size of the cycle we
1648 * broke. Commits C and D were "lost" from A's chain.
1649 *
1650 * If we instead cut D->B, then the depth of A is correct at 3.
1651 * We keep all commits in the chain that we examined.
1652 */
1653 cur->dfs_state = DFS_ACTIVE;
1654 if (cur->delta->dfs_state == DFS_ACTIVE) {
1655 drop_reused_delta(cur);
1656 cur->dfs_state = DFS_DONE;
1657 break;
1658 }
1659 }
1660
1661 /*
1662 * And now that we've gone all the way to the bottom of the chain, we
1663 * need to clear the active flags and set the depth fields as
1664 * appropriate. Unlike the loop above, which can quit when it drops a
1665 * delta, we need to keep going to look for more depth cuts. So we need
1666 * an extra "next" pointer to keep going after we reset cur->delta.
1667 */
1668 for (cur = entry; cur; cur = next) {
1669 next = cur->delta;
1670
1671 /*
1672 * We should have a chain of zero or more ACTIVE states down to
1673 * a final DONE. We can quit after the DONE, because either it
1674 * has no bases, or we've already handled them in a previous
1675 * call.
1676 */
1677 if (cur->dfs_state == DFS_DONE)
1678 break;
1679 else if (cur->dfs_state != DFS_ACTIVE)
1680 die("BUG: confusing delta dfs state in second pass: %d",
1681 cur->dfs_state);
1682
1683 /*
1684 * If the total_depth is more than depth, then we need to snip
1685 * the chain into two or more smaller chains that don't exceed
1686 * the maximum depth. Most of the resulting chains will contain
1687 * (depth + 1) entries (i.e., depth deltas plus one base), and
1688 * the last chain (i.e., the one containing entry) will contain
1689 * whatever entries are left over, namely
1690 * (total_depth % (depth + 1)) of them.
1691 *
1692 * Since we are iterating towards decreasing depth, we need to
1693 * decrement total_depth as we go, and we need to write to the
1694 * entry what its final depth will be after all of the
1695 * snipping. Since we're snipping into chains of length (depth
1696 * + 1) entries, the final depth of an entry will be its
1697 * original depth modulo (depth + 1). Any time we encounter an
1698 * entry whose final depth is supposed to be zero, we snip it
1699 * from its delta base, thereby making it so.
1700 */
1701 cur->depth = (total_depth--) % (depth + 1);
1702 if (!cur->depth)
1703 drop_reused_delta(cur);
1704
1705 cur->dfs_state = DFS_DONE;
1706 }
1707 }
1708
1709 static void get_object_details(void)
1710 {
1711 uint32_t i;
1712 struct object_entry **sorted_by_offset;
1713
1714 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1715 for (i = 0; i < to_pack.nr_objects; i++)
1716 sorted_by_offset[i] = to_pack.objects + i;
1717 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1718
1719 for (i = 0; i < to_pack.nr_objects; i++) {
1720 struct object_entry *entry = sorted_by_offset[i];
1721 check_object(entry);
1722 if (big_file_threshold < entry->size)
1723 entry->no_try_delta = 1;
1724 }
1725
1726 /*
1727 * This must happen in a second pass, since we rely on the delta
1728 * information for the whole list being completed.
1729 */
1730 for (i = 0; i < to_pack.nr_objects; i++)
1731 break_delta_chains(&to_pack.objects[i]);
1732
1733 free(sorted_by_offset);
1734 }
1735
1736 /*
1737 * We search for deltas in a list sorted by type, by filename hash, and then
1738 * by size, so that we see progressively smaller and smaller files.
1739 * That's because we prefer deltas to be from the bigger file
1740 * to the smaller -- deletes are potentially cheaper, but perhaps
1741 * more importantly, the bigger file is likely the more recent
1742 * one. The deepest deltas are therefore the oldest objects which are
1743 * less susceptible to be accessed often.
1744 */
1745 static int type_size_sort(const void *_a, const void *_b)
1746 {
1747 const struct object_entry *a = *(struct object_entry **)_a;
1748 const struct object_entry *b = *(struct object_entry **)_b;
1749
1750 if (a->type > b->type)
1751 return -1;
1752 if (a->type < b->type)
1753 return 1;
1754 if (a->hash > b->hash)
1755 return -1;
1756 if (a->hash < b->hash)
1757 return 1;
1758 if (a->preferred_base > b->preferred_base)
1759 return -1;
1760 if (a->preferred_base < b->preferred_base)
1761 return 1;
1762 if (a->size > b->size)
1763 return -1;
1764 if (a->size < b->size)
1765 return 1;
1766 return a < b ? -1 : (a > b); /* newest first */
1767 }
1768
1769 struct unpacked {
1770 struct object_entry *entry;
1771 void *data;
1772 struct delta_index *index;
1773 unsigned depth;
1774 };
1775
1776 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1777 unsigned long delta_size)
1778 {
1779 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1780 return 0;
1781
1782 if (delta_size < cache_max_small_delta_size)
1783 return 1;
1784
1785 /* cache delta, if objects are large enough compared to delta size */
1786 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1787 return 1;
1788
1789 return 0;
1790 }
1791
1792 #ifndef NO_PTHREADS
1793
1794 static pthread_mutex_t read_mutex;
1795 #define read_lock() pthread_mutex_lock(&read_mutex)
1796 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1797
1798 static pthread_mutex_t cache_mutex;
1799 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1800 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1801
1802 static pthread_mutex_t progress_mutex;
1803 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1804 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1805
1806 #else
1807
1808 #define read_lock() (void)0
1809 #define read_unlock() (void)0
1810 #define cache_lock() (void)0
1811 #define cache_unlock() (void)0
1812 #define progress_lock() (void)0
1813 #define progress_unlock() (void)0
1814
1815 #endif
1816
1817 static int try_delta(struct unpacked *trg, struct unpacked *src,
1818 unsigned max_depth, unsigned long *mem_usage)
1819 {
1820 struct object_entry *trg_entry = trg->entry;
1821 struct object_entry *src_entry = src->entry;
1822 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1823 unsigned ref_depth;
1824 enum object_type type;
1825 void *delta_buf;
1826
1827 /* Don't bother doing diffs between different types */
1828 if (trg_entry->type != src_entry->type)
1829 return -1;
1830
1831 /*
1832 * We do not bother to try a delta that we discarded on an
1833 * earlier try, but only when reusing delta data. Note that
1834 * src_entry that is marked as the preferred_base should always
1835 * be considered, as even if we produce a suboptimal delta against
1836 * it, we will still save the transfer cost, as we already know
1837 * the other side has it and we won't send src_entry at all.
1838 */
1839 if (reuse_delta && trg_entry->in_pack &&
1840 trg_entry->in_pack == src_entry->in_pack &&
1841 !src_entry->preferred_base &&
1842 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1843 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1844 return 0;
1845
1846 /* Let's not bust the allowed depth. */
1847 if (src->depth >= max_depth)
1848 return 0;
1849
1850 /* Now some size filtering heuristics. */
1851 trg_size = trg_entry->size;
1852 if (!trg_entry->delta) {
1853 max_size = trg_size/2 - 20;
1854 ref_depth = 1;
1855 } else {
1856 max_size = trg_entry->delta_size;
1857 ref_depth = trg->depth;
1858 }
1859 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1860 (max_depth - ref_depth + 1);
1861 if (max_size == 0)
1862 return 0;
1863 src_size = src_entry->size;
1864 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1865 if (sizediff >= max_size)
1866 return 0;
1867 if (trg_size < src_size / 32)
1868 return 0;
1869
1870 /* Load data if not already done */
1871 if (!trg->data) {
1872 read_lock();
1873 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
1874 read_unlock();
1875 if (!trg->data)
1876 die("object %s cannot be read",
1877 oid_to_hex(&trg_entry->idx.oid));
1878 if (sz != trg_size)
1879 die("object %s inconsistent object length (%lu vs %lu)",
1880 oid_to_hex(&trg_entry->idx.oid), sz,
1881 trg_size);
1882 *mem_usage += sz;
1883 }
1884 if (!src->data) {
1885 read_lock();
1886 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
1887 read_unlock();
1888 if (!src->data) {
1889 if (src_entry->preferred_base) {
1890 static int warned = 0;
1891 if (!warned++)
1892 warning("object %s cannot be read",
1893 oid_to_hex(&src_entry->idx.oid));
1894 /*
1895 * Those objects are not included in the
1896 * resulting pack. Be resilient and ignore
1897 * them if they can't be read, in case the
1898 * pack could be created nevertheless.
1899 */
1900 return 0;
1901 }
1902 die("object %s cannot be read",
1903 oid_to_hex(&src_entry->idx.oid));
1904 }
1905 if (sz != src_size)
1906 die("object %s inconsistent object length (%lu vs %lu)",
1907 oid_to_hex(&src_entry->idx.oid), sz,
1908 src_size);
1909 *mem_usage += sz;
1910 }
1911 if (!src->index) {
1912 src->index = create_delta_index(src->data, src_size);
1913 if (!src->index) {
1914 static int warned = 0;
1915 if (!warned++)
1916 warning("suboptimal pack - out of memory");
1917 return 0;
1918 }
1919 *mem_usage += sizeof_delta_index(src->index);
1920 }
1921
1922 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1923 if (!delta_buf)
1924 return 0;
1925
1926 if (trg_entry->delta) {
1927 /* Prefer only shallower same-sized deltas. */
1928 if (delta_size == trg_entry->delta_size &&
1929 src->depth + 1 >= trg->depth) {
1930 free(delta_buf);
1931 return 0;
1932 }
1933 }
1934
1935 /*
1936 * Handle memory allocation outside of the cache
1937 * accounting lock. Compiler will optimize the strangeness
1938 * away when NO_PTHREADS is defined.
1939 */
1940 free(trg_entry->delta_data);
1941 cache_lock();
1942 if (trg_entry->delta_data) {
1943 delta_cache_size -= trg_entry->delta_size;
1944 trg_entry->delta_data = NULL;
1945 }
1946 if (delta_cacheable(src_size, trg_size, delta_size)) {
1947 delta_cache_size += delta_size;
1948 cache_unlock();
1949 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1950 } else {
1951 cache_unlock();
1952 free(delta_buf);
1953 }
1954
1955 trg_entry->delta = src_entry;
1956 trg_entry->delta_size = delta_size;
1957 trg->depth = src->depth + 1;
1958
1959 return 1;
1960 }
1961
1962 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1963 {
1964 struct object_entry *child = me->delta_child;
1965 unsigned int m = n;
1966 while (child) {
1967 unsigned int c = check_delta_limit(child, n + 1);
1968 if (m < c)
1969 m = c;
1970 child = child->delta_sibling;
1971 }
1972 return m;
1973 }
1974
1975 static unsigned long free_unpacked(struct unpacked *n)
1976 {
1977 unsigned long freed_mem = sizeof_delta_index(n->index);
1978 free_delta_index(n->index);
1979 n->index = NULL;
1980 if (n->data) {
1981 freed_mem += n->entry->size;
1982 FREE_AND_NULL(n->data);
1983 }
1984 n->entry = NULL;
1985 n->depth = 0;
1986 return freed_mem;
1987 }
1988
1989 static void find_deltas(struct object_entry **list, unsigned *list_size,
1990 int window, int depth, unsigned *processed)
1991 {
1992 uint32_t i, idx = 0, count = 0;
1993 struct unpacked *array;
1994 unsigned long mem_usage = 0;
1995
1996 array = xcalloc(window, sizeof(struct unpacked));
1997
1998 for (;;) {
1999 struct object_entry *entry;
2000 struct unpacked *n = array + idx;
2001 int j, max_depth, best_base = -1;
2002
2003 progress_lock();
2004 if (!*list_size) {
2005 progress_unlock();
2006 break;
2007 }
2008 entry = *list++;
2009 (*list_size)--;
2010 if (!entry->preferred_base) {
2011 (*processed)++;
2012 display_progress(progress_state, *processed);
2013 }
2014 progress_unlock();
2015
2016 mem_usage -= free_unpacked(n);
2017 n->entry = entry;
2018
2019 while (window_memory_limit &&
2020 mem_usage > window_memory_limit &&
2021 count > 1) {
2022 uint32_t tail = (idx + window - count) % window;
2023 mem_usage -= free_unpacked(array + tail);
2024 count--;
2025 }
2026
2027 /* We do not compute delta to *create* objects we are not
2028 * going to pack.
2029 */
2030 if (entry->preferred_base)
2031 goto next;
2032
2033 /*
2034 * If the current object is at pack edge, take the depth the
2035 * objects that depend on the current object into account
2036 * otherwise they would become too deep.
2037 */
2038 max_depth = depth;
2039 if (entry->delta_child) {
2040 max_depth -= check_delta_limit(entry, 0);
2041 if (max_depth <= 0)
2042 goto next;
2043 }
2044
2045 j = window;
2046 while (--j > 0) {
2047 int ret;
2048 uint32_t other_idx = idx + j;
2049 struct unpacked *m;
2050 if (other_idx >= window)
2051 other_idx -= window;
2052 m = array + other_idx;
2053 if (!m->entry)
2054 break;
2055 ret = try_delta(n, m, max_depth, &mem_usage);
2056 if (ret < 0)
2057 break;
2058 else if (ret > 0)
2059 best_base = other_idx;
2060 }
2061
2062 /*
2063 * If we decided to cache the delta data, then it is best
2064 * to compress it right away. First because we have to do
2065 * it anyway, and doing it here while we're threaded will
2066 * save a lot of time in the non threaded write phase,
2067 * as well as allow for caching more deltas within
2068 * the same cache size limit.
2069 * ...
2070 * But only if not writing to stdout, since in that case
2071 * the network is most likely throttling writes anyway,
2072 * and therefore it is best to go to the write phase ASAP
2073 * instead, as we can afford spending more time compressing
2074 * between writes at that moment.
2075 */
2076 if (entry->delta_data && !pack_to_stdout) {
2077 entry->z_delta_size = do_compress(&entry->delta_data,
2078 entry->delta_size);
2079 cache_lock();
2080 delta_cache_size -= entry->delta_size;
2081 delta_cache_size += entry->z_delta_size;
2082 cache_unlock();
2083 }
2084
2085 /* if we made n a delta, and if n is already at max
2086 * depth, leaving it in the window is pointless. we
2087 * should evict it first.
2088 */
2089 if (entry->delta && max_depth <= n->depth)
2090 continue;
2091
2092 /*
2093 * Move the best delta base up in the window, after the
2094 * currently deltified object, to keep it longer. It will
2095 * be the first base object to be attempted next.
2096 */
2097 if (entry->delta) {
2098 struct unpacked swap = array[best_base];
2099 int dist = (window + idx - best_base) % window;
2100 int dst = best_base;
2101 while (dist--) {
2102 int src = (dst + 1) % window;
2103 array[dst] = array[src];
2104 dst = src;
2105 }
2106 array[dst] = swap;
2107 }
2108
2109 next:
2110 idx++;
2111 if (count + 1 < window)
2112 count++;
2113 if (idx >= window)
2114 idx = 0;
2115 }
2116
2117 for (i = 0; i < window; ++i) {
2118 free_delta_index(array[i].index);
2119 free(array[i].data);
2120 }
2121 free(array);
2122 }
2123
2124 #ifndef NO_PTHREADS
2125
2126 static void try_to_free_from_threads(size_t size)
2127 {
2128 read_lock();
2129 release_pack_memory(size);
2130 read_unlock();
2131 }
2132
2133 static try_to_free_t old_try_to_free_routine;
2134
2135 /*
2136 * The main thread waits on the condition that (at least) one of the workers
2137 * has stopped working (which is indicated in the .working member of
2138 * struct thread_params).
2139 * When a work thread has completed its work, it sets .working to 0 and
2140 * signals the main thread and waits on the condition that .data_ready
2141 * becomes 1.
2142 */
2143
2144 struct thread_params {
2145 pthread_t thread;
2146 struct object_entry **list;
2147 unsigned list_size;
2148 unsigned remaining;
2149 int window;
2150 int depth;
2151 int working;
2152 int data_ready;
2153 pthread_mutex_t mutex;
2154 pthread_cond_t cond;
2155 unsigned *processed;
2156 };
2157
2158 static pthread_cond_t progress_cond;
2159
2160 /*
2161 * Mutex and conditional variable can't be statically-initialized on Windows.
2162 */
2163 static void init_threaded_search(void)
2164 {
2165 init_recursive_mutex(&read_mutex);
2166 pthread_mutex_init(&cache_mutex, NULL);
2167 pthread_mutex_init(&progress_mutex, NULL);
2168 pthread_cond_init(&progress_cond, NULL);
2169 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2170 }
2171
2172 static void cleanup_threaded_search(void)
2173 {
2174 set_try_to_free_routine(old_try_to_free_routine);
2175 pthread_cond_destroy(&progress_cond);
2176 pthread_mutex_destroy(&read_mutex);
2177 pthread_mutex_destroy(&cache_mutex);
2178 pthread_mutex_destroy(&progress_mutex);
2179 }
2180
2181 static void *threaded_find_deltas(void *arg)
2182 {
2183 struct thread_params *me = arg;
2184
2185 progress_lock();
2186 while (me->remaining) {
2187 progress_unlock();
2188
2189 find_deltas(me->list, &me->remaining,
2190 me->window, me->depth, me->processed);
2191
2192 progress_lock();
2193 me->working = 0;
2194 pthread_cond_signal(&progress_cond);
2195 progress_unlock();
2196
2197 /*
2198 * We must not set ->data_ready before we wait on the
2199 * condition because the main thread may have set it to 1
2200 * before we get here. In order to be sure that new
2201 * work is available if we see 1 in ->data_ready, it
2202 * was initialized to 0 before this thread was spawned
2203 * and we reset it to 0 right away.
2204 */
2205 pthread_mutex_lock(&me->mutex);
2206 while (!me->data_ready)
2207 pthread_cond_wait(&me->cond, &me->mutex);
2208 me->data_ready = 0;
2209 pthread_mutex_unlock(&me->mutex);
2210
2211 progress_lock();
2212 }
2213 progress_unlock();
2214 /* leave ->working 1 so that this doesn't get more work assigned */
2215 return NULL;
2216 }
2217
2218 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2219 int window, int depth, unsigned *processed)
2220 {
2221 struct thread_params *p;
2222 int i, ret, active_threads = 0;
2223
2224 init_threaded_search();
2225
2226 if (delta_search_threads <= 1) {
2227 find_deltas(list, &list_size, window, depth, processed);
2228 cleanup_threaded_search();
2229 return;
2230 }
2231 if (progress > pack_to_stdout)
2232 fprintf(stderr, "Delta compression using up to %d threads.\n",
2233 delta_search_threads);
2234 p = xcalloc(delta_search_threads, sizeof(*p));
2235
2236 /* Partition the work amongst work threads. */
2237 for (i = 0; i < delta_search_threads; i++) {
2238 unsigned sub_size = list_size / (delta_search_threads - i);
2239
2240 /* don't use too small segments or no deltas will be found */
2241 if (sub_size < 2*window && i+1 < delta_search_threads)
2242 sub_size = 0;
2243
2244 p[i].window = window;
2245 p[i].depth = depth;
2246 p[i].processed = processed;
2247 p[i].working = 1;
2248 p[i].data_ready = 0;
2249
2250 /* try to split chunks on "path" boundaries */
2251 while (sub_size && sub_size < list_size &&
2252 list[sub_size]->hash &&
2253 list[sub_size]->hash == list[sub_size-1]->hash)
2254 sub_size++;
2255
2256 p[i].list = list;
2257 p[i].list_size = sub_size;
2258 p[i].remaining = sub_size;
2259
2260 list += sub_size;
2261 list_size -= sub_size;
2262 }
2263
2264 /* Start work threads. */
2265 for (i = 0; i < delta_search_threads; i++) {
2266 if (!p[i].list_size)
2267 continue;
2268 pthread_mutex_init(&p[i].mutex, NULL);
2269 pthread_cond_init(&p[i].cond, NULL);
2270 ret = pthread_create(&p[i].thread, NULL,
2271 threaded_find_deltas, &p[i]);
2272 if (ret)
2273 die("unable to create thread: %s", strerror(ret));
2274 active_threads++;
2275 }
2276
2277 /*
2278 * Now let's wait for work completion. Each time a thread is done
2279 * with its work, we steal half of the remaining work from the
2280 * thread with the largest number of unprocessed objects and give
2281 * it to that newly idle thread. This ensure good load balancing
2282 * until the remaining object list segments are simply too short
2283 * to be worth splitting anymore.
2284 */
2285 while (active_threads) {
2286 struct thread_params *target = NULL;
2287 struct thread_params *victim = NULL;
2288 unsigned sub_size = 0;
2289
2290 progress_lock();
2291 for (;;) {
2292 for (i = 0; !target && i < delta_search_threads; i++)
2293 if (!p[i].working)
2294 target = &p[i];
2295 if (target)
2296 break;
2297 pthread_cond_wait(&progress_cond, &progress_mutex);
2298 }
2299
2300 for (i = 0; i < delta_search_threads; i++)
2301 if (p[i].remaining > 2*window &&
2302 (!victim || victim->remaining < p[i].remaining))
2303 victim = &p[i];
2304 if (victim) {
2305 sub_size = victim->remaining / 2;
2306 list = victim->list + victim->list_size - sub_size;
2307 while (sub_size && list[0]->hash &&
2308 list[0]->hash == list[-1]->hash) {
2309 list++;
2310 sub_size--;
2311 }
2312 if (!sub_size) {
2313 /*
2314 * It is possible for some "paths" to have
2315 * so many objects that no hash boundary
2316 * might be found. Let's just steal the
2317 * exact half in that case.
2318 */
2319 sub_size = victim->remaining / 2;
2320 list -= sub_size;
2321 }
2322 target->list = list;
2323 victim->list_size -= sub_size;
2324 victim->remaining -= sub_size;
2325 }
2326 target->list_size = sub_size;
2327 target->remaining = sub_size;
2328 target->working = 1;
2329 progress_unlock();
2330
2331 pthread_mutex_lock(&target->mutex);
2332 target->data_ready = 1;
2333 pthread_cond_signal(&target->cond);
2334 pthread_mutex_unlock(&target->mutex);
2335
2336 if (!sub_size) {
2337 pthread_join(target->thread, NULL);
2338 pthread_cond_destroy(&target->cond);
2339 pthread_mutex_destroy(&target->mutex);
2340 active_threads--;
2341 }
2342 }
2343 cleanup_threaded_search();
2344 free(p);
2345 }
2346
2347 #else
2348 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2349 #endif
2350
2351 static void add_tag_chain(const struct object_id *oid)
2352 {
2353 struct tag *tag;
2354
2355 /*
2356 * We catch duplicates already in add_object_entry(), but we'd
2357 * prefer to do this extra check to avoid having to parse the
2358 * tag at all if we already know that it's being packed (e.g., if
2359 * it was included via bitmaps, we would not have parsed it
2360 * previously).
2361 */
2362 if (packlist_find(&to_pack, oid->hash, NULL))
2363 return;
2364
2365 tag = lookup_tag(oid);
2366 while (1) {
2367 if (!tag || parse_tag(tag) || !tag->tagged)
2368 die("unable to pack objects reachable from tag %s",
2369 oid_to_hex(oid));
2370
2371 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2372
2373 if (tag->tagged->type != OBJ_TAG)
2374 return;
2375
2376 tag = (struct tag *)tag->tagged;
2377 }
2378 }
2379
2380 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2381 {
2382 struct object_id peeled;
2383
2384 if (starts_with(path, "refs/tags/") && /* is a tag? */
2385 !peel_ref(path, &peeled) && /* peelable? */
2386 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2387 add_tag_chain(oid);
2388 return 0;
2389 }
2390
2391 static void prepare_pack(int window, int depth)
2392 {
2393 struct object_entry **delta_list;
2394 uint32_t i, nr_deltas;
2395 unsigned n;
2396
2397 get_object_details();
2398
2399 /*
2400 * If we're locally repacking then we need to be doubly careful
2401 * from now on in order to make sure no stealth corruption gets
2402 * propagated to the new pack. Clients receiving streamed packs
2403 * should validate everything they get anyway so no need to incur
2404 * the additional cost here in that case.
2405 */
2406 if (!pack_to_stdout)
2407 do_check_packed_object_crc = 1;
2408
2409 if (!to_pack.nr_objects || !window || !depth)
2410 return;
2411
2412 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2413 nr_deltas = n = 0;
2414
2415 for (i = 0; i < to_pack.nr_objects; i++) {
2416 struct object_entry *entry = to_pack.objects + i;
2417
2418 if (entry->delta)
2419 /* This happens if we decided to reuse existing
2420 * delta from a pack. "reuse_delta &&" is implied.
2421 */
2422 continue;
2423
2424 if (entry->size < 50)
2425 continue;
2426
2427 if (entry->no_try_delta)
2428 continue;
2429
2430 if (!entry->preferred_base) {
2431 nr_deltas++;
2432 if (entry->type < 0)
2433 die("unable to get type of object %s",
2434 oid_to_hex(&entry->idx.oid));
2435 } else {
2436 if (entry->type < 0) {
2437 /*
2438 * This object is not found, but we
2439 * don't have to include it anyway.
2440 */
2441 continue;
2442 }
2443 }
2444
2445 delta_list[n++] = entry;
2446 }
2447
2448 if (nr_deltas && n > 1) {
2449 unsigned nr_done = 0;
2450 if (progress)
2451 progress_state = start_progress(_("Compressing objects"),
2452 nr_deltas);
2453 QSORT(delta_list, n, type_size_sort);
2454 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2455 stop_progress(&progress_state);
2456 if (nr_done != nr_deltas)
2457 die("inconsistency with delta count");
2458 }
2459 free(delta_list);
2460 }
2461
2462 static int git_pack_config(const char *k, const char *v, void *cb)
2463 {
2464 if (!strcmp(k, "pack.window")) {
2465 window = git_config_int(k, v);
2466 return 0;
2467 }
2468 if (!strcmp(k, "pack.windowmemory")) {
2469 window_memory_limit = git_config_ulong(k, v);
2470 return 0;
2471 }
2472 if (!strcmp(k, "pack.depth")) {
2473 depth = git_config_int(k, v);
2474 return 0;
2475 }
2476 if (!strcmp(k, "pack.deltacachesize")) {
2477 max_delta_cache_size = git_config_int(k, v);
2478 return 0;
2479 }
2480 if (!strcmp(k, "pack.deltacachelimit")) {
2481 cache_max_small_delta_size = git_config_int(k, v);
2482 return 0;
2483 }
2484 if (!strcmp(k, "pack.writebitmaphashcache")) {
2485 if (git_config_bool(k, v))
2486 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2487 else
2488 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2489 }
2490 if (!strcmp(k, "pack.usebitmaps")) {
2491 use_bitmap_index_default = git_config_bool(k, v);
2492 return 0;
2493 }
2494 if (!strcmp(k, "pack.threads")) {
2495 delta_search_threads = git_config_int(k, v);
2496 if (delta_search_threads < 0)
2497 die("invalid number of threads specified (%d)",
2498 delta_search_threads);
2499 #ifdef NO_PTHREADS
2500 if (delta_search_threads != 1) {
2501 warning("no threads support, ignoring %s", k);
2502 delta_search_threads = 0;
2503 }
2504 #endif
2505 return 0;
2506 }
2507 if (!strcmp(k, "pack.indexversion")) {
2508 pack_idx_opts.version = git_config_int(k, v);
2509 if (pack_idx_opts.version > 2)
2510 die("bad pack.indexversion=%"PRIu32,
2511 pack_idx_opts.version);
2512 return 0;
2513 }
2514 return git_default_config(k, v, cb);
2515 }
2516
2517 static void read_object_list_from_stdin(void)
2518 {
2519 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2520 struct object_id oid;
2521 const char *p;
2522
2523 for (;;) {
2524 if (!fgets(line, sizeof(line), stdin)) {
2525 if (feof(stdin))
2526 break;
2527 if (!ferror(stdin))
2528 die("fgets returned NULL, not EOF, not error!");
2529 if (errno != EINTR)
2530 die_errno("fgets");
2531 clearerr(stdin);
2532 continue;
2533 }
2534 if (line[0] == '-') {
2535 if (get_oid_hex(line+1, &oid))
2536 die("expected edge object ID, got garbage:\n %s",
2537 line);
2538 add_preferred_base(&oid);
2539 continue;
2540 }
2541 if (parse_oid_hex(line, &oid, &p))
2542 die("expected object ID, got garbage:\n %s", line);
2543
2544 add_preferred_base_object(p + 1);
2545 add_object_entry(&oid, 0, p + 1, 0);
2546 }
2547 }
2548
2549 /* Remember to update object flag allocation in object.h */
2550 #define OBJECT_ADDED (1u<<20)
2551
2552 static void show_commit(struct commit *commit, void *data)
2553 {
2554 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2555 commit->object.flags |= OBJECT_ADDED;
2556
2557 if (write_bitmap_index)
2558 index_commit_for_bitmap(commit);
2559 }
2560
2561 static void show_object(struct object *obj, const char *name, void *data)
2562 {
2563 add_preferred_base_object(name);
2564 add_object_entry(&obj->oid, obj->type, name, 0);
2565 obj->flags |= OBJECT_ADDED;
2566 }
2567
2568 static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2569 {
2570 assert(arg_missing_action == MA_ALLOW_ANY);
2571
2572 /*
2573 * Quietly ignore ALL missing objects. This avoids problems with
2574 * staging them now and getting an odd error later.
2575 */
2576 if (!has_object_file(&obj->oid))
2577 return;
2578
2579 show_object(obj, name, data);
2580 }
2581
2582 static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2583 {
2584 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2585
2586 /*
2587 * Quietly ignore EXPECTED missing objects. This avoids problems with
2588 * staging them now and getting an odd error later.
2589 */
2590 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2591 return;
2592
2593 show_object(obj, name, data);
2594 }
2595
2596 static int option_parse_missing_action(const struct option *opt,
2597 const char *arg, int unset)
2598 {
2599 assert(arg);
2600 assert(!unset);
2601
2602 if (!strcmp(arg, "error")) {
2603 arg_missing_action = MA_ERROR;
2604 fn_show_object = show_object;
2605 return 0;
2606 }
2607
2608 if (!strcmp(arg, "allow-any")) {
2609 arg_missing_action = MA_ALLOW_ANY;
2610 fetch_if_missing = 0;
2611 fn_show_object = show_object__ma_allow_any;
2612 return 0;
2613 }
2614
2615 if (!strcmp(arg, "allow-promisor")) {
2616 arg_missing_action = MA_ALLOW_PROMISOR;
2617 fetch_if_missing = 0;
2618 fn_show_object = show_object__ma_allow_promisor;
2619 return 0;
2620 }
2621
2622 die(_("invalid value for --missing"));
2623 return 0;
2624 }
2625
2626 static void show_edge(struct commit *commit)
2627 {
2628 add_preferred_base(&commit->object.oid);
2629 }
2630
2631 struct in_pack_object {
2632 off_t offset;
2633 struct object *object;
2634 };
2635
2636 struct in_pack {
2637 unsigned int alloc;
2638 unsigned int nr;
2639 struct in_pack_object *array;
2640 };
2641
2642 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2643 {
2644 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2645 in_pack->array[in_pack->nr].object = object;
2646 in_pack->nr++;
2647 }
2648
2649 /*
2650 * Compare the objects in the offset order, in order to emulate the
2651 * "git rev-list --objects" output that produced the pack originally.
2652 */
2653 static int ofscmp(const void *a_, const void *b_)
2654 {
2655 struct in_pack_object *a = (struct in_pack_object *)a_;
2656 struct in_pack_object *b = (struct in_pack_object *)b_;
2657
2658 if (a->offset < b->offset)
2659 return -1;
2660 else if (a->offset > b->offset)
2661 return 1;
2662 else
2663 return oidcmp(&a->object->oid, &b->object->oid);
2664 }
2665
2666 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2667 {
2668 struct packed_git *p;
2669 struct in_pack in_pack;
2670 uint32_t i;
2671
2672 memset(&in_pack, 0, sizeof(in_pack));
2673
2674 for (p = get_packed_git(the_repository); p; p = p->next) {
2675 struct object_id oid;
2676 struct object *o;
2677
2678 if (!p->pack_local || p->pack_keep)
2679 continue;
2680 if (open_pack_index(p))
2681 die("cannot open pack index");
2682
2683 ALLOC_GROW(in_pack.array,
2684 in_pack.nr + p->num_objects,
2685 in_pack.alloc);
2686
2687 for (i = 0; i < p->num_objects; i++) {
2688 nth_packed_object_oid(&oid, p, i);
2689 o = lookup_unknown_object(oid.hash);
2690 if (!(o->flags & OBJECT_ADDED))
2691 mark_in_pack_object(o, p, &in_pack);
2692 o->flags |= OBJECT_ADDED;
2693 }
2694 }
2695
2696 if (in_pack.nr) {
2697 QSORT(in_pack.array, in_pack.nr, ofscmp);
2698 for (i = 0; i < in_pack.nr; i++) {
2699 struct object *o = in_pack.array[i].object;
2700 add_object_entry(&o->oid, o->type, "", 0);
2701 }
2702 }
2703 free(in_pack.array);
2704 }
2705
2706 static int add_loose_object(const struct object_id *oid, const char *path,
2707 void *data)
2708 {
2709 enum object_type type = oid_object_info(oid, NULL);
2710
2711 if (type < 0) {
2712 warning("loose object at %s could not be examined", path);
2713 return 0;
2714 }
2715
2716 add_object_entry(oid, type, "", 0);
2717 return 0;
2718 }
2719
2720 /*
2721 * We actually don't even have to worry about reachability here.
2722 * add_object_entry will weed out duplicates, so we just add every
2723 * loose object we find.
2724 */
2725 static void add_unreachable_loose_objects(void)
2726 {
2727 for_each_loose_file_in_objdir(get_object_directory(),
2728 add_loose_object,
2729 NULL, NULL, NULL);
2730 }
2731
2732 static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2733 {
2734 static struct packed_git *last_found = (void *)1;
2735 struct packed_git *p;
2736
2737 p = (last_found != (void *)1) ? last_found :
2738 get_packed_git(the_repository);
2739
2740 while (p) {
2741 if ((!p->pack_local || p->pack_keep) &&
2742 find_pack_entry_one(oid->hash, p)) {
2743 last_found = p;
2744 return 1;
2745 }
2746 if (p == last_found)
2747 p = get_packed_git(the_repository);
2748 else
2749 p = p->next;
2750 if (p == last_found)
2751 p = p->next;
2752 }
2753 return 0;
2754 }
2755
2756 /*
2757 * Store a list of sha1s that are should not be discarded
2758 * because they are either written too recently, or are
2759 * reachable from another object that was.
2760 *
2761 * This is filled by get_object_list.
2762 */
2763 static struct oid_array recent_objects;
2764
2765 static int loosened_object_can_be_discarded(const struct object_id *oid,
2766 timestamp_t mtime)
2767 {
2768 if (!unpack_unreachable_expiration)
2769 return 0;
2770 if (mtime > unpack_unreachable_expiration)
2771 return 0;
2772 if (oid_array_lookup(&recent_objects, oid) >= 0)
2773 return 0;
2774 return 1;
2775 }
2776
2777 static void loosen_unused_packed_objects(struct rev_info *revs)
2778 {
2779 struct packed_git *p;
2780 uint32_t i;
2781 struct object_id oid;
2782
2783 for (p = get_packed_git(the_repository); p; p = p->next) {
2784 if (!p->pack_local || p->pack_keep)
2785 continue;
2786
2787 if (open_pack_index(p))
2788 die("cannot open pack index");
2789
2790 for (i = 0; i < p->num_objects; i++) {
2791 nth_packed_object_oid(&oid, p, i);
2792 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2793 !has_sha1_pack_kept_or_nonlocal(&oid) &&
2794 !loosened_object_can_be_discarded(&oid, p->mtime))
2795 if (force_object_loose(&oid, p->mtime))
2796 die("unable to force loose object");
2797 }
2798 }
2799 }
2800
2801 /*
2802 * This tracks any options which pack-reuse code expects to be on, or which a
2803 * reader of the pack might not understand, and which would therefore prevent
2804 * blind reuse of what we have on disk.
2805 */
2806 static int pack_options_allow_reuse(void)
2807 {
2808 return pack_to_stdout &&
2809 allow_ofs_delta &&
2810 !ignore_packed_keep &&
2811 (!local || !have_non_local_packs) &&
2812 !incremental;
2813 }
2814
2815 static int get_object_list_from_bitmap(struct rev_info *revs)
2816 {
2817 if (prepare_bitmap_walk(revs) < 0)
2818 return -1;
2819
2820 if (pack_options_allow_reuse() &&
2821 !reuse_partial_packfile_from_bitmap(
2822 &reuse_packfile,
2823 &reuse_packfile_objects,
2824 &reuse_packfile_offset)) {
2825 assert(reuse_packfile_objects);
2826 nr_result += reuse_packfile_objects;
2827 display_progress(progress_state, nr_result);
2828 }
2829
2830 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2831 return 0;
2832 }
2833
2834 static void record_recent_object(struct object *obj,
2835 const char *name,
2836 void *data)
2837 {
2838 oid_array_append(&recent_objects, &obj->oid);
2839 }
2840
2841 static void record_recent_commit(struct commit *commit, void *data)
2842 {
2843 oid_array_append(&recent_objects, &commit->object.oid);
2844 }
2845
2846 static void get_object_list(int ac, const char **av)
2847 {
2848 struct rev_info revs;
2849 char line[1000];
2850 int flags = 0;
2851
2852 init_revisions(&revs, NULL);
2853 save_commit_buffer = 0;
2854 setup_revisions(ac, av, &revs, NULL);
2855
2856 /* make sure shallows are read */
2857 is_repository_shallow();
2858
2859 while (fgets(line, sizeof(line), stdin) != NULL) {
2860 int len = strlen(line);
2861 if (len && line[len - 1] == '\n')
2862 line[--len] = 0;
2863 if (!len)
2864 break;
2865 if (*line == '-') {
2866 if (!strcmp(line, "--not")) {
2867 flags ^= UNINTERESTING;
2868 write_bitmap_index = 0;
2869 continue;
2870 }
2871 if (starts_with(line, "--shallow ")) {
2872 struct object_id oid;
2873 if (get_oid_hex(line + 10, &oid))
2874 die("not an SHA-1 '%s'", line + 10);
2875 register_shallow(&oid);
2876 use_bitmap_index = 0;
2877 continue;
2878 }
2879 die("not a rev '%s'", line);
2880 }
2881 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2882 die("bad revision '%s'", line);
2883 }
2884
2885 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2886 return;
2887
2888 if (prepare_revision_walk(&revs))
2889 die("revision walk setup failed");
2890 mark_edges_uninteresting(&revs, show_edge);
2891
2892 if (!fn_show_object)
2893 fn_show_object = show_object;
2894 traverse_commit_list_filtered(&filter_options, &revs,
2895 show_commit, fn_show_object, NULL,
2896 NULL);
2897
2898 if (unpack_unreachable_expiration) {
2899 revs.ignore_missing_links = 1;
2900 if (add_unseen_recent_objects_to_traversal(&revs,
2901 unpack_unreachable_expiration))
2902 die("unable to add recent objects");
2903 if (prepare_revision_walk(&revs))
2904 die("revision walk setup failed");
2905 traverse_commit_list(&revs, record_recent_commit,
2906 record_recent_object, NULL);
2907 }
2908
2909 if (keep_unreachable)
2910 add_objects_in_unpacked_packs(&revs);
2911 if (pack_loose_unreachable)
2912 add_unreachable_loose_objects();
2913 if (unpack_unreachable)
2914 loosen_unused_packed_objects(&revs);
2915
2916 oid_array_clear(&recent_objects);
2917 }
2918
2919 static int option_parse_index_version(const struct option *opt,
2920 const char *arg, int unset)
2921 {
2922 char *c;
2923 const char *val = arg;
2924 pack_idx_opts.version = strtoul(val, &c, 10);
2925 if (pack_idx_opts.version > 2)
2926 die(_("unsupported index version %s"), val);
2927 if (*c == ',' && c[1])
2928 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2929 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2930 die(_("bad index version '%s'"), val);
2931 return 0;
2932 }
2933
2934 static int option_parse_unpack_unreachable(const struct option *opt,
2935 const char *arg, int unset)
2936 {
2937 if (unset) {
2938 unpack_unreachable = 0;
2939 unpack_unreachable_expiration = 0;
2940 }
2941 else {
2942 unpack_unreachable = 1;
2943 if (arg)
2944 unpack_unreachable_expiration = approxidate(arg);
2945 }
2946 return 0;
2947 }
2948
2949 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2950 {
2951 int use_internal_rev_list = 0;
2952 int thin = 0;
2953 int shallow = 0;
2954 int all_progress_implied = 0;
2955 struct argv_array rp = ARGV_ARRAY_INIT;
2956 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2957 int rev_list_index = 0;
2958 struct option pack_objects_options[] = {
2959 OPT_SET_INT('q', "quiet", &progress,
2960 N_("do not show progress meter"), 0),
2961 OPT_SET_INT(0, "progress", &progress,
2962 N_("show progress meter"), 1),
2963 OPT_SET_INT(0, "all-progress", &progress,
2964 N_("show progress meter during object writing phase"), 2),
2965 OPT_BOOL(0, "all-progress-implied",
2966 &all_progress_implied,
2967 N_("similar to --all-progress when progress meter is shown")),
2968 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2969 N_("write the pack index file in the specified idx format version"),
2970 0, option_parse_index_version },
2971 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2972 N_("maximum size of each output pack file")),
2973 OPT_BOOL(0, "local", &local,
2974 N_("ignore borrowed objects from alternate object store")),
2975 OPT_BOOL(0, "incremental", &incremental,
2976 N_("ignore packed objects")),
2977 OPT_INTEGER(0, "window", &window,
2978 N_("limit pack window by objects")),
2979 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2980 N_("limit pack window by memory in addition to object limit")),
2981 OPT_INTEGER(0, "depth", &depth,
2982 N_("maximum length of delta chain allowed in the resulting pack")),
2983 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2984 N_("reuse existing deltas")),
2985 OPT_BOOL(0, "reuse-object", &reuse_object,
2986 N_("reuse existing objects")),
2987 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2988 N_("use OFS_DELTA objects")),
2989 OPT_INTEGER(0, "threads", &delta_search_threads,
2990 N_("use threads when searching for best delta matches")),
2991 OPT_BOOL(0, "non-empty", &non_empty,
2992 N_("do not create an empty pack output")),
2993 OPT_BOOL(0, "revs", &use_internal_rev_list,
2994 N_("read revision arguments from standard input")),
2995 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2996 N_("limit the objects to those that are not yet packed"),
2997 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2998 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2999 N_("include objects reachable from any reference"),
3000 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3001 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
3002 N_("include objects referred by reflog entries"),
3003 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3004 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
3005 N_("include objects referred to by the index"),
3006 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3007 OPT_BOOL(0, "stdout", &pack_to_stdout,
3008 N_("output pack to stdout")),
3009 OPT_BOOL(0, "include-tag", &include_tag,
3010 N_("include tag objects that refer to objects to be packed")),
3011 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3012 N_("keep unreachable objects")),
3013 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3014 N_("pack loose unreachable objects")),
3015 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3016 N_("unpack unreachable objects newer than <time>"),
3017 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3018 OPT_BOOL(0, "thin", &thin,
3019 N_("create thin packs")),
3020 OPT_BOOL(0, "shallow", &shallow,
3021 N_("create packs suitable for shallow fetches")),
3022 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
3023 N_("ignore packs that have companion .keep file")),
3024 OPT_INTEGER(0, "compression", &pack_compression_level,
3025 N_("pack compression level")),
3026 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3027 N_("do not hide commits by grafts"), 0),
3028 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3029 N_("use a bitmap index if available to speed up counting objects")),
3030 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3031 N_("write a bitmap index together with the pack index")),
3032 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3033 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3034 N_("handling for missing objects"), PARSE_OPT_NONEG,
3035 option_parse_missing_action },
3036 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3037 N_("do not pack objects in promisor packfiles")),
3038 OPT_END(),
3039 };
3040
3041 check_replace_refs = 0;
3042
3043 reset_pack_idx_option(&pack_idx_opts);
3044 git_config(git_pack_config, NULL);
3045
3046 progress = isatty(2);
3047 argc = parse_options(argc, argv, prefix, pack_objects_options,
3048 pack_usage, 0);
3049
3050 if (argc) {
3051 base_name = argv[0];
3052 argc--;
3053 }
3054 if (pack_to_stdout != !base_name || argc)
3055 usage_with_options(pack_usage, pack_objects_options);
3056
3057 argv_array_push(&rp, "pack-objects");
3058 if (thin) {
3059 use_internal_rev_list = 1;
3060 argv_array_push(&rp, shallow
3061 ? "--objects-edge-aggressive"
3062 : "--objects-edge");
3063 } else
3064 argv_array_push(&rp, "--objects");
3065
3066 if (rev_list_all) {
3067 use_internal_rev_list = 1;
3068 argv_array_push(&rp, "--all");
3069 }
3070 if (rev_list_reflog) {
3071 use_internal_rev_list = 1;
3072 argv_array_push(&rp, "--reflog");
3073 }
3074 if (rev_list_index) {
3075 use_internal_rev_list = 1;
3076 argv_array_push(&rp, "--indexed-objects");
3077 }
3078 if (rev_list_unpacked) {
3079 use_internal_rev_list = 1;
3080 argv_array_push(&rp, "--unpacked");
3081 }
3082
3083 if (exclude_promisor_objects) {
3084 use_internal_rev_list = 1;
3085 fetch_if_missing = 0;
3086 argv_array_push(&rp, "--exclude-promisor-objects");
3087 }
3088
3089 if (!reuse_object)
3090 reuse_delta = 0;
3091 if (pack_compression_level == -1)
3092 pack_compression_level = Z_DEFAULT_COMPRESSION;
3093 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3094 die("bad pack compression level %d", pack_compression_level);
3095
3096 if (!delta_search_threads) /* --threads=0 means autodetect */
3097 delta_search_threads = online_cpus();
3098
3099 #ifdef NO_PTHREADS
3100 if (delta_search_threads != 1)
3101 warning("no threads support, ignoring --threads");
3102 #endif
3103 if (!pack_to_stdout && !pack_size_limit)
3104 pack_size_limit = pack_size_limit_cfg;
3105 if (pack_to_stdout && pack_size_limit)
3106 die("--max-pack-size cannot be used to build a pack for transfer.");
3107 if (pack_size_limit && pack_size_limit < 1024*1024) {
3108 warning("minimum pack size limit is 1 MiB");
3109 pack_size_limit = 1024*1024;
3110 }
3111
3112 if (!pack_to_stdout && thin)
3113 die("--thin cannot be used to build an indexable pack.");
3114
3115 if (keep_unreachable && unpack_unreachable)
3116 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3117 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3118 unpack_unreachable_expiration = 0;
3119
3120 if (filter_options.choice) {
3121 if (!pack_to_stdout)
3122 die("cannot use --filter without --stdout.");
3123 use_bitmap_index = 0;
3124 }
3125
3126 /*
3127 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3128 *
3129 * - to produce good pack (with bitmap index not-yet-packed objects are
3130 * packed in suboptimal order).
3131 *
3132 * - to use more robust pack-generation codepath (avoiding possible
3133 * bugs in bitmap code and possible bitmap index corruption).
3134 */
3135 if (!pack_to_stdout)
3136 use_bitmap_index_default = 0;
3137
3138 if (use_bitmap_index < 0)
3139 use_bitmap_index = use_bitmap_index_default;
3140
3141 /* "hard" reasons not to use bitmaps; these just won't work at all */
3142 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3143 use_bitmap_index = 0;
3144
3145 if (pack_to_stdout || !rev_list_all)
3146 write_bitmap_index = 0;
3147
3148 if (progress && all_progress_implied)
3149 progress = 2;
3150
3151 if (ignore_packed_keep) {
3152 struct packed_git *p;
3153 for (p = get_packed_git(the_repository); p; p = p->next)
3154 if (p->pack_local && p->pack_keep)
3155 break;
3156 if (!p) /* no keep-able packs found */
3157 ignore_packed_keep = 0;
3158 }
3159 if (local) {
3160 /*
3161 * unlike ignore_packed_keep above, we do not want to
3162 * unset "local" based on looking at packs, as it
3163 * also covers non-local objects
3164 */
3165 struct packed_git *p;
3166 for (p = get_packed_git(the_repository); p; p = p->next) {
3167 if (!p->pack_local) {
3168 have_non_local_packs = 1;
3169 break;
3170 }
3171 }
3172 }
3173
3174 if (progress)
3175 progress_state = start_progress(_("Counting objects"), 0);
3176 if (!use_internal_rev_list)
3177 read_object_list_from_stdin();
3178 else {
3179 get_object_list(rp.argc, rp.argv);
3180 argv_array_clear(&rp);
3181 }
3182 cleanup_preferred_base();
3183 if (include_tag && nr_result)
3184 for_each_ref(add_ref_tag, NULL);
3185 stop_progress(&progress_state);
3186
3187 if (non_empty && !nr_result)
3188 return 0;
3189 if (nr_result)
3190 prepare_pack(window, depth);
3191 write_pack_file();
3192 if (progress)
3193 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3194 " reused %"PRIu32" (delta %"PRIu32")\n",
3195 written, written_delta, reused, reused_delta);
3196 return 0;
3197 }