Merge branch 'km/t1400-modernization'
[git/git.git] / builtin / pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "pack-objects.h"
18 #include "progress.h"
19 #include "refs.h"
20 #include "streaming.h"
21 #include "thread-utils.h"
22 #include "pack-bitmap.h"
23 #include "reachable.h"
24 #include "sha1-array.h"
25 #include "argv-array.h"
26 #include "mru.h"
27
28 static const char *pack_usage[] = {
29 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
30 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
31 NULL
32 };
33
34 /*
35 * Objects we are going to pack are collected in the `to_pack` structure.
36 * It contains an array (dynamically expanded) of the object data, and a map
37 * that can resolve SHA1s to their position in the array.
38 */
39 static struct packing_data to_pack;
40
41 static struct pack_idx_entry **written_list;
42 static uint32_t nr_result, nr_written;
43
44 static int non_empty;
45 static int reuse_delta = 1, reuse_object = 1;
46 static int keep_unreachable, unpack_unreachable, include_tag;
47 static unsigned long unpack_unreachable_expiration;
48 static int pack_loose_unreachable;
49 static int local;
50 static int have_non_local_packs;
51 static int incremental;
52 static int ignore_packed_keep;
53 static int allow_ofs_delta;
54 static struct pack_idx_option pack_idx_opts;
55 static const char *base_name;
56 static int progress = 1;
57 static int window = 10;
58 static unsigned long pack_size_limit;
59 static int depth = 50;
60 static int delta_search_threads;
61 static int pack_to_stdout;
62 static int num_preferred_base;
63 static struct progress *progress_state;
64
65 static struct packed_git *reuse_packfile;
66 static uint32_t reuse_packfile_objects;
67 static off_t reuse_packfile_offset;
68
69 static int use_bitmap_index_default = 1;
70 static int use_bitmap_index = -1;
71 static int write_bitmap_index;
72 static uint16_t write_bitmap_options;
73
74 static unsigned long delta_cache_size = 0;
75 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
76 static unsigned long cache_max_small_delta_size = 1000;
77
78 static unsigned long window_memory_limit = 0;
79
80 /*
81 * stats
82 */
83 static uint32_t written, written_delta;
84 static uint32_t reused, reused_delta;
85
86 /*
87 * Indexed commits
88 */
89 static struct commit **indexed_commits;
90 static unsigned int indexed_commits_nr;
91 static unsigned int indexed_commits_alloc;
92
93 static void index_commit_for_bitmap(struct commit *commit)
94 {
95 if (indexed_commits_nr >= indexed_commits_alloc) {
96 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
97 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
98 }
99
100 indexed_commits[indexed_commits_nr++] = commit;
101 }
102
103 static void *get_delta(struct object_entry *entry)
104 {
105 unsigned long size, base_size, delta_size;
106 void *buf, *base_buf, *delta_buf;
107 enum object_type type;
108
109 buf = read_sha1_file(entry->idx.sha1, &type, &size);
110 if (!buf)
111 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
112 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
113 if (!base_buf)
114 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
115 delta_buf = diff_delta(base_buf, base_size,
116 buf, size, &delta_size, 0);
117 if (!delta_buf || delta_size != entry->delta_size)
118 die("delta size changed");
119 free(buf);
120 free(base_buf);
121 return delta_buf;
122 }
123
124 static unsigned long do_compress(void **pptr, unsigned long size)
125 {
126 git_zstream stream;
127 void *in, *out;
128 unsigned long maxsize;
129
130 git_deflate_init(&stream, pack_compression_level);
131 maxsize = git_deflate_bound(&stream, size);
132
133 in = *pptr;
134 out = xmalloc(maxsize);
135 *pptr = out;
136
137 stream.next_in = in;
138 stream.avail_in = size;
139 stream.next_out = out;
140 stream.avail_out = maxsize;
141 while (git_deflate(&stream, Z_FINISH) == Z_OK)
142 ; /* nothing */
143 git_deflate_end(&stream);
144
145 free(in);
146 return stream.total_out;
147 }
148
149 static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
150 const unsigned char *sha1)
151 {
152 git_zstream stream;
153 unsigned char ibuf[1024 * 16];
154 unsigned char obuf[1024 * 16];
155 unsigned long olen = 0;
156
157 git_deflate_init(&stream, pack_compression_level);
158
159 for (;;) {
160 ssize_t readlen;
161 int zret = Z_OK;
162 readlen = read_istream(st, ibuf, sizeof(ibuf));
163 if (readlen == -1)
164 die(_("unable to read %s"), sha1_to_hex(sha1));
165
166 stream.next_in = ibuf;
167 stream.avail_in = readlen;
168 while ((stream.avail_in || readlen == 0) &&
169 (zret == Z_OK || zret == Z_BUF_ERROR)) {
170 stream.next_out = obuf;
171 stream.avail_out = sizeof(obuf);
172 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
173 sha1write(f, obuf, stream.next_out - obuf);
174 olen += stream.next_out - obuf;
175 }
176 if (stream.avail_in)
177 die(_("deflate error (%d)"), zret);
178 if (readlen == 0) {
179 if (zret != Z_STREAM_END)
180 die(_("deflate error (%d)"), zret);
181 break;
182 }
183 }
184 git_deflate_end(&stream);
185 return olen;
186 }
187
188 /*
189 * we are going to reuse the existing object data as is. make
190 * sure it is not corrupt.
191 */
192 static int check_pack_inflate(struct packed_git *p,
193 struct pack_window **w_curs,
194 off_t offset,
195 off_t len,
196 unsigned long expect)
197 {
198 git_zstream stream;
199 unsigned char fakebuf[4096], *in;
200 int st;
201
202 memset(&stream, 0, sizeof(stream));
203 git_inflate_init(&stream);
204 do {
205 in = use_pack(p, w_curs, offset, &stream.avail_in);
206 stream.next_in = in;
207 stream.next_out = fakebuf;
208 stream.avail_out = sizeof(fakebuf);
209 st = git_inflate(&stream, Z_FINISH);
210 offset += stream.next_in - in;
211 } while (st == Z_OK || st == Z_BUF_ERROR);
212 git_inflate_end(&stream);
213 return (st == Z_STREAM_END &&
214 stream.total_out == expect &&
215 stream.total_in == len) ? 0 : -1;
216 }
217
218 static void copy_pack_data(struct sha1file *f,
219 struct packed_git *p,
220 struct pack_window **w_curs,
221 off_t offset,
222 off_t len)
223 {
224 unsigned char *in;
225 unsigned long avail;
226
227 while (len) {
228 in = use_pack(p, w_curs, offset, &avail);
229 if (avail > len)
230 avail = (unsigned long)len;
231 sha1write(f, in, avail);
232 offset += avail;
233 len -= avail;
234 }
235 }
236
237 /* Return 0 if we will bust the pack-size limit */
238 static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
239 unsigned long limit, int usable_delta)
240 {
241 unsigned long size, datalen;
242 unsigned char header[MAX_PACK_OBJECT_HEADER],
243 dheader[MAX_PACK_OBJECT_HEADER];
244 unsigned hdrlen;
245 enum object_type type;
246 void *buf;
247 struct git_istream *st = NULL;
248
249 if (!usable_delta) {
250 if (entry->type == OBJ_BLOB &&
251 entry->size > big_file_threshold &&
252 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
253 buf = NULL;
254 else {
255 buf = read_sha1_file(entry->idx.sha1, &type, &size);
256 if (!buf)
257 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
258 }
259 /*
260 * make sure no cached delta data remains from a
261 * previous attempt before a pack split occurred.
262 */
263 free(entry->delta_data);
264 entry->delta_data = NULL;
265 entry->z_delta_size = 0;
266 } else if (entry->delta_data) {
267 size = entry->delta_size;
268 buf = entry->delta_data;
269 entry->delta_data = NULL;
270 type = (allow_ofs_delta && entry->delta->idx.offset) ?
271 OBJ_OFS_DELTA : OBJ_REF_DELTA;
272 } else {
273 buf = get_delta(entry);
274 size = entry->delta_size;
275 type = (allow_ofs_delta && entry->delta->idx.offset) ?
276 OBJ_OFS_DELTA : OBJ_REF_DELTA;
277 }
278
279 if (st) /* large blob case, just assume we don't compress well */
280 datalen = size;
281 else if (entry->z_delta_size)
282 datalen = entry->z_delta_size;
283 else
284 datalen = do_compress(&buf, size);
285
286 /*
287 * The object header is a byte of 'type' followed by zero or
288 * more bytes of length.
289 */
290 hdrlen = encode_in_pack_object_header(header, sizeof(header),
291 type, size);
292
293 if (type == OBJ_OFS_DELTA) {
294 /*
295 * Deltas with relative base contain an additional
296 * encoding of the relative offset for the delta
297 * base from this object's position in the pack.
298 */
299 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
300 unsigned pos = sizeof(dheader) - 1;
301 dheader[pos] = ofs & 127;
302 while (ofs >>= 7)
303 dheader[--pos] = 128 | (--ofs & 127);
304 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
305 if (st)
306 close_istream(st);
307 free(buf);
308 return 0;
309 }
310 sha1write(f, header, hdrlen);
311 sha1write(f, dheader + pos, sizeof(dheader) - pos);
312 hdrlen += sizeof(dheader) - pos;
313 } else if (type == OBJ_REF_DELTA) {
314 /*
315 * Deltas with a base reference contain
316 * an additional 20 bytes for the base sha1.
317 */
318 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
319 if (st)
320 close_istream(st);
321 free(buf);
322 return 0;
323 }
324 sha1write(f, header, hdrlen);
325 sha1write(f, entry->delta->idx.sha1, 20);
326 hdrlen += 20;
327 } else {
328 if (limit && hdrlen + datalen + 20 >= limit) {
329 if (st)
330 close_istream(st);
331 free(buf);
332 return 0;
333 }
334 sha1write(f, header, hdrlen);
335 }
336 if (st) {
337 datalen = write_large_blob_data(st, f, entry->idx.sha1);
338 close_istream(st);
339 } else {
340 sha1write(f, buf, datalen);
341 free(buf);
342 }
343
344 return hdrlen + datalen;
345 }
346
347 /* Return 0 if we will bust the pack-size limit */
348 static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
349 unsigned long limit, int usable_delta)
350 {
351 struct packed_git *p = entry->in_pack;
352 struct pack_window *w_curs = NULL;
353 struct revindex_entry *revidx;
354 off_t offset;
355 enum object_type type = entry->type;
356 off_t datalen;
357 unsigned char header[MAX_PACK_OBJECT_HEADER],
358 dheader[MAX_PACK_OBJECT_HEADER];
359 unsigned hdrlen;
360
361 if (entry->delta)
362 type = (allow_ofs_delta && entry->delta->idx.offset) ?
363 OBJ_OFS_DELTA : OBJ_REF_DELTA;
364 hdrlen = encode_in_pack_object_header(header, sizeof(header),
365 type, entry->size);
366
367 offset = entry->in_pack_offset;
368 revidx = find_pack_revindex(p, offset);
369 datalen = revidx[1].offset - offset;
370 if (!pack_to_stdout && p->index_version > 1 &&
371 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
372 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
373 unuse_pack(&w_curs);
374 return write_no_reuse_object(f, entry, limit, usable_delta);
375 }
376
377 offset += entry->in_pack_header_size;
378 datalen -= entry->in_pack_header_size;
379
380 if (!pack_to_stdout && p->index_version == 1 &&
381 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
382 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
383 unuse_pack(&w_curs);
384 return write_no_reuse_object(f, entry, limit, usable_delta);
385 }
386
387 if (type == OBJ_OFS_DELTA) {
388 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
389 unsigned pos = sizeof(dheader) - 1;
390 dheader[pos] = ofs & 127;
391 while (ofs >>= 7)
392 dheader[--pos] = 128 | (--ofs & 127);
393 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
394 unuse_pack(&w_curs);
395 return 0;
396 }
397 sha1write(f, header, hdrlen);
398 sha1write(f, dheader + pos, sizeof(dheader) - pos);
399 hdrlen += sizeof(dheader) - pos;
400 reused_delta++;
401 } else if (type == OBJ_REF_DELTA) {
402 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
403 unuse_pack(&w_curs);
404 return 0;
405 }
406 sha1write(f, header, hdrlen);
407 sha1write(f, entry->delta->idx.sha1, 20);
408 hdrlen += 20;
409 reused_delta++;
410 } else {
411 if (limit && hdrlen + datalen + 20 >= limit) {
412 unuse_pack(&w_curs);
413 return 0;
414 }
415 sha1write(f, header, hdrlen);
416 }
417 copy_pack_data(f, p, &w_curs, offset, datalen);
418 unuse_pack(&w_curs);
419 reused++;
420 return hdrlen + datalen;
421 }
422
423 /* Return 0 if we will bust the pack-size limit */
424 static off_t write_object(struct sha1file *f,
425 struct object_entry *entry,
426 off_t write_offset)
427 {
428 unsigned long limit;
429 off_t len;
430 int usable_delta, to_reuse;
431
432 if (!pack_to_stdout)
433 crc32_begin(f);
434
435 /* apply size limit if limited packsize and not first object */
436 if (!pack_size_limit || !nr_written)
437 limit = 0;
438 else if (pack_size_limit <= write_offset)
439 /*
440 * the earlier object did not fit the limit; avoid
441 * mistaking this with unlimited (i.e. limit = 0).
442 */
443 limit = 1;
444 else
445 limit = pack_size_limit - write_offset;
446
447 if (!entry->delta)
448 usable_delta = 0; /* no delta */
449 else if (!pack_size_limit)
450 usable_delta = 1; /* unlimited packfile */
451 else if (entry->delta->idx.offset == (off_t)-1)
452 usable_delta = 0; /* base was written to another pack */
453 else if (entry->delta->idx.offset)
454 usable_delta = 1; /* base already exists in this pack */
455 else
456 usable_delta = 0; /* base could end up in another pack */
457
458 if (!reuse_object)
459 to_reuse = 0; /* explicit */
460 else if (!entry->in_pack)
461 to_reuse = 0; /* can't reuse what we don't have */
462 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
463 /* check_object() decided it for us ... */
464 to_reuse = usable_delta;
465 /* ... but pack split may override that */
466 else if (entry->type != entry->in_pack_type)
467 to_reuse = 0; /* pack has delta which is unusable */
468 else if (entry->delta)
469 to_reuse = 0; /* we want to pack afresh */
470 else
471 to_reuse = 1; /* we have it in-pack undeltified,
472 * and we do not need to deltify it.
473 */
474
475 if (!to_reuse)
476 len = write_no_reuse_object(f, entry, limit, usable_delta);
477 else
478 len = write_reuse_object(f, entry, limit, usable_delta);
479 if (!len)
480 return 0;
481
482 if (usable_delta)
483 written_delta++;
484 written++;
485 if (!pack_to_stdout)
486 entry->idx.crc32 = crc32_end(f);
487 return len;
488 }
489
490 enum write_one_status {
491 WRITE_ONE_SKIP = -1, /* already written */
492 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
493 WRITE_ONE_WRITTEN = 1, /* normal */
494 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
495 };
496
497 static enum write_one_status write_one(struct sha1file *f,
498 struct object_entry *e,
499 off_t *offset)
500 {
501 off_t size;
502 int recursing;
503
504 /*
505 * we set offset to 1 (which is an impossible value) to mark
506 * the fact that this object is involved in "write its base
507 * first before writing a deltified object" recursion.
508 */
509 recursing = (e->idx.offset == 1);
510 if (recursing) {
511 warning("recursive delta detected for object %s",
512 sha1_to_hex(e->idx.sha1));
513 return WRITE_ONE_RECURSIVE;
514 } else if (e->idx.offset || e->preferred_base) {
515 /* offset is non zero if object is written already. */
516 return WRITE_ONE_SKIP;
517 }
518
519 /* if we are deltified, write out base object first. */
520 if (e->delta) {
521 e->idx.offset = 1; /* now recurse */
522 switch (write_one(f, e->delta, offset)) {
523 case WRITE_ONE_RECURSIVE:
524 /* we cannot depend on this one */
525 e->delta = NULL;
526 break;
527 default:
528 break;
529 case WRITE_ONE_BREAK:
530 e->idx.offset = recursing;
531 return WRITE_ONE_BREAK;
532 }
533 }
534
535 e->idx.offset = *offset;
536 size = write_object(f, e, *offset);
537 if (!size) {
538 e->idx.offset = recursing;
539 return WRITE_ONE_BREAK;
540 }
541 written_list[nr_written++] = &e->idx;
542
543 /* make sure off_t is sufficiently large not to wrap */
544 if (signed_add_overflows(*offset, size))
545 die("pack too large for current definition of off_t");
546 *offset += size;
547 return WRITE_ONE_WRITTEN;
548 }
549
550 static int mark_tagged(const char *path, const struct object_id *oid, int flag,
551 void *cb_data)
552 {
553 unsigned char peeled[20];
554 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
555
556 if (entry)
557 entry->tagged = 1;
558 if (!peel_ref(path, peeled)) {
559 entry = packlist_find(&to_pack, peeled, NULL);
560 if (entry)
561 entry->tagged = 1;
562 }
563 return 0;
564 }
565
566 static inline void add_to_write_order(struct object_entry **wo,
567 unsigned int *endp,
568 struct object_entry *e)
569 {
570 if (e->filled)
571 return;
572 wo[(*endp)++] = e;
573 e->filled = 1;
574 }
575
576 static void add_descendants_to_write_order(struct object_entry **wo,
577 unsigned int *endp,
578 struct object_entry *e)
579 {
580 int add_to_order = 1;
581 while (e) {
582 if (add_to_order) {
583 struct object_entry *s;
584 /* add this node... */
585 add_to_write_order(wo, endp, e);
586 /* all its siblings... */
587 for (s = e->delta_sibling; s; s = s->delta_sibling) {
588 add_to_write_order(wo, endp, s);
589 }
590 }
591 /* drop down a level to add left subtree nodes if possible */
592 if (e->delta_child) {
593 add_to_order = 1;
594 e = e->delta_child;
595 } else {
596 add_to_order = 0;
597 /* our sibling might have some children, it is next */
598 if (e->delta_sibling) {
599 e = e->delta_sibling;
600 continue;
601 }
602 /* go back to our parent node */
603 e = e->delta;
604 while (e && !e->delta_sibling) {
605 /* we're on the right side of a subtree, keep
606 * going up until we can go right again */
607 e = e->delta;
608 }
609 if (!e) {
610 /* done- we hit our original root node */
611 return;
612 }
613 /* pass it off to sibling at this level */
614 e = e->delta_sibling;
615 }
616 };
617 }
618
619 static void add_family_to_write_order(struct object_entry **wo,
620 unsigned int *endp,
621 struct object_entry *e)
622 {
623 struct object_entry *root;
624
625 for (root = e; root->delta; root = root->delta)
626 ; /* nothing */
627 add_descendants_to_write_order(wo, endp, root);
628 }
629
630 static struct object_entry **compute_write_order(void)
631 {
632 unsigned int i, wo_end, last_untagged;
633
634 struct object_entry **wo;
635 struct object_entry *objects = to_pack.objects;
636
637 for (i = 0; i < to_pack.nr_objects; i++) {
638 objects[i].tagged = 0;
639 objects[i].filled = 0;
640 objects[i].delta_child = NULL;
641 objects[i].delta_sibling = NULL;
642 }
643
644 /*
645 * Fully connect delta_child/delta_sibling network.
646 * Make sure delta_sibling is sorted in the original
647 * recency order.
648 */
649 for (i = to_pack.nr_objects; i > 0;) {
650 struct object_entry *e = &objects[--i];
651 if (!e->delta)
652 continue;
653 /* Mark me as the first child */
654 e->delta_sibling = e->delta->delta_child;
655 e->delta->delta_child = e;
656 }
657
658 /*
659 * Mark objects that are at the tip of tags.
660 */
661 for_each_tag_ref(mark_tagged, NULL);
662
663 /*
664 * Give the objects in the original recency order until
665 * we see a tagged tip.
666 */
667 ALLOC_ARRAY(wo, to_pack.nr_objects);
668 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
669 if (objects[i].tagged)
670 break;
671 add_to_write_order(wo, &wo_end, &objects[i]);
672 }
673 last_untagged = i;
674
675 /*
676 * Then fill all the tagged tips.
677 */
678 for (; i < to_pack.nr_objects; i++) {
679 if (objects[i].tagged)
680 add_to_write_order(wo, &wo_end, &objects[i]);
681 }
682
683 /*
684 * And then all remaining commits and tags.
685 */
686 for (i = last_untagged; i < to_pack.nr_objects; i++) {
687 if (objects[i].type != OBJ_COMMIT &&
688 objects[i].type != OBJ_TAG)
689 continue;
690 add_to_write_order(wo, &wo_end, &objects[i]);
691 }
692
693 /*
694 * And then all the trees.
695 */
696 for (i = last_untagged; i < to_pack.nr_objects; i++) {
697 if (objects[i].type != OBJ_TREE)
698 continue;
699 add_to_write_order(wo, &wo_end, &objects[i]);
700 }
701
702 /*
703 * Finally all the rest in really tight order
704 */
705 for (i = last_untagged; i < to_pack.nr_objects; i++) {
706 if (!objects[i].filled)
707 add_family_to_write_order(wo, &wo_end, &objects[i]);
708 }
709
710 if (wo_end != to_pack.nr_objects)
711 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
712
713 return wo;
714 }
715
716 static off_t write_reused_pack(struct sha1file *f)
717 {
718 unsigned char buffer[8192];
719 off_t to_write, total;
720 int fd;
721
722 if (!is_pack_valid(reuse_packfile))
723 die("packfile is invalid: %s", reuse_packfile->pack_name);
724
725 fd = git_open(reuse_packfile->pack_name);
726 if (fd < 0)
727 die_errno("unable to open packfile for reuse: %s",
728 reuse_packfile->pack_name);
729
730 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
731 die_errno("unable to seek in reused packfile");
732
733 if (reuse_packfile_offset < 0)
734 reuse_packfile_offset = reuse_packfile->pack_size - 20;
735
736 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
737
738 while (to_write) {
739 int read_pack = xread(fd, buffer, sizeof(buffer));
740
741 if (read_pack <= 0)
742 die_errno("unable to read from reused packfile");
743
744 if (read_pack > to_write)
745 read_pack = to_write;
746
747 sha1write(f, buffer, read_pack);
748 to_write -= read_pack;
749
750 /*
751 * We don't know the actual number of objects written,
752 * only how many bytes written, how many bytes total, and
753 * how many objects total. So we can fake it by pretending all
754 * objects we are writing are the same size. This gives us a
755 * smooth progress meter, and at the end it matches the true
756 * answer.
757 */
758 written = reuse_packfile_objects *
759 (((double)(total - to_write)) / total);
760 display_progress(progress_state, written);
761 }
762
763 close(fd);
764 written = reuse_packfile_objects;
765 display_progress(progress_state, written);
766 return reuse_packfile_offset - sizeof(struct pack_header);
767 }
768
769 static const char no_split_warning[] = N_(
770 "disabling bitmap writing, packs are split due to pack.packSizeLimit"
771 );
772
773 static void write_pack_file(void)
774 {
775 uint32_t i = 0, j;
776 struct sha1file *f;
777 off_t offset;
778 uint32_t nr_remaining = nr_result;
779 time_t last_mtime = 0;
780 struct object_entry **write_order;
781
782 if (progress > pack_to_stdout)
783 progress_state = start_progress(_("Writing objects"), nr_result);
784 ALLOC_ARRAY(written_list, to_pack.nr_objects);
785 write_order = compute_write_order();
786
787 do {
788 unsigned char sha1[20];
789 char *pack_tmp_name = NULL;
790
791 if (pack_to_stdout)
792 f = sha1fd_throughput(1, "<stdout>", progress_state);
793 else
794 f = create_tmp_packfile(&pack_tmp_name);
795
796 offset = write_pack_header(f, nr_remaining);
797
798 if (reuse_packfile) {
799 off_t packfile_size;
800 assert(pack_to_stdout);
801
802 packfile_size = write_reused_pack(f);
803 offset += packfile_size;
804 }
805
806 nr_written = 0;
807 for (; i < to_pack.nr_objects; i++) {
808 struct object_entry *e = write_order[i];
809 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
810 break;
811 display_progress(progress_state, written);
812 }
813
814 /*
815 * Did we write the wrong # entries in the header?
816 * If so, rewrite it like in fast-import
817 */
818 if (pack_to_stdout) {
819 sha1close(f, sha1, CSUM_CLOSE);
820 } else if (nr_written == nr_remaining) {
821 sha1close(f, sha1, CSUM_FSYNC);
822 } else {
823 int fd = sha1close(f, sha1, 0);
824 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
825 nr_written, sha1, offset);
826 close(fd);
827 if (write_bitmap_index) {
828 warning(_(no_split_warning));
829 write_bitmap_index = 0;
830 }
831 }
832
833 if (!pack_to_stdout) {
834 struct stat st;
835 struct strbuf tmpname = STRBUF_INIT;
836
837 /*
838 * Packs are runtime accessed in their mtime
839 * order since newer packs are more likely to contain
840 * younger objects. So if we are creating multiple
841 * packs then we should modify the mtime of later ones
842 * to preserve this property.
843 */
844 if (stat(pack_tmp_name, &st) < 0) {
845 warning_errno("failed to stat %s", pack_tmp_name);
846 } else if (!last_mtime) {
847 last_mtime = st.st_mtime;
848 } else {
849 struct utimbuf utb;
850 utb.actime = st.st_atime;
851 utb.modtime = --last_mtime;
852 if (utime(pack_tmp_name, &utb) < 0)
853 warning_errno("failed utime() on %s", pack_tmp_name);
854 }
855
856 strbuf_addf(&tmpname, "%s-", base_name);
857
858 if (write_bitmap_index) {
859 bitmap_writer_set_checksum(sha1);
860 bitmap_writer_build_type_index(written_list, nr_written);
861 }
862
863 finish_tmp_packfile(&tmpname, pack_tmp_name,
864 written_list, nr_written,
865 &pack_idx_opts, sha1);
866
867 if (write_bitmap_index) {
868 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
869
870 stop_progress(&progress_state);
871
872 bitmap_writer_show_progress(progress);
873 bitmap_writer_reuse_bitmaps(&to_pack);
874 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
875 bitmap_writer_build(&to_pack);
876 bitmap_writer_finish(written_list, nr_written,
877 tmpname.buf, write_bitmap_options);
878 write_bitmap_index = 0;
879 }
880
881 strbuf_release(&tmpname);
882 free(pack_tmp_name);
883 puts(sha1_to_hex(sha1));
884 }
885
886 /* mark written objects as written to previous pack */
887 for (j = 0; j < nr_written; j++) {
888 written_list[j]->offset = (off_t)-1;
889 }
890 nr_remaining -= nr_written;
891 } while (nr_remaining && i < to_pack.nr_objects);
892
893 free(written_list);
894 free(write_order);
895 stop_progress(&progress_state);
896 if (written != nr_result)
897 die("wrote %"PRIu32" objects while expecting %"PRIu32,
898 written, nr_result);
899 }
900
901 static int no_try_delta(const char *path)
902 {
903 static struct attr_check *check;
904
905 if (!check)
906 check = attr_check_initl("delta", NULL);
907 if (git_check_attr(path, check))
908 return 0;
909 if (ATTR_FALSE(check->items[0].value))
910 return 1;
911 return 0;
912 }
913
914 /*
915 * When adding an object, check whether we have already added it
916 * to our packing list. If so, we can skip. However, if we are
917 * being asked to excludei t, but the previous mention was to include
918 * it, make sure to adjust its flags and tweak our numbers accordingly.
919 *
920 * As an optimization, we pass out the index position where we would have
921 * found the item, since that saves us from having to look it up again a
922 * few lines later when we want to add the new entry.
923 */
924 static int have_duplicate_entry(const unsigned char *sha1,
925 int exclude,
926 uint32_t *index_pos)
927 {
928 struct object_entry *entry;
929
930 entry = packlist_find(&to_pack, sha1, index_pos);
931 if (!entry)
932 return 0;
933
934 if (exclude) {
935 if (!entry->preferred_base)
936 nr_result--;
937 entry->preferred_base = 1;
938 }
939
940 return 1;
941 }
942
943 static int want_found_object(int exclude, struct packed_git *p)
944 {
945 if (exclude)
946 return 1;
947 if (incremental)
948 return 0;
949
950 /*
951 * When asked to do --local (do not include an object that appears in a
952 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
953 * an object that appears in a pack marked with .keep), finding a pack
954 * that matches the criteria is sufficient for us to decide to omit it.
955 * However, even if this pack does not satisfy the criteria, we need to
956 * make sure no copy of this object appears in _any_ pack that makes us
957 * to omit the object, so we need to check all the packs.
958 *
959 * We can however first check whether these options can possible matter;
960 * if they do not matter we know we want the object in generated pack.
961 * Otherwise, we signal "-1" at the end to tell the caller that we do
962 * not know either way, and it needs to check more packs.
963 */
964 if (!ignore_packed_keep &&
965 (!local || !have_non_local_packs))
966 return 1;
967
968 if (local && !p->pack_local)
969 return 0;
970 if (ignore_packed_keep && p->pack_local && p->pack_keep)
971 return 0;
972
973 /* we don't know yet; keep looking for more packs */
974 return -1;
975 }
976
977 /*
978 * Check whether we want the object in the pack (e.g., we do not want
979 * objects found in non-local stores if the "--local" option was used).
980 *
981 * If the caller already knows an existing pack it wants to take the object
982 * from, that is passed in *found_pack and *found_offset; otherwise this
983 * function finds if there is any pack that has the object and returns the pack
984 * and its offset in these variables.
985 */
986 static int want_object_in_pack(const unsigned char *sha1,
987 int exclude,
988 struct packed_git **found_pack,
989 off_t *found_offset)
990 {
991 struct mru_entry *entry;
992 int want;
993
994 if (!exclude && local && has_loose_object_nonlocal(sha1))
995 return 0;
996
997 /*
998 * If we already know the pack object lives in, start checks from that
999 * pack - in the usual case when neither --local was given nor .keep files
1000 * are present we will determine the answer right now.
1001 */
1002 if (*found_pack) {
1003 want = want_found_object(exclude, *found_pack);
1004 if (want != -1)
1005 return want;
1006 }
1007
1008 for (entry = packed_git_mru->head; entry; entry = entry->next) {
1009 struct packed_git *p = entry->item;
1010 off_t offset;
1011
1012 if (p == *found_pack)
1013 offset = *found_offset;
1014 else
1015 offset = find_pack_entry_one(sha1, p);
1016
1017 if (offset) {
1018 if (!*found_pack) {
1019 if (!is_pack_valid(p))
1020 continue;
1021 *found_offset = offset;
1022 *found_pack = p;
1023 }
1024 want = want_found_object(exclude, p);
1025 if (!exclude && want > 0)
1026 mru_mark(packed_git_mru, entry);
1027 if (want != -1)
1028 return want;
1029 }
1030 }
1031
1032 return 1;
1033 }
1034
1035 static void create_object_entry(const unsigned char *sha1,
1036 enum object_type type,
1037 uint32_t hash,
1038 int exclude,
1039 int no_try_delta,
1040 uint32_t index_pos,
1041 struct packed_git *found_pack,
1042 off_t found_offset)
1043 {
1044 struct object_entry *entry;
1045
1046 entry = packlist_alloc(&to_pack, sha1, index_pos);
1047 entry->hash = hash;
1048 if (type)
1049 entry->type = type;
1050 if (exclude)
1051 entry->preferred_base = 1;
1052 else
1053 nr_result++;
1054 if (found_pack) {
1055 entry->in_pack = found_pack;
1056 entry->in_pack_offset = found_offset;
1057 }
1058
1059 entry->no_try_delta = no_try_delta;
1060 }
1061
1062 static const char no_closure_warning[] = N_(
1063 "disabling bitmap writing, as some objects are not being packed"
1064 );
1065
1066 static int add_object_entry(const unsigned char *sha1, enum object_type type,
1067 const char *name, int exclude)
1068 {
1069 struct packed_git *found_pack = NULL;
1070 off_t found_offset = 0;
1071 uint32_t index_pos;
1072
1073 if (have_duplicate_entry(sha1, exclude, &index_pos))
1074 return 0;
1075
1076 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1077 /* The pack is missing an object, so it will not have closure */
1078 if (write_bitmap_index) {
1079 warning(_(no_closure_warning));
1080 write_bitmap_index = 0;
1081 }
1082 return 0;
1083 }
1084
1085 create_object_entry(sha1, type, pack_name_hash(name),
1086 exclude, name && no_try_delta(name),
1087 index_pos, found_pack, found_offset);
1088
1089 display_progress(progress_state, nr_result);
1090 return 1;
1091 }
1092
1093 static int add_object_entry_from_bitmap(const unsigned char *sha1,
1094 enum object_type type,
1095 int flags, uint32_t name_hash,
1096 struct packed_git *pack, off_t offset)
1097 {
1098 uint32_t index_pos;
1099
1100 if (have_duplicate_entry(sha1, 0, &index_pos))
1101 return 0;
1102
1103 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1104 return 0;
1105
1106 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1107
1108 display_progress(progress_state, nr_result);
1109 return 1;
1110 }
1111
1112 struct pbase_tree_cache {
1113 unsigned char sha1[20];
1114 int ref;
1115 int temporary;
1116 void *tree_data;
1117 unsigned long tree_size;
1118 };
1119
1120 static struct pbase_tree_cache *(pbase_tree_cache[256]);
1121 static int pbase_tree_cache_ix(const unsigned char *sha1)
1122 {
1123 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1124 }
1125 static int pbase_tree_cache_ix_incr(int ix)
1126 {
1127 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1128 }
1129
1130 static struct pbase_tree {
1131 struct pbase_tree *next;
1132 /* This is a phony "cache" entry; we are not
1133 * going to evict it or find it through _get()
1134 * mechanism -- this is for the toplevel node that
1135 * would almost always change with any commit.
1136 */
1137 struct pbase_tree_cache pcache;
1138 } *pbase_tree;
1139
1140 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1141 {
1142 struct pbase_tree_cache *ent, *nent;
1143 void *data;
1144 unsigned long size;
1145 enum object_type type;
1146 int neigh;
1147 int my_ix = pbase_tree_cache_ix(sha1);
1148 int available_ix = -1;
1149
1150 /* pbase-tree-cache acts as a limited hashtable.
1151 * your object will be found at your index or within a few
1152 * slots after that slot if it is cached.
1153 */
1154 for (neigh = 0; neigh < 8; neigh++) {
1155 ent = pbase_tree_cache[my_ix];
1156 if (ent && !hashcmp(ent->sha1, sha1)) {
1157 ent->ref++;
1158 return ent;
1159 }
1160 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1161 ((0 <= available_ix) &&
1162 (!ent && pbase_tree_cache[available_ix])))
1163 available_ix = my_ix;
1164 if (!ent)
1165 break;
1166 my_ix = pbase_tree_cache_ix_incr(my_ix);
1167 }
1168
1169 /* Did not find one. Either we got a bogus request or
1170 * we need to read and perhaps cache.
1171 */
1172 data = read_sha1_file(sha1, &type, &size);
1173 if (!data)
1174 return NULL;
1175 if (type != OBJ_TREE) {
1176 free(data);
1177 return NULL;
1178 }
1179
1180 /* We need to either cache or return a throwaway copy */
1181
1182 if (available_ix < 0)
1183 ent = NULL;
1184 else {
1185 ent = pbase_tree_cache[available_ix];
1186 my_ix = available_ix;
1187 }
1188
1189 if (!ent) {
1190 nent = xmalloc(sizeof(*nent));
1191 nent->temporary = (available_ix < 0);
1192 }
1193 else {
1194 /* evict and reuse */
1195 free(ent->tree_data);
1196 nent = ent;
1197 }
1198 hashcpy(nent->sha1, sha1);
1199 nent->tree_data = data;
1200 nent->tree_size = size;
1201 nent->ref = 1;
1202 if (!nent->temporary)
1203 pbase_tree_cache[my_ix] = nent;
1204 return nent;
1205 }
1206
1207 static void pbase_tree_put(struct pbase_tree_cache *cache)
1208 {
1209 if (!cache->temporary) {
1210 cache->ref--;
1211 return;
1212 }
1213 free(cache->tree_data);
1214 free(cache);
1215 }
1216
1217 static int name_cmp_len(const char *name)
1218 {
1219 int i;
1220 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1221 ;
1222 return i;
1223 }
1224
1225 static void add_pbase_object(struct tree_desc *tree,
1226 const char *name,
1227 int cmplen,
1228 const char *fullname)
1229 {
1230 struct name_entry entry;
1231 int cmp;
1232
1233 while (tree_entry(tree,&entry)) {
1234 if (S_ISGITLINK(entry.mode))
1235 continue;
1236 cmp = tree_entry_len(&entry) != cmplen ? 1 :
1237 memcmp(name, entry.path, cmplen);
1238 if (cmp > 0)
1239 continue;
1240 if (cmp < 0)
1241 return;
1242 if (name[cmplen] != '/') {
1243 add_object_entry(entry.oid->hash,
1244 object_type(entry.mode),
1245 fullname, 1);
1246 return;
1247 }
1248 if (S_ISDIR(entry.mode)) {
1249 struct tree_desc sub;
1250 struct pbase_tree_cache *tree;
1251 const char *down = name+cmplen+1;
1252 int downlen = name_cmp_len(down);
1253
1254 tree = pbase_tree_get(entry.oid->hash);
1255 if (!tree)
1256 return;
1257 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1258
1259 add_pbase_object(&sub, down, downlen, fullname);
1260 pbase_tree_put(tree);
1261 }
1262 }
1263 }
1264
1265 static unsigned *done_pbase_paths;
1266 static int done_pbase_paths_num;
1267 static int done_pbase_paths_alloc;
1268 static int done_pbase_path_pos(unsigned hash)
1269 {
1270 int lo = 0;
1271 int hi = done_pbase_paths_num;
1272 while (lo < hi) {
1273 int mi = (hi + lo) / 2;
1274 if (done_pbase_paths[mi] == hash)
1275 return mi;
1276 if (done_pbase_paths[mi] < hash)
1277 hi = mi;
1278 else
1279 lo = mi + 1;
1280 }
1281 return -lo-1;
1282 }
1283
1284 static int check_pbase_path(unsigned hash)
1285 {
1286 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1287 if (0 <= pos)
1288 return 1;
1289 pos = -pos - 1;
1290 ALLOC_GROW(done_pbase_paths,
1291 done_pbase_paths_num + 1,
1292 done_pbase_paths_alloc);
1293 done_pbase_paths_num++;
1294 if (pos < done_pbase_paths_num)
1295 memmove(done_pbase_paths + pos + 1,
1296 done_pbase_paths + pos,
1297 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1298 done_pbase_paths[pos] = hash;
1299 return 0;
1300 }
1301
1302 static void add_preferred_base_object(const char *name)
1303 {
1304 struct pbase_tree *it;
1305 int cmplen;
1306 unsigned hash = pack_name_hash(name);
1307
1308 if (!num_preferred_base || check_pbase_path(hash))
1309 return;
1310
1311 cmplen = name_cmp_len(name);
1312 for (it = pbase_tree; it; it = it->next) {
1313 if (cmplen == 0) {
1314 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1315 }
1316 else {
1317 struct tree_desc tree;
1318 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1319 add_pbase_object(&tree, name, cmplen, name);
1320 }
1321 }
1322 }
1323
1324 static void add_preferred_base(unsigned char *sha1)
1325 {
1326 struct pbase_tree *it;
1327 void *data;
1328 unsigned long size;
1329 unsigned char tree_sha1[20];
1330
1331 if (window <= num_preferred_base++)
1332 return;
1333
1334 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1335 if (!data)
1336 return;
1337
1338 for (it = pbase_tree; it; it = it->next) {
1339 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1340 free(data);
1341 return;
1342 }
1343 }
1344
1345 it = xcalloc(1, sizeof(*it));
1346 it->next = pbase_tree;
1347 pbase_tree = it;
1348
1349 hashcpy(it->pcache.sha1, tree_sha1);
1350 it->pcache.tree_data = data;
1351 it->pcache.tree_size = size;
1352 }
1353
1354 static void cleanup_preferred_base(void)
1355 {
1356 struct pbase_tree *it;
1357 unsigned i;
1358
1359 it = pbase_tree;
1360 pbase_tree = NULL;
1361 while (it) {
1362 struct pbase_tree *this = it;
1363 it = this->next;
1364 free(this->pcache.tree_data);
1365 free(this);
1366 }
1367
1368 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1369 if (!pbase_tree_cache[i])
1370 continue;
1371 free(pbase_tree_cache[i]->tree_data);
1372 free(pbase_tree_cache[i]);
1373 pbase_tree_cache[i] = NULL;
1374 }
1375
1376 free(done_pbase_paths);
1377 done_pbase_paths = NULL;
1378 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1379 }
1380
1381 static void check_object(struct object_entry *entry)
1382 {
1383 if (entry->in_pack) {
1384 struct packed_git *p = entry->in_pack;
1385 struct pack_window *w_curs = NULL;
1386 const unsigned char *base_ref = NULL;
1387 struct object_entry *base_entry;
1388 unsigned long used, used_0;
1389 unsigned long avail;
1390 off_t ofs;
1391 unsigned char *buf, c;
1392
1393 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1394
1395 /*
1396 * We want in_pack_type even if we do not reuse delta
1397 * since non-delta representations could still be reused.
1398 */
1399 used = unpack_object_header_buffer(buf, avail,
1400 &entry->in_pack_type,
1401 &entry->size);
1402 if (used == 0)
1403 goto give_up;
1404
1405 /*
1406 * Determine if this is a delta and if so whether we can
1407 * reuse it or not. Otherwise let's find out as cheaply as
1408 * possible what the actual type and size for this object is.
1409 */
1410 switch (entry->in_pack_type) {
1411 default:
1412 /* Not a delta hence we've already got all we need. */
1413 entry->type = entry->in_pack_type;
1414 entry->in_pack_header_size = used;
1415 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1416 goto give_up;
1417 unuse_pack(&w_curs);
1418 return;
1419 case OBJ_REF_DELTA:
1420 if (reuse_delta && !entry->preferred_base)
1421 base_ref = use_pack(p, &w_curs,
1422 entry->in_pack_offset + used, NULL);
1423 entry->in_pack_header_size = used + 20;
1424 break;
1425 case OBJ_OFS_DELTA:
1426 buf = use_pack(p, &w_curs,
1427 entry->in_pack_offset + used, NULL);
1428 used_0 = 0;
1429 c = buf[used_0++];
1430 ofs = c & 127;
1431 while (c & 128) {
1432 ofs += 1;
1433 if (!ofs || MSB(ofs, 7)) {
1434 error("delta base offset overflow in pack for %s",
1435 sha1_to_hex(entry->idx.sha1));
1436 goto give_up;
1437 }
1438 c = buf[used_0++];
1439 ofs = (ofs << 7) + (c & 127);
1440 }
1441 ofs = entry->in_pack_offset - ofs;
1442 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1443 error("delta base offset out of bound for %s",
1444 sha1_to_hex(entry->idx.sha1));
1445 goto give_up;
1446 }
1447 if (reuse_delta && !entry->preferred_base) {
1448 struct revindex_entry *revidx;
1449 revidx = find_pack_revindex(p, ofs);
1450 if (!revidx)
1451 goto give_up;
1452 base_ref = nth_packed_object_sha1(p, revidx->nr);
1453 }
1454 entry->in_pack_header_size = used + used_0;
1455 break;
1456 }
1457
1458 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1459 /*
1460 * If base_ref was set above that means we wish to
1461 * reuse delta data, and we even found that base
1462 * in the list of objects we want to pack. Goodie!
1463 *
1464 * Depth value does not matter - find_deltas() will
1465 * never consider reused delta as the base object to
1466 * deltify other objects against, in order to avoid
1467 * circular deltas.
1468 */
1469 entry->type = entry->in_pack_type;
1470 entry->delta = base_entry;
1471 entry->delta_size = entry->size;
1472 entry->delta_sibling = base_entry->delta_child;
1473 base_entry->delta_child = entry;
1474 unuse_pack(&w_curs);
1475 return;
1476 }
1477
1478 if (entry->type) {
1479 /*
1480 * This must be a delta and we already know what the
1481 * final object type is. Let's extract the actual
1482 * object size from the delta header.
1483 */
1484 entry->size = get_size_from_delta(p, &w_curs,
1485 entry->in_pack_offset + entry->in_pack_header_size);
1486 if (entry->size == 0)
1487 goto give_up;
1488 unuse_pack(&w_curs);
1489 return;
1490 }
1491
1492 /*
1493 * No choice but to fall back to the recursive delta walk
1494 * with sha1_object_info() to find about the object type
1495 * at this point...
1496 */
1497 give_up:
1498 unuse_pack(&w_curs);
1499 }
1500
1501 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1502 /*
1503 * The error condition is checked in prepare_pack(). This is
1504 * to permit a missing preferred base object to be ignored
1505 * as a preferred base. Doing so can result in a larger
1506 * pack file, but the transfer will still take place.
1507 */
1508 }
1509
1510 static int pack_offset_sort(const void *_a, const void *_b)
1511 {
1512 const struct object_entry *a = *(struct object_entry **)_a;
1513 const struct object_entry *b = *(struct object_entry **)_b;
1514
1515 /* avoid filesystem trashing with loose objects */
1516 if (!a->in_pack && !b->in_pack)
1517 return hashcmp(a->idx.sha1, b->idx.sha1);
1518
1519 if (a->in_pack < b->in_pack)
1520 return -1;
1521 if (a->in_pack > b->in_pack)
1522 return 1;
1523 return a->in_pack_offset < b->in_pack_offset ? -1 :
1524 (a->in_pack_offset > b->in_pack_offset);
1525 }
1526
1527 /*
1528 * Drop an on-disk delta we were planning to reuse. Naively, this would
1529 * just involve blanking out the "delta" field, but we have to deal
1530 * with some extra book-keeping:
1531 *
1532 * 1. Removing ourselves from the delta_sibling linked list.
1533 *
1534 * 2. Updating our size/type to the non-delta representation. These were
1535 * either not recorded initially (size) or overwritten with the delta type
1536 * (type) when check_object() decided to reuse the delta.
1537 *
1538 * 3. Resetting our delta depth, as we are now a base object.
1539 */
1540 static void drop_reused_delta(struct object_entry *entry)
1541 {
1542 struct object_entry **p = &entry->delta->delta_child;
1543 struct object_info oi = OBJECT_INFO_INIT;
1544
1545 while (*p) {
1546 if (*p == entry)
1547 *p = (*p)->delta_sibling;
1548 else
1549 p = &(*p)->delta_sibling;
1550 }
1551 entry->delta = NULL;
1552 entry->depth = 0;
1553
1554 oi.sizep = &entry->size;
1555 oi.typep = &entry->type;
1556 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1557 /*
1558 * We failed to get the info from this pack for some reason;
1559 * fall back to sha1_object_info, which may find another copy.
1560 * And if that fails, the error will be recorded in entry->type
1561 * and dealt with in prepare_pack().
1562 */
1563 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1564 }
1565 }
1566
1567 /*
1568 * Follow the chain of deltas from this entry onward, throwing away any links
1569 * that cause us to hit a cycle (as determined by the DFS state flags in
1570 * the entries).
1571 *
1572 * We also detect too-long reused chains that would violate our --depth
1573 * limit.
1574 */
1575 static void break_delta_chains(struct object_entry *entry)
1576 {
1577 /*
1578 * The actual depth of each object we will write is stored as an int,
1579 * as it cannot exceed our int "depth" limit. But before we break
1580 * changes based no that limit, we may potentially go as deep as the
1581 * number of objects, which is elsewhere bounded to a uint32_t.
1582 */
1583 uint32_t total_depth;
1584 struct object_entry *cur, *next;
1585
1586 for (cur = entry, total_depth = 0;
1587 cur;
1588 cur = cur->delta, total_depth++) {
1589 if (cur->dfs_state == DFS_DONE) {
1590 /*
1591 * We've already seen this object and know it isn't
1592 * part of a cycle. We do need to append its depth
1593 * to our count.
1594 */
1595 total_depth += cur->depth;
1596 break;
1597 }
1598
1599 /*
1600 * We break cycles before looping, so an ACTIVE state (or any
1601 * other cruft which made its way into the state variable)
1602 * is a bug.
1603 */
1604 if (cur->dfs_state != DFS_NONE)
1605 die("BUG: confusing delta dfs state in first pass: %d",
1606 cur->dfs_state);
1607
1608 /*
1609 * Now we know this is the first time we've seen the object. If
1610 * it's not a delta, we're done traversing, but we'll mark it
1611 * done to save time on future traversals.
1612 */
1613 if (!cur->delta) {
1614 cur->dfs_state = DFS_DONE;
1615 break;
1616 }
1617
1618 /*
1619 * Mark ourselves as active and see if the next step causes
1620 * us to cycle to another active object. It's important to do
1621 * this _before_ we loop, because it impacts where we make the
1622 * cut, and thus how our total_depth counter works.
1623 * E.g., We may see a partial loop like:
1624 *
1625 * A -> B -> C -> D -> B
1626 *
1627 * Cutting B->C breaks the cycle. But now the depth of A is
1628 * only 1, and our total_depth counter is at 3. The size of the
1629 * error is always one less than the size of the cycle we
1630 * broke. Commits C and D were "lost" from A's chain.
1631 *
1632 * If we instead cut D->B, then the depth of A is correct at 3.
1633 * We keep all commits in the chain that we examined.
1634 */
1635 cur->dfs_state = DFS_ACTIVE;
1636 if (cur->delta->dfs_state == DFS_ACTIVE) {
1637 drop_reused_delta(cur);
1638 cur->dfs_state = DFS_DONE;
1639 break;
1640 }
1641 }
1642
1643 /*
1644 * And now that we've gone all the way to the bottom of the chain, we
1645 * need to clear the active flags and set the depth fields as
1646 * appropriate. Unlike the loop above, which can quit when it drops a
1647 * delta, we need to keep going to look for more depth cuts. So we need
1648 * an extra "next" pointer to keep going after we reset cur->delta.
1649 */
1650 for (cur = entry; cur; cur = next) {
1651 next = cur->delta;
1652
1653 /*
1654 * We should have a chain of zero or more ACTIVE states down to
1655 * a final DONE. We can quit after the DONE, because either it
1656 * has no bases, or we've already handled them in a previous
1657 * call.
1658 */
1659 if (cur->dfs_state == DFS_DONE)
1660 break;
1661 else if (cur->dfs_state != DFS_ACTIVE)
1662 die("BUG: confusing delta dfs state in second pass: %d",
1663 cur->dfs_state);
1664
1665 /*
1666 * If the total_depth is more than depth, then we need to snip
1667 * the chain into two or more smaller chains that don't exceed
1668 * the maximum depth. Most of the resulting chains will contain
1669 * (depth + 1) entries (i.e., depth deltas plus one base), and
1670 * the last chain (i.e., the one containing entry) will contain
1671 * whatever entries are left over, namely
1672 * (total_depth % (depth + 1)) of them.
1673 *
1674 * Since we are iterating towards decreasing depth, we need to
1675 * decrement total_depth as we go, and we need to write to the
1676 * entry what its final depth will be after all of the
1677 * snipping. Since we're snipping into chains of length (depth
1678 * + 1) entries, the final depth of an entry will be its
1679 * original depth modulo (depth + 1). Any time we encounter an
1680 * entry whose final depth is supposed to be zero, we snip it
1681 * from its delta base, thereby making it so.
1682 */
1683 cur->depth = (total_depth--) % (depth + 1);
1684 if (!cur->depth)
1685 drop_reused_delta(cur);
1686
1687 cur->dfs_state = DFS_DONE;
1688 }
1689 }
1690
1691 static void get_object_details(void)
1692 {
1693 uint32_t i;
1694 struct object_entry **sorted_by_offset;
1695
1696 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1697 for (i = 0; i < to_pack.nr_objects; i++)
1698 sorted_by_offset[i] = to_pack.objects + i;
1699 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1700
1701 for (i = 0; i < to_pack.nr_objects; i++) {
1702 struct object_entry *entry = sorted_by_offset[i];
1703 check_object(entry);
1704 if (big_file_threshold < entry->size)
1705 entry->no_try_delta = 1;
1706 }
1707
1708 /*
1709 * This must happen in a second pass, since we rely on the delta
1710 * information for the whole list being completed.
1711 */
1712 for (i = 0; i < to_pack.nr_objects; i++)
1713 break_delta_chains(&to_pack.objects[i]);
1714
1715 free(sorted_by_offset);
1716 }
1717
1718 /*
1719 * We search for deltas in a list sorted by type, by filename hash, and then
1720 * by size, so that we see progressively smaller and smaller files.
1721 * That's because we prefer deltas to be from the bigger file
1722 * to the smaller -- deletes are potentially cheaper, but perhaps
1723 * more importantly, the bigger file is likely the more recent
1724 * one. The deepest deltas are therefore the oldest objects which are
1725 * less susceptible to be accessed often.
1726 */
1727 static int type_size_sort(const void *_a, const void *_b)
1728 {
1729 const struct object_entry *a = *(struct object_entry **)_a;
1730 const struct object_entry *b = *(struct object_entry **)_b;
1731
1732 if (a->type > b->type)
1733 return -1;
1734 if (a->type < b->type)
1735 return 1;
1736 if (a->hash > b->hash)
1737 return -1;
1738 if (a->hash < b->hash)
1739 return 1;
1740 if (a->preferred_base > b->preferred_base)
1741 return -1;
1742 if (a->preferred_base < b->preferred_base)
1743 return 1;
1744 if (a->size > b->size)
1745 return -1;
1746 if (a->size < b->size)
1747 return 1;
1748 return a < b ? -1 : (a > b); /* newest first */
1749 }
1750
1751 struct unpacked {
1752 struct object_entry *entry;
1753 void *data;
1754 struct delta_index *index;
1755 unsigned depth;
1756 };
1757
1758 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1759 unsigned long delta_size)
1760 {
1761 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1762 return 0;
1763
1764 if (delta_size < cache_max_small_delta_size)
1765 return 1;
1766
1767 /* cache delta, if objects are large enough compared to delta size */
1768 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1769 return 1;
1770
1771 return 0;
1772 }
1773
1774 #ifndef NO_PTHREADS
1775
1776 static pthread_mutex_t read_mutex;
1777 #define read_lock() pthread_mutex_lock(&read_mutex)
1778 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1779
1780 static pthread_mutex_t cache_mutex;
1781 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1782 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1783
1784 static pthread_mutex_t progress_mutex;
1785 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1786 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1787
1788 #else
1789
1790 #define read_lock() (void)0
1791 #define read_unlock() (void)0
1792 #define cache_lock() (void)0
1793 #define cache_unlock() (void)0
1794 #define progress_lock() (void)0
1795 #define progress_unlock() (void)0
1796
1797 #endif
1798
1799 static int try_delta(struct unpacked *trg, struct unpacked *src,
1800 unsigned max_depth, unsigned long *mem_usage)
1801 {
1802 struct object_entry *trg_entry = trg->entry;
1803 struct object_entry *src_entry = src->entry;
1804 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1805 unsigned ref_depth;
1806 enum object_type type;
1807 void *delta_buf;
1808
1809 /* Don't bother doing diffs between different types */
1810 if (trg_entry->type != src_entry->type)
1811 return -1;
1812
1813 /*
1814 * We do not bother to try a delta that we discarded on an
1815 * earlier try, but only when reusing delta data. Note that
1816 * src_entry that is marked as the preferred_base should always
1817 * be considered, as even if we produce a suboptimal delta against
1818 * it, we will still save the transfer cost, as we already know
1819 * the other side has it and we won't send src_entry at all.
1820 */
1821 if (reuse_delta && trg_entry->in_pack &&
1822 trg_entry->in_pack == src_entry->in_pack &&
1823 !src_entry->preferred_base &&
1824 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1825 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1826 return 0;
1827
1828 /* Let's not bust the allowed depth. */
1829 if (src->depth >= max_depth)
1830 return 0;
1831
1832 /* Now some size filtering heuristics. */
1833 trg_size = trg_entry->size;
1834 if (!trg_entry->delta) {
1835 max_size = trg_size/2 - 20;
1836 ref_depth = 1;
1837 } else {
1838 max_size = trg_entry->delta_size;
1839 ref_depth = trg->depth;
1840 }
1841 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1842 (max_depth - ref_depth + 1);
1843 if (max_size == 0)
1844 return 0;
1845 src_size = src_entry->size;
1846 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1847 if (sizediff >= max_size)
1848 return 0;
1849 if (trg_size < src_size / 32)
1850 return 0;
1851
1852 /* Load data if not already done */
1853 if (!trg->data) {
1854 read_lock();
1855 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1856 read_unlock();
1857 if (!trg->data)
1858 die("object %s cannot be read",
1859 sha1_to_hex(trg_entry->idx.sha1));
1860 if (sz != trg_size)
1861 die("object %s inconsistent object length (%lu vs %lu)",
1862 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1863 *mem_usage += sz;
1864 }
1865 if (!src->data) {
1866 read_lock();
1867 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1868 read_unlock();
1869 if (!src->data) {
1870 if (src_entry->preferred_base) {
1871 static int warned = 0;
1872 if (!warned++)
1873 warning("object %s cannot be read",
1874 sha1_to_hex(src_entry->idx.sha1));
1875 /*
1876 * Those objects are not included in the
1877 * resulting pack. Be resilient and ignore
1878 * them if they can't be read, in case the
1879 * pack could be created nevertheless.
1880 */
1881 return 0;
1882 }
1883 die("object %s cannot be read",
1884 sha1_to_hex(src_entry->idx.sha1));
1885 }
1886 if (sz != src_size)
1887 die("object %s inconsistent object length (%lu vs %lu)",
1888 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1889 *mem_usage += sz;
1890 }
1891 if (!src->index) {
1892 src->index = create_delta_index(src->data, src_size);
1893 if (!src->index) {
1894 static int warned = 0;
1895 if (!warned++)
1896 warning("suboptimal pack - out of memory");
1897 return 0;
1898 }
1899 *mem_usage += sizeof_delta_index(src->index);
1900 }
1901
1902 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1903 if (!delta_buf)
1904 return 0;
1905
1906 if (trg_entry->delta) {
1907 /* Prefer only shallower same-sized deltas. */
1908 if (delta_size == trg_entry->delta_size &&
1909 src->depth + 1 >= trg->depth) {
1910 free(delta_buf);
1911 return 0;
1912 }
1913 }
1914
1915 /*
1916 * Handle memory allocation outside of the cache
1917 * accounting lock. Compiler will optimize the strangeness
1918 * away when NO_PTHREADS is defined.
1919 */
1920 free(trg_entry->delta_data);
1921 cache_lock();
1922 if (trg_entry->delta_data) {
1923 delta_cache_size -= trg_entry->delta_size;
1924 trg_entry->delta_data = NULL;
1925 }
1926 if (delta_cacheable(src_size, trg_size, delta_size)) {
1927 delta_cache_size += delta_size;
1928 cache_unlock();
1929 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1930 } else {
1931 cache_unlock();
1932 free(delta_buf);
1933 }
1934
1935 trg_entry->delta = src_entry;
1936 trg_entry->delta_size = delta_size;
1937 trg->depth = src->depth + 1;
1938
1939 return 1;
1940 }
1941
1942 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1943 {
1944 struct object_entry *child = me->delta_child;
1945 unsigned int m = n;
1946 while (child) {
1947 unsigned int c = check_delta_limit(child, n + 1);
1948 if (m < c)
1949 m = c;
1950 child = child->delta_sibling;
1951 }
1952 return m;
1953 }
1954
1955 static unsigned long free_unpacked(struct unpacked *n)
1956 {
1957 unsigned long freed_mem = sizeof_delta_index(n->index);
1958 free_delta_index(n->index);
1959 n->index = NULL;
1960 if (n->data) {
1961 freed_mem += n->entry->size;
1962 free(n->data);
1963 n->data = NULL;
1964 }
1965 n->entry = NULL;
1966 n->depth = 0;
1967 return freed_mem;
1968 }
1969
1970 static void find_deltas(struct object_entry **list, unsigned *list_size,
1971 int window, int depth, unsigned *processed)
1972 {
1973 uint32_t i, idx = 0, count = 0;
1974 struct unpacked *array;
1975 unsigned long mem_usage = 0;
1976
1977 array = xcalloc(window, sizeof(struct unpacked));
1978
1979 for (;;) {
1980 struct object_entry *entry;
1981 struct unpacked *n = array + idx;
1982 int j, max_depth, best_base = -1;
1983
1984 progress_lock();
1985 if (!*list_size) {
1986 progress_unlock();
1987 break;
1988 }
1989 entry = *list++;
1990 (*list_size)--;
1991 if (!entry->preferred_base) {
1992 (*processed)++;
1993 display_progress(progress_state, *processed);
1994 }
1995 progress_unlock();
1996
1997 mem_usage -= free_unpacked(n);
1998 n->entry = entry;
1999
2000 while (window_memory_limit &&
2001 mem_usage > window_memory_limit &&
2002 count > 1) {
2003 uint32_t tail = (idx + window - count) % window;
2004 mem_usage -= free_unpacked(array + tail);
2005 count--;
2006 }
2007
2008 /* We do not compute delta to *create* objects we are not
2009 * going to pack.
2010 */
2011 if (entry->preferred_base)
2012 goto next;
2013
2014 /*
2015 * If the current object is at pack edge, take the depth the
2016 * objects that depend on the current object into account
2017 * otherwise they would become too deep.
2018 */
2019 max_depth = depth;
2020 if (entry->delta_child) {
2021 max_depth -= check_delta_limit(entry, 0);
2022 if (max_depth <= 0)
2023 goto next;
2024 }
2025
2026 j = window;
2027 while (--j > 0) {
2028 int ret;
2029 uint32_t other_idx = idx + j;
2030 struct unpacked *m;
2031 if (other_idx >= window)
2032 other_idx -= window;
2033 m = array + other_idx;
2034 if (!m->entry)
2035 break;
2036 ret = try_delta(n, m, max_depth, &mem_usage);
2037 if (ret < 0)
2038 break;
2039 else if (ret > 0)
2040 best_base = other_idx;
2041 }
2042
2043 /*
2044 * If we decided to cache the delta data, then it is best
2045 * to compress it right away. First because we have to do
2046 * it anyway, and doing it here while we're threaded will
2047 * save a lot of time in the non threaded write phase,
2048 * as well as allow for caching more deltas within
2049 * the same cache size limit.
2050 * ...
2051 * But only if not writing to stdout, since in that case
2052 * the network is most likely throttling writes anyway,
2053 * and therefore it is best to go to the write phase ASAP
2054 * instead, as we can afford spending more time compressing
2055 * between writes at that moment.
2056 */
2057 if (entry->delta_data && !pack_to_stdout) {
2058 entry->z_delta_size = do_compress(&entry->delta_data,
2059 entry->delta_size);
2060 cache_lock();
2061 delta_cache_size -= entry->delta_size;
2062 delta_cache_size += entry->z_delta_size;
2063 cache_unlock();
2064 }
2065
2066 /* if we made n a delta, and if n is already at max
2067 * depth, leaving it in the window is pointless. we
2068 * should evict it first.
2069 */
2070 if (entry->delta && max_depth <= n->depth)
2071 continue;
2072
2073 /*
2074 * Move the best delta base up in the window, after the
2075 * currently deltified object, to keep it longer. It will
2076 * be the first base object to be attempted next.
2077 */
2078 if (entry->delta) {
2079 struct unpacked swap = array[best_base];
2080 int dist = (window + idx - best_base) % window;
2081 int dst = best_base;
2082 while (dist--) {
2083 int src = (dst + 1) % window;
2084 array[dst] = array[src];
2085 dst = src;
2086 }
2087 array[dst] = swap;
2088 }
2089
2090 next:
2091 idx++;
2092 if (count + 1 < window)
2093 count++;
2094 if (idx >= window)
2095 idx = 0;
2096 }
2097
2098 for (i = 0; i < window; ++i) {
2099 free_delta_index(array[i].index);
2100 free(array[i].data);
2101 }
2102 free(array);
2103 }
2104
2105 #ifndef NO_PTHREADS
2106
2107 static void try_to_free_from_threads(size_t size)
2108 {
2109 read_lock();
2110 release_pack_memory(size);
2111 read_unlock();
2112 }
2113
2114 static try_to_free_t old_try_to_free_routine;
2115
2116 /*
2117 * The main thread waits on the condition that (at least) one of the workers
2118 * has stopped working (which is indicated in the .working member of
2119 * struct thread_params).
2120 * When a work thread has completed its work, it sets .working to 0 and
2121 * signals the main thread and waits on the condition that .data_ready
2122 * becomes 1.
2123 */
2124
2125 struct thread_params {
2126 pthread_t thread;
2127 struct object_entry **list;
2128 unsigned list_size;
2129 unsigned remaining;
2130 int window;
2131 int depth;
2132 int working;
2133 int data_ready;
2134 pthread_mutex_t mutex;
2135 pthread_cond_t cond;
2136 unsigned *processed;
2137 };
2138
2139 static pthread_cond_t progress_cond;
2140
2141 /*
2142 * Mutex and conditional variable can't be statically-initialized on Windows.
2143 */
2144 static void init_threaded_search(void)
2145 {
2146 init_recursive_mutex(&read_mutex);
2147 pthread_mutex_init(&cache_mutex, NULL);
2148 pthread_mutex_init(&progress_mutex, NULL);
2149 pthread_cond_init(&progress_cond, NULL);
2150 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2151 }
2152
2153 static void cleanup_threaded_search(void)
2154 {
2155 set_try_to_free_routine(old_try_to_free_routine);
2156 pthread_cond_destroy(&progress_cond);
2157 pthread_mutex_destroy(&read_mutex);
2158 pthread_mutex_destroy(&cache_mutex);
2159 pthread_mutex_destroy(&progress_mutex);
2160 }
2161
2162 static void *threaded_find_deltas(void *arg)
2163 {
2164 struct thread_params *me = arg;
2165
2166 while (me->remaining) {
2167 find_deltas(me->list, &me->remaining,
2168 me->window, me->depth, me->processed);
2169
2170 progress_lock();
2171 me->working = 0;
2172 pthread_cond_signal(&progress_cond);
2173 progress_unlock();
2174
2175 /*
2176 * We must not set ->data_ready before we wait on the
2177 * condition because the main thread may have set it to 1
2178 * before we get here. In order to be sure that new
2179 * work is available if we see 1 in ->data_ready, it
2180 * was initialized to 0 before this thread was spawned
2181 * and we reset it to 0 right away.
2182 */
2183 pthread_mutex_lock(&me->mutex);
2184 while (!me->data_ready)
2185 pthread_cond_wait(&me->cond, &me->mutex);
2186 me->data_ready = 0;
2187 pthread_mutex_unlock(&me->mutex);
2188 }
2189 /* leave ->working 1 so that this doesn't get more work assigned */
2190 return NULL;
2191 }
2192
2193 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2194 int window, int depth, unsigned *processed)
2195 {
2196 struct thread_params *p;
2197 int i, ret, active_threads = 0;
2198
2199 init_threaded_search();
2200
2201 if (delta_search_threads <= 1) {
2202 find_deltas(list, &list_size, window, depth, processed);
2203 cleanup_threaded_search();
2204 return;
2205 }
2206 if (progress > pack_to_stdout)
2207 fprintf(stderr, "Delta compression using up to %d threads.\n",
2208 delta_search_threads);
2209 p = xcalloc(delta_search_threads, sizeof(*p));
2210
2211 /* Partition the work amongst work threads. */
2212 for (i = 0; i < delta_search_threads; i++) {
2213 unsigned sub_size = list_size / (delta_search_threads - i);
2214
2215 /* don't use too small segments or no deltas will be found */
2216 if (sub_size < 2*window && i+1 < delta_search_threads)
2217 sub_size = 0;
2218
2219 p[i].window = window;
2220 p[i].depth = depth;
2221 p[i].processed = processed;
2222 p[i].working = 1;
2223 p[i].data_ready = 0;
2224
2225 /* try to split chunks on "path" boundaries */
2226 while (sub_size && sub_size < list_size &&
2227 list[sub_size]->hash &&
2228 list[sub_size]->hash == list[sub_size-1]->hash)
2229 sub_size++;
2230
2231 p[i].list = list;
2232 p[i].list_size = sub_size;
2233 p[i].remaining = sub_size;
2234
2235 list += sub_size;
2236 list_size -= sub_size;
2237 }
2238
2239 /* Start work threads. */
2240 for (i = 0; i < delta_search_threads; i++) {
2241 if (!p[i].list_size)
2242 continue;
2243 pthread_mutex_init(&p[i].mutex, NULL);
2244 pthread_cond_init(&p[i].cond, NULL);
2245 ret = pthread_create(&p[i].thread, NULL,
2246 threaded_find_deltas, &p[i]);
2247 if (ret)
2248 die("unable to create thread: %s", strerror(ret));
2249 active_threads++;
2250 }
2251
2252 /*
2253 * Now let's wait for work completion. Each time a thread is done
2254 * with its work, we steal half of the remaining work from the
2255 * thread with the largest number of unprocessed objects and give
2256 * it to that newly idle thread. This ensure good load balancing
2257 * until the remaining object list segments are simply too short
2258 * to be worth splitting anymore.
2259 */
2260 while (active_threads) {
2261 struct thread_params *target = NULL;
2262 struct thread_params *victim = NULL;
2263 unsigned sub_size = 0;
2264
2265 progress_lock();
2266 for (;;) {
2267 for (i = 0; !target && i < delta_search_threads; i++)
2268 if (!p[i].working)
2269 target = &p[i];
2270 if (target)
2271 break;
2272 pthread_cond_wait(&progress_cond, &progress_mutex);
2273 }
2274
2275 for (i = 0; i < delta_search_threads; i++)
2276 if (p[i].remaining > 2*window &&
2277 (!victim || victim->remaining < p[i].remaining))
2278 victim = &p[i];
2279 if (victim) {
2280 sub_size = victim->remaining / 2;
2281 list = victim->list + victim->list_size - sub_size;
2282 while (sub_size && list[0]->hash &&
2283 list[0]->hash == list[-1]->hash) {
2284 list++;
2285 sub_size--;
2286 }
2287 if (!sub_size) {
2288 /*
2289 * It is possible for some "paths" to have
2290 * so many objects that no hash boundary
2291 * might be found. Let's just steal the
2292 * exact half in that case.
2293 */
2294 sub_size = victim->remaining / 2;
2295 list -= sub_size;
2296 }
2297 target->list = list;
2298 victim->list_size -= sub_size;
2299 victim->remaining -= sub_size;
2300 }
2301 target->list_size = sub_size;
2302 target->remaining = sub_size;
2303 target->working = 1;
2304 progress_unlock();
2305
2306 pthread_mutex_lock(&target->mutex);
2307 target->data_ready = 1;
2308 pthread_cond_signal(&target->cond);
2309 pthread_mutex_unlock(&target->mutex);
2310
2311 if (!sub_size) {
2312 pthread_join(target->thread, NULL);
2313 pthread_cond_destroy(&target->cond);
2314 pthread_mutex_destroy(&target->mutex);
2315 active_threads--;
2316 }
2317 }
2318 cleanup_threaded_search();
2319 free(p);
2320 }
2321
2322 #else
2323 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
2324 #endif
2325
2326 static void add_tag_chain(const struct object_id *oid)
2327 {
2328 struct tag *tag;
2329
2330 /*
2331 * We catch duplicates already in add_object_entry(), but we'd
2332 * prefer to do this extra check to avoid having to parse the
2333 * tag at all if we already know that it's being packed (e.g., if
2334 * it was included via bitmaps, we would not have parsed it
2335 * previously).
2336 */
2337 if (packlist_find(&to_pack, oid->hash, NULL))
2338 return;
2339
2340 tag = lookup_tag(oid->hash);
2341 while (1) {
2342 if (!tag || parse_tag(tag) || !tag->tagged)
2343 die("unable to pack objects reachable from tag %s",
2344 oid_to_hex(oid));
2345
2346 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2347
2348 if (tag->tagged->type != OBJ_TAG)
2349 return;
2350
2351 tag = (struct tag *)tag->tagged;
2352 }
2353 }
2354
2355 static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2356 {
2357 struct object_id peeled;
2358
2359 if (starts_with(path, "refs/tags/") && /* is a tag? */
2360 !peel_ref(path, peeled.hash) && /* peelable? */
2361 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2362 add_tag_chain(oid);
2363 return 0;
2364 }
2365
2366 static void prepare_pack(int window, int depth)
2367 {
2368 struct object_entry **delta_list;
2369 uint32_t i, nr_deltas;
2370 unsigned n;
2371
2372 get_object_details();
2373
2374 /*
2375 * If we're locally repacking then we need to be doubly careful
2376 * from now on in order to make sure no stealth corruption gets
2377 * propagated to the new pack. Clients receiving streamed packs
2378 * should validate everything they get anyway so no need to incur
2379 * the additional cost here in that case.
2380 */
2381 if (!pack_to_stdout)
2382 do_check_packed_object_crc = 1;
2383
2384 if (!to_pack.nr_objects || !window || !depth)
2385 return;
2386
2387 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2388 nr_deltas = n = 0;
2389
2390 for (i = 0; i < to_pack.nr_objects; i++) {
2391 struct object_entry *entry = to_pack.objects + i;
2392
2393 if (entry->delta)
2394 /* This happens if we decided to reuse existing
2395 * delta from a pack. "reuse_delta &&" is implied.
2396 */
2397 continue;
2398
2399 if (entry->size < 50)
2400 continue;
2401
2402 if (entry->no_try_delta)
2403 continue;
2404
2405 if (!entry->preferred_base) {
2406 nr_deltas++;
2407 if (entry->type < 0)
2408 die("unable to get type of object %s",
2409 sha1_to_hex(entry->idx.sha1));
2410 } else {
2411 if (entry->type < 0) {
2412 /*
2413 * This object is not found, but we
2414 * don't have to include it anyway.
2415 */
2416 continue;
2417 }
2418 }
2419
2420 delta_list[n++] = entry;
2421 }
2422
2423 if (nr_deltas && n > 1) {
2424 unsigned nr_done = 0;
2425 if (progress)
2426 progress_state = start_progress(_("Compressing objects"),
2427 nr_deltas);
2428 QSORT(delta_list, n, type_size_sort);
2429 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2430 stop_progress(&progress_state);
2431 if (nr_done != nr_deltas)
2432 die("inconsistency with delta count");
2433 }
2434 free(delta_list);
2435 }
2436
2437 static int git_pack_config(const char *k, const char *v, void *cb)
2438 {
2439 if (!strcmp(k, "pack.window")) {
2440 window = git_config_int(k, v);
2441 return 0;
2442 }
2443 if (!strcmp(k, "pack.windowmemory")) {
2444 window_memory_limit = git_config_ulong(k, v);
2445 return 0;
2446 }
2447 if (!strcmp(k, "pack.depth")) {
2448 depth = git_config_int(k, v);
2449 return 0;
2450 }
2451 if (!strcmp(k, "pack.deltacachesize")) {
2452 max_delta_cache_size = git_config_int(k, v);
2453 return 0;
2454 }
2455 if (!strcmp(k, "pack.deltacachelimit")) {
2456 cache_max_small_delta_size = git_config_int(k, v);
2457 return 0;
2458 }
2459 if (!strcmp(k, "pack.writebitmaphashcache")) {
2460 if (git_config_bool(k, v))
2461 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2462 else
2463 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2464 }
2465 if (!strcmp(k, "pack.usebitmaps")) {
2466 use_bitmap_index_default = git_config_bool(k, v);
2467 return 0;
2468 }
2469 if (!strcmp(k, "pack.threads")) {
2470 delta_search_threads = git_config_int(k, v);
2471 if (delta_search_threads < 0)
2472 die("invalid number of threads specified (%d)",
2473 delta_search_threads);
2474 #ifdef NO_PTHREADS
2475 if (delta_search_threads != 1)
2476 warning("no threads support, ignoring %s", k);
2477 #endif
2478 return 0;
2479 }
2480 if (!strcmp(k, "pack.indexversion")) {
2481 pack_idx_opts.version = git_config_int(k, v);
2482 if (pack_idx_opts.version > 2)
2483 die("bad pack.indexversion=%"PRIu32,
2484 pack_idx_opts.version);
2485 return 0;
2486 }
2487 return git_default_config(k, v, cb);
2488 }
2489
2490 static void read_object_list_from_stdin(void)
2491 {
2492 char line[40 + 1 + PATH_MAX + 2];
2493 unsigned char sha1[20];
2494
2495 for (;;) {
2496 if (!fgets(line, sizeof(line), stdin)) {
2497 if (feof(stdin))
2498 break;
2499 if (!ferror(stdin))
2500 die("fgets returned NULL, not EOF, not error!");
2501 if (errno != EINTR)
2502 die_errno("fgets");
2503 clearerr(stdin);
2504 continue;
2505 }
2506 if (line[0] == '-') {
2507 if (get_sha1_hex(line+1, sha1))
2508 die("expected edge sha1, got garbage:\n %s",
2509 line);
2510 add_preferred_base(sha1);
2511 continue;
2512 }
2513 if (get_sha1_hex(line, sha1))
2514 die("expected sha1, got garbage:\n %s", line);
2515
2516 add_preferred_base_object(line+41);
2517 add_object_entry(sha1, 0, line+41, 0);
2518 }
2519 }
2520
2521 #define OBJECT_ADDED (1u<<20)
2522
2523 static void show_commit(struct commit *commit, void *data)
2524 {
2525 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2526 commit->object.flags |= OBJECT_ADDED;
2527
2528 if (write_bitmap_index)
2529 index_commit_for_bitmap(commit);
2530 }
2531
2532 static void show_object(struct object *obj, const char *name, void *data)
2533 {
2534 add_preferred_base_object(name);
2535 add_object_entry(obj->oid.hash, obj->type, name, 0);
2536 obj->flags |= OBJECT_ADDED;
2537 }
2538
2539 static void show_edge(struct commit *commit)
2540 {
2541 add_preferred_base(commit->object.oid.hash);
2542 }
2543
2544 struct in_pack_object {
2545 off_t offset;
2546 struct object *object;
2547 };
2548
2549 struct in_pack {
2550 int alloc;
2551 int nr;
2552 struct in_pack_object *array;
2553 };
2554
2555 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2556 {
2557 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2558 in_pack->array[in_pack->nr].object = object;
2559 in_pack->nr++;
2560 }
2561
2562 /*
2563 * Compare the objects in the offset order, in order to emulate the
2564 * "git rev-list --objects" output that produced the pack originally.
2565 */
2566 static int ofscmp(const void *a_, const void *b_)
2567 {
2568 struct in_pack_object *a = (struct in_pack_object *)a_;
2569 struct in_pack_object *b = (struct in_pack_object *)b_;
2570
2571 if (a->offset < b->offset)
2572 return -1;
2573 else if (a->offset > b->offset)
2574 return 1;
2575 else
2576 return oidcmp(&a->object->oid, &b->object->oid);
2577 }
2578
2579 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2580 {
2581 struct packed_git *p;
2582 struct in_pack in_pack;
2583 uint32_t i;
2584
2585 memset(&in_pack, 0, sizeof(in_pack));
2586
2587 for (p = packed_git; p; p = p->next) {
2588 const unsigned char *sha1;
2589 struct object *o;
2590
2591 if (!p->pack_local || p->pack_keep)
2592 continue;
2593 if (open_pack_index(p))
2594 die("cannot open pack index");
2595
2596 ALLOC_GROW(in_pack.array,
2597 in_pack.nr + p->num_objects,
2598 in_pack.alloc);
2599
2600 for (i = 0; i < p->num_objects; i++) {
2601 sha1 = nth_packed_object_sha1(p, i);
2602 o = lookup_unknown_object(sha1);
2603 if (!(o->flags & OBJECT_ADDED))
2604 mark_in_pack_object(o, p, &in_pack);
2605 o->flags |= OBJECT_ADDED;
2606 }
2607 }
2608
2609 if (in_pack.nr) {
2610 QSORT(in_pack.array, in_pack.nr, ofscmp);
2611 for (i = 0; i < in_pack.nr; i++) {
2612 struct object *o = in_pack.array[i].object;
2613 add_object_entry(o->oid.hash, o->type, "", 0);
2614 }
2615 }
2616 free(in_pack.array);
2617 }
2618
2619 static int add_loose_object(const struct object_id *oid, const char *path,
2620 void *data)
2621 {
2622 enum object_type type = sha1_object_info(oid->hash, NULL);
2623
2624 if (type < 0) {
2625 warning("loose object at %s could not be examined", path);
2626 return 0;
2627 }
2628
2629 add_object_entry(oid->hash, type, "", 0);
2630 return 0;
2631 }
2632
2633 /*
2634 * We actually don't even have to worry about reachability here.
2635 * add_object_entry will weed out duplicates, so we just add every
2636 * loose object we find.
2637 */
2638 static void add_unreachable_loose_objects(void)
2639 {
2640 for_each_loose_file_in_objdir(get_object_directory(),
2641 add_loose_object,
2642 NULL, NULL, NULL);
2643 }
2644
2645 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2646 {
2647 static struct packed_git *last_found = (void *)1;
2648 struct packed_git *p;
2649
2650 p = (last_found != (void *)1) ? last_found : packed_git;
2651
2652 while (p) {
2653 if ((!p->pack_local || p->pack_keep) &&
2654 find_pack_entry_one(sha1, p)) {
2655 last_found = p;
2656 return 1;
2657 }
2658 if (p == last_found)
2659 p = packed_git;
2660 else
2661 p = p->next;
2662 if (p == last_found)
2663 p = p->next;
2664 }
2665 return 0;
2666 }
2667
2668 /*
2669 * Store a list of sha1s that are should not be discarded
2670 * because they are either written too recently, or are
2671 * reachable from another object that was.
2672 *
2673 * This is filled by get_object_list.
2674 */
2675 static struct oid_array recent_objects;
2676
2677 static int loosened_object_can_be_discarded(const struct object_id *oid,
2678 unsigned long mtime)
2679 {
2680 if (!unpack_unreachable_expiration)
2681 return 0;
2682 if (mtime > unpack_unreachable_expiration)
2683 return 0;
2684 if (oid_array_lookup(&recent_objects, oid) >= 0)
2685 return 0;
2686 return 1;
2687 }
2688
2689 static void loosen_unused_packed_objects(struct rev_info *revs)
2690 {
2691 struct packed_git *p;
2692 uint32_t i;
2693 struct object_id oid;
2694
2695 for (p = packed_git; p; p = p->next) {
2696 if (!p->pack_local || p->pack_keep)
2697 continue;
2698
2699 if (open_pack_index(p))
2700 die("cannot open pack index");
2701
2702 for (i = 0; i < p->num_objects; i++) {
2703 nth_packed_object_oid(&oid, p, i);
2704 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2705 !has_sha1_pack_kept_or_nonlocal(oid.hash) &&
2706 !loosened_object_can_be_discarded(&oid, p->mtime))
2707 if (force_object_loose(oid.hash, p->mtime))
2708 die("unable to force loose object");
2709 }
2710 }
2711 }
2712
2713 /*
2714 * This tracks any options which pack-reuse code expects to be on, or which a
2715 * reader of the pack might not understand, and which would therefore prevent
2716 * blind reuse of what we have on disk.
2717 */
2718 static int pack_options_allow_reuse(void)
2719 {
2720 return pack_to_stdout && allow_ofs_delta;
2721 }
2722
2723 static int get_object_list_from_bitmap(struct rev_info *revs)
2724 {
2725 if (prepare_bitmap_walk(revs) < 0)
2726 return -1;
2727
2728 if (pack_options_allow_reuse() &&
2729 !reuse_partial_packfile_from_bitmap(
2730 &reuse_packfile,
2731 &reuse_packfile_objects,
2732 &reuse_packfile_offset)) {
2733 assert(reuse_packfile_objects);
2734 nr_result += reuse_packfile_objects;
2735 display_progress(progress_state, nr_result);
2736 }
2737
2738 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2739 return 0;
2740 }
2741
2742 static void record_recent_object(struct object *obj,
2743 const char *name,
2744 void *data)
2745 {
2746 oid_array_append(&recent_objects, &obj->oid);
2747 }
2748
2749 static void record_recent_commit(struct commit *commit, void *data)
2750 {
2751 oid_array_append(&recent_objects, &commit->object.oid);
2752 }
2753
2754 static void get_object_list(int ac, const char **av)
2755 {
2756 struct rev_info revs;
2757 char line[1000];
2758 int flags = 0;
2759
2760 init_revisions(&revs, NULL);
2761 save_commit_buffer = 0;
2762 setup_revisions(ac, av, &revs, NULL);
2763
2764 /* make sure shallows are read */
2765 is_repository_shallow();
2766
2767 while (fgets(line, sizeof(line), stdin) != NULL) {
2768 int len = strlen(line);
2769 if (len && line[len - 1] == '\n')
2770 line[--len] = 0;
2771 if (!len)
2772 break;
2773 if (*line == '-') {
2774 if (!strcmp(line, "--not")) {
2775 flags ^= UNINTERESTING;
2776 write_bitmap_index = 0;
2777 continue;
2778 }
2779 if (starts_with(line, "--shallow ")) {
2780 unsigned char sha1[20];
2781 if (get_sha1_hex(line + 10, sha1))
2782 die("not an SHA-1 '%s'", line + 10);
2783 register_shallow(sha1);
2784 use_bitmap_index = 0;
2785 continue;
2786 }
2787 die("not a rev '%s'", line);
2788 }
2789 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2790 die("bad revision '%s'", line);
2791 }
2792
2793 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2794 return;
2795
2796 if (prepare_revision_walk(&revs))
2797 die("revision walk setup failed");
2798 mark_edges_uninteresting(&revs, show_edge);
2799 traverse_commit_list(&revs, show_commit, show_object, NULL);
2800
2801 if (unpack_unreachable_expiration) {
2802 revs.ignore_missing_links = 1;
2803 if (add_unseen_recent_objects_to_traversal(&revs,
2804 unpack_unreachable_expiration))
2805 die("unable to add recent objects");
2806 if (prepare_revision_walk(&revs))
2807 die("revision walk setup failed");
2808 traverse_commit_list(&revs, record_recent_commit,
2809 record_recent_object, NULL);
2810 }
2811
2812 if (keep_unreachable)
2813 add_objects_in_unpacked_packs(&revs);
2814 if (pack_loose_unreachable)
2815 add_unreachable_loose_objects();
2816 if (unpack_unreachable)
2817 loosen_unused_packed_objects(&revs);
2818
2819 oid_array_clear(&recent_objects);
2820 }
2821
2822 static int option_parse_index_version(const struct option *opt,
2823 const char *arg, int unset)
2824 {
2825 char *c;
2826 const char *val = arg;
2827 pack_idx_opts.version = strtoul(val, &c, 10);
2828 if (pack_idx_opts.version > 2)
2829 die(_("unsupported index version %s"), val);
2830 if (*c == ',' && c[1])
2831 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2832 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2833 die(_("bad index version '%s'"), val);
2834 return 0;
2835 }
2836
2837 static int option_parse_unpack_unreachable(const struct option *opt,
2838 const char *arg, int unset)
2839 {
2840 if (unset) {
2841 unpack_unreachable = 0;
2842 unpack_unreachable_expiration = 0;
2843 }
2844 else {
2845 unpack_unreachable = 1;
2846 if (arg)
2847 unpack_unreachable_expiration = approxidate(arg);
2848 }
2849 return 0;
2850 }
2851
2852 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2853 {
2854 int use_internal_rev_list = 0;
2855 int thin = 0;
2856 int shallow = 0;
2857 int all_progress_implied = 0;
2858 struct argv_array rp = ARGV_ARRAY_INIT;
2859 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2860 int rev_list_index = 0;
2861 struct option pack_objects_options[] = {
2862 OPT_SET_INT('q', "quiet", &progress,
2863 N_("do not show progress meter"), 0),
2864 OPT_SET_INT(0, "progress", &progress,
2865 N_("show progress meter"), 1),
2866 OPT_SET_INT(0, "all-progress", &progress,
2867 N_("show progress meter during object writing phase"), 2),
2868 OPT_BOOL(0, "all-progress-implied",
2869 &all_progress_implied,
2870 N_("similar to --all-progress when progress meter is shown")),
2871 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2872 N_("write the pack index file in the specified idx format version"),
2873 0, option_parse_index_version },
2874 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2875 N_("maximum size of each output pack file")),
2876 OPT_BOOL(0, "local", &local,
2877 N_("ignore borrowed objects from alternate object store")),
2878 OPT_BOOL(0, "incremental", &incremental,
2879 N_("ignore packed objects")),
2880 OPT_INTEGER(0, "window", &window,
2881 N_("limit pack window by objects")),
2882 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2883 N_("limit pack window by memory in addition to object limit")),
2884 OPT_INTEGER(0, "depth", &depth,
2885 N_("maximum length of delta chain allowed in the resulting pack")),
2886 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2887 N_("reuse existing deltas")),
2888 OPT_BOOL(0, "reuse-object", &reuse_object,
2889 N_("reuse existing objects")),
2890 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2891 N_("use OFS_DELTA objects")),
2892 OPT_INTEGER(0, "threads", &delta_search_threads,
2893 N_("use threads when searching for best delta matches")),
2894 OPT_BOOL(0, "non-empty", &non_empty,
2895 N_("do not create an empty pack output")),
2896 OPT_BOOL(0, "revs", &use_internal_rev_list,
2897 N_("read revision arguments from standard input")),
2898 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2899 N_("limit the objects to those that are not yet packed"),
2900 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2901 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2902 N_("include objects reachable from any reference"),
2903 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2904 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2905 N_("include objects referred by reflog entries"),
2906 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2907 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2908 N_("include objects referred to by the index"),
2909 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2910 OPT_BOOL(0, "stdout", &pack_to_stdout,
2911 N_("output pack to stdout")),
2912 OPT_BOOL(0, "include-tag", &include_tag,
2913 N_("include tag objects that refer to objects to be packed")),
2914 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2915 N_("keep unreachable objects")),
2916 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2917 N_("pack loose unreachable objects")),
2918 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2919 N_("unpack unreachable objects newer than <time>"),
2920 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2921 OPT_BOOL(0, "thin", &thin,
2922 N_("create thin packs")),
2923 OPT_BOOL(0, "shallow", &shallow,
2924 N_("create packs suitable for shallow fetches")),
2925 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2926 N_("ignore packs that have companion .keep file")),
2927 OPT_INTEGER(0, "compression", &pack_compression_level,
2928 N_("pack compression level")),
2929 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2930 N_("do not hide commits by grafts"), 0),
2931 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2932 N_("use a bitmap index if available to speed up counting objects")),
2933 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2934 N_("write a bitmap index together with the pack index")),
2935 OPT_END(),
2936 };
2937
2938 check_replace_refs = 0;
2939
2940 reset_pack_idx_option(&pack_idx_opts);
2941 git_config(git_pack_config, NULL);
2942
2943 progress = isatty(2);
2944 argc = parse_options(argc, argv, prefix, pack_objects_options,
2945 pack_usage, 0);
2946
2947 if (argc) {
2948 base_name = argv[0];
2949 argc--;
2950 }
2951 if (pack_to_stdout != !base_name || argc)
2952 usage_with_options(pack_usage, pack_objects_options);
2953
2954 argv_array_push(&rp, "pack-objects");
2955 if (thin) {
2956 use_internal_rev_list = 1;
2957 argv_array_push(&rp, shallow
2958 ? "--objects-edge-aggressive"
2959 : "--objects-edge");
2960 } else
2961 argv_array_push(&rp, "--objects");
2962
2963 if (rev_list_all) {
2964 use_internal_rev_list = 1;
2965 argv_array_push(&rp, "--all");
2966 }
2967 if (rev_list_reflog) {
2968 use_internal_rev_list = 1;
2969 argv_array_push(&rp, "--reflog");
2970 }
2971 if (rev_list_index) {
2972 use_internal_rev_list = 1;
2973 argv_array_push(&rp, "--indexed-objects");
2974 }
2975 if (rev_list_unpacked) {
2976 use_internal_rev_list = 1;
2977 argv_array_push(&rp, "--unpacked");
2978 }
2979
2980 if (!reuse_object)
2981 reuse_delta = 0;
2982 if (pack_compression_level == -1)
2983 pack_compression_level = Z_DEFAULT_COMPRESSION;
2984 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2985 die("bad pack compression level %d", pack_compression_level);
2986
2987 if (!delta_search_threads) /* --threads=0 means autodetect */
2988 delta_search_threads = online_cpus();
2989
2990 #ifdef NO_PTHREADS
2991 if (delta_search_threads != 1)
2992 warning("no threads support, ignoring --threads");
2993 #endif
2994 if (!pack_to_stdout && !pack_size_limit)
2995 pack_size_limit = pack_size_limit_cfg;
2996 if (pack_to_stdout && pack_size_limit)
2997 die("--max-pack-size cannot be used to build a pack for transfer.");
2998 if (pack_size_limit && pack_size_limit < 1024*1024) {
2999 warning("minimum pack size limit is 1 MiB");
3000 pack_size_limit = 1024*1024;
3001 }
3002
3003 if (!pack_to_stdout && thin)
3004 die("--thin cannot be used to build an indexable pack.");
3005
3006 if (keep_unreachable && unpack_unreachable)
3007 die("--keep-unreachable and --unpack-unreachable are incompatible.");
3008 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3009 unpack_unreachable_expiration = 0;
3010
3011 /*
3012 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3013 *
3014 * - to produce good pack (with bitmap index not-yet-packed objects are
3015 * packed in suboptimal order).
3016 *
3017 * - to use more robust pack-generation codepath (avoiding possible
3018 * bugs in bitmap code and possible bitmap index corruption).
3019 */
3020 if (!pack_to_stdout)
3021 use_bitmap_index_default = 0;
3022
3023 if (use_bitmap_index < 0)
3024 use_bitmap_index = use_bitmap_index_default;
3025
3026 /* "hard" reasons not to use bitmaps; these just won't work at all */
3027 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3028 use_bitmap_index = 0;
3029
3030 if (pack_to_stdout || !rev_list_all)
3031 write_bitmap_index = 0;
3032
3033 if (progress && all_progress_implied)
3034 progress = 2;
3035
3036 prepare_packed_git();
3037 if (ignore_packed_keep) {
3038 struct packed_git *p;
3039 for (p = packed_git; p; p = p->next)
3040 if (p->pack_local && p->pack_keep)
3041 break;
3042 if (!p) /* no keep-able packs found */
3043 ignore_packed_keep = 0;
3044 }
3045 if (local) {
3046 /*
3047 * unlike ignore_packed_keep above, we do not want to
3048 * unset "local" based on looking at packs, as it
3049 * also covers non-local objects
3050 */
3051 struct packed_git *p;
3052 for (p = packed_git; p; p = p->next) {
3053 if (!p->pack_local) {
3054 have_non_local_packs = 1;
3055 break;
3056 }
3057 }
3058 }
3059
3060 if (progress)
3061 progress_state = start_progress(_("Counting objects"), 0);
3062 if (!use_internal_rev_list)
3063 read_object_list_from_stdin();
3064 else {
3065 get_object_list(rp.argc, rp.argv);
3066 argv_array_clear(&rp);
3067 }
3068 cleanup_preferred_base();
3069 if (include_tag && nr_result)
3070 for_each_ref(add_ref_tag, NULL);
3071 stop_progress(&progress_state);
3072
3073 if (non_empty && !nr_result)
3074 return 0;
3075 if (nr_result)
3076 prepare_pack(window, depth);
3077 write_pack_file();
3078 if (progress)
3079 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3080 " reused %"PRIu32" (delta %"PRIu32")\n",
3081 written, written_delta, reused, reused_delta);
3082 return 0;
3083 }