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