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