git repack: keep commits hidden by a graft
[git/git.git] / builtin-pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d 2#include "cache.h"
a74db82e 3#include "attr.h"
c323ac7d 4#include "object.h"
8e440259
PE
5#include "blob.h"
6#include "commit.h"
7#include "tag.h"
8#include "tree.h"
c323ac7d 9#include "delta.h"
a733cb60 10#include "pack.h"
3449f8c4 11#include "pack-revindex.h"
c38138cd 12#include "csum-file.h"
1b0c7174 13#include "tree-walk.h"
b5d97e6b
JH
14#include "diff.h"
15#include "revision.h"
16#include "list-objects.h"
96a02f8f 17#include "progress.h"
f0a24aa5 18#include "refs.h"
c323ac7d 19
8ecce684 20#ifdef THREADED_DELTA_SEARCH
833e3df1 21#include "thread-utils.h"
8ecce684
NP
22#include <pthread.h>
23#endif
24
231f240b 25static const char pack_usage[] = "\
f18d244a 26git pack-objects [{ -q | --progress | --all-progress }] \n\
a97773ce
BD
27 [--max-pack-size=N] [--local] [--incremental] \n\
28 [--window=N] [--window-memory=N] [--depth=N] \n\
fa736f72 29 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
367f4a43 30 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
ca11b212
NP
31 [--stdout | base-name] [--include-tag] \n\
32 [--keep-unreachable | --unpack-unreachable] \n\
f0a24aa5 33 [<ref-list | <object-list]";
c323ac7d 34
c323ac7d 35struct object_entry {
aa7e44bf 36 struct pack_idx_entry idx;
3f9ac8d2 37 unsigned long size; /* uncompressed size */
3f9ac8d2 38 struct packed_git *in_pack; /* already in pack */
6777a59f 39 off_t in_pack_offset;
a3fbf4df 40 struct object_entry *delta; /* delta base object */
82e5a82f 41 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
42 struct object_entry *delta_sibling; /* other deltified objects who
43 * uses the same base as me
44 */
074b2eea 45 void *delta_data; /* cached delta (uncompressed) */
a3fbf4df 46 unsigned long delta_size; /* delta data size (uncompressed) */
ed4a9031 47 unsigned long z_delta_size; /* delta data size (compressed) */
5a235b5e 48 unsigned int hash; /* name hint hash */
a3fbf4df
NP
49 enum object_type type;
50 enum object_type in_pack_type; /* could be delta */
51 unsigned char in_pack_header_size;
52 unsigned char preferred_base; /* we do not pack this, but is available
a74db82e 53 * to be used as the base object to delta
a3fbf4df
NP
54 * objects against.
55 */
a74db82e 56 unsigned char no_try_delta;
c323ac7d
LT
57};
58
3f9ac8d2 59/*
82e5a82f 60 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
61 * expanded). nr_objects & nr_alloc controls this array. They are stored
62 * in the order we see -- typically rev-list --objects order that gives us
63 * nice "minimum seek" order.
3f9ac8d2 64 */
e4d58311 65static struct object_entry *objects;
79814f42 66static struct pack_idx_entry **written_list;
d01fb92f 67static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
3f9ac8d2 68
96f1e58f 69static int non_empty;
a7de7130 70static int reuse_delta = 1, reuse_object = 1;
e5e9714a 71static int keep_unreachable, unpack_unreachable, include_tag;
96f1e58f
DR
72static int local;
73static int incremental;
e96fb9b8 74static int ignore_packed_keep;
be6b1914 75static int allow_ofs_delta;
d01fb92f 76static const char *base_name;
024701f1 77static int progress = 1;
4812a93a 78static int window = 10;
2b84b5a8 79static uint32_t pack_size_limit, pack_size_limit_cfg;
618e613a 80static int depth = 50;
43cc2b42 81static int delta_search_threads;
df6d6101 82static int pack_to_stdout;
8d1d8f83 83static int num_preferred_base;
dc6a0757 84static struct progress *progress_state;
960ccca6
DH
85static int pack_compression_level = Z_DEFAULT_COMPRESSION;
86static int pack_compression_seen;
c323ac7d 87
074b2eea
MK
88static unsigned long delta_cache_size = 0;
89static unsigned long max_delta_cache_size = 0;
e3dfddb3 90static unsigned long cache_max_small_delta_size = 1000;
074b2eea 91
a97773ce
BD
92static unsigned long window_memory_limit = 0;
93
3f9ac8d2
JH
94/*
95 * The object names in objects array are hashed with this hashtable,
e4d58311 96 * to help looking up the entry by object name.
3f9ac8d2
JH
97 * This hashtable is built after all the objects are seen.
98 */
96f1e58f
DR
99static int *object_ix;
100static int object_ix_hashsz;
3f9ac8d2 101
3f9ac8d2
JH
102/*
103 * stats
104 */
7cadf491
SP
105static uint32_t written, written_delta;
106static uint32_t reused, reused_delta;
3f9ac8d2 107
780e6e73 108
3613f9b4 109static void *get_delta(struct object_entry *entry)
c323ac7d 110{
3613f9b4
NP
111 unsigned long size, base_size, delta_size;
112 void *buf, *base_buf, *delta_buf;
21666f1a 113 enum object_type type;
c323ac7d 114
3613f9b4
NP
115 buf = read_sha1_file(entry->idx.sha1, &type, &size);
116 if (!buf)
117 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
118 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
119 if (!base_buf)
aa7e44bf 120 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
3613f9b4 121 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 122 buf, size, &delta_size, 0);
3613f9b4 123 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 124 die("delta size changed");
3613f9b4
NP
125 free(buf);
126 free(base_buf);
c323ac7d
LT
127 return delta_buf;
128}
129
30ebb40a
NP
130static unsigned long do_compress(void **pptr, unsigned long size)
131{
132 z_stream stream;
133 void *in, *out;
134 unsigned long maxsize;
135
136 memset(&stream, 0, sizeof(stream));
137 deflateInit(&stream, pack_compression_level);
138 maxsize = deflateBound(&stream, size);
139
140 in = *pptr;
141 out = xmalloc(maxsize);
142 *pptr = out;
143
144 stream.next_in = in;
145 stream.avail_in = size;
146 stream.next_out = out;
147 stream.avail_out = maxsize;
148 while (deflate(&stream, Z_FINISH) == Z_OK)
149 ; /* nothing */
150 deflateEnd(&stream);
151
152 free(in);
153 return stream.total_out;
154}
155
a733cb60
LT
156/*
157 * The per-object header is a pretty dense thing, which is
158 * - first byte: low four bits are "size", then three bits of "type",
159 * and the high bit is "size continues".
160 * - each byte afterwards: low seven bits are size continuation,
161 * with the high bit being "size continues"
162 */
163static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
164{
01247d87 165 int n = 1;
a733cb60
LT
166 unsigned char c;
167
eb32d236 168 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
169 die("bad type %d", type);
170
01247d87
LT
171 c = (type << 4) | (size & 15);
172 size >>= 4;
173 while (size) {
a733cb60 174 *hdr++ = c | 0x80;
01247d87
LT
175 c = size & 0x7f;
176 size >>= 7;
177 n++;
a733cb60
LT
178 }
179 *hdr = c;
180 return n;
181}
182
780e6e73
NP
183/*
184 * we are going to reuse the existing object data as is. make
185 * sure it is not corrupt.
186 */
079afb18
SP
187static int check_pack_inflate(struct packed_git *p,
188 struct pack_window **w_curs,
6777a59f
SP
189 off_t offset,
190 off_t len,
079afb18
SP
191 unsigned long expect)
192{
193 z_stream stream;
194 unsigned char fakebuf[4096], *in;
195 int st;
196
197 memset(&stream, 0, sizeof(stream));
39c68542 198 git_inflate_init(&stream);
079afb18
SP
199 do {
200 in = use_pack(p, w_curs, offset, &stream.avail_in);
201 stream.next_in = in;
202 stream.next_out = fakebuf;
203 stream.avail_out = sizeof(fakebuf);
39c68542 204 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
205 offset += stream.next_in - in;
206 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 207 git_inflate_end(&stream);
079afb18
SP
208 return (st == Z_STREAM_END &&
209 stream.total_out == expect &&
210 stream.total_in == len) ? 0 : -1;
211}
212
213static void copy_pack_data(struct sha1file *f,
214 struct packed_git *p,
215 struct pack_window **w_curs,
6777a59f
SP
216 off_t offset,
217 off_t len)
079afb18
SP
218{
219 unsigned char *in;
220 unsigned int avail;
221
222 while (len) {
223 in = use_pack(p, w_curs, offset, &avail);
224 if (avail > len)
6777a59f 225 avail = (unsigned int)len;
079afb18
SP
226 sha1write(f, in, avail);
227 offset += avail;
228 len -= avail;
229 }
230}
231
d7dd0223 232static unsigned long write_object(struct sha1file *f,
17b08f2c
DH
233 struct object_entry *entry,
234 off_t write_offset)
c323ac7d 235{
30ebb40a 236 unsigned long size, limit, datalen;
3f9ac8d2 237 void *buf;
2c5ef824 238 unsigned char header[10], dheader[10];
6777a59f 239 unsigned hdrlen;
2c5ef824
NP
240 enum object_type type;
241 int usable_delta, to_reuse;
c323ac7d 242
78d1e84f
NP
243 if (!pack_to_stdout)
244 crc32_begin(f);
245
2c5ef824
NP
246 type = entry->type;
247
248 /* write limit if limited packsize and not first object */
a1e4760f
NP
249 if (!pack_size_limit || !nr_written)
250 limit = 0;
251 else if (pack_size_limit <= write_offset)
252 /*
253 * the earlier object did not fit the limit; avoid
254 * mistaking this with unlimited (i.e. limit = 0).
255 */
256 limit = 1;
257 else
258 limit = pack_size_limit - write_offset;
2c5ef824
NP
259
260 if (!entry->delta)
261 usable_delta = 0; /* no delta */
262 else if (!pack_size_limit)
263 usable_delta = 1; /* unlimited packfile */
264 else if (entry->delta->idx.offset == (off_t)-1)
265 usable_delta = 0; /* base was written to another pack */
266 else if (entry->delta->idx.offset)
267 usable_delta = 1; /* base already exists in this pack */
268 else
269 usable_delta = 0; /* base could end up in another pack */
270
a7de7130 271 if (!reuse_object)
fa736f72
NP
272 to_reuse = 0; /* explicit */
273 else if (!entry->in_pack)
ab7cd7bb 274 to_reuse = 0; /* can't reuse what we don't have */
2c5ef824 275 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
17b08f2c
DH
276 /* check_object() decided it for us ... */
277 to_reuse = usable_delta;
278 /* ... but pack split may override that */
2c5ef824 279 else if (type != entry->in_pack_type)
ab7cd7bb
JH
280 to_reuse = 0; /* pack has delta which is unusable */
281 else if (entry->delta)
282 to_reuse = 0; /* we want to pack afresh */
283 else
284 to_reuse = 1; /* we have it in-pack undeltified,
285 * and we do not need to deltify it.
286 */
287
df6d6101 288 if (!to_reuse) {
64bd76b1 289 no_reuse:
5476a8ad 290 if (!usable_delta) {
2c5ef824 291 buf = read_sha1_file(entry->idx.sha1, &type, &size);
5476a8ad 292 if (!buf)
aa7e44bf 293 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
ed4a9031
NP
294 /*
295 * make sure no cached delta data remains from a
3ea3c215 296 * previous attempt before a pack split occurred.
ed4a9031
NP
297 */
298 free(entry->delta_data);
299 entry->delta_data = NULL;
300 entry->z_delta_size = 0;
5476a8ad 301 } else if (entry->delta_data) {
3f9ac8d2 302 size = entry->delta_size;
5476a8ad
NP
303 buf = entry->delta_data;
304 entry->delta_data = NULL;
2c5ef824 305 type = (allow_ofs_delta && entry->delta->idx.offset) ?
be6b1914 306 OBJ_OFS_DELTA : OBJ_REF_DELTA;
17b08f2c 307 } else {
3613f9b4 308 buf = get_delta(entry);
5476a8ad 309 size = entry->delta_size;
2c5ef824 310 type = (allow_ofs_delta && entry->delta->idx.offset) ?
5476a8ad 311 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2 312 }
ed4a9031
NP
313
314 if (entry->z_delta_size)
315 datalen = entry->z_delta_size;
316 else
317 datalen = do_compress(&buf, size);
8c3c7b2a 318
3f9ac8d2
JH
319 /*
320 * The object header is a byte of 'type' followed by zero or
be6b1914 321 * more bytes of length.
3f9ac8d2 322 */
2c5ef824 323 hdrlen = encode_header(type, size, header);
3f9ac8d2 324
2c5ef824 325 if (type == OBJ_OFS_DELTA) {
be6b1914
NP
326 /*
327 * Deltas with relative base contain an additional
328 * encoding of the relative offset for the delta
329 * base from this object's position in the pack.
330 */
aa7e44bf 331 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
332 unsigned pos = sizeof(dheader) - 1;
333 dheader[pos] = ofs & 127;
be6b1914 334 while (ofs >>= 7)
17b08f2c
DH
335 dheader[--pos] = 128 | (--ofs & 127);
336 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
17b08f2c
DH
337 free(buf);
338 return 0;
339 }
340 sha1write(f, header, hdrlen);
341 sha1write(f, dheader + pos, sizeof(dheader) - pos);
342 hdrlen += sizeof(dheader) - pos;
2c5ef824 343 } else if (type == OBJ_REF_DELTA) {
be6b1914
NP
344 /*
345 * Deltas with a base reference contain
346 * an additional 20 bytes for the base sha1.
347 */
17b08f2c 348 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
17b08f2c
DH
349 free(buf);
350 return 0;
351 }
352 sha1write(f, header, hdrlen);
aa7e44bf 353 sha1write(f, entry->delta->idx.sha1, 20);
3f9ac8d2 354 hdrlen += 20;
17b08f2c
DH
355 } else {
356 if (limit && hdrlen + datalen + 20 >= limit) {
17b08f2c
DH
357 free(buf);
358 return 0;
359 }
360 sha1write(f, header, hdrlen);
3f9ac8d2 361 }
30ebb40a 362 sha1write(f, buf, datalen);
3f9ac8d2 363 free(buf);
c323ac7d 364 }
3f9ac8d2
JH
365 else {
366 struct packed_git *p = entry->in_pack;
03e79c88 367 struct pack_window *w_curs = NULL;
91ecbeca 368 struct revindex_entry *revidx;
6777a59f 369 off_t offset;
3f9ac8d2 370
64bd76b1 371 if (entry->delta)
2c5ef824 372 type = (allow_ofs_delta && entry->delta->idx.offset) ?
780e6e73 373 OBJ_OFS_DELTA : OBJ_REF_DELTA;
2c5ef824 374 hdrlen = encode_header(type, entry->size, header);
64bd76b1 375
91ecbeca 376 offset = entry->in_pack_offset;
3449f8c4 377 revidx = find_pack_revindex(p, offset);
91ecbeca
NP
378 datalen = revidx[1].offset - offset;
379 if (!pack_to_stdout && p->index_version > 1 &&
64bd76b1
NP
380 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
381 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
382 unuse_pack(&w_curs);
383 goto no_reuse;
384 }
385
91ecbeca
NP
386 offset += entry->in_pack_header_size;
387 datalen -= entry->in_pack_header_size;
64bd76b1
NP
388 if (!pack_to_stdout && p->index_version == 1 &&
389 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
390 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
391 unuse_pack(&w_curs);
392 goto no_reuse;
393 }
394
2c5ef824 395 if (type == OBJ_OFS_DELTA) {
aa7e44bf 396 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
397 unsigned pos = sizeof(dheader) - 1;
398 dheader[pos] = ofs & 127;
399 while (ofs >>= 7)
400 dheader[--pos] = 128 | (--ofs & 127);
59dd9ed1
NP
401 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
402 unuse_pack(&w_curs);
17b08f2c 403 return 0;
59dd9ed1 404 }
17b08f2c
DH
405 sha1write(f, header, hdrlen);
406 sha1write(f, dheader + pos, sizeof(dheader) - pos);
407 hdrlen += sizeof(dheader) - pos;
64bd76b1 408 reused_delta++;
2c5ef824 409 } else if (type == OBJ_REF_DELTA) {
59dd9ed1
NP
410 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
411 unuse_pack(&w_curs);
17b08f2c 412 return 0;
59dd9ed1 413 }
17b08f2c 414 sha1write(f, header, hdrlen);
aa7e44bf 415 sha1write(f, entry->delta->idx.sha1, 20);
17b08f2c 416 hdrlen += 20;
64bd76b1 417 reused_delta++;
17b08f2c 418 } else {
59dd9ed1
NP
419 if (limit && hdrlen + datalen + 20 >= limit) {
420 unuse_pack(&w_curs);
17b08f2c 421 return 0;
59dd9ed1 422 }
17b08f2c
DH
423 sha1write(f, header, hdrlen);
424 }
079afb18 425 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 426 unuse_pack(&w_curs);
3f9ac8d2 427 reused++;
a733cb60 428 }
17b08f2c 429 if (usable_delta)
ab7cd7bb 430 written_delta++;
3f9ac8d2 431 written++;
78d1e84f 432 if (!pack_to_stdout)
aa7e44bf 433 entry->idx.crc32 = crc32_end(f);
c323ac7d
LT
434 return hdrlen + datalen;
435}
436
6ed7f25e 437static int write_one(struct sha1file *f,
9d5ab962 438 struct object_entry *e,
6ed7f25e 439 off_t *offset)
9d5ab962 440{
d7dd0223
NP
441 unsigned long size;
442
443 /* offset is non zero if object is written already. */
aa7e44bf 444 if (e->idx.offset || e->preferred_base)
6ed7f25e 445 return 1;
d7dd0223
NP
446
447 /* if we are deltified, write out base object first. */
6ed7f25e
NP
448 if (e->delta && !write_one(f, e->delta, offset))
449 return 0;
d7dd0223 450
6ed7f25e
NP
451 e->idx.offset = *offset;
452 size = write_object(f, e, *offset);
17b08f2c 453 if (!size) {
aa7e44bf 454 e->idx.offset = 0;
17b08f2c
DH
455 return 0;
456 }
79814f42 457 written_list[nr_written++] = &e->idx;
d7dd0223
NP
458
459 /* make sure off_t is sufficiently large not to wrap */
6ed7f25e 460 if (*offset > *offset + size)
d7dd0223 461 die("pack too large for current definition of off_t");
6ed7f25e
NP
462 *offset += size;
463 return 1;
9d5ab962
JH
464}
465
aa7e44bf 466/* forward declaration for write_pack_file */
d01fb92f
DH
467static int adjust_perm(const char *path, mode_t mode);
468
469static void write_pack_file(void)
c323ac7d 470{
ebe27b13 471 uint32_t i = 0, j;
d22b9290 472 struct sha1file *f;
6ed7f25e 473 off_t offset;
a733cb60 474 struct pack_header hdr;
ebe27b13 475 uint32_t nr_remaining = nr_result;
f746bae8 476 time_t last_mtime = 0;
81a216a5 477
bcd7954e 478 if (progress > pack_to_stdout)
dc6a0757 479 progress_state = start_progress("Writing objects", nr_result);
79814f42 480 written_list = xmalloc(nr_objects * sizeof(*written_list));
183bdb2c 481
ebe27b13 482 do {
aa7e44bf 483 unsigned char sha1[20];
7ba502c4 484 char *pack_tmp_name = NULL;
aa7e44bf 485
ebe27b13 486 if (pack_to_stdout) {
2a128d63 487 f = sha1fd_throughput(1, "<stdout>", progress_state);
ebe27b13 488 } else {
7ba502c4
NP
489 char tmpname[PATH_MAX];
490 int fd;
6e180cdc
JH
491 fd = odb_mkstemp(tmpname, sizeof(tmpname),
492 "pack/tmp_pack_XXXXXX");
ebe27b13
DH
493 pack_tmp_name = xstrdup(tmpname);
494 f = sha1fd(fd, pack_tmp_name);
495 }
496
497 hdr.hdr_signature = htonl(PACK_SIGNATURE);
498 hdr.hdr_version = htonl(PACK_VERSION);
499 hdr.hdr_entries = htonl(nr_remaining);
500 sha1write(f, &hdr, sizeof(hdr));
501 offset = sizeof(hdr);
502 nr_written = 0;
503 for (; i < nr_objects; i++) {
6ed7f25e 504 if (!write_one(f, objects + i, &offset))
ebe27b13 505 break;
4d4fcc54 506 display_progress(progress_state, written);
ebe27b13 507 }
c553ca25 508
ebe27b13
DH
509 /*
510 * Did we write the wrong # entries in the header?
511 * If so, rewrite it like in fast-import
512 */
54352bb2
LT
513 if (pack_to_stdout) {
514 sha1close(f, sha1, CSUM_CLOSE);
515 } else if (nr_written == nr_remaining) {
516 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 517 } else {
ac0463ed 518 int fd = sha1close(f, sha1, 0);
abeb40e5 519 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 520 nr_written, sha1, offset);
7ba502c4 521 close(fd);
ebe27b13
DH
522 }
523
524 if (!pack_to_stdout) {
d01fb92f 525 mode_t mode = umask(0);
f746bae8 526 struct stat st;
7ba502c4 527 char *idx_tmp_name, tmpname[PATH_MAX];
d01fb92f
DH
528
529 umask(mode);
530 mode = 0444 & ~mode;
531
79814f42
NP
532 idx_tmp_name = write_idx_file(NULL, written_list,
533 nr_written, sha1);
f746bae8 534
d01fb92f 535 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
aa7e44bf 536 base_name, sha1_to_hex(sha1));
c74faea1 537 free_pack_by_name(tmpname);
d01fb92f
DH
538 if (adjust_perm(pack_tmp_name, mode))
539 die("unable to make temporary pack file readable: %s",
540 strerror(errno));
541 if (rename(pack_tmp_name, tmpname))
542 die("unable to rename temporary pack file: %s",
543 strerror(errno));
f746bae8
NP
544
545 /*
546 * Packs are runtime accessed in their mtime
547 * order since newer packs are more likely to contain
548 * younger objects. So if we are creating multiple
549 * packs then we should modify the mtime of later ones
550 * to preserve this property.
551 */
552 if (stat(tmpname, &st) < 0) {
553 warning("failed to stat %s: %s",
554 tmpname, strerror(errno));
555 } else if (!last_mtime) {
556 last_mtime = st.st_mtime;
557 } else {
558 struct utimbuf utb;
559 utb.actime = st.st_atime;
560 utb.modtime = --last_mtime;
561 if (utime(tmpname, &utb) < 0)
562 warning("failed utime() on %s: %s",
563 tmpname, strerror(errno));
564 }
565
d01fb92f 566 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 567 base_name, sha1_to_hex(sha1));
d01fb92f
DH
568 if (adjust_perm(idx_tmp_name, mode))
569 die("unable to make temporary index file readable: %s",
570 strerror(errno));
571 if (rename(idx_tmp_name, tmpname))
572 die("unable to rename temporary index file: %s",
573 strerror(errno));
f746bae8 574
7ba502c4
NP
575 free(idx_tmp_name);
576 free(pack_tmp_name);
aa7e44bf 577 puts(sha1_to_hex(sha1));
ebe27b13
DH
578 }
579
580 /* mark written objects as written to previous pack */
581 for (j = 0; j < nr_written; j++) {
79814f42 582 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
583 }
584 nr_remaining -= nr_written;
585 } while (nr_remaining && i < nr_objects);
586
587 free(written_list);
4d4fcc54 588 stop_progress(&progress_state);
67c08ce1 589 if (written != nr_result)
6e1c2344
RJ
590 die("wrote %"PRIu32" objects while expecting %"PRIu32,
591 written, nr_result);
01c12a23
DH
592 /*
593 * We have scanned through [0 ... i). Since we have written
594 * the correct number of objects, the remaining [i ... nr_objects)
595 * items must be either already written (due to out-of-order delta base)
596 * or a preferred base. Count those which are neither and complain if any.
597 */
ebe27b13
DH
598 for (j = 0; i < nr_objects; i++) {
599 struct object_entry *e = objects + i;
aa7e44bf 600 j += !e->idx.offset && !e->preferred_base;
d01fb92f 601 }
ebe27b13 602 if (j)
6e1c2344
RJ
603 die("wrote %"PRIu32" objects as expected but %"PRIu32
604 " unwritten", written, j);
c323ac7d
LT
605}
606
7a979d99
JH
607static int locate_object_entry_hash(const unsigned char *sha1)
608{
609 int i;
610 unsigned int ui;
611 memcpy(&ui, sha1, sizeof(unsigned int));
612 i = ui % object_ix_hashsz;
613 while (0 < object_ix[i]) {
aa7e44bf 614 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
615 return i;
616 if (++i == object_ix_hashsz)
617 i = 0;
618 }
619 return -1 - i;
620}
621
622static struct object_entry *locate_object_entry(const unsigned char *sha1)
623{
624 int i;
625
626 if (!object_ix_hashsz)
627 return NULL;
628
629 i = locate_object_entry_hash(sha1);
630 if (0 <= i)
631 return &objects[object_ix[i]-1];
632 return NULL;
633}
634
635static void rehash_objects(void)
c323ac7d 636{
7cadf491 637 uint32_t i;
7a979d99
JH
638 struct object_entry *oe;
639
640 object_ix_hashsz = nr_objects * 3;
641 if (object_ix_hashsz < 1024)
642 object_ix_hashsz = 1024;
643 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 644 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 645 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 646 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
647 if (0 <= ix)
648 continue;
649 ix = -1 - ix;
650 object_ix[ix] = i + 1;
651 }
652}
653
ce0bd642 654static unsigned name_hash(const char *name)
1d6b38cc 655{
48fb7deb 656 unsigned c, hash = 0;
ce0bd642 657
bc32fed5
JH
658 if (!name)
659 return 0;
660
eeef7135 661 /*
ce0bd642
LT
662 * This effectively just creates a sortable number from the
663 * last sixteen non-whitespace characters. Last characters
664 * count "most", so things that end in ".c" sort together.
eeef7135 665 */
ce0bd642
LT
666 while ((c = *name++) != 0) {
667 if (isspace(c))
668 continue;
669 hash = (hash >> 2) + (c << 24);
670 }
1d6b38cc
JH
671 return hash;
672}
673
a74db82e
JH
674static void setup_delta_attr_check(struct git_attr_check *check)
675{
676 static struct git_attr *attr_delta;
677
678 if (!attr_delta)
679 attr_delta = git_attr("delta", 5);
680
681 check[0].attr = attr_delta;
682}
683
684static int no_try_delta(const char *path)
685{
686 struct git_attr_check check[1];
687
688 setup_delta_attr_check(check);
689 if (git_checkattr(path, ARRAY_SIZE(check), check))
690 return 0;
691 if (ATTR_FALSE(check->value))
692 return 1;
693 return 0;
694}
695
5c49c116 696static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 697 const char *name, int exclude)
c323ac7d 698{
c323ac7d 699 struct object_entry *entry;
29b734e4 700 struct packed_git *p, *found_pack = NULL;
6777a59f 701 off_t found_offset = 0;
29b734e4 702 int ix;
bc32fed5 703 unsigned hash = name_hash(name);
29b734e4
NP
704
705 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
706 if (ix >= 0) {
707 if (exclude) {
708 entry = objects + object_ix[ix] - 1;
81a216a5
NP
709 if (!entry->preferred_base)
710 nr_result--;
29b734e4
NP
711 entry->preferred_base = 1;
712 }
713 return 0;
714 }
c323ac7d 715
daae0625
BC
716 if (!exclude && local && has_loose_object_nonlocal(sha1))
717 return 0;
718
5c49c116
NP
719 for (p = packed_git; p; p = p->next) {
720 off_t offset = find_pack_entry_one(sha1, p);
721 if (offset) {
722 if (!found_pack) {
723 found_offset = offset;
724 found_pack = p;
64560374 725 }
5c49c116
NP
726 if (exclude)
727 break;
728 if (incremental)
729 return 0;
730 if (local && !p->pack_local)
731 return 0;
e96fb9b8
BC
732 if (ignore_packed_keep && p->pack_local && p->pack_keep)
733 return 0;
64560374
LT
734 }
735 }
eb019375 736
29b734e4
NP
737 if (nr_objects >= nr_alloc) {
738 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 739 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 740 }
29b734e4
NP
741
742 entry = objects + nr_objects++;
c323ac7d 743 memset(entry, 0, sizeof(*entry));
aa7e44bf 744 hashcpy(entry->idx.sha1, sha1);
27225f2e 745 entry->hash = hash;
5c49c116
NP
746 if (type)
747 entry->type = type;
29b734e4
NP
748 if (exclude)
749 entry->preferred_base = 1;
81a216a5
NP
750 else
751 nr_result++;
29b734e4
NP
752 if (found_pack) {
753 entry->in_pack = found_pack;
754 entry->in_pack_offset = found_offset;
755 }
7a979d99
JH
756
757 if (object_ix_hashsz * 3 <= nr_objects * 4)
758 rehash_objects();
29b734e4
NP
759 else
760 object_ix[-1 - ix] = nr_objects;
7a979d99 761
4d4fcc54 762 display_progress(progress_state, nr_objects);
29b734e4 763
a74db82e
JH
764 if (name && no_try_delta(name))
765 entry->no_try_delta = 1;
766
29b734e4 767 return 1;
c323ac7d
LT
768}
769
5379a5c5
JH
770struct pbase_tree_cache {
771 unsigned char sha1[20];
772 int ref;
773 int temporary;
774 void *tree_data;
775 unsigned long tree_size;
776};
777
778static struct pbase_tree_cache *(pbase_tree_cache[256]);
779static int pbase_tree_cache_ix(const unsigned char *sha1)
780{
781 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
782}
783static int pbase_tree_cache_ix_incr(int ix)
784{
785 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
786}
787
788static struct pbase_tree {
789 struct pbase_tree *next;
790 /* This is a phony "cache" entry; we are not
791 * going to evict it nor find it through _get()
792 * mechanism -- this is for the toplevel node that
793 * would almost always change with any commit.
794 */
795 struct pbase_tree_cache pcache;
796} *pbase_tree;
797
798static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
799{
800 struct pbase_tree_cache *ent, *nent;
801 void *data;
802 unsigned long size;
21666f1a 803 enum object_type type;
5379a5c5
JH
804 int neigh;
805 int my_ix = pbase_tree_cache_ix(sha1);
806 int available_ix = -1;
807
808 /* pbase-tree-cache acts as a limited hashtable.
809 * your object will be found at your index or within a few
810 * slots after that slot if it is cached.
811 */
812 for (neigh = 0; neigh < 8; neigh++) {
813 ent = pbase_tree_cache[my_ix];
a89fccd2 814 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
815 ent->ref++;
816 return ent;
817 }
818 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
819 ((0 <= available_ix) &&
820 (!ent && pbase_tree_cache[available_ix])))
821 available_ix = my_ix;
822 if (!ent)
823 break;
824 my_ix = pbase_tree_cache_ix_incr(my_ix);
825 }
826
827 /* Did not find one. Either we got a bogus request or
828 * we need to read and perhaps cache.
829 */
21666f1a 830 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
831 if (!data)
832 return NULL;
21666f1a 833 if (type != OBJ_TREE) {
5379a5c5
JH
834 free(data);
835 return NULL;
836 }
837
838 /* We need to either cache or return a throwaway copy */
839
840 if (available_ix < 0)
841 ent = NULL;
842 else {
843 ent = pbase_tree_cache[available_ix];
844 my_ix = available_ix;
845 }
846
847 if (!ent) {
848 nent = xmalloc(sizeof(*nent));
849 nent->temporary = (available_ix < 0);
850 }
851 else {
852 /* evict and reuse */
853 free(ent->tree_data);
854 nent = ent;
855 }
e702496e 856 hashcpy(nent->sha1, sha1);
5379a5c5
JH
857 nent->tree_data = data;
858 nent->tree_size = size;
859 nent->ref = 1;
860 if (!nent->temporary)
861 pbase_tree_cache[my_ix] = nent;
862 return nent;
863}
864
865static void pbase_tree_put(struct pbase_tree_cache *cache)
866{
867 if (!cache->temporary) {
868 cache->ref--;
869 return;
870 }
871 free(cache->tree_data);
872 free(cache);
873}
874
875static int name_cmp_len(const char *name)
876{
877 int i;
878 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
879 ;
880 return i;
881}
882
883static void add_pbase_object(struct tree_desc *tree,
5379a5c5 884 const char *name,
ce0bd642
LT
885 int cmplen,
886 const char *fullname)
3f9ac8d2 887{
4c068a98 888 struct name_entry entry;
8a5a8d6c 889 int cmp;
4c068a98
LT
890
891 while (tree_entry(tree,&entry)) {
1211be6b
LT
892 if (S_ISGITLINK(entry.mode))
893 continue;
8a5a8d6c
NP
894 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
895 memcmp(name, entry.path, cmplen);
896 if (cmp > 0)
7a979d99 897 continue;
8a5a8d6c
NP
898 if (cmp < 0)
899 return;
5379a5c5 900 if (name[cmplen] != '/') {
5c49c116 901 add_object_entry(entry.sha1,
4d1012c3 902 object_type(entry.mode),
bc32fed5 903 fullname, 1);
5379a5c5
JH
904 return;
905 }
8a5a8d6c 906 if (S_ISDIR(entry.mode)) {
7a979d99 907 struct tree_desc sub;
5379a5c5
JH
908 struct pbase_tree_cache *tree;
909 const char *down = name+cmplen+1;
910 int downlen = name_cmp_len(down);
911
4c068a98 912 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
913 if (!tree)
914 return;
6fda5e51 915 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 916
ce0bd642 917 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
918 pbase_tree_put(tree);
919 }
920 }
921}
1d6b38cc 922
5379a5c5
JH
923static unsigned *done_pbase_paths;
924static int done_pbase_paths_num;
925static int done_pbase_paths_alloc;
926static int done_pbase_path_pos(unsigned hash)
927{
928 int lo = 0;
929 int hi = done_pbase_paths_num;
930 while (lo < hi) {
931 int mi = (hi + lo) / 2;
932 if (done_pbase_paths[mi] == hash)
933 return mi;
934 if (done_pbase_paths[mi] < hash)
935 hi = mi;
936 else
937 lo = mi + 1;
938 }
939 return -lo-1;
940}
941
942static int check_pbase_path(unsigned hash)
943{
944 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
945 if (0 <= pos)
946 return 1;
947 pos = -pos - 1;
948 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
949 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
950 done_pbase_paths = xrealloc(done_pbase_paths,
951 done_pbase_paths_alloc *
952 sizeof(unsigned));
953 }
954 done_pbase_paths_num++;
955 if (pos < done_pbase_paths_num)
956 memmove(done_pbase_paths + pos + 1,
957 done_pbase_paths + pos,
958 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
959 done_pbase_paths[pos] = hash;
960 return 0;
961}
962
bc32fed5 963static void add_preferred_base_object(const char *name)
5379a5c5
JH
964{
965 struct pbase_tree *it;
8a5a8d6c 966 int cmplen;
bc32fed5 967 unsigned hash = name_hash(name);
5379a5c5 968
8a5a8d6c 969 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
970 return;
971
8a5a8d6c 972 cmplen = name_cmp_len(name);
5379a5c5
JH
973 for (it = pbase_tree; it; it = it->next) {
974 if (cmplen == 0) {
bc32fed5 975 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
976 }
977 else {
978 struct tree_desc tree;
6fda5e51 979 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 980 add_pbase_object(&tree, name, cmplen, name);
7a979d99 981 }
3f9ac8d2 982 }
3f9ac8d2
JH
983}
984
7a979d99 985static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 986{
5379a5c5
JH
987 struct pbase_tree *it;
988 void *data;
989 unsigned long size;
990 unsigned char tree_sha1[20];
1d6b38cc 991
8d1d8f83
JH
992 if (window <= num_preferred_base++)
993 return;
994
5379a5c5
JH
995 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
996 if (!data)
7a979d99 997 return;
5379a5c5
JH
998
999 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1000 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1001 free(data);
1002 return;
1003 }
1004 }
1005
1006 it = xcalloc(1, sizeof(*it));
1007 it->next = pbase_tree;
1008 pbase_tree = it;
1009
e702496e 1010 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1011 it->pcache.tree_data = data;
1012 it->pcache.tree_size = size;
3f9ac8d2
JH
1013}
1014
c323ac7d
LT
1015static void check_object(struct object_entry *entry)
1016{
5c49c116 1017 if (entry->in_pack) {
780e6e73 1018 struct packed_git *p = entry->in_pack;
03e79c88 1019 struct pack_window *w_curs = NULL;
5c49c116
NP
1020 const unsigned char *base_ref = NULL;
1021 struct object_entry *base_entry;
1022 unsigned long used, used_0;
6777a59f 1023 unsigned int avail;
5c49c116
NP
1024 off_t ofs;
1025 unsigned char *buf, c;
780e6e73 1026
6777a59f 1027 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1028
5c49c116 1029 /*
fa736f72
NP
1030 * We want in_pack_type even if we do not reuse delta
1031 * since non-delta representations could still be reused.
ab7cd7bb 1032 */
09ded04b 1033 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1034 &entry->in_pack_type,
1035 &entry->size);
03d66015
NP
1036 if (used == 0)
1037 goto give_up;
ab7cd7bb 1038
5c49c116
NP
1039 /*
1040 * Determine if this is a delta and if so whether we can
1041 * reuse it or not. Otherwise let's find out as cheaply as
1042 * possible what the actual type and size for this object is.
3f9ac8d2 1043 */
5c49c116
NP
1044 switch (entry->in_pack_type) {
1045 default:
1046 /* Not a delta hence we've already got all we need. */
1047 entry->type = entry->in_pack_type;
1048 entry->in_pack_header_size = used;
03d66015
NP
1049 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1050 goto give_up;
5c49c116
NP
1051 unuse_pack(&w_curs);
1052 return;
1053 case OBJ_REF_DELTA:
a7de7130 1054 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1055 base_ref = use_pack(p, &w_curs,
1056 entry->in_pack_offset + used, NULL);
1057 entry->in_pack_header_size = used + 20;
1058 break;
1059 case OBJ_OFS_DELTA:
1060 buf = use_pack(p, &w_curs,
1061 entry->in_pack_offset + used, NULL);
1062 used_0 = 0;
1063 c = buf[used_0++];
1064 ofs = c & 127;
1065 while (c & 128) {
1066 ofs += 1;
03d66015
NP
1067 if (!ofs || MSB(ofs, 7)) {
1068 error("delta base offset overflow in pack for %s",
1069 sha1_to_hex(entry->idx.sha1));
1070 goto give_up;
1071 }
5c49c116
NP
1072 c = buf[used_0++];
1073 ofs = (ofs << 7) + (c & 127);
780e6e73 1074 }
5c49c116 1075 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1076 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1077 error("delta base offset out of bound for %s",
1078 sha1_to_hex(entry->idx.sha1));
1079 goto give_up;
1080 }
a7de7130 1081 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1082 struct revindex_entry *revidx;
1083 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1084 if (!revidx)
1085 goto give_up;
3449f8c4
NP
1086 base_ref = nth_packed_object_sha1(p, revidx->nr);
1087 }
5c49c116
NP
1088 entry->in_pack_header_size = used + used_0;
1089 break;
780e6e73 1090 }
780e6e73 1091
5c49c116
NP
1092 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1093 /*
1094 * If base_ref was set above that means we wish to
1095 * reuse delta data, and we even found that base
1096 * in the list of objects we want to pack. Goodie!
1097 *
1098 * Depth value does not matter - find_deltas() will
1099 * never consider reused delta as the base object to
1100 * deltify other objects against, in order to avoid
1101 * circular deltas.
3f9ac8d2 1102 */
780e6e73 1103 entry->type = entry->in_pack_type;
5c49c116 1104 entry->delta = base_entry;
64bd76b1 1105 entry->delta_size = entry->size;
15b4d577
JH
1106 entry->delta_sibling = base_entry->delta_child;
1107 base_entry->delta_child = entry;
5c49c116
NP
1108 unuse_pack(&w_curs);
1109 return;
1110 }
ab7cd7bb 1111
5c49c116
NP
1112 if (entry->type) {
1113 /*
1114 * This must be a delta and we already know what the
1115 * final object type is. Let's extract the actual
1116 * object size from the delta header.
1117 */
1118 entry->size = get_size_from_delta(p, &w_curs,
1119 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1120 if (entry->size == 0)
1121 goto give_up;
5c49c116 1122 unuse_pack(&w_curs);
3f9ac8d2
JH
1123 return;
1124 }
5c49c116
NP
1125
1126 /*
1127 * No choice but to fall back to the recursive delta walk
1128 * with sha1_object_info() to find about the object type
1129 * at this point...
1130 */
03d66015 1131 give_up:
5c49c116 1132 unuse_pack(&w_curs);
36e4d74a 1133 }
3f9ac8d2 1134
aa7e44bf 1135 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1136 /*
1137 * The error condition is checked in prepare_pack(). This is
1138 * to permit a missing preferred base object to be ignored
1139 * as a preferred base. Doing so can result in a larger
1140 * pack file, but the transfer will still take place.
1141 */
3f9ac8d2
JH
1142}
1143
5c49c116
NP
1144static int pack_offset_sort(const void *_a, const void *_b)
1145{
1146 const struct object_entry *a = *(struct object_entry **)_a;
1147 const struct object_entry *b = *(struct object_entry **)_b;
1148
1149 /* avoid filesystem trashing with loose objects */
1150 if (!a->in_pack && !b->in_pack)
aa7e44bf 1151 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1152
1153 if (a->in_pack < b->in_pack)
1154 return -1;
1155 if (a->in_pack > b->in_pack)
1156 return 1;
1157 return a->in_pack_offset < b->in_pack_offset ? -1 :
1158 (a->in_pack_offset > b->in_pack_offset);
1159}
1160
c323ac7d
LT
1161static void get_object_details(void)
1162{
7cadf491 1163 uint32_t i;
5c49c116
NP
1164 struct object_entry **sorted_by_offset;
1165
1166 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1167 for (i = 0; i < nr_objects; i++)
1168 sorted_by_offset[i] = objects + i;
1169 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1170
5c49c116
NP
1171 for (i = 0; i < nr_objects; i++)
1172 check_object(sorted_by_offset[i]);
3449f8c4 1173
5c49c116 1174 free(sorted_by_offset);
c323ac7d
LT
1175}
1176
b904166c
NP
1177/*
1178 * We search for deltas in a list sorted by type, by filename hash, and then
1179 * by size, so that we see progressively smaller and smaller files.
1180 * That's because we prefer deltas to be from the bigger file
1181 * to the smaller -- deletes are potentially cheaper, but perhaps
1182 * more importantly, the bigger file is likely the more recent
1183 * one. The deepest deltas are therefore the oldest objects which are
1184 * less susceptible to be accessed often.
1185 */
9668cf59 1186static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1187{
9668cf59
NP
1188 const struct object_entry *a = *(struct object_entry **)_a;
1189 const struct object_entry *b = *(struct object_entry **)_b;
1190
c323ac7d 1191 if (a->type > b->type)
27225f2e 1192 return -1;
b904166c 1193 if (a->type < b->type)
27225f2e 1194 return 1;
b904166c 1195 if (a->hash > b->hash)
7a979d99 1196 return -1;
b904166c 1197 if (a->hash < b->hash)
7a979d99 1198 return 1;
b904166c 1199 if (a->preferred_base > b->preferred_base)
c323ac7d 1200 return -1;
b904166c
NP
1201 if (a->preferred_base < b->preferred_base)
1202 return 1;
c323ac7d 1203 if (a->size > b->size)
b904166c
NP
1204 return -1;
1205 if (a->size < b->size)
c323ac7d 1206 return 1;
b904166c 1207 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1208}
1209
1210struct unpacked {
1211 struct object_entry *entry;
1212 void *data;
f6c7081a 1213 struct delta_index *index;
5a235b5e 1214 unsigned depth;
c323ac7d
LT
1215};
1216
d250626c
NP
1217static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1218 unsigned long delta_size)
074b2eea
MK
1219{
1220 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1221 return 0;
1222
e3dfddb3
MK
1223 if (delta_size < cache_max_small_delta_size)
1224 return 1;
1225
074b2eea
MK
1226 /* cache delta, if objects are large enough compared to delta size */
1227 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1228 return 1;
1229
1230 return 0;
1231}
1232
8ecce684
NP
1233#ifdef THREADED_DELTA_SEARCH
1234
1235static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1236#define read_lock() pthread_mutex_lock(&read_mutex)
1237#define read_unlock() pthread_mutex_unlock(&read_mutex)
1238
3c701839
NP
1239static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1240#define cache_lock() pthread_mutex_lock(&cache_mutex)
1241#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1242
8ecce684
NP
1243static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1244#define progress_lock() pthread_mutex_lock(&progress_mutex)
1245#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1246
1247#else
1248
e1ef8673
JH
1249#define read_lock() (void)0
1250#define read_unlock() (void)0
1251#define cache_lock() (void)0
1252#define cache_unlock() (void)0
1253#define progress_lock() (void)0
1254#define progress_unlock() (void)0
8ecce684
NP
1255
1256#endif
1257
f6c7081a 1258static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1259 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1260{
f6c7081a
NP
1261 struct object_entry *trg_entry = trg->entry;
1262 struct object_entry *src_entry = src->entry;
560b25a8 1263 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1264 unsigned ref_depth;
21666f1a 1265 enum object_type type;
c323ac7d
LT
1266 void *delta_buf;
1267
1268 /* Don't bother doing diffs between different types */
f6c7081a 1269 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1270 return -1;
1271
51d1e83f
LT
1272 /*
1273 * We do not bother to try a delta that we discarded
8dbbd14e 1274 * on an earlier try, but only when reusing delta data.
51d1e83f 1275 */
a7de7130 1276 if (reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1277 trg_entry->in_pack == src_entry->in_pack &&
1278 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1279 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1280 return 0;
1281
898b14ce 1282 /* Let's not bust the allowed depth. */
5a235b5e 1283 if (src->depth >= max_depth)
d116a45a 1284 return 0;
c323ac7d 1285
c3b06a69 1286 /* Now some size filtering heuristics. */
560b25a8 1287 trg_size = trg_entry->size;
c83f032e
NP
1288 if (!trg_entry->delta) {
1289 max_size = trg_size/2 - 20;
1290 ref_depth = 1;
1291 } else {
1292 max_size = trg_entry->delta_size;
5a235b5e 1293 ref_depth = trg->depth;
c83f032e 1294 }
720fe22d 1295 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1296 (max_depth - ref_depth + 1);
c3b06a69
NP
1297 if (max_size == 0)
1298 return 0;
f6c7081a 1299 src_size = src_entry->size;
560b25a8 1300 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1301 if (sizediff >= max_size)
f527cb8c 1302 return 0;
a1dab41a
BD
1303 if (trg_size < src_size / 32)
1304 return 0;
f6c7081a 1305
560b25a8
NP
1306 /* Load data if not already done */
1307 if (!trg->data) {
8ecce684 1308 read_lock();
aa7e44bf 1309 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1310 read_unlock();
2e3404c3
JH
1311 if (!trg->data)
1312 die("object %s cannot be read",
1313 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1314 if (sz != trg_size)
1315 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1316 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1317 *mem_usage += sz;
560b25a8
NP
1318 }
1319 if (!src->data) {
8ecce684 1320 read_lock();
aa7e44bf 1321 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1322 read_unlock();
2e3404c3
JH
1323 if (!src->data)
1324 die("object %s cannot be read",
1325 sha1_to_hex(src_entry->idx.sha1));
560b25a8
NP
1326 if (sz != src_size)
1327 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1328 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1329 *mem_usage += sz;
560b25a8
NP
1330 }
1331 if (!src->index) {
1332 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1333 if (!src->index) {
1334 static int warned = 0;
1335 if (!warned++)
1336 warning("suboptimal pack - out of memory");
1337 return 0;
1338 }
ef0316fc 1339 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1340 }
1341
1342 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1343 if (!delta_buf)
75c42d8c 1344 return 0;
f6c7081a 1345
9e2d57a0 1346 if (trg_entry->delta) {
848d732c
BD
1347 /* Prefer only shallower same-sized deltas. */
1348 if (delta_size == trg_entry->delta_size &&
5a235b5e 1349 src->depth + 1 >= trg->depth) {
848d732c
BD
1350 free(delta_buf);
1351 return 0;
1352 }
074b2eea 1353 }
9e2d57a0 1354
3c701839
NP
1355 /*
1356 * Handle memory allocation outside of the cache
1357 * accounting lock. Compiler will optimize the strangeness
1358 * away when THREADED_DELTA_SEARCH is not defined.
1359 */
8e0f7003 1360 free(trg_entry->delta_data);
3c701839 1361 cache_lock();
9e2d57a0
NP
1362 if (trg_entry->delta_data) {
1363 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1364 trg_entry->delta_data = NULL;
1365 }
d250626c 1366 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1367 delta_cache_size += delta_size;
3c701839
NP
1368 cache_unlock();
1369 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1370 } else {
1371 cache_unlock();
074b2eea 1372 free(delta_buf);
3c701839
NP
1373 }
1374
b7a28f78
NP
1375 trg_entry->delta = src_entry;
1376 trg_entry->delta_size = delta_size;
1377 trg->depth = src->depth + 1;
1378
f6c7081a 1379 return 1;
c323ac7d
LT
1380}
1381
898b14ce 1382static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1383{
898b14ce
NP
1384 struct object_entry *child = me->delta_child;
1385 unsigned int m = n;
1386 while (child) {
1387 unsigned int c = check_delta_limit(child, n + 1);
1388 if (m < c)
1389 m = c;
1390 child = child->delta_sibling;
1391 }
1392 return m;
b2504a0d
NP
1393}
1394
75ad235c 1395static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1396{
ef0316fc 1397 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1398 free_delta_index(n->index);
1399 n->index = NULL;
1400 if (n->data) {
ef0316fc 1401 freed_mem += n->entry->size;
a97773ce
BD
1402 free(n->data);
1403 n->data = NULL;
a97773ce
BD
1404 }
1405 n->entry = NULL;
7d7baa5e 1406 n->depth = 0;
ef0316fc 1407 return freed_mem;
a97773ce
BD
1408}
1409
384b32c0 1410static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1411 int window, int depth, unsigned *processed)
c323ac7d 1412{
384b32c0 1413 uint32_t i, idx = 0, count = 0;
7cadf491 1414 struct unpacked *array;
ef0316fc 1415 unsigned long mem_usage = 0;
c323ac7d 1416
19d4b416 1417 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1418
384b32c0 1419 for (;;) {
421b488a 1420 struct object_entry *entry;
c323ac7d 1421 struct unpacked *n = array + idx;
ef0316fc 1422 int j, max_depth, best_base = -1;
c323ac7d 1423
384b32c0
NP
1424 progress_lock();
1425 if (!*list_size) {
1426 progress_unlock();
1427 break;
1428 }
421b488a 1429 entry = *list++;
384b32c0
NP
1430 (*list_size)--;
1431 if (!entry->preferred_base) {
1432 (*processed)++;
1433 display_progress(progress_state, *processed);
1434 }
1435 progress_unlock();
1436
ef0316fc 1437 mem_usage -= free_unpacked(n);
c323ac7d 1438 n->entry = entry;
ab7cd7bb 1439
a97773ce 1440 while (window_memory_limit &&
ef0316fc 1441 mem_usage > window_memory_limit &&
a97773ce
BD
1442 count > 1) {
1443 uint32_t tail = (idx + window - count) % window;
75ad235c 1444 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1445 count--;
1446 }
1447
75d39853
NP
1448 /* We do not compute delta to *create* objects we are not
1449 * going to pack.
1450 */
1451 if (entry->preferred_base)
1452 goto next;
1453
898b14ce
NP
1454 /*
1455 * If the current object is at pack edge, take the depth the
1456 * objects that depend on the current object into account
1457 * otherwise they would become too deep.
1458 */
1459 max_depth = depth;
1460 if (entry->delta_child) {
1461 max_depth -= check_delta_limit(entry, 0);
1462 if (max_depth <= 0)
1463 goto next;
1464 }
1465
78817c15
LT
1466 j = window;
1467 while (--j > 0) {
77639870 1468 int ret;
7cadf491 1469 uint32_t other_idx = idx + j;
c323ac7d 1470 struct unpacked *m;
78817c15
LT
1471 if (other_idx >= window)
1472 other_idx -= window;
c323ac7d
LT
1473 m = array + other_idx;
1474 if (!m->entry)
1475 break;
ef0316fc 1476 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1477 if (ret < 0)
c323ac7d 1478 break;
77639870
JH
1479 else if (ret > 0)
1480 best_base = other_idx;
c323ac7d 1481 }
898b14ce 1482
ed4a9031
NP
1483 /*
1484 * If we decided to cache the delta data, then it is best
1485 * to compress it right away. First because we have to do
1486 * it anyway, and doing it here while we're threaded will
1487 * save a lot of time in the non threaded write phase,
1488 * as well as allow for caching more deltas within
1489 * the same cache size limit.
1490 * ...
1491 * But only if not writing to stdout, since in that case
1492 * the network is most likely throttling writes anyway,
1493 * and therefore it is best to go to the write phase ASAP
1494 * instead, as we can afford spending more time compressing
1495 * between writes at that moment.
1496 */
1497 if (entry->delta_data && !pack_to_stdout) {
1498 entry->z_delta_size = do_compress(&entry->delta_data,
1499 entry->delta_size);
1500 cache_lock();
1501 delta_cache_size -= entry->delta_size;
1502 delta_cache_size += entry->z_delta_size;
1503 cache_unlock();
1504 }
1505
70ca1a3f
JH
1506 /* if we made n a delta, and if n is already at max
1507 * depth, leaving it in the window is pointless. we
1508 * should evict it first.
70ca1a3f 1509 */
70baf5d4 1510 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1511 continue;
ff45715c 1512
77639870
JH
1513 /*
1514 * Move the best delta base up in the window, after the
1515 * currently deltified object, to keep it longer. It will
1516 * be the first base object to be attempted next.
1517 */
1518 if (entry->delta) {
1519 struct unpacked swap = array[best_base];
1520 int dist = (window + idx - best_base) % window;
1521 int dst = best_base;
1522 while (dist--) {
1523 int src = (dst + 1) % window;
1524 array[dst] = array[src];
1525 dst = src;
1526 }
1527 array[dst] = swap;
1528 }
1529
898b14ce 1530 next:
521a4f4c 1531 idx++;
a97773ce
BD
1532 if (count + 1 < window)
1533 count++;
521a4f4c
LT
1534 if (idx >= window)
1535 idx = 0;
384b32c0 1536 }
adee7bdf 1537
f6c7081a 1538 for (i = 0; i < window; ++i) {
ff45715c 1539 free_delta_index(array[i].index);
adee7bdf 1540 free(array[i].data);
f6c7081a 1541 }
adee7bdf 1542 free(array);
c323ac7d
LT
1543}
1544
8ecce684
NP
1545#ifdef THREADED_DELTA_SEARCH
1546
50f22ada
JS
1547/*
1548 * The main thread waits on the condition that (at least) one of the workers
1549 * has stopped working (which is indicated in the .working member of
1550 * struct thread_params).
1551 * When a work thread has completed its work, it sets .working to 0 and
1552 * signals the main thread and waits on the condition that .data_ready
1553 * becomes 1.
1554 */
1555
8ecce684
NP
1556struct thread_params {
1557 pthread_t thread;
1558 struct object_entry **list;
1559 unsigned list_size;
384b32c0 1560 unsigned remaining;
8ecce684
NP
1561 int window;
1562 int depth;
50f22ada
JS
1563 int working;
1564 int data_ready;
1565 pthread_mutex_t mutex;
1566 pthread_cond_t cond;
8ecce684
NP
1567 unsigned *processed;
1568};
1569
50f22ada 1570static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
c2a33679 1571
8ecce684
NP
1572static void *threaded_find_deltas(void *arg)
1573{
c2a33679
NP
1574 struct thread_params *me = arg;
1575
50f22ada 1576 while (me->remaining) {
384b32c0 1577 find_deltas(me->list, &me->remaining,
c2a33679 1578 me->window, me->depth, me->processed);
50f22ada
JS
1579
1580 progress_lock();
1581 me->working = 0;
1582 pthread_cond_signal(&progress_cond);
1583 progress_unlock();
1584
1585 /*
1586 * We must not set ->data_ready before we wait on the
1587 * condition because the main thread may have set it to 1
1588 * before we get here. In order to be sure that new
1589 * work is available if we see 1 in ->data_ready, it
1590 * was initialized to 0 before this thread was spawned
1591 * and we reset it to 0 right away.
1592 */
1593 pthread_mutex_lock(&me->mutex);
1594 while (!me->data_ready)
1595 pthread_cond_wait(&me->cond, &me->mutex);
1596 me->data_ready = 0;
1597 pthread_mutex_unlock(&me->mutex);
c2a33679 1598 }
50f22ada
JS
1599 /* leave ->working 1 so that this doesn't get more work assigned */
1600 return NULL;
8ecce684
NP
1601}
1602
8ecce684
NP
1603static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1604 int window, int depth, unsigned *processed)
1605{
50f22ada 1606 struct thread_params p[delta_search_threads];
384b32c0 1607 int i, ret, active_threads = 0;
c2a33679 1608
367f4a43 1609 if (delta_search_threads <= 1) {
384b32c0 1610 find_deltas(list, &list_size, window, depth, processed);
367f4a43
NP
1611 return;
1612 }
43cc2b42 1613 if (progress > pack_to_stdout)
b6c29915 1614 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1615 delta_search_threads);
367f4a43 1616
50f22ada 1617 /* Partition the work amongst work threads. */
367f4a43 1618 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1619 unsigned sub_size = list_size / (delta_search_threads - i);
1620
bf874896
NP
1621 /* don't use too small segments or no deltas will be found */
1622 if (sub_size < 2*window && i+1 < delta_search_threads)
1623 sub_size = 0;
1624
8ecce684
NP
1625 p[i].window = window;
1626 p[i].depth = depth;
1627 p[i].processed = processed;
50f22ada
JS
1628 p[i].working = 1;
1629 p[i].data_ready = 0;
c2a33679 1630
59921b4b 1631 /* try to split chunks on "path" boundaries */
6fc74703
NP
1632 while (sub_size && sub_size < list_size &&
1633 list[sub_size]->hash &&
384b32c0
NP
1634 list[sub_size]->hash == list[sub_size-1]->hash)
1635 sub_size++;
1636
50f22ada
JS
1637 p[i].list = list;
1638 p[i].list_size = sub_size;
1639 p[i].remaining = sub_size;
59921b4b 1640
384b32c0
NP
1641 list += sub_size;
1642 list_size -= sub_size;
1643 }
1644
50f22ada
JS
1645 /* Start work threads. */
1646 for (i = 0; i < delta_search_threads; i++) {
1647 if (!p[i].list_size)
1648 continue;
68e6a4f8
JS
1649 pthread_mutex_init(&p[i].mutex, NULL);
1650 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1651 ret = pthread_create(&p[i].thread, NULL,
1652 threaded_find_deltas, &p[i]);
1653 if (ret)
1654 die("unable to create thread: %s", strerror(ret));
1655 active_threads++;
1656 }
1657
384b32c0
NP
1658 /*
1659 * Now let's wait for work completion. Each time a thread is done
1660 * with its work, we steal half of the remaining work from the
1661 * thread with the largest number of unprocessed objects and give
1662 * it to that newly idle thread. This ensure good load balancing
1663 * until the remaining object list segments are simply too short
1664 * to be worth splitting anymore.
1665 */
50f22ada
JS
1666 while (active_threads) {
1667 struct thread_params *target = NULL;
384b32c0
NP
1668 struct thread_params *victim = NULL;
1669 unsigned sub_size = 0;
384b32c0
NP
1670
1671 progress_lock();
50f22ada
JS
1672 for (;;) {
1673 for (i = 0; !target && i < delta_search_threads; i++)
1674 if (!p[i].working)
1675 target = &p[i];
1676 if (target)
1677 break;
1678 pthread_cond_wait(&progress_cond, &progress_mutex);
1679 }
1680
384b32c0
NP
1681 for (i = 0; i < delta_search_threads; i++)
1682 if (p[i].remaining > 2*window &&
1683 (!victim || victim->remaining < p[i].remaining))
1684 victim = &p[i];
1685 if (victim) {
1686 sub_size = victim->remaining / 2;
1687 list = victim->list + victim->list_size - sub_size;
1688 while (sub_size && list[0]->hash &&
1689 list[0]->hash == list[-1]->hash) {
1690 list++;
1691 sub_size--;
1692 }
eb9688ff
NP
1693 if (!sub_size) {
1694 /*
1695 * It is possible for some "paths" to have
1696 * so many objects that no hash boundary
1697 * might be found. Let's just steal the
1698 * exact half in that case.
1699 */
1700 sub_size = victim->remaining / 2;
1701 list -= sub_size;
1702 }
384b32c0
NP
1703 target->list = list;
1704 victim->list_size -= sub_size;
1705 victim->remaining -= sub_size;
1706 }
384b32c0
NP
1707 target->list_size = sub_size;
1708 target->remaining = sub_size;
50f22ada
JS
1709 target->working = 1;
1710 progress_unlock();
1711
1712 pthread_mutex_lock(&target->mutex);
1713 target->data_ready = 1;
1714 pthread_cond_signal(&target->cond);
1715 pthread_mutex_unlock(&target->mutex);
c2a33679 1716
384b32c0 1717 if (!sub_size) {
b81d9af7 1718 pthread_join(target->thread, NULL);
50f22ada
JS
1719 pthread_cond_destroy(&target->cond);
1720 pthread_mutex_destroy(&target->mutex);
384b32c0 1721 active_threads--;
c2a33679 1722 }
50f22ada 1723 }
8ecce684
NP
1724}
1725
1726#else
384b32c0 1727#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
1728#endif
1729
f0a24aa5
SP
1730static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1731{
1732 unsigned char peeled[20];
1733
1734 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1735 !peel_ref(path, peeled) && /* peelable? */
1736 !is_null_sha1(peeled) && /* annotated tag? */
1737 locate_object_entry(peeled)) /* object packed? */
1738 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1739 return 0;
1740}
1741
f3123c4a
JH
1742static void prepare_pack(int window, int depth)
1743{
9668cf59 1744 struct object_entry **delta_list;
6e1c2344
RJ
1745 uint32_t i, nr_deltas;
1746 unsigned n;
9668cf59 1747
3f9ac8d2 1748 get_object_details();
9668cf59 1749
0e8189e2
NP
1750 /*
1751 * If we're locally repacking then we need to be doubly careful
1752 * from now on in order to make sure no stealth corruption gets
1753 * propagated to the new pack. Clients receiving streamed packs
1754 * should validate everything they get anyway so no need to incur
1755 * the additional cost here in that case.
1756 */
1757 if (!pack_to_stdout)
1758 do_check_packed_object_crc = 1;
1759
75d39853 1760 if (!nr_objects || !window || !depth)
9668cf59
NP
1761 return;
1762
1763 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1764 nr_deltas = n = 0;
1765
1766 for (i = 0; i < nr_objects; i++) {
1767 struct object_entry *entry = objects + i;
1768
1769 if (entry->delta)
1770 /* This happens if we decided to reuse existing
a7de7130 1771 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
1772 */
1773 continue;
1774
1775 if (entry->size < 50)
1776 continue;
1777
1778 if (entry->no_try_delta)
1779 continue;
1780
6d6f9cdd 1781 if (!entry->preferred_base) {
75d39853 1782 nr_deltas++;
6d6f9cdd
SP
1783 if (entry->type < 0)
1784 die("unable to get type of object %s",
1785 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
1786 } else {
1787 if (entry->type < 0) {
1788 /*
1789 * This object is not found, but we
1790 * don't have to include it anyway.
1791 */
1792 continue;
1793 }
6d6f9cdd 1794 }
75d39853
NP
1795
1796 delta_list[n++] = entry;
1797 }
1798
2f8b8947 1799 if (nr_deltas && n > 1) {
e334977d
NP
1800 unsigned nr_done = 0;
1801 if (progress)
dc6a0757
NP
1802 progress_state = start_progress("Compressing objects",
1803 nr_deltas);
75d39853 1804 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1805 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1806 stop_progress(&progress_state);
e334977d
NP
1807 if (nr_done != nr_deltas)
1808 die("inconsistency with delta count");
75d39853 1809 }
9668cf59 1810 free(delta_list);
f3123c4a
JH
1811}
1812
ef90d6d4 1813static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a
JK
1814{
1815 if(!strcmp(k, "pack.window")) {
1816 window = git_config_int(k, v);
1817 return 0;
1818 }
a97773ce
BD
1819 if (!strcmp(k, "pack.windowmemory")) {
1820 window_memory_limit = git_config_ulong(k, v);
1821 return 0;
1822 }
1823 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1824 depth = git_config_int(k, v);
1825 return 0;
1826 }
960ccca6
DH
1827 if (!strcmp(k, "pack.compression")) {
1828 int level = git_config_int(k, v);
1829 if (level == -1)
1830 level = Z_DEFAULT_COMPRESSION;
1831 else if (level < 0 || level > Z_BEST_COMPRESSION)
1832 die("bad pack compression level %d", level);
1833 pack_compression_level = level;
1834 pack_compression_seen = 1;
1835 return 0;
1836 }
074b2eea
MK
1837 if (!strcmp(k, "pack.deltacachesize")) {
1838 max_delta_cache_size = git_config_int(k, v);
1839 return 0;
1840 }
e3dfddb3
MK
1841 if (!strcmp(k, "pack.deltacachelimit")) {
1842 cache_max_small_delta_size = git_config_int(k, v);
1843 return 0;
1844 }
693b86ff
NP
1845 if (!strcmp(k, "pack.threads")) {
1846 delta_search_threads = git_config_int(k, v);
833e3df1 1847 if (delta_search_threads < 0)
693b86ff
NP
1848 die("invalid number of threads specified (%d)",
1849 delta_search_threads);
1850#ifndef THREADED_DELTA_SEARCH
833e3df1 1851 if (delta_search_threads != 1)
693b86ff
NP
1852 warning("no threads support, ignoring %s", k);
1853#endif
1854 return 0;
1855 }
4d00bda2
NP
1856 if (!strcmp(k, "pack.indexversion")) {
1857 pack_idx_default_version = git_config_int(k, v);
1858 if (pack_idx_default_version > 2)
6e1c2344
RJ
1859 die("bad pack.indexversion=%"PRIu32,
1860 pack_idx_default_version);
4d00bda2
NP
1861 return 0;
1862 }
2b84b5a8
JS
1863 if (!strcmp(k, "pack.packsizelimit")) {
1864 pack_size_limit_cfg = git_config_ulong(k, v);
1865 return 0;
1866 }
ef90d6d4 1867 return git_default_config(k, v, cb);
4812a93a
JK
1868}
1869
b5d97e6b 1870static void read_object_list_from_stdin(void)
c323ac7d 1871{
b5d97e6b
JH
1872 char line[40 + 1 + PATH_MAX + 2];
1873 unsigned char sha1[20];
b2504a0d 1874
da93d12b 1875 for (;;) {
da93d12b
LT
1876 if (!fgets(line, sizeof(line), stdin)) {
1877 if (feof(stdin))
1878 break;
1879 if (!ferror(stdin))
1880 die("fgets returned NULL, not EOF, not error!");
687dd75c
JH
1881 if (errno != EINTR)
1882 die("fgets: %s", strerror(errno));
1883 clearerr(stdin);
1884 continue;
da93d12b 1885 }
7a979d99
JH
1886 if (line[0] == '-') {
1887 if (get_sha1_hex(line+1, sha1))
1888 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1889 line);
8d1d8f83 1890 add_preferred_base(sha1);
7a979d99 1891 continue;
21fcd1bd 1892 }
c323ac7d 1893 if (get_sha1_hex(line, sha1))
ef07618f 1894 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1895
bc32fed5
JH
1896 add_preferred_base_object(line+41);
1897 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1898 }
b5d97e6b
JH
1899}
1900
08cdfb13
JH
1901#define OBJECT_ADDED (1u<<20)
1902
11c211fa 1903static void show_commit(struct commit *commit, void *data)
b5d97e6b 1904{
bc32fed5 1905 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1906 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1907}
1908
cf2ab916 1909static void show_object(struct object *obj, const struct name_path *path, const char *last)
b5d97e6b 1910{
cf2ab916
LT
1911 char *name = path_name(path, last);
1912
8d2dfc49
LT
1913 add_preferred_base_object(name);
1914 add_object_entry(obj->sha1, obj->type, name, 0);
1915 obj->flags |= OBJECT_ADDED;
1916
1917 /*
1918 * We will have generated the hash from the name,
1919 * but not saved a pointer to it - we can free it
1920 */
1921 free((char *)name);
b5d97e6b
JH
1922}
1923
8d1d8f83
JH
1924static void show_edge(struct commit *commit)
1925{
1926 add_preferred_base(commit->object.sha1);
1927}
1928
08cdfb13
JH
1929struct in_pack_object {
1930 off_t offset;
1931 struct object *object;
1932};
1933
1934struct in_pack {
1935 int alloc;
1936 int nr;
1937 struct in_pack_object *array;
1938};
1939
1940static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1941{
1942 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1943 in_pack->array[in_pack->nr].object = object;
1944 in_pack->nr++;
1945}
1946
1947/*
1948 * Compare the objects in the offset order, in order to emulate the
f18d244a 1949 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
1950 */
1951static int ofscmp(const void *a_, const void *b_)
1952{
1953 struct in_pack_object *a = (struct in_pack_object *)a_;
1954 struct in_pack_object *b = (struct in_pack_object *)b_;
1955
1956 if (a->offset < b->offset)
1957 return -1;
1958 else if (a->offset > b->offset)
1959 return 1;
1960 else
1961 return hashcmp(a->object->sha1, b->object->sha1);
1962}
1963
1964static void add_objects_in_unpacked_packs(struct rev_info *revs)
1965{
1966 struct packed_git *p;
1967 struct in_pack in_pack;
1968 uint32_t i;
1969
1970 memset(&in_pack, 0, sizeof(in_pack));
1971
1972 for (p = packed_git; p; p = p->next) {
1973 const unsigned char *sha1;
1974 struct object *o;
1975
79bc4c71 1976 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
1977 continue;
1978 if (open_pack_index(p))
1979 die("cannot open pack index");
1980
1981 ALLOC_GROW(in_pack.array,
1982 in_pack.nr + p->num_objects,
1983 in_pack.alloc);
1984
1985 for (i = 0; i < p->num_objects; i++) {
1986 sha1 = nth_packed_object_sha1(p, i);
1987 o = lookup_unknown_object(sha1);
1988 if (!(o->flags & OBJECT_ADDED))
1989 mark_in_pack_object(o, p, &in_pack);
1990 o->flags |= OBJECT_ADDED;
1991 }
1992 }
1993
1994 if (in_pack.nr) {
1995 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1996 ofscmp);
1997 for (i = 0; i < in_pack.nr; i++) {
1998 struct object *o = in_pack.array[i].object;
1999 add_object_entry(o->sha1, o->type, "", 0);
2000 }
2001 }
2002 free(in_pack.array);
2003}
2004
094085e3
BC
2005static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2006{
2007 static struct packed_git *last_found = (void *)1;
2008 struct packed_git *p;
2009
2010 p = (last_found != (void *)1) ? last_found : packed_git;
2011
2012 while (p) {
2013 if ((!p->pack_local || p->pack_keep) &&
2014 find_pack_entry_one(sha1, p)) {
2015 last_found = p;
2016 return 1;
2017 }
2018 if (p == last_found)
2019 p = packed_git;
2020 else
2021 p = p->next;
2022 if (p == last_found)
2023 p = p->next;
2024 }
2025 return 0;
2026}
2027
ca11b212
NP
2028static void loosen_unused_packed_objects(struct rev_info *revs)
2029{
2030 struct packed_git *p;
2031 uint32_t i;
2032 const unsigned char *sha1;
2033
2034 for (p = packed_git; p; p = p->next) {
79bc4c71 2035 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2036 continue;
2037
2038 if (open_pack_index(p))
2039 die("cannot open pack index");
2040
2041 for (i = 0; i < p->num_objects; i++) {
2042 sha1 = nth_packed_object_sha1(p, i);
094085e3
BC
2043 if (!locate_object_entry(sha1) &&
2044 !has_sha1_pack_kept_or_nonlocal(sha1))
ca11b212
NP
2045 if (force_object_loose(sha1, p->mtime))
2046 die("unable to force loose object");
2047 }
2048 }
2049}
2050
8d1d8f83 2051static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2052{
2053 struct rev_info revs;
2054 char line[1000];
b5d97e6b
JH
2055 int flags = 0;
2056
b5d97e6b
JH
2057 init_revisions(&revs, NULL);
2058 save_commit_buffer = 0;
b5d97e6b
JH
2059 setup_revisions(ac, av, &revs, NULL);
2060
b5d97e6b
JH
2061 while (fgets(line, sizeof(line), stdin) != NULL) {
2062 int len = strlen(line);
872c930d 2063 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2064 line[--len] = 0;
2065 if (!len)
2066 break;
2067 if (*line == '-') {
2068 if (!strcmp(line, "--not")) {
2069 flags ^= UNINTERESTING;
2070 continue;
2071 }
2072 die("not a rev '%s'", line);
2073 }
2074 if (handle_revision_arg(line, &revs, flags, 1))
2075 die("bad revision '%s'", line);
2076 }
2077
3d51e1b5
MK
2078 if (prepare_revision_walk(&revs))
2079 die("revision walk setup failed");
8d1d8f83 2080 mark_edges_uninteresting(revs.commits, &revs, show_edge);
11c211fa 2081 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13
JH
2082
2083 if (keep_unreachable)
2084 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2085 if (unpack_unreachable)
2086 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2087}
2088
b6b32ccb
JH
2089static int adjust_perm(const char *path, mode_t mode)
2090{
2091 if (chmod(path, mode))
2092 return -1;
2093 return adjust_shared_perm(path);
2094}
2095
b5d97e6b
JH
2096int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2097{
b5d97e6b 2098 int use_internal_rev_list = 0;
8d1d8f83 2099 int thin = 0;
7cadf491 2100 uint32_t i;
ffa84ffb
RD
2101 const char **rp_av;
2102 int rp_ac_alloc = 64;
8d1d8f83
JH
2103 int rp_ac;
2104
ffa84ffb
RD
2105 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2106
8d1d8f83
JH
2107 rp_av[0] = "pack-objects";
2108 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2109 rp_ac = 2;
b5d97e6b 2110
ef90d6d4 2111 git_config(git_pack_config, NULL);
960ccca6
DH
2112 if (!pack_compression_seen && core_compression_seen)
2113 pack_compression_level = core_compression_level;
b5d97e6b
JH
2114
2115 progress = isatty(2);
2116 for (i = 1; i < argc; i++) {
2117 const char *arg = argv[i];
2118
2119 if (*arg != '-')
2120 break;
2121
2122 if (!strcmp("--non-empty", arg)) {
2123 non_empty = 1;
2124 continue;
2125 }
2126 if (!strcmp("--local", arg)) {
2127 local = 1;
2128 continue;
2129 }
b5d97e6b
JH
2130 if (!strcmp("--incremental", arg)) {
2131 incremental = 1;
2132 continue;
2133 }
e96fb9b8
BC
2134 if (!strcmp("--honor-pack-keep", arg)) {
2135 ignore_packed_keep = 1;
2136 continue;
2137 }
960ccca6
DH
2138 if (!prefixcmp(arg, "--compression=")) {
2139 char *end;
2140 int level = strtoul(arg+14, &end, 0);
2141 if (!arg[14] || *end)
2142 usage(pack_usage);
2143 if (level == -1)
2144 level = Z_DEFAULT_COMPRESSION;
2145 else if (level < 0 || level > Z_BEST_COMPRESSION)
2146 die("bad pack compression level %d", level);
2147 pack_compression_level = level;
2148 continue;
2149 }
6b94b1a0
DH
2150 if (!prefixcmp(arg, "--max-pack-size=")) {
2151 char *end;
2b84b5a8 2152 pack_size_limit_cfg = 0;
6b94b1a0
DH
2153 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2154 if (!arg[16] || *end)
2155 usage(pack_usage);
2156 continue;
2157 }
599065a3 2158 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2159 char *end;
2160 window = strtoul(arg+9, &end, 0);
2161 if (!arg[9] || *end)
2162 usage(pack_usage);
2163 continue;
2164 }
a97773ce
BD
2165 if (!prefixcmp(arg, "--window-memory=")) {
2166 if (!git_parse_ulong(arg+16, &window_memory_limit))
2167 usage(pack_usage);
2168 continue;
2169 }
367f4a43
NP
2170 if (!prefixcmp(arg, "--threads=")) {
2171 char *end;
2172 delta_search_threads = strtoul(arg+10, &end, 0);
833e3df1 2173 if (!arg[10] || *end || delta_search_threads < 0)
367f4a43
NP
2174 usage(pack_usage);
2175#ifndef THREADED_DELTA_SEARCH
833e3df1 2176 if (delta_search_threads != 1)
367f4a43
NP
2177 warning("no threads support, "
2178 "ignoring %s", arg);
2179#endif
2180 continue;
2181 }
599065a3 2182 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2183 char *end;
2184 depth = strtoul(arg+8, &end, 0);
2185 if (!arg[8] || *end)
2186 usage(pack_usage);
2187 continue;
2188 }
2189 if (!strcmp("--progress", arg)) {
2190 progress = 1;
2191 continue;
2192 }
231f240b
NP
2193 if (!strcmp("--all-progress", arg)) {
2194 progress = 2;
2195 continue;
2196 }
b5d97e6b
JH
2197 if (!strcmp("-q", arg)) {
2198 progress = 0;
2199 continue;
2200 }
2201 if (!strcmp("--no-reuse-delta", arg)) {
a7de7130 2202 reuse_delta = 0;
b5d97e6b
JH
2203 continue;
2204 }
fa736f72 2205 if (!strcmp("--no-reuse-object", arg)) {
a7de7130 2206 reuse_object = reuse_delta = 0;
fa736f72
NP
2207 continue;
2208 }
be6b1914 2209 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2210 allow_ofs_delta = 1;
be6b1914
NP
2211 continue;
2212 }
b5d97e6b
JH
2213 if (!strcmp("--stdout", arg)) {
2214 pack_to_stdout = 1;
2215 continue;
2216 }
2217 if (!strcmp("--revs", arg)) {
2218 use_internal_rev_list = 1;
2219 continue;
2220 }
08cdfb13
JH
2221 if (!strcmp("--keep-unreachable", arg)) {
2222 keep_unreachable = 1;
2223 continue;
2224 }
ca11b212
NP
2225 if (!strcmp("--unpack-unreachable", arg)) {
2226 unpack_unreachable = 1;
2227 continue;
2228 }
f0a24aa5
SP
2229 if (!strcmp("--include-tag", arg)) {
2230 include_tag = 1;
2231 continue;
2232 }
8d1d8f83 2233 if (!strcmp("--unpacked", arg) ||
63049292 2234 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2235 !strcmp("--all", arg)) {
2236 use_internal_rev_list = 1;
ffa84ffb
RD
2237 if (rp_ac >= rp_ac_alloc - 1) {
2238 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2239 rp_av = xrealloc(rp_av,
2240 rp_ac_alloc * sizeof(*rp_av));
2241 }
8d1d8f83 2242 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2243 continue;
2244 }
8d1d8f83
JH
2245 if (!strcmp("--thin", arg)) {
2246 use_internal_rev_list = 1;
2247 thin = 1;
2248 rp_av[1] = "--objects-edge";
b5d97e6b 2249 continue;
4ba7d711
NP
2250 }
2251 if (!prefixcmp(arg, "--index-version=")) {
2252 char *c;
aa7e44bf
GB
2253 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2254 if (pack_idx_default_version > 2)
4ba7d711
NP
2255 die("bad %s", arg);
2256 if (*c == ',')
aa7e44bf
GB
2257 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2258 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2259 die("bad %s", arg);
2260 continue;
b5d97e6b 2261 }
7f3140cd
JS
2262 if (!strcmp(arg, "--keep-true-parents")) {
2263 grafts_replace_parents = 0;
2264 continue;
2265 }
b5d97e6b
JH
2266 usage(pack_usage);
2267 }
2268
2269 /* Traditionally "pack-objects [options] base extra" failed;
2270 * we would however want to take refs parameter that would
2271 * have been given to upstream rev-list ourselves, which means
2272 * we somehow want to say what the base name is. So the
2273 * syntax would be:
2274 *
2275 * pack-objects [options] base <refs...>
2276 *
2277 * in other words, we would treat the first non-option as the
2278 * base_name and send everything else to the internal revision
2279 * walker.
2280 */
2281
2282 if (!pack_to_stdout)
2283 base_name = argv[i++];
2284
2285 if (pack_to_stdout != !base_name)
2286 usage(pack_usage);
2287
2b84b5a8
JS
2288 if (!pack_to_stdout && !pack_size_limit)
2289 pack_size_limit = pack_size_limit_cfg;
2290
01c12a23
DH
2291 if (pack_to_stdout && pack_size_limit)
2292 die("--max-pack-size cannot be used to build a pack for transfer.");
2293
8d1d8f83
JH
2294 if (!pack_to_stdout && thin)
2295 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2296
ca11b212
NP
2297 if (keep_unreachable && unpack_unreachable)
2298 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2299
833e3df1
AE
2300#ifdef THREADED_DELTA_SEARCH
2301 if (!delta_search_threads) /* --threads=0 means autodetect */
2302 delta_search_threads = online_cpus();
2303#endif
2304
b5d97e6b
JH
2305 prepare_packed_git();
2306
13aaf148 2307 if (progress)
dc6a0757 2308 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2309 if (!use_internal_rev_list)
2310 read_object_list_from_stdin();
8d1d8f83
JH
2311 else {
2312 rp_av[rp_ac] = NULL;
2313 get_object_list(rp_ac, rp_av);
2314 }
f0a24aa5
SP
2315 if (include_tag && nr_result)
2316 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2317 stop_progress(&progress_state);
96a02f8f 2318
f0b0af1b 2319 if (non_empty && !nr_result)
1c4a2912 2320 return 0;
f7ae6a93
NP
2321 if (nr_result)
2322 prepare_pack(window, depth);
d01fb92f 2323 write_pack_file();
ab7cd7bb 2324 if (progress)
6e1c2344
RJ
2325 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2326 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2327 written, written_delta, reused, reused_delta);
c323ac7d
LT
2328 return 0;
2329}