Make xmalloc and xrealloc thread-safe
[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
7eb151d6 20#ifndef NO_PTHREADS
833e3df1 21#include "thread-utils.h"
8ecce684
NP
22#include <pthread.h>
23#endif
24
812bdbc3
TF
25static const char pack_usage[] =
26 "git pack-objects [{ -q | --progress | --all-progress }]\n"
4f366275 27 " [--all-progress-implied]\n"
812bdbc3
TF
28 " [--max-pack-size=N] [--local] [--incremental]\n"
29 " [--window=N] [--window-memory=N] [--depth=N]\n"
30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
31 " [--threads=N] [--non-empty] [--revs [--unpacked | --all]*]\n"
32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
33 " [--keep-unreachable | --unpack-unreachable \n"
34 " [<ref-list | <object-list]";
c323ac7d 35
c323ac7d 36struct object_entry {
aa7e44bf 37 struct pack_idx_entry idx;
3f9ac8d2 38 unsigned long size; /* uncompressed size */
3f9ac8d2 39 struct packed_git *in_pack; /* already in pack */
6777a59f 40 off_t in_pack_offset;
a3fbf4df 41 struct object_entry *delta; /* delta base object */
82e5a82f 42 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
43 struct object_entry *delta_sibling; /* other deltified objects who
44 * uses the same base as me
45 */
074b2eea 46 void *delta_data; /* cached delta (uncompressed) */
a3fbf4df 47 unsigned long delta_size; /* delta data size (uncompressed) */
ed4a9031 48 unsigned long z_delta_size; /* delta data size (compressed) */
5a235b5e 49 unsigned int hash; /* name hint hash */
a3fbf4df
NP
50 enum object_type type;
51 enum object_type in_pack_type; /* could be delta */
52 unsigned char in_pack_header_size;
53 unsigned char preferred_base; /* we do not pack this, but is available
a74db82e 54 * to be used as the base object to delta
a3fbf4df
NP
55 * objects against.
56 */
a74db82e 57 unsigned char no_try_delta;
c323ac7d
LT
58};
59
3f9ac8d2 60/*
82e5a82f 61 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
62 * expanded). nr_objects & nr_alloc controls this array. They are stored
63 * in the order we see -- typically rev-list --objects order that gives us
64 * nice "minimum seek" order.
3f9ac8d2 65 */
e4d58311 66static struct object_entry *objects;
79814f42 67static struct pack_idx_entry **written_list;
d01fb92f 68static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
3f9ac8d2 69
96f1e58f 70static int non_empty;
a7de7130 71static int reuse_delta = 1, reuse_object = 1;
e5e9714a 72static int keep_unreachable, unpack_unreachable, include_tag;
96f1e58f
DR
73static int local;
74static int incremental;
e96fb9b8 75static int ignore_packed_keep;
be6b1914 76static int allow_ofs_delta;
d01fb92f 77static const char *base_name;
024701f1 78static int progress = 1;
4812a93a 79static int window = 10;
07cf0f24 80static unsigned long pack_size_limit, pack_size_limit_cfg;
618e613a 81static int depth = 50;
43cc2b42 82static int delta_search_threads;
df6d6101 83static int pack_to_stdout;
8d1d8f83 84static int num_preferred_base;
dc6a0757 85static struct progress *progress_state;
960ccca6
DH
86static int pack_compression_level = Z_DEFAULT_COMPRESSION;
87static int pack_compression_seen;
c323ac7d 88
074b2eea 89static unsigned long delta_cache_size = 0;
5749b0b2 90static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
e3dfddb3 91static unsigned long cache_max_small_delta_size = 1000;
074b2eea 92
a97773ce
BD
93static unsigned long window_memory_limit = 0;
94
3f9ac8d2
JH
95/*
96 * The object names in objects array are hashed with this hashtable,
e4d58311 97 * to help looking up the entry by object name.
3f9ac8d2
JH
98 * This hashtable is built after all the objects are seen.
99 */
96f1e58f
DR
100static int *object_ix;
101static int object_ix_hashsz;
3f9ac8d2 102
3f9ac8d2
JH
103/*
104 * stats
105 */
7cadf491
SP
106static uint32_t written, written_delta;
107static uint32_t reused, reused_delta;
3f9ac8d2 108
780e6e73 109
3613f9b4 110static void *get_delta(struct object_entry *entry)
c323ac7d 111{
3613f9b4
NP
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
21666f1a 114 enum object_type type;
c323ac7d 115
3613f9b4
NP
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
117 if (!buf)
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
120 if (!base_buf)
aa7e44bf 121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
3613f9b4 122 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 123 buf, size, &delta_size, 0);
3613f9b4 124 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 125 die("delta size changed");
3613f9b4
NP
126 free(buf);
127 free(base_buf);
c323ac7d
LT
128 return delta_buf;
129}
130
30ebb40a
NP
131static unsigned long do_compress(void **pptr, unsigned long size)
132{
133 z_stream stream;
134 void *in, *out;
135 unsigned long maxsize;
136
137 memset(&stream, 0, sizeof(stream));
138 deflateInit(&stream, pack_compression_level);
139 maxsize = deflateBound(&stream, size);
140
141 in = *pptr;
142 out = xmalloc(maxsize);
143 *pptr = out;
144
145 stream.next_in = in;
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (deflate(&stream, Z_FINISH) == Z_OK)
150 ; /* nothing */
151 deflateEnd(&stream);
152
153 free(in);
154 return stream.total_out;
155}
156
a733cb60
LT
157/*
158 * The per-object header is a pretty dense thing, which is
159 * - first byte: low four bits are "size", then three bits of "type",
160 * and the high bit is "size continues".
161 * - each byte afterwards: low seven bits are size continuation,
162 * with the high bit being "size continues"
163 */
164static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
165{
01247d87 166 int n = 1;
a733cb60
LT
167 unsigned char c;
168
eb32d236 169 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
170 die("bad type %d", type);
171
01247d87
LT
172 c = (type << 4) | (size & 15);
173 size >>= 4;
174 while (size) {
a733cb60 175 *hdr++ = c | 0x80;
01247d87
LT
176 c = size & 0x7f;
177 size >>= 7;
178 n++;
a733cb60
LT
179 }
180 *hdr = c;
181 return n;
182}
183
780e6e73
NP
184/*
185 * we are going to reuse the existing object data as is. make
186 * sure it is not corrupt.
187 */
079afb18
SP
188static int check_pack_inflate(struct packed_git *p,
189 struct pack_window **w_curs,
6777a59f
SP
190 off_t offset,
191 off_t len,
079afb18
SP
192 unsigned long expect)
193{
194 z_stream stream;
195 unsigned char fakebuf[4096], *in;
196 int st;
197
198 memset(&stream, 0, sizeof(stream));
39c68542 199 git_inflate_init(&stream);
079afb18
SP
200 do {
201 in = use_pack(p, w_curs, offset, &stream.avail_in);
202 stream.next_in = in;
203 stream.next_out = fakebuf;
204 stream.avail_out = sizeof(fakebuf);
39c68542 205 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
206 offset += stream.next_in - in;
207 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 208 git_inflate_end(&stream);
079afb18
SP
209 return (st == Z_STREAM_END &&
210 stream.total_out == expect &&
211 stream.total_in == len) ? 0 : -1;
212}
213
214static void copy_pack_data(struct sha1file *f,
215 struct packed_git *p,
216 struct pack_window **w_curs,
6777a59f
SP
217 off_t offset,
218 off_t len)
079afb18
SP
219{
220 unsigned char *in;
221 unsigned int avail;
222
223 while (len) {
224 in = use_pack(p, w_curs, offset, &avail);
225 if (avail > len)
6777a59f 226 avail = (unsigned int)len;
079afb18
SP
227 sha1write(f, in, avail);
228 offset += avail;
229 len -= avail;
230 }
231}
232
d7dd0223 233static unsigned long write_object(struct sha1file *f,
17b08f2c
DH
234 struct object_entry *entry,
235 off_t write_offset)
c323ac7d 236{
30ebb40a 237 unsigned long size, limit, datalen;
3f9ac8d2 238 void *buf;
2c5ef824 239 unsigned char header[10], dheader[10];
6777a59f 240 unsigned hdrlen;
2c5ef824
NP
241 enum object_type type;
242 int usable_delta, to_reuse;
c323ac7d 243
78d1e84f
NP
244 if (!pack_to_stdout)
245 crc32_begin(f);
246
2c5ef824
NP
247 type = entry->type;
248
a2430dde 249 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
250 if (!pack_size_limit || !nr_written)
251 limit = 0;
252 else if (pack_size_limit <= write_offset)
253 /*
254 * the earlier object did not fit the limit; avoid
255 * mistaking this with unlimited (i.e. limit = 0).
256 */
257 limit = 1;
258 else
259 limit = pack_size_limit - write_offset;
2c5ef824
NP
260
261 if (!entry->delta)
262 usable_delta = 0; /* no delta */
263 else if (!pack_size_limit)
264 usable_delta = 1; /* unlimited packfile */
265 else if (entry->delta->idx.offset == (off_t)-1)
266 usable_delta = 0; /* base was written to another pack */
267 else if (entry->delta->idx.offset)
268 usable_delta = 1; /* base already exists in this pack */
269 else
270 usable_delta = 0; /* base could end up in another pack */
271
a7de7130 272 if (!reuse_object)
fa736f72
NP
273 to_reuse = 0; /* explicit */
274 else if (!entry->in_pack)
ab7cd7bb 275 to_reuse = 0; /* can't reuse what we don't have */
2c5ef824 276 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
17b08f2c
DH
277 /* check_object() decided it for us ... */
278 to_reuse = usable_delta;
279 /* ... but pack split may override that */
2c5ef824 280 else if (type != entry->in_pack_type)
ab7cd7bb
JH
281 to_reuse = 0; /* pack has delta which is unusable */
282 else if (entry->delta)
283 to_reuse = 0; /* we want to pack afresh */
284 else
285 to_reuse = 1; /* we have it in-pack undeltified,
286 * and we do not need to deltify it.
287 */
288
df6d6101 289 if (!to_reuse) {
64bd76b1 290 no_reuse:
5476a8ad 291 if (!usable_delta) {
2c5ef824 292 buf = read_sha1_file(entry->idx.sha1, &type, &size);
5476a8ad 293 if (!buf)
aa7e44bf 294 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
ed4a9031
NP
295 /*
296 * make sure no cached delta data remains from a
3ea3c215 297 * previous attempt before a pack split occurred.
ed4a9031
NP
298 */
299 free(entry->delta_data);
300 entry->delta_data = NULL;
301 entry->z_delta_size = 0;
5476a8ad 302 } else if (entry->delta_data) {
3f9ac8d2 303 size = entry->delta_size;
5476a8ad
NP
304 buf = entry->delta_data;
305 entry->delta_data = NULL;
2c5ef824 306 type = (allow_ofs_delta && entry->delta->idx.offset) ?
be6b1914 307 OBJ_OFS_DELTA : OBJ_REF_DELTA;
17b08f2c 308 } else {
3613f9b4 309 buf = get_delta(entry);
5476a8ad 310 size = entry->delta_size;
2c5ef824 311 type = (allow_ofs_delta && entry->delta->idx.offset) ?
5476a8ad 312 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2 313 }
ed4a9031
NP
314
315 if (entry->z_delta_size)
316 datalen = entry->z_delta_size;
317 else
318 datalen = do_compress(&buf, size);
8c3c7b2a 319
3f9ac8d2
JH
320 /*
321 * The object header is a byte of 'type' followed by zero or
be6b1914 322 * more bytes of length.
3f9ac8d2 323 */
2c5ef824 324 hdrlen = encode_header(type, size, header);
3f9ac8d2 325
2c5ef824 326 if (type == OBJ_OFS_DELTA) {
be6b1914
NP
327 /*
328 * Deltas with relative base contain an additional
329 * encoding of the relative offset for the delta
330 * base from this object's position in the pack.
331 */
aa7e44bf 332 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
333 unsigned pos = sizeof(dheader) - 1;
334 dheader[pos] = ofs & 127;
be6b1914 335 while (ofs >>= 7)
17b08f2c
DH
336 dheader[--pos] = 128 | (--ofs & 127);
337 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
17b08f2c
DH
338 free(buf);
339 return 0;
340 }
341 sha1write(f, header, hdrlen);
342 sha1write(f, dheader + pos, sizeof(dheader) - pos);
343 hdrlen += sizeof(dheader) - pos;
2c5ef824 344 } else if (type == OBJ_REF_DELTA) {
be6b1914
NP
345 /*
346 * Deltas with a base reference contain
347 * an additional 20 bytes for the base sha1.
348 */
17b08f2c 349 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
17b08f2c
DH
350 free(buf);
351 return 0;
352 }
353 sha1write(f, header, hdrlen);
aa7e44bf 354 sha1write(f, entry->delta->idx.sha1, 20);
3f9ac8d2 355 hdrlen += 20;
17b08f2c
DH
356 } else {
357 if (limit && hdrlen + datalen + 20 >= limit) {
17b08f2c
DH
358 free(buf);
359 return 0;
360 }
361 sha1write(f, header, hdrlen);
3f9ac8d2 362 }
30ebb40a 363 sha1write(f, buf, datalen);
3f9ac8d2 364 free(buf);
c323ac7d 365 }
3f9ac8d2
JH
366 else {
367 struct packed_git *p = entry->in_pack;
03e79c88 368 struct pack_window *w_curs = NULL;
91ecbeca 369 struct revindex_entry *revidx;
6777a59f 370 off_t offset;
3f9ac8d2 371
64bd76b1 372 if (entry->delta)
2c5ef824 373 type = (allow_ofs_delta && entry->delta->idx.offset) ?
780e6e73 374 OBJ_OFS_DELTA : OBJ_REF_DELTA;
2c5ef824 375 hdrlen = encode_header(type, entry->size, header);
64bd76b1 376
91ecbeca 377 offset = entry->in_pack_offset;
3449f8c4 378 revidx = find_pack_revindex(p, offset);
91ecbeca
NP
379 datalen = revidx[1].offset - offset;
380 if (!pack_to_stdout && p->index_version > 1 &&
64bd76b1
NP
381 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
382 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
383 unuse_pack(&w_curs);
384 goto no_reuse;
385 }
386
91ecbeca
NP
387 offset += entry->in_pack_header_size;
388 datalen -= entry->in_pack_header_size;
64bd76b1
NP
389 if (!pack_to_stdout && p->index_version == 1 &&
390 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
391 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
392 unuse_pack(&w_curs);
393 goto no_reuse;
394 }
395
2c5ef824 396 if (type == OBJ_OFS_DELTA) {
aa7e44bf 397 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
398 unsigned pos = sizeof(dheader) - 1;
399 dheader[pos] = ofs & 127;
400 while (ofs >>= 7)
401 dheader[--pos] = 128 | (--ofs & 127);
59dd9ed1
NP
402 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
403 unuse_pack(&w_curs);
17b08f2c 404 return 0;
59dd9ed1 405 }
17b08f2c
DH
406 sha1write(f, header, hdrlen);
407 sha1write(f, dheader + pos, sizeof(dheader) - pos);
408 hdrlen += sizeof(dheader) - pos;
64bd76b1 409 reused_delta++;
2c5ef824 410 } else if (type == OBJ_REF_DELTA) {
59dd9ed1
NP
411 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
412 unuse_pack(&w_curs);
17b08f2c 413 return 0;
59dd9ed1 414 }
17b08f2c 415 sha1write(f, header, hdrlen);
aa7e44bf 416 sha1write(f, entry->delta->idx.sha1, 20);
17b08f2c 417 hdrlen += 20;
64bd76b1 418 reused_delta++;
17b08f2c 419 } else {
59dd9ed1
NP
420 if (limit && hdrlen + datalen + 20 >= limit) {
421 unuse_pack(&w_curs);
17b08f2c 422 return 0;
59dd9ed1 423 }
17b08f2c
DH
424 sha1write(f, header, hdrlen);
425 }
079afb18 426 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 427 unuse_pack(&w_curs);
3f9ac8d2 428 reused++;
a733cb60 429 }
17b08f2c 430 if (usable_delta)
ab7cd7bb 431 written_delta++;
3f9ac8d2 432 written++;
78d1e84f 433 if (!pack_to_stdout)
aa7e44bf 434 entry->idx.crc32 = crc32_end(f);
c323ac7d
LT
435 return hdrlen + datalen;
436}
437
6ed7f25e 438static int write_one(struct sha1file *f,
9d5ab962 439 struct object_entry *e,
6ed7f25e 440 off_t *offset)
9d5ab962 441{
d7dd0223
NP
442 unsigned long size;
443
444 /* offset is non zero if object is written already. */
aa7e44bf 445 if (e->idx.offset || e->preferred_base)
a2430dde 446 return -1;
d7dd0223 447
720c9f7b
NP
448 /* if we are deltified, write out base object first. */
449 if (e->delta && !write_one(f, e->delta, offset))
450 return 0;
d7dd0223 451
6ed7f25e
NP
452 e->idx.offset = *offset;
453 size = write_object(f, e, *offset);
17b08f2c 454 if (!size) {
aa7e44bf 455 e->idx.offset = 0;
17b08f2c
DH
456 return 0;
457 }
79814f42 458 written_list[nr_written++] = &e->idx;
d7dd0223
NP
459
460 /* make sure off_t is sufficiently large not to wrap */
6ed7f25e 461 if (*offset > *offset + size)
d7dd0223 462 die("pack too large for current definition of off_t");
6ed7f25e
NP
463 *offset += size;
464 return 1;
9d5ab962
JH
465}
466
d01fb92f 467static void write_pack_file(void)
c323ac7d 468{
ebe27b13 469 uint32_t i = 0, j;
d22b9290 470 struct sha1file *f;
6ed7f25e 471 off_t offset;
a733cb60 472 struct pack_header hdr;
ebe27b13 473 uint32_t nr_remaining = nr_result;
f746bae8 474 time_t last_mtime = 0;
81a216a5 475
bcd7954e 476 if (progress > pack_to_stdout)
dc6a0757 477 progress_state = start_progress("Writing objects", nr_result);
79814f42 478 written_list = xmalloc(nr_objects * sizeof(*written_list));
183bdb2c 479
ebe27b13 480 do {
aa7e44bf 481 unsigned char sha1[20];
7ba502c4 482 char *pack_tmp_name = NULL;
aa7e44bf 483
ebe27b13 484 if (pack_to_stdout) {
2a128d63 485 f = sha1fd_throughput(1, "<stdout>", progress_state);
ebe27b13 486 } else {
7ba502c4
NP
487 char tmpname[PATH_MAX];
488 int fd;
6e180cdc
JH
489 fd = odb_mkstemp(tmpname, sizeof(tmpname),
490 "pack/tmp_pack_XXXXXX");
ebe27b13
DH
491 pack_tmp_name = xstrdup(tmpname);
492 f = sha1fd(fd, pack_tmp_name);
493 }
494
495 hdr.hdr_signature = htonl(PACK_SIGNATURE);
496 hdr.hdr_version = htonl(PACK_VERSION);
497 hdr.hdr_entries = htonl(nr_remaining);
498 sha1write(f, &hdr, sizeof(hdr));
499 offset = sizeof(hdr);
500 nr_written = 0;
720c9f7b
NP
501 for (; i < nr_objects; i++) {
502 if (!write_one(f, objects + i, &offset))
503 break;
504 display_progress(progress_state, written);
505 }
c553ca25 506
ebe27b13
DH
507 /*
508 * Did we write the wrong # entries in the header?
509 * If so, rewrite it like in fast-import
510 */
54352bb2
LT
511 if (pack_to_stdout) {
512 sha1close(f, sha1, CSUM_CLOSE);
513 } else if (nr_written == nr_remaining) {
514 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 515 } else {
ac0463ed 516 int fd = sha1close(f, sha1, 0);
abeb40e5 517 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 518 nr_written, sha1, offset);
7ba502c4 519 close(fd);
ebe27b13
DH
520 }
521
522 if (!pack_to_stdout) {
f746bae8 523 struct stat st;
3bb72562
LT
524 const char *idx_tmp_name;
525 char tmpname[PATH_MAX];
d01fb92f 526
79814f42
NP
527 idx_tmp_name = write_idx_file(NULL, written_list,
528 nr_written, sha1);
f746bae8 529
d01fb92f 530 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
aa7e44bf 531 base_name, sha1_to_hex(sha1));
c74faea1 532 free_pack_by_name(tmpname);
f80c7ae8 533 if (adjust_shared_perm(pack_tmp_name))
d824cbba 534 die_errno("unable to make temporary pack file readable");
d01fb92f 535 if (rename(pack_tmp_name, tmpname))
d824cbba 536 die_errno("unable to rename temporary pack file");
f746bae8
NP
537
538 /*
539 * Packs are runtime accessed in their mtime
540 * order since newer packs are more likely to contain
541 * younger objects. So if we are creating multiple
542 * packs then we should modify the mtime of later ones
543 * to preserve this property.
544 */
545 if (stat(tmpname, &st) < 0) {
546 warning("failed to stat %s: %s",
547 tmpname, strerror(errno));
548 } else if (!last_mtime) {
549 last_mtime = st.st_mtime;
550 } else {
551 struct utimbuf utb;
552 utb.actime = st.st_atime;
553 utb.modtime = --last_mtime;
554 if (utime(tmpname, &utb) < 0)
555 warning("failed utime() on %s: %s",
556 tmpname, strerror(errno));
557 }
558
d01fb92f 559 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 560 base_name, sha1_to_hex(sha1));
f80c7ae8 561 if (adjust_shared_perm(idx_tmp_name))
d824cbba 562 die_errno("unable to make temporary index file readable");
d01fb92f 563 if (rename(idx_tmp_name, tmpname))
d824cbba 564 die_errno("unable to rename temporary index file");
f746bae8 565
3bb72562 566 free((void *) idx_tmp_name);
7ba502c4 567 free(pack_tmp_name);
aa7e44bf 568 puts(sha1_to_hex(sha1));
ebe27b13
DH
569 }
570
571 /* mark written objects as written to previous pack */
572 for (j = 0; j < nr_written; j++) {
79814f42 573 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
574 }
575 nr_remaining -= nr_written;
720c9f7b 576 } while (nr_remaining && i < nr_objects);
ebe27b13
DH
577
578 free(written_list);
4d4fcc54 579 stop_progress(&progress_state);
67c08ce1 580 if (written != nr_result)
6e1c2344
RJ
581 die("wrote %"PRIu32" objects while expecting %"PRIu32,
582 written, nr_result);
c323ac7d
LT
583}
584
7a979d99
JH
585static int locate_object_entry_hash(const unsigned char *sha1)
586{
587 int i;
588 unsigned int ui;
589 memcpy(&ui, sha1, sizeof(unsigned int));
590 i = ui % object_ix_hashsz;
591 while (0 < object_ix[i]) {
aa7e44bf 592 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
593 return i;
594 if (++i == object_ix_hashsz)
595 i = 0;
596 }
597 return -1 - i;
598}
599
600static struct object_entry *locate_object_entry(const unsigned char *sha1)
601{
602 int i;
603
604 if (!object_ix_hashsz)
605 return NULL;
606
607 i = locate_object_entry_hash(sha1);
608 if (0 <= i)
609 return &objects[object_ix[i]-1];
610 return NULL;
611}
612
613static void rehash_objects(void)
c323ac7d 614{
7cadf491 615 uint32_t i;
7a979d99
JH
616 struct object_entry *oe;
617
618 object_ix_hashsz = nr_objects * 3;
619 if (object_ix_hashsz < 1024)
620 object_ix_hashsz = 1024;
621 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 622 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 623 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 624 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
625 if (0 <= ix)
626 continue;
627 ix = -1 - ix;
628 object_ix[ix] = i + 1;
629 }
630}
631
ce0bd642 632static unsigned name_hash(const char *name)
1d6b38cc 633{
48fb7deb 634 unsigned c, hash = 0;
ce0bd642 635
bc32fed5
JH
636 if (!name)
637 return 0;
638
eeef7135 639 /*
ce0bd642
LT
640 * This effectively just creates a sortable number from the
641 * last sixteen non-whitespace characters. Last characters
642 * count "most", so things that end in ".c" sort together.
eeef7135 643 */
ce0bd642
LT
644 while ((c = *name++) != 0) {
645 if (isspace(c))
646 continue;
647 hash = (hash >> 2) + (c << 24);
648 }
1d6b38cc
JH
649 return hash;
650}
651
a74db82e
JH
652static void setup_delta_attr_check(struct git_attr_check *check)
653{
654 static struct git_attr *attr_delta;
655
656 if (!attr_delta)
7fb0eaa2 657 attr_delta = git_attr("delta");
a74db82e
JH
658
659 check[0].attr = attr_delta;
660}
661
662static int no_try_delta(const char *path)
663{
664 struct git_attr_check check[1];
665
666 setup_delta_attr_check(check);
667 if (git_checkattr(path, ARRAY_SIZE(check), check))
668 return 0;
669 if (ATTR_FALSE(check->value))
670 return 1;
671 return 0;
672}
673
5c49c116 674static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 675 const char *name, int exclude)
c323ac7d 676{
c323ac7d 677 struct object_entry *entry;
29b734e4 678 struct packed_git *p, *found_pack = NULL;
6777a59f 679 off_t found_offset = 0;
29b734e4 680 int ix;
bc32fed5 681 unsigned hash = name_hash(name);
29b734e4
NP
682
683 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
684 if (ix >= 0) {
685 if (exclude) {
686 entry = objects + object_ix[ix] - 1;
81a216a5
NP
687 if (!entry->preferred_base)
688 nr_result--;
29b734e4
NP
689 entry->preferred_base = 1;
690 }
691 return 0;
692 }
c323ac7d 693
daae0625
BC
694 if (!exclude && local && has_loose_object_nonlocal(sha1))
695 return 0;
696
5c49c116
NP
697 for (p = packed_git; p; p = p->next) {
698 off_t offset = find_pack_entry_one(sha1, p);
699 if (offset) {
700 if (!found_pack) {
701 found_offset = offset;
702 found_pack = p;
64560374 703 }
5c49c116
NP
704 if (exclude)
705 break;
706 if (incremental)
707 return 0;
708 if (local && !p->pack_local)
709 return 0;
e96fb9b8
BC
710 if (ignore_packed_keep && p->pack_local && p->pack_keep)
711 return 0;
64560374
LT
712 }
713 }
eb019375 714
29b734e4
NP
715 if (nr_objects >= nr_alloc) {
716 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 717 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 718 }
29b734e4
NP
719
720 entry = objects + nr_objects++;
c323ac7d 721 memset(entry, 0, sizeof(*entry));
aa7e44bf 722 hashcpy(entry->idx.sha1, sha1);
27225f2e 723 entry->hash = hash;
5c49c116
NP
724 if (type)
725 entry->type = type;
29b734e4
NP
726 if (exclude)
727 entry->preferred_base = 1;
81a216a5
NP
728 else
729 nr_result++;
29b734e4
NP
730 if (found_pack) {
731 entry->in_pack = found_pack;
732 entry->in_pack_offset = found_offset;
733 }
7a979d99
JH
734
735 if (object_ix_hashsz * 3 <= nr_objects * 4)
736 rehash_objects();
29b734e4
NP
737 else
738 object_ix[-1 - ix] = nr_objects;
7a979d99 739
4d4fcc54 740 display_progress(progress_state, nr_objects);
29b734e4 741
a74db82e
JH
742 if (name && no_try_delta(name))
743 entry->no_try_delta = 1;
744
29b734e4 745 return 1;
c323ac7d
LT
746}
747
5379a5c5
JH
748struct pbase_tree_cache {
749 unsigned char sha1[20];
750 int ref;
751 int temporary;
752 void *tree_data;
753 unsigned long tree_size;
754};
755
756static struct pbase_tree_cache *(pbase_tree_cache[256]);
757static int pbase_tree_cache_ix(const unsigned char *sha1)
758{
759 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
760}
761static int pbase_tree_cache_ix_incr(int ix)
762{
763 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
764}
765
766static struct pbase_tree {
767 struct pbase_tree *next;
768 /* This is a phony "cache" entry; we are not
769 * going to evict it nor find it through _get()
770 * mechanism -- this is for the toplevel node that
771 * would almost always change with any commit.
772 */
773 struct pbase_tree_cache pcache;
774} *pbase_tree;
775
776static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
777{
778 struct pbase_tree_cache *ent, *nent;
779 void *data;
780 unsigned long size;
21666f1a 781 enum object_type type;
5379a5c5
JH
782 int neigh;
783 int my_ix = pbase_tree_cache_ix(sha1);
784 int available_ix = -1;
785
786 /* pbase-tree-cache acts as a limited hashtable.
787 * your object will be found at your index or within a few
788 * slots after that slot if it is cached.
789 */
790 for (neigh = 0; neigh < 8; neigh++) {
791 ent = pbase_tree_cache[my_ix];
a89fccd2 792 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
793 ent->ref++;
794 return ent;
795 }
796 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
797 ((0 <= available_ix) &&
798 (!ent && pbase_tree_cache[available_ix])))
799 available_ix = my_ix;
800 if (!ent)
801 break;
802 my_ix = pbase_tree_cache_ix_incr(my_ix);
803 }
804
805 /* Did not find one. Either we got a bogus request or
806 * we need to read and perhaps cache.
807 */
21666f1a 808 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
809 if (!data)
810 return NULL;
21666f1a 811 if (type != OBJ_TREE) {
5379a5c5
JH
812 free(data);
813 return NULL;
814 }
815
816 /* We need to either cache or return a throwaway copy */
817
818 if (available_ix < 0)
819 ent = NULL;
820 else {
821 ent = pbase_tree_cache[available_ix];
822 my_ix = available_ix;
823 }
824
825 if (!ent) {
826 nent = xmalloc(sizeof(*nent));
827 nent->temporary = (available_ix < 0);
828 }
829 else {
830 /* evict and reuse */
831 free(ent->tree_data);
832 nent = ent;
833 }
e702496e 834 hashcpy(nent->sha1, sha1);
5379a5c5
JH
835 nent->tree_data = data;
836 nent->tree_size = size;
837 nent->ref = 1;
838 if (!nent->temporary)
839 pbase_tree_cache[my_ix] = nent;
840 return nent;
841}
842
843static void pbase_tree_put(struct pbase_tree_cache *cache)
844{
845 if (!cache->temporary) {
846 cache->ref--;
847 return;
848 }
849 free(cache->tree_data);
850 free(cache);
851}
852
853static int name_cmp_len(const char *name)
854{
855 int i;
856 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
857 ;
858 return i;
859}
860
861static void add_pbase_object(struct tree_desc *tree,
5379a5c5 862 const char *name,
ce0bd642
LT
863 int cmplen,
864 const char *fullname)
3f9ac8d2 865{
4c068a98 866 struct name_entry entry;
8a5a8d6c 867 int cmp;
4c068a98
LT
868
869 while (tree_entry(tree,&entry)) {
1211be6b
LT
870 if (S_ISGITLINK(entry.mode))
871 continue;
8a5a8d6c
NP
872 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
873 memcmp(name, entry.path, cmplen);
874 if (cmp > 0)
7a979d99 875 continue;
8a5a8d6c
NP
876 if (cmp < 0)
877 return;
5379a5c5 878 if (name[cmplen] != '/') {
5c49c116 879 add_object_entry(entry.sha1,
4d1012c3 880 object_type(entry.mode),
bc32fed5 881 fullname, 1);
5379a5c5
JH
882 return;
883 }
8a5a8d6c 884 if (S_ISDIR(entry.mode)) {
7a979d99 885 struct tree_desc sub;
5379a5c5
JH
886 struct pbase_tree_cache *tree;
887 const char *down = name+cmplen+1;
888 int downlen = name_cmp_len(down);
889
4c068a98 890 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
891 if (!tree)
892 return;
6fda5e51 893 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 894
ce0bd642 895 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
896 pbase_tree_put(tree);
897 }
898 }
899}
1d6b38cc 900
5379a5c5
JH
901static unsigned *done_pbase_paths;
902static int done_pbase_paths_num;
903static int done_pbase_paths_alloc;
904static int done_pbase_path_pos(unsigned hash)
905{
906 int lo = 0;
907 int hi = done_pbase_paths_num;
908 while (lo < hi) {
909 int mi = (hi + lo) / 2;
910 if (done_pbase_paths[mi] == hash)
911 return mi;
912 if (done_pbase_paths[mi] < hash)
913 hi = mi;
914 else
915 lo = mi + 1;
916 }
917 return -lo-1;
918}
919
920static int check_pbase_path(unsigned hash)
921{
922 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
923 if (0 <= pos)
924 return 1;
925 pos = -pos - 1;
926 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
927 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
928 done_pbase_paths = xrealloc(done_pbase_paths,
929 done_pbase_paths_alloc *
930 sizeof(unsigned));
931 }
932 done_pbase_paths_num++;
933 if (pos < done_pbase_paths_num)
934 memmove(done_pbase_paths + pos + 1,
935 done_pbase_paths + pos,
936 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
937 done_pbase_paths[pos] = hash;
938 return 0;
939}
940
bc32fed5 941static void add_preferred_base_object(const char *name)
5379a5c5
JH
942{
943 struct pbase_tree *it;
8a5a8d6c 944 int cmplen;
bc32fed5 945 unsigned hash = name_hash(name);
5379a5c5 946
8a5a8d6c 947 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
948 return;
949
8a5a8d6c 950 cmplen = name_cmp_len(name);
5379a5c5
JH
951 for (it = pbase_tree; it; it = it->next) {
952 if (cmplen == 0) {
bc32fed5 953 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
954 }
955 else {
956 struct tree_desc tree;
6fda5e51 957 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 958 add_pbase_object(&tree, name, cmplen, name);
7a979d99 959 }
3f9ac8d2 960 }
3f9ac8d2
JH
961}
962
7a979d99 963static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 964{
5379a5c5
JH
965 struct pbase_tree *it;
966 void *data;
967 unsigned long size;
968 unsigned char tree_sha1[20];
1d6b38cc 969
8d1d8f83
JH
970 if (window <= num_preferred_base++)
971 return;
972
5379a5c5
JH
973 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
974 if (!data)
7a979d99 975 return;
5379a5c5
JH
976
977 for (it = pbase_tree; it; it = it->next) {
a89fccd2 978 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
979 free(data);
980 return;
981 }
982 }
983
984 it = xcalloc(1, sizeof(*it));
985 it->next = pbase_tree;
986 pbase_tree = it;
987
e702496e 988 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
989 it->pcache.tree_data = data;
990 it->pcache.tree_size = size;
3f9ac8d2
JH
991}
992
0ef95f72
NP
993static void cleanup_preferred_base(void)
994{
995 struct pbase_tree *it;
996 unsigned i;
997
998 it = pbase_tree;
999 pbase_tree = NULL;
1000 while (it) {
1001 struct pbase_tree *this = it;
1002 it = this->next;
1003 free(this->pcache.tree_data);
1004 free(this);
1005 }
1006
1007 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1008 if (!pbase_tree_cache[i])
1009 continue;
1010 free(pbase_tree_cache[i]->tree_data);
1011 free(pbase_tree_cache[i]);
1012 pbase_tree_cache[i] = NULL;
1013 }
1014
1015 free(done_pbase_paths);
1016 done_pbase_paths = NULL;
1017 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1018}
1019
c323ac7d
LT
1020static void check_object(struct object_entry *entry)
1021{
5c49c116 1022 if (entry->in_pack) {
780e6e73 1023 struct packed_git *p = entry->in_pack;
03e79c88 1024 struct pack_window *w_curs = NULL;
5c49c116
NP
1025 const unsigned char *base_ref = NULL;
1026 struct object_entry *base_entry;
1027 unsigned long used, used_0;
6777a59f 1028 unsigned int avail;
5c49c116
NP
1029 off_t ofs;
1030 unsigned char *buf, c;
780e6e73 1031
6777a59f 1032 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1033
5c49c116 1034 /*
fa736f72
NP
1035 * We want in_pack_type even if we do not reuse delta
1036 * since non-delta representations could still be reused.
ab7cd7bb 1037 */
09ded04b 1038 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1039 &entry->in_pack_type,
1040 &entry->size);
03d66015
NP
1041 if (used == 0)
1042 goto give_up;
ab7cd7bb 1043
5c49c116
NP
1044 /*
1045 * Determine if this is a delta and if so whether we can
1046 * reuse it or not. Otherwise let's find out as cheaply as
1047 * possible what the actual type and size for this object is.
3f9ac8d2 1048 */
5c49c116
NP
1049 switch (entry->in_pack_type) {
1050 default:
1051 /* Not a delta hence we've already got all we need. */
1052 entry->type = entry->in_pack_type;
1053 entry->in_pack_header_size = used;
03d66015
NP
1054 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1055 goto give_up;
5c49c116
NP
1056 unuse_pack(&w_curs);
1057 return;
1058 case OBJ_REF_DELTA:
a7de7130 1059 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1060 base_ref = use_pack(p, &w_curs,
1061 entry->in_pack_offset + used, NULL);
1062 entry->in_pack_header_size = used + 20;
1063 break;
1064 case OBJ_OFS_DELTA:
1065 buf = use_pack(p, &w_curs,
1066 entry->in_pack_offset + used, NULL);
1067 used_0 = 0;
1068 c = buf[used_0++];
1069 ofs = c & 127;
1070 while (c & 128) {
1071 ofs += 1;
03d66015
NP
1072 if (!ofs || MSB(ofs, 7)) {
1073 error("delta base offset overflow in pack for %s",
1074 sha1_to_hex(entry->idx.sha1));
1075 goto give_up;
1076 }
5c49c116
NP
1077 c = buf[used_0++];
1078 ofs = (ofs << 7) + (c & 127);
780e6e73 1079 }
5c49c116 1080 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1081 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1082 error("delta base offset out of bound for %s",
1083 sha1_to_hex(entry->idx.sha1));
1084 goto give_up;
1085 }
a7de7130 1086 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1087 struct revindex_entry *revidx;
1088 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1089 if (!revidx)
1090 goto give_up;
3449f8c4
NP
1091 base_ref = nth_packed_object_sha1(p, revidx->nr);
1092 }
5c49c116
NP
1093 entry->in_pack_header_size = used + used_0;
1094 break;
780e6e73 1095 }
780e6e73 1096
5c49c116
NP
1097 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1098 /*
1099 * If base_ref was set above that means we wish to
1100 * reuse delta data, and we even found that base
1101 * in the list of objects we want to pack. Goodie!
1102 *
1103 * Depth value does not matter - find_deltas() will
1104 * never consider reused delta as the base object to
1105 * deltify other objects against, in order to avoid
1106 * circular deltas.
3f9ac8d2 1107 */
780e6e73 1108 entry->type = entry->in_pack_type;
5c49c116 1109 entry->delta = base_entry;
64bd76b1 1110 entry->delta_size = entry->size;
15b4d577
JH
1111 entry->delta_sibling = base_entry->delta_child;
1112 base_entry->delta_child = entry;
5c49c116
NP
1113 unuse_pack(&w_curs);
1114 return;
1115 }
ab7cd7bb 1116
5c49c116
NP
1117 if (entry->type) {
1118 /*
1119 * This must be a delta and we already know what the
1120 * final object type is. Let's extract the actual
1121 * object size from the delta header.
1122 */
1123 entry->size = get_size_from_delta(p, &w_curs,
1124 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1125 if (entry->size == 0)
1126 goto give_up;
5c49c116 1127 unuse_pack(&w_curs);
3f9ac8d2
JH
1128 return;
1129 }
5c49c116
NP
1130
1131 /*
1132 * No choice but to fall back to the recursive delta walk
1133 * with sha1_object_info() to find about the object type
1134 * at this point...
1135 */
03d66015 1136 give_up:
5c49c116 1137 unuse_pack(&w_curs);
36e4d74a 1138 }
3f9ac8d2 1139
aa7e44bf 1140 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1141 /*
1142 * The error condition is checked in prepare_pack(). This is
1143 * to permit a missing preferred base object to be ignored
1144 * as a preferred base. Doing so can result in a larger
1145 * pack file, but the transfer will still take place.
1146 */
3f9ac8d2
JH
1147}
1148
5c49c116
NP
1149static int pack_offset_sort(const void *_a, const void *_b)
1150{
1151 const struct object_entry *a = *(struct object_entry **)_a;
1152 const struct object_entry *b = *(struct object_entry **)_b;
1153
1154 /* avoid filesystem trashing with loose objects */
1155 if (!a->in_pack && !b->in_pack)
aa7e44bf 1156 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1157
1158 if (a->in_pack < b->in_pack)
1159 return -1;
1160 if (a->in_pack > b->in_pack)
1161 return 1;
1162 return a->in_pack_offset < b->in_pack_offset ? -1 :
1163 (a->in_pack_offset > b->in_pack_offset);
1164}
1165
c323ac7d
LT
1166static void get_object_details(void)
1167{
7cadf491 1168 uint32_t i;
5c49c116
NP
1169 struct object_entry **sorted_by_offset;
1170
1171 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1172 for (i = 0; i < nr_objects; i++)
1173 sorted_by_offset[i] = objects + i;
1174 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1175
5c49c116
NP
1176 for (i = 0; i < nr_objects; i++)
1177 check_object(sorted_by_offset[i]);
3449f8c4 1178
5c49c116 1179 free(sorted_by_offset);
c323ac7d
LT
1180}
1181
b904166c
NP
1182/*
1183 * We search for deltas in a list sorted by type, by filename hash, and then
1184 * by size, so that we see progressively smaller and smaller files.
1185 * That's because we prefer deltas to be from the bigger file
1186 * to the smaller -- deletes are potentially cheaper, but perhaps
1187 * more importantly, the bigger file is likely the more recent
1188 * one. The deepest deltas are therefore the oldest objects which are
1189 * less susceptible to be accessed often.
1190 */
9668cf59 1191static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1192{
9668cf59
NP
1193 const struct object_entry *a = *(struct object_entry **)_a;
1194 const struct object_entry *b = *(struct object_entry **)_b;
1195
c323ac7d 1196 if (a->type > b->type)
27225f2e 1197 return -1;
b904166c 1198 if (a->type < b->type)
27225f2e 1199 return 1;
b904166c 1200 if (a->hash > b->hash)
7a979d99 1201 return -1;
b904166c 1202 if (a->hash < b->hash)
7a979d99 1203 return 1;
b904166c 1204 if (a->preferred_base > b->preferred_base)
c323ac7d 1205 return -1;
b904166c
NP
1206 if (a->preferred_base < b->preferred_base)
1207 return 1;
c323ac7d 1208 if (a->size > b->size)
b904166c
NP
1209 return -1;
1210 if (a->size < b->size)
c323ac7d 1211 return 1;
b904166c 1212 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1213}
1214
1215struct unpacked {
1216 struct object_entry *entry;
1217 void *data;
f6c7081a 1218 struct delta_index *index;
5a235b5e 1219 unsigned depth;
c323ac7d
LT
1220};
1221
d250626c
NP
1222static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1223 unsigned long delta_size)
074b2eea
MK
1224{
1225 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1226 return 0;
1227
e3dfddb3
MK
1228 if (delta_size < cache_max_small_delta_size)
1229 return 1;
1230
074b2eea
MK
1231 /* cache delta, if objects are large enough compared to delta size */
1232 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1233 return 1;
1234
1235 return 0;
1236}
1237
7eb151d6 1238#ifndef NO_PTHREADS
8ecce684 1239
44626dc7 1240static pthread_mutex_t read_mutex;
8ecce684
NP
1241#define read_lock() pthread_mutex_lock(&read_mutex)
1242#define read_unlock() pthread_mutex_unlock(&read_mutex)
1243
44626dc7 1244static pthread_mutex_t cache_mutex;
3c701839
NP
1245#define cache_lock() pthread_mutex_lock(&cache_mutex)
1246#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1247
44626dc7 1248static pthread_mutex_t progress_mutex;
8ecce684
NP
1249#define progress_lock() pthread_mutex_lock(&progress_mutex)
1250#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1251
1252#else
1253
e1ef8673
JH
1254#define read_lock() (void)0
1255#define read_unlock() (void)0
1256#define cache_lock() (void)0
1257#define cache_unlock() (void)0
1258#define progress_lock() (void)0
1259#define progress_unlock() (void)0
8ecce684
NP
1260
1261#endif
1262
f6c7081a 1263static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1264 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1265{
f6c7081a
NP
1266 struct object_entry *trg_entry = trg->entry;
1267 struct object_entry *src_entry = src->entry;
560b25a8 1268 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1269 unsigned ref_depth;
21666f1a 1270 enum object_type type;
c323ac7d
LT
1271 void *delta_buf;
1272
1273 /* Don't bother doing diffs between different types */
f6c7081a 1274 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1275 return -1;
1276
51d1e83f
LT
1277 /*
1278 * We do not bother to try a delta that we discarded
8dbbd14e 1279 * on an earlier try, but only when reusing delta data.
51d1e83f 1280 */
a7de7130 1281 if (reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1282 trg_entry->in_pack == src_entry->in_pack &&
1283 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1284 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1285 return 0;
1286
898b14ce 1287 /* Let's not bust the allowed depth. */
5a235b5e 1288 if (src->depth >= max_depth)
d116a45a 1289 return 0;
c323ac7d 1290
c3b06a69 1291 /* Now some size filtering heuristics. */
560b25a8 1292 trg_size = trg_entry->size;
c83f032e
NP
1293 if (!trg_entry->delta) {
1294 max_size = trg_size/2 - 20;
1295 ref_depth = 1;
1296 } else {
1297 max_size = trg_entry->delta_size;
5a235b5e 1298 ref_depth = trg->depth;
c83f032e 1299 }
720fe22d 1300 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1301 (max_depth - ref_depth + 1);
c3b06a69
NP
1302 if (max_size == 0)
1303 return 0;
f6c7081a 1304 src_size = src_entry->size;
560b25a8 1305 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1306 if (sizediff >= max_size)
f527cb8c 1307 return 0;
a1dab41a
BD
1308 if (trg_size < src_size / 32)
1309 return 0;
f6c7081a 1310
560b25a8
NP
1311 /* Load data if not already done */
1312 if (!trg->data) {
8ecce684 1313 read_lock();
aa7e44bf 1314 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1315 read_unlock();
2e3404c3
JH
1316 if (!trg->data)
1317 die("object %s cannot be read",
1318 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1319 if (sz != trg_size)
1320 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1321 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1322 *mem_usage += sz;
560b25a8
NP
1323 }
1324 if (!src->data) {
8ecce684 1325 read_lock();
aa7e44bf 1326 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1327 read_unlock();
2e3404c3
JH
1328 if (!src->data)
1329 die("object %s cannot be read",
1330 sha1_to_hex(src_entry->idx.sha1));
560b25a8
NP
1331 if (sz != src_size)
1332 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1333 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1334 *mem_usage += sz;
560b25a8
NP
1335 }
1336 if (!src->index) {
1337 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1338 if (!src->index) {
1339 static int warned = 0;
1340 if (!warned++)
1341 warning("suboptimal pack - out of memory");
1342 return 0;
1343 }
ef0316fc 1344 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1345 }
1346
1347 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1348 if (!delta_buf)
75c42d8c 1349 return 0;
f6c7081a 1350
9e2d57a0 1351 if (trg_entry->delta) {
848d732c
BD
1352 /* Prefer only shallower same-sized deltas. */
1353 if (delta_size == trg_entry->delta_size &&
5a235b5e 1354 src->depth + 1 >= trg->depth) {
848d732c
BD
1355 free(delta_buf);
1356 return 0;
1357 }
074b2eea 1358 }
9e2d57a0 1359
3c701839
NP
1360 /*
1361 * Handle memory allocation outside of the cache
1362 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1363 * away when NO_PTHREADS is defined.
3c701839 1364 */
8e0f7003 1365 free(trg_entry->delta_data);
3c701839 1366 cache_lock();
9e2d57a0
NP
1367 if (trg_entry->delta_data) {
1368 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1369 trg_entry->delta_data = NULL;
1370 }
d250626c 1371 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1372 delta_cache_size += delta_size;
3c701839
NP
1373 cache_unlock();
1374 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1375 } else {
1376 cache_unlock();
074b2eea 1377 free(delta_buf);
3c701839
NP
1378 }
1379
b7a28f78
NP
1380 trg_entry->delta = src_entry;
1381 trg_entry->delta_size = delta_size;
1382 trg->depth = src->depth + 1;
1383
f6c7081a 1384 return 1;
c323ac7d
LT
1385}
1386
898b14ce 1387static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1388{
898b14ce
NP
1389 struct object_entry *child = me->delta_child;
1390 unsigned int m = n;
1391 while (child) {
1392 unsigned int c = check_delta_limit(child, n + 1);
1393 if (m < c)
1394 m = c;
1395 child = child->delta_sibling;
1396 }
1397 return m;
b2504a0d
NP
1398}
1399
75ad235c 1400static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1401{
ef0316fc 1402 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1403 free_delta_index(n->index);
1404 n->index = NULL;
1405 if (n->data) {
ef0316fc 1406 freed_mem += n->entry->size;
a97773ce
BD
1407 free(n->data);
1408 n->data = NULL;
a97773ce
BD
1409 }
1410 n->entry = NULL;
7d7baa5e 1411 n->depth = 0;
ef0316fc 1412 return freed_mem;
a97773ce
BD
1413}
1414
384b32c0 1415static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1416 int window, int depth, unsigned *processed)
c323ac7d 1417{
384b32c0 1418 uint32_t i, idx = 0, count = 0;
7cadf491 1419 struct unpacked *array;
ef0316fc 1420 unsigned long mem_usage = 0;
c323ac7d 1421
19d4b416 1422 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1423
384b32c0 1424 for (;;) {
421b488a 1425 struct object_entry *entry;
c323ac7d 1426 struct unpacked *n = array + idx;
ef0316fc 1427 int j, max_depth, best_base = -1;
c323ac7d 1428
384b32c0
NP
1429 progress_lock();
1430 if (!*list_size) {
1431 progress_unlock();
1432 break;
1433 }
421b488a 1434 entry = *list++;
384b32c0
NP
1435 (*list_size)--;
1436 if (!entry->preferred_base) {
1437 (*processed)++;
1438 display_progress(progress_state, *processed);
1439 }
1440 progress_unlock();
1441
ef0316fc 1442 mem_usage -= free_unpacked(n);
c323ac7d 1443 n->entry = entry;
ab7cd7bb 1444
a97773ce 1445 while (window_memory_limit &&
ef0316fc 1446 mem_usage > window_memory_limit &&
a97773ce
BD
1447 count > 1) {
1448 uint32_t tail = (idx + window - count) % window;
75ad235c 1449 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1450 count--;
1451 }
1452
75d39853
NP
1453 /* We do not compute delta to *create* objects we are not
1454 * going to pack.
1455 */
1456 if (entry->preferred_base)
1457 goto next;
1458
898b14ce
NP
1459 /*
1460 * If the current object is at pack edge, take the depth the
1461 * objects that depend on the current object into account
1462 * otherwise they would become too deep.
1463 */
1464 max_depth = depth;
1465 if (entry->delta_child) {
1466 max_depth -= check_delta_limit(entry, 0);
1467 if (max_depth <= 0)
1468 goto next;
1469 }
1470
78817c15
LT
1471 j = window;
1472 while (--j > 0) {
77639870 1473 int ret;
7cadf491 1474 uint32_t other_idx = idx + j;
c323ac7d 1475 struct unpacked *m;
78817c15
LT
1476 if (other_idx >= window)
1477 other_idx -= window;
c323ac7d
LT
1478 m = array + other_idx;
1479 if (!m->entry)
1480 break;
ef0316fc 1481 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1482 if (ret < 0)
c323ac7d 1483 break;
77639870
JH
1484 else if (ret > 0)
1485 best_base = other_idx;
c323ac7d 1486 }
898b14ce 1487
ed4a9031
NP
1488 /*
1489 * If we decided to cache the delta data, then it is best
1490 * to compress it right away. First because we have to do
1491 * it anyway, and doing it here while we're threaded will
1492 * save a lot of time in the non threaded write phase,
1493 * as well as allow for caching more deltas within
1494 * the same cache size limit.
1495 * ...
1496 * But only if not writing to stdout, since in that case
1497 * the network is most likely throttling writes anyway,
1498 * and therefore it is best to go to the write phase ASAP
1499 * instead, as we can afford spending more time compressing
1500 * between writes at that moment.
1501 */
1502 if (entry->delta_data && !pack_to_stdout) {
1503 entry->z_delta_size = do_compress(&entry->delta_data,
1504 entry->delta_size);
1505 cache_lock();
1506 delta_cache_size -= entry->delta_size;
1507 delta_cache_size += entry->z_delta_size;
1508 cache_unlock();
1509 }
1510
70ca1a3f
JH
1511 /* if we made n a delta, and if n is already at max
1512 * depth, leaving it in the window is pointless. we
1513 * should evict it first.
70ca1a3f 1514 */
70baf5d4 1515 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1516 continue;
ff45715c 1517
77639870
JH
1518 /*
1519 * Move the best delta base up in the window, after the
1520 * currently deltified object, to keep it longer. It will
1521 * be the first base object to be attempted next.
1522 */
1523 if (entry->delta) {
1524 struct unpacked swap = array[best_base];
1525 int dist = (window + idx - best_base) % window;
1526 int dst = best_base;
1527 while (dist--) {
1528 int src = (dst + 1) % window;
1529 array[dst] = array[src];
1530 dst = src;
1531 }
1532 array[dst] = swap;
1533 }
1534
898b14ce 1535 next:
521a4f4c 1536 idx++;
a97773ce
BD
1537 if (count + 1 < window)
1538 count++;
521a4f4c
LT
1539 if (idx >= window)
1540 idx = 0;
384b32c0 1541 }
adee7bdf 1542
f6c7081a 1543 for (i = 0; i < window; ++i) {
ff45715c 1544 free_delta_index(array[i].index);
adee7bdf 1545 free(array[i].data);
f6c7081a 1546 }
adee7bdf 1547 free(array);
c323ac7d
LT
1548}
1549
7eb151d6 1550#ifndef NO_PTHREADS
8ecce684 1551
a9a74636
NP
1552static void try_to_free_from_threads(size_t size)
1553{
1554 read_lock();
1555 release_pack_memory(size, -1);
1556 read_unlock();
1557}
1558
50f22ada
JS
1559/*
1560 * The main thread waits on the condition that (at least) one of the workers
1561 * has stopped working (which is indicated in the .working member of
1562 * struct thread_params).
1563 * When a work thread has completed its work, it sets .working to 0 and
1564 * signals the main thread and waits on the condition that .data_ready
1565 * becomes 1.
1566 */
1567
8ecce684
NP
1568struct thread_params {
1569 pthread_t thread;
1570 struct object_entry **list;
1571 unsigned list_size;
384b32c0 1572 unsigned remaining;
8ecce684
NP
1573 int window;
1574 int depth;
50f22ada
JS
1575 int working;
1576 int data_ready;
1577 pthread_mutex_t mutex;
1578 pthread_cond_t cond;
8ecce684
NP
1579 unsigned *processed;
1580};
1581
44626dc7
AH
1582static pthread_cond_t progress_cond;
1583
1584/*
1585 * Mutex and conditional variable can't be statically-initialized on Windows.
1586 */
1587static void init_threaded_search(void)
1588{
1589 pthread_mutex_init(&read_mutex, NULL);
1590 pthread_mutex_init(&cache_mutex, NULL);
1591 pthread_mutex_init(&progress_mutex, NULL);
1592 pthread_cond_init(&progress_cond, NULL);
a9a74636 1593 set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
1594}
1595
1596static void cleanup_threaded_search(void)
1597{
a9a74636 1598 set_try_to_free_routine(NULL);
44626dc7
AH
1599 pthread_cond_destroy(&progress_cond);
1600 pthread_mutex_destroy(&read_mutex);
1601 pthread_mutex_destroy(&cache_mutex);
1602 pthread_mutex_destroy(&progress_mutex);
1603}
c2a33679 1604
8ecce684
NP
1605static void *threaded_find_deltas(void *arg)
1606{
c2a33679
NP
1607 struct thread_params *me = arg;
1608
50f22ada 1609 while (me->remaining) {
384b32c0 1610 find_deltas(me->list, &me->remaining,
c2a33679 1611 me->window, me->depth, me->processed);
50f22ada
JS
1612
1613 progress_lock();
1614 me->working = 0;
1615 pthread_cond_signal(&progress_cond);
1616 progress_unlock();
1617
1618 /*
1619 * We must not set ->data_ready before we wait on the
1620 * condition because the main thread may have set it to 1
1621 * before we get here. In order to be sure that new
1622 * work is available if we see 1 in ->data_ready, it
1623 * was initialized to 0 before this thread was spawned
1624 * and we reset it to 0 right away.
1625 */
1626 pthread_mutex_lock(&me->mutex);
1627 while (!me->data_ready)
1628 pthread_cond_wait(&me->cond, &me->mutex);
1629 me->data_ready = 0;
1630 pthread_mutex_unlock(&me->mutex);
c2a33679 1631 }
50f22ada
JS
1632 /* leave ->working 1 so that this doesn't get more work assigned */
1633 return NULL;
8ecce684
NP
1634}
1635
8ecce684
NP
1636static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1637 int window, int depth, unsigned *processed)
1638{
dcda3614 1639 struct thread_params *p;
384b32c0 1640 int i, ret, active_threads = 0;
c2a33679 1641
44626dc7
AH
1642 init_threaded_search();
1643
ef055571
NP
1644 if (!delta_search_threads) /* --threads=0 means autodetect */
1645 delta_search_threads = online_cpus();
367f4a43 1646 if (delta_search_threads <= 1) {
384b32c0 1647 find_deltas(list, &list_size, window, depth, processed);
44626dc7 1648 cleanup_threaded_search();
367f4a43
NP
1649 return;
1650 }
43cc2b42 1651 if (progress > pack_to_stdout)
b6c29915 1652 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1653 delta_search_threads);
dcda3614 1654 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 1655
50f22ada 1656 /* Partition the work amongst work threads. */
367f4a43 1657 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1658 unsigned sub_size = list_size / (delta_search_threads - i);
1659
bf874896
NP
1660 /* don't use too small segments or no deltas will be found */
1661 if (sub_size < 2*window && i+1 < delta_search_threads)
1662 sub_size = 0;
1663
8ecce684
NP
1664 p[i].window = window;
1665 p[i].depth = depth;
1666 p[i].processed = processed;
50f22ada
JS
1667 p[i].working = 1;
1668 p[i].data_ready = 0;
c2a33679 1669
59921b4b 1670 /* try to split chunks on "path" boundaries */
6fc74703
NP
1671 while (sub_size && sub_size < list_size &&
1672 list[sub_size]->hash &&
384b32c0
NP
1673 list[sub_size]->hash == list[sub_size-1]->hash)
1674 sub_size++;
1675
50f22ada
JS
1676 p[i].list = list;
1677 p[i].list_size = sub_size;
1678 p[i].remaining = sub_size;
59921b4b 1679
384b32c0
NP
1680 list += sub_size;
1681 list_size -= sub_size;
1682 }
1683
50f22ada
JS
1684 /* Start work threads. */
1685 for (i = 0; i < delta_search_threads; i++) {
1686 if (!p[i].list_size)
1687 continue;
68e6a4f8
JS
1688 pthread_mutex_init(&p[i].mutex, NULL);
1689 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1690 ret = pthread_create(&p[i].thread, NULL,
1691 threaded_find_deltas, &p[i]);
1692 if (ret)
1693 die("unable to create thread: %s", strerror(ret));
1694 active_threads++;
1695 }
1696
384b32c0
NP
1697 /*
1698 * Now let's wait for work completion. Each time a thread is done
1699 * with its work, we steal half of the remaining work from the
1700 * thread with the largest number of unprocessed objects and give
1701 * it to that newly idle thread. This ensure good load balancing
1702 * until the remaining object list segments are simply too short
1703 * to be worth splitting anymore.
1704 */
50f22ada
JS
1705 while (active_threads) {
1706 struct thread_params *target = NULL;
384b32c0
NP
1707 struct thread_params *victim = NULL;
1708 unsigned sub_size = 0;
384b32c0
NP
1709
1710 progress_lock();
50f22ada
JS
1711 for (;;) {
1712 for (i = 0; !target && i < delta_search_threads; i++)
1713 if (!p[i].working)
1714 target = &p[i];
1715 if (target)
1716 break;
1717 pthread_cond_wait(&progress_cond, &progress_mutex);
1718 }
1719
384b32c0
NP
1720 for (i = 0; i < delta_search_threads; i++)
1721 if (p[i].remaining > 2*window &&
1722 (!victim || victim->remaining < p[i].remaining))
1723 victim = &p[i];
1724 if (victim) {
1725 sub_size = victim->remaining / 2;
1726 list = victim->list + victim->list_size - sub_size;
1727 while (sub_size && list[0]->hash &&
1728 list[0]->hash == list[-1]->hash) {
1729 list++;
1730 sub_size--;
1731 }
eb9688ff
NP
1732 if (!sub_size) {
1733 /*
1734 * It is possible for some "paths" to have
1735 * so many objects that no hash boundary
1736 * might be found. Let's just steal the
1737 * exact half in that case.
1738 */
1739 sub_size = victim->remaining / 2;
1740 list -= sub_size;
1741 }
384b32c0
NP
1742 target->list = list;
1743 victim->list_size -= sub_size;
1744 victim->remaining -= sub_size;
1745 }
384b32c0
NP
1746 target->list_size = sub_size;
1747 target->remaining = sub_size;
50f22ada
JS
1748 target->working = 1;
1749 progress_unlock();
1750
1751 pthread_mutex_lock(&target->mutex);
1752 target->data_ready = 1;
1753 pthread_cond_signal(&target->cond);
1754 pthread_mutex_unlock(&target->mutex);
c2a33679 1755
384b32c0 1756 if (!sub_size) {
b81d9af7 1757 pthread_join(target->thread, NULL);
50f22ada
JS
1758 pthread_cond_destroy(&target->cond);
1759 pthread_mutex_destroy(&target->mutex);
384b32c0 1760 active_threads--;
c2a33679 1761 }
50f22ada 1762 }
44626dc7 1763 cleanup_threaded_search();
dcda3614 1764 free(p);
8ecce684
NP
1765}
1766
1767#else
384b32c0 1768#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
1769#endif
1770
f0a24aa5
SP
1771static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1772{
1773 unsigned char peeled[20];
1774
1775 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1776 !peel_ref(path, peeled) && /* peelable? */
1777 !is_null_sha1(peeled) && /* annotated tag? */
1778 locate_object_entry(peeled)) /* object packed? */
1779 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1780 return 0;
1781}
1782
f3123c4a
JH
1783static void prepare_pack(int window, int depth)
1784{
9668cf59 1785 struct object_entry **delta_list;
6e1c2344
RJ
1786 uint32_t i, nr_deltas;
1787 unsigned n;
9668cf59 1788
3f9ac8d2 1789 get_object_details();
9668cf59 1790
0e8189e2
NP
1791 /*
1792 * If we're locally repacking then we need to be doubly careful
1793 * from now on in order to make sure no stealth corruption gets
1794 * propagated to the new pack. Clients receiving streamed packs
1795 * should validate everything they get anyway so no need to incur
1796 * the additional cost here in that case.
1797 */
1798 if (!pack_to_stdout)
1799 do_check_packed_object_crc = 1;
1800
75d39853 1801 if (!nr_objects || !window || !depth)
9668cf59
NP
1802 return;
1803
1804 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1805 nr_deltas = n = 0;
1806
1807 for (i = 0; i < nr_objects; i++) {
1808 struct object_entry *entry = objects + i;
1809
1810 if (entry->delta)
1811 /* This happens if we decided to reuse existing
a7de7130 1812 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
1813 */
1814 continue;
1815
1816 if (entry->size < 50)
1817 continue;
1818
1819 if (entry->no_try_delta)
1820 continue;
1821
6d6f9cdd 1822 if (!entry->preferred_base) {
75d39853 1823 nr_deltas++;
6d6f9cdd
SP
1824 if (entry->type < 0)
1825 die("unable to get type of object %s",
1826 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
1827 } else {
1828 if (entry->type < 0) {
1829 /*
1830 * This object is not found, but we
1831 * don't have to include it anyway.
1832 */
1833 continue;
1834 }
6d6f9cdd 1835 }
75d39853
NP
1836
1837 delta_list[n++] = entry;
1838 }
1839
2f8b8947 1840 if (nr_deltas && n > 1) {
e334977d
NP
1841 unsigned nr_done = 0;
1842 if (progress)
dc6a0757
NP
1843 progress_state = start_progress("Compressing objects",
1844 nr_deltas);
75d39853 1845 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1846 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1847 stop_progress(&progress_state);
e334977d
NP
1848 if (nr_done != nr_deltas)
1849 die("inconsistency with delta count");
75d39853 1850 }
9668cf59 1851 free(delta_list);
f3123c4a
JH
1852}
1853
ef90d6d4 1854static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 1855{
eeefa7c9 1856 if (!strcmp(k, "pack.window")) {
4812a93a
JK
1857 window = git_config_int(k, v);
1858 return 0;
1859 }
a97773ce
BD
1860 if (!strcmp(k, "pack.windowmemory")) {
1861 window_memory_limit = git_config_ulong(k, v);
1862 return 0;
1863 }
1864 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1865 depth = git_config_int(k, v);
1866 return 0;
1867 }
960ccca6
DH
1868 if (!strcmp(k, "pack.compression")) {
1869 int level = git_config_int(k, v);
1870 if (level == -1)
1871 level = Z_DEFAULT_COMPRESSION;
1872 else if (level < 0 || level > Z_BEST_COMPRESSION)
1873 die("bad pack compression level %d", level);
1874 pack_compression_level = level;
1875 pack_compression_seen = 1;
1876 return 0;
1877 }
074b2eea
MK
1878 if (!strcmp(k, "pack.deltacachesize")) {
1879 max_delta_cache_size = git_config_int(k, v);
1880 return 0;
1881 }
e3dfddb3
MK
1882 if (!strcmp(k, "pack.deltacachelimit")) {
1883 cache_max_small_delta_size = git_config_int(k, v);
1884 return 0;
1885 }
693b86ff
NP
1886 if (!strcmp(k, "pack.threads")) {
1887 delta_search_threads = git_config_int(k, v);
833e3df1 1888 if (delta_search_threads < 0)
693b86ff
NP
1889 die("invalid number of threads specified (%d)",
1890 delta_search_threads);
7eb151d6 1891#ifdef NO_PTHREADS
833e3df1 1892 if (delta_search_threads != 1)
693b86ff
NP
1893 warning("no threads support, ignoring %s", k);
1894#endif
1895 return 0;
1896 }
4d00bda2
NP
1897 if (!strcmp(k, "pack.indexversion")) {
1898 pack_idx_default_version = git_config_int(k, v);
1899 if (pack_idx_default_version > 2)
6e1c2344
RJ
1900 die("bad pack.indexversion=%"PRIu32,
1901 pack_idx_default_version);
4d00bda2
NP
1902 return 0;
1903 }
2b84b5a8
JS
1904 if (!strcmp(k, "pack.packsizelimit")) {
1905 pack_size_limit_cfg = git_config_ulong(k, v);
1906 return 0;
1907 }
ef90d6d4 1908 return git_default_config(k, v, cb);
4812a93a
JK
1909}
1910
b5d97e6b 1911static void read_object_list_from_stdin(void)
c323ac7d 1912{
b5d97e6b
JH
1913 char line[40 + 1 + PATH_MAX + 2];
1914 unsigned char sha1[20];
b2504a0d 1915
da93d12b 1916 for (;;) {
da93d12b
LT
1917 if (!fgets(line, sizeof(line), stdin)) {
1918 if (feof(stdin))
1919 break;
1920 if (!ferror(stdin))
1921 die("fgets returned NULL, not EOF, not error!");
687dd75c 1922 if (errno != EINTR)
d824cbba 1923 die_errno("fgets");
687dd75c
JH
1924 clearerr(stdin);
1925 continue;
da93d12b 1926 }
7a979d99
JH
1927 if (line[0] == '-') {
1928 if (get_sha1_hex(line+1, sha1))
1929 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1930 line);
8d1d8f83 1931 add_preferred_base(sha1);
7a979d99 1932 continue;
21fcd1bd 1933 }
c323ac7d 1934 if (get_sha1_hex(line, sha1))
ef07618f 1935 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1936
bc32fed5
JH
1937 add_preferred_base_object(line+41);
1938 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1939 }
b5d97e6b
JH
1940}
1941
08cdfb13
JH
1942#define OBJECT_ADDED (1u<<20)
1943
11c211fa 1944static void show_commit(struct commit *commit, void *data)
b5d97e6b 1945{
bc32fed5 1946 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1947 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1948}
1949
cf2ab916 1950static void show_object(struct object *obj, const struct name_path *path, const char *last)
b5d97e6b 1951{
cf2ab916
LT
1952 char *name = path_name(path, last);
1953
8d2dfc49
LT
1954 add_preferred_base_object(name);
1955 add_object_entry(obj->sha1, obj->type, name, 0);
1956 obj->flags |= OBJECT_ADDED;
1957
1958 /*
1959 * We will have generated the hash from the name,
1960 * but not saved a pointer to it - we can free it
1961 */
1962 free((char *)name);
b5d97e6b
JH
1963}
1964
8d1d8f83
JH
1965static void show_edge(struct commit *commit)
1966{
1967 add_preferred_base(commit->object.sha1);
1968}
1969
08cdfb13
JH
1970struct in_pack_object {
1971 off_t offset;
1972 struct object *object;
1973};
1974
1975struct in_pack {
1976 int alloc;
1977 int nr;
1978 struct in_pack_object *array;
1979};
1980
1981static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1982{
1983 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1984 in_pack->array[in_pack->nr].object = object;
1985 in_pack->nr++;
1986}
1987
1988/*
1989 * Compare the objects in the offset order, in order to emulate the
f18d244a 1990 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
1991 */
1992static int ofscmp(const void *a_, const void *b_)
1993{
1994 struct in_pack_object *a = (struct in_pack_object *)a_;
1995 struct in_pack_object *b = (struct in_pack_object *)b_;
1996
1997 if (a->offset < b->offset)
1998 return -1;
1999 else if (a->offset > b->offset)
2000 return 1;
2001 else
2002 return hashcmp(a->object->sha1, b->object->sha1);
2003}
2004
2005static void add_objects_in_unpacked_packs(struct rev_info *revs)
2006{
2007 struct packed_git *p;
2008 struct in_pack in_pack;
2009 uint32_t i;
2010
2011 memset(&in_pack, 0, sizeof(in_pack));
2012
2013 for (p = packed_git; p; p = p->next) {
2014 const unsigned char *sha1;
2015 struct object *o;
2016
79bc4c71 2017 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2018 continue;
2019 if (open_pack_index(p))
2020 die("cannot open pack index");
2021
2022 ALLOC_GROW(in_pack.array,
2023 in_pack.nr + p->num_objects,
2024 in_pack.alloc);
2025
2026 for (i = 0; i < p->num_objects; i++) {
2027 sha1 = nth_packed_object_sha1(p, i);
2028 o = lookup_unknown_object(sha1);
2029 if (!(o->flags & OBJECT_ADDED))
2030 mark_in_pack_object(o, p, &in_pack);
2031 o->flags |= OBJECT_ADDED;
2032 }
2033 }
2034
2035 if (in_pack.nr) {
2036 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2037 ofscmp);
2038 for (i = 0; i < in_pack.nr; i++) {
2039 struct object *o = in_pack.array[i].object;
2040 add_object_entry(o->sha1, o->type, "", 0);
2041 }
2042 }
2043 free(in_pack.array);
2044}
2045
094085e3
BC
2046static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2047{
2048 static struct packed_git *last_found = (void *)1;
2049 struct packed_git *p;
2050
2051 p = (last_found != (void *)1) ? last_found : packed_git;
2052
2053 while (p) {
2054 if ((!p->pack_local || p->pack_keep) &&
2055 find_pack_entry_one(sha1, p)) {
2056 last_found = p;
2057 return 1;
2058 }
2059 if (p == last_found)
2060 p = packed_git;
2061 else
2062 p = p->next;
2063 if (p == last_found)
2064 p = p->next;
2065 }
2066 return 0;
2067}
2068
ca11b212
NP
2069static void loosen_unused_packed_objects(struct rev_info *revs)
2070{
2071 struct packed_git *p;
2072 uint32_t i;
2073 const unsigned char *sha1;
2074
2075 for (p = packed_git; p; p = p->next) {
79bc4c71 2076 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2077 continue;
2078
2079 if (open_pack_index(p))
2080 die("cannot open pack index");
2081
2082 for (i = 0; i < p->num_objects; i++) {
2083 sha1 = nth_packed_object_sha1(p, i);
094085e3
BC
2084 if (!locate_object_entry(sha1) &&
2085 !has_sha1_pack_kept_or_nonlocal(sha1))
ca11b212
NP
2086 if (force_object_loose(sha1, p->mtime))
2087 die("unable to force loose object");
2088 }
2089 }
2090}
2091
8d1d8f83 2092static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2093{
2094 struct rev_info revs;
2095 char line[1000];
b5d97e6b
JH
2096 int flags = 0;
2097
b5d97e6b
JH
2098 init_revisions(&revs, NULL);
2099 save_commit_buffer = 0;
b5d97e6b
JH
2100 setup_revisions(ac, av, &revs, NULL);
2101
b5d97e6b
JH
2102 while (fgets(line, sizeof(line), stdin) != NULL) {
2103 int len = strlen(line);
872c930d 2104 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2105 line[--len] = 0;
2106 if (!len)
2107 break;
2108 if (*line == '-') {
2109 if (!strcmp(line, "--not")) {
2110 flags ^= UNINTERESTING;
2111 continue;
2112 }
2113 die("not a rev '%s'", line);
2114 }
2115 if (handle_revision_arg(line, &revs, flags, 1))
2116 die("bad revision '%s'", line);
2117 }
2118
3d51e1b5
MK
2119 if (prepare_revision_walk(&revs))
2120 die("revision walk setup failed");
8d1d8f83 2121 mark_edges_uninteresting(revs.commits, &revs, show_edge);
11c211fa 2122 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13
JH
2123
2124 if (keep_unreachable)
2125 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2126 if (unpack_unreachable)
2127 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2128}
2129
2130int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2131{
b5d97e6b 2132 int use_internal_rev_list = 0;
8d1d8f83 2133 int thin = 0;
4f366275 2134 int all_progress_implied = 0;
7cadf491 2135 uint32_t i;
ffa84ffb
RD
2136 const char **rp_av;
2137 int rp_ac_alloc = 64;
8d1d8f83
JH
2138 int rp_ac;
2139
dae556bd
CC
2140 read_replace_refs = 0;
2141
ffa84ffb
RD
2142 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2143
8d1d8f83
JH
2144 rp_av[0] = "pack-objects";
2145 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2146 rp_ac = 2;
b5d97e6b 2147
ef90d6d4 2148 git_config(git_pack_config, NULL);
960ccca6
DH
2149 if (!pack_compression_seen && core_compression_seen)
2150 pack_compression_level = core_compression_level;
b5d97e6b
JH
2151
2152 progress = isatty(2);
2153 for (i = 1; i < argc; i++) {
2154 const char *arg = argv[i];
2155
2156 if (*arg != '-')
2157 break;
2158
2159 if (!strcmp("--non-empty", arg)) {
2160 non_empty = 1;
2161 continue;
2162 }
2163 if (!strcmp("--local", arg)) {
2164 local = 1;
2165 continue;
2166 }
b5d97e6b
JH
2167 if (!strcmp("--incremental", arg)) {
2168 incremental = 1;
2169 continue;
2170 }
e96fb9b8
BC
2171 if (!strcmp("--honor-pack-keep", arg)) {
2172 ignore_packed_keep = 1;
2173 continue;
2174 }
960ccca6
DH
2175 if (!prefixcmp(arg, "--compression=")) {
2176 char *end;
2177 int level = strtoul(arg+14, &end, 0);
2178 if (!arg[14] || *end)
2179 usage(pack_usage);
2180 if (level == -1)
2181 level = Z_DEFAULT_COMPRESSION;
2182 else if (level < 0 || level > Z_BEST_COMPRESSION)
2183 die("bad pack compression level %d", level);
2184 pack_compression_level = level;
2185 continue;
2186 }
6b94b1a0 2187 if (!prefixcmp(arg, "--max-pack-size=")) {
2b84b5a8 2188 pack_size_limit_cfg = 0;
07cf0f24 2189 if (!git_parse_ulong(arg+16, &pack_size_limit))
6b94b1a0
DH
2190 usage(pack_usage);
2191 continue;
2192 }
599065a3 2193 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2194 char *end;
2195 window = strtoul(arg+9, &end, 0);
2196 if (!arg[9] || *end)
2197 usage(pack_usage);
2198 continue;
2199 }
a97773ce
BD
2200 if (!prefixcmp(arg, "--window-memory=")) {
2201 if (!git_parse_ulong(arg+16, &window_memory_limit))
2202 usage(pack_usage);
2203 continue;
2204 }
367f4a43
NP
2205 if (!prefixcmp(arg, "--threads=")) {
2206 char *end;
2207 delta_search_threads = strtoul(arg+10, &end, 0);
833e3df1 2208 if (!arg[10] || *end || delta_search_threads < 0)
367f4a43 2209 usage(pack_usage);
7eb151d6 2210#ifdef NO_PTHREADS
833e3df1 2211 if (delta_search_threads != 1)
367f4a43
NP
2212 warning("no threads support, "
2213 "ignoring %s", arg);
2214#endif
2215 continue;
2216 }
599065a3 2217 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2218 char *end;
2219 depth = strtoul(arg+8, &end, 0);
2220 if (!arg[8] || *end)
2221 usage(pack_usage);
2222 continue;
2223 }
2224 if (!strcmp("--progress", arg)) {
2225 progress = 1;
2226 continue;
2227 }
231f240b
NP
2228 if (!strcmp("--all-progress", arg)) {
2229 progress = 2;
2230 continue;
2231 }
4f366275
NP
2232 if (!strcmp("--all-progress-implied", arg)) {
2233 all_progress_implied = 1;
2234 continue;
2235 }
b5d97e6b
JH
2236 if (!strcmp("-q", arg)) {
2237 progress = 0;
2238 continue;
2239 }
2240 if (!strcmp("--no-reuse-delta", arg)) {
a7de7130 2241 reuse_delta = 0;
b5d97e6b
JH
2242 continue;
2243 }
fa736f72 2244 if (!strcmp("--no-reuse-object", arg)) {
a7de7130 2245 reuse_object = reuse_delta = 0;
fa736f72
NP
2246 continue;
2247 }
be6b1914 2248 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2249 allow_ofs_delta = 1;
be6b1914
NP
2250 continue;
2251 }
b5d97e6b
JH
2252 if (!strcmp("--stdout", arg)) {
2253 pack_to_stdout = 1;
2254 continue;
2255 }
2256 if (!strcmp("--revs", arg)) {
2257 use_internal_rev_list = 1;
2258 continue;
2259 }
08cdfb13
JH
2260 if (!strcmp("--keep-unreachable", arg)) {
2261 keep_unreachable = 1;
2262 continue;
2263 }
ca11b212
NP
2264 if (!strcmp("--unpack-unreachable", arg)) {
2265 unpack_unreachable = 1;
2266 continue;
2267 }
f0a24aa5
SP
2268 if (!strcmp("--include-tag", arg)) {
2269 include_tag = 1;
2270 continue;
2271 }
8d1d8f83 2272 if (!strcmp("--unpacked", arg) ||
63049292 2273 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2274 !strcmp("--all", arg)) {
2275 use_internal_rev_list = 1;
ffa84ffb
RD
2276 if (rp_ac >= rp_ac_alloc - 1) {
2277 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2278 rp_av = xrealloc(rp_av,
2279 rp_ac_alloc * sizeof(*rp_av));
2280 }
8d1d8f83 2281 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2282 continue;
2283 }
8d1d8f83
JH
2284 if (!strcmp("--thin", arg)) {
2285 use_internal_rev_list = 1;
2286 thin = 1;
2287 rp_av[1] = "--objects-edge";
b5d97e6b 2288 continue;
4ba7d711
NP
2289 }
2290 if (!prefixcmp(arg, "--index-version=")) {
2291 char *c;
aa7e44bf
GB
2292 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2293 if (pack_idx_default_version > 2)
4ba7d711
NP
2294 die("bad %s", arg);
2295 if (*c == ',')
aa7e44bf
GB
2296 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2297 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2298 die("bad %s", arg);
2299 continue;
b5d97e6b 2300 }
7f3140cd
JS
2301 if (!strcmp(arg, "--keep-true-parents")) {
2302 grafts_replace_parents = 0;
2303 continue;
2304 }
b5d97e6b
JH
2305 usage(pack_usage);
2306 }
2307
2308 /* Traditionally "pack-objects [options] base extra" failed;
2309 * we would however want to take refs parameter that would
2310 * have been given to upstream rev-list ourselves, which means
2311 * we somehow want to say what the base name is. So the
2312 * syntax would be:
2313 *
2314 * pack-objects [options] base <refs...>
2315 *
2316 * in other words, we would treat the first non-option as the
2317 * base_name and send everything else to the internal revision
2318 * walker.
2319 */
2320
2321 if (!pack_to_stdout)
2322 base_name = argv[i++];
2323
2324 if (pack_to_stdout != !base_name)
2325 usage(pack_usage);
2326
2b84b5a8
JS
2327 if (!pack_to_stdout && !pack_size_limit)
2328 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
2329 if (pack_to_stdout && pack_size_limit)
2330 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
2331 if (pack_size_limit && pack_size_limit < 1024*1024) {
2332 warning("minimum pack size limit is 1 MiB");
2333 pack_size_limit = 1024*1024;
2334 }
01c12a23 2335
8d1d8f83
JH
2336 if (!pack_to_stdout && thin)
2337 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2338
ca11b212
NP
2339 if (keep_unreachable && unpack_unreachable)
2340 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2341
4f366275
NP
2342 if (progress && all_progress_implied)
2343 progress = 2;
2344
b5d97e6b
JH
2345 prepare_packed_git();
2346
13aaf148 2347 if (progress)
dc6a0757 2348 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2349 if (!use_internal_rev_list)
2350 read_object_list_from_stdin();
8d1d8f83
JH
2351 else {
2352 rp_av[rp_ac] = NULL;
2353 get_object_list(rp_ac, rp_av);
2354 }
0ef95f72 2355 cleanup_preferred_base();
f0a24aa5
SP
2356 if (include_tag && nr_result)
2357 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2358 stop_progress(&progress_state);
96a02f8f 2359
f0b0af1b 2360 if (non_empty && !nr_result)
1c4a2912 2361 return 0;
f7ae6a93
NP
2362 if (nr_result)
2363 prepare_pack(window, depth);
d01fb92f 2364 write_pack_file();
ab7cd7bb 2365 if (progress)
6e1c2344
RJ
2366 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2367 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2368 written, written_delta, reused, reused_delta);
c323ac7d
LT
2369 return 0;
2370}