pack-objects: move thread autodetection closer to relevant code
[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
812bdbc3
TF
25static const char pack_usage[] =
26 "git pack-objects [{ -q | --progress | --all-progress }]\n"
27 " [--max-pack-size=N] [--local] [--incremental]\n"
28 " [--window=N] [--window-memory=N] [--depth=N]\n"
29 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
30 " [--threads=N] [--non-empty] [--revs [--unpacked | --all]*]\n"
31 " [--reflog] [--stdout | base-name] [--include-tag]\n"
32 " [--keep-unreachable | --unpack-unreachable \n"
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 88static unsigned long delta_cache_size = 0;
5749b0b2 89static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
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 538 if (adjust_perm(pack_tmp_name, mode))
d824cbba 539 die_errno("unable to make temporary pack file readable");
d01fb92f 540 if (rename(pack_tmp_name, tmpname))
d824cbba 541 die_errno("unable to rename temporary pack file");
f746bae8
NP
542
543 /*
544 * Packs are runtime accessed in their mtime
545 * order since newer packs are more likely to contain
546 * younger objects. So if we are creating multiple
547 * packs then we should modify the mtime of later ones
548 * to preserve this property.
549 */
550 if (stat(tmpname, &st) < 0) {
551 warning("failed to stat %s: %s",
552 tmpname, strerror(errno));
553 } else if (!last_mtime) {
554 last_mtime = st.st_mtime;
555 } else {
556 struct utimbuf utb;
557 utb.actime = st.st_atime;
558 utb.modtime = --last_mtime;
559 if (utime(tmpname, &utb) < 0)
560 warning("failed utime() on %s: %s",
561 tmpname, strerror(errno));
562 }
563
d01fb92f 564 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 565 base_name, sha1_to_hex(sha1));
d01fb92f 566 if (adjust_perm(idx_tmp_name, mode))
d824cbba 567 die_errno("unable to make temporary index file readable");
d01fb92f 568 if (rename(idx_tmp_name, tmpname))
d824cbba 569 die_errno("unable to rename temporary index file");
f746bae8 570
7ba502c4
NP
571 free(idx_tmp_name);
572 free(pack_tmp_name);
aa7e44bf 573 puts(sha1_to_hex(sha1));
ebe27b13
DH
574 }
575
576 /* mark written objects as written to previous pack */
577 for (j = 0; j < nr_written; j++) {
79814f42 578 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
579 }
580 nr_remaining -= nr_written;
581 } while (nr_remaining && i < nr_objects);
582
583 free(written_list);
4d4fcc54 584 stop_progress(&progress_state);
67c08ce1 585 if (written != nr_result)
6e1c2344
RJ
586 die("wrote %"PRIu32" objects while expecting %"PRIu32,
587 written, nr_result);
01c12a23
DH
588 /*
589 * We have scanned through [0 ... i). Since we have written
590 * the correct number of objects, the remaining [i ... nr_objects)
591 * items must be either already written (due to out-of-order delta base)
592 * or a preferred base. Count those which are neither and complain if any.
593 */
ebe27b13
DH
594 for (j = 0; i < nr_objects; i++) {
595 struct object_entry *e = objects + i;
aa7e44bf 596 j += !e->idx.offset && !e->preferred_base;
d01fb92f 597 }
ebe27b13 598 if (j)
6e1c2344
RJ
599 die("wrote %"PRIu32" objects as expected but %"PRIu32
600 " unwritten", written, j);
c323ac7d
LT
601}
602
7a979d99
JH
603static int locate_object_entry_hash(const unsigned char *sha1)
604{
605 int i;
606 unsigned int ui;
607 memcpy(&ui, sha1, sizeof(unsigned int));
608 i = ui % object_ix_hashsz;
609 while (0 < object_ix[i]) {
aa7e44bf 610 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
611 return i;
612 if (++i == object_ix_hashsz)
613 i = 0;
614 }
615 return -1 - i;
616}
617
618static struct object_entry *locate_object_entry(const unsigned char *sha1)
619{
620 int i;
621
622 if (!object_ix_hashsz)
623 return NULL;
624
625 i = locate_object_entry_hash(sha1);
626 if (0 <= i)
627 return &objects[object_ix[i]-1];
628 return NULL;
629}
630
631static void rehash_objects(void)
c323ac7d 632{
7cadf491 633 uint32_t i;
7a979d99
JH
634 struct object_entry *oe;
635
636 object_ix_hashsz = nr_objects * 3;
637 if (object_ix_hashsz < 1024)
638 object_ix_hashsz = 1024;
639 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 640 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 641 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 642 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
643 if (0 <= ix)
644 continue;
645 ix = -1 - ix;
646 object_ix[ix] = i + 1;
647 }
648}
649
ce0bd642 650static unsigned name_hash(const char *name)
1d6b38cc 651{
48fb7deb 652 unsigned c, hash = 0;
ce0bd642 653
bc32fed5
JH
654 if (!name)
655 return 0;
656
eeef7135 657 /*
ce0bd642
LT
658 * This effectively just creates a sortable number from the
659 * last sixteen non-whitespace characters. Last characters
660 * count "most", so things that end in ".c" sort together.
eeef7135 661 */
ce0bd642
LT
662 while ((c = *name++) != 0) {
663 if (isspace(c))
664 continue;
665 hash = (hash >> 2) + (c << 24);
666 }
1d6b38cc
JH
667 return hash;
668}
669
a74db82e
JH
670static void setup_delta_attr_check(struct git_attr_check *check)
671{
672 static struct git_attr *attr_delta;
673
674 if (!attr_delta)
675 attr_delta = git_attr("delta", 5);
676
677 check[0].attr = attr_delta;
678}
679
680static int no_try_delta(const char *path)
681{
682 struct git_attr_check check[1];
683
684 setup_delta_attr_check(check);
685 if (git_checkattr(path, ARRAY_SIZE(check), check))
686 return 0;
687 if (ATTR_FALSE(check->value))
688 return 1;
689 return 0;
690}
691
5c49c116 692static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 693 const char *name, int exclude)
c323ac7d 694{
c323ac7d 695 struct object_entry *entry;
29b734e4 696 struct packed_git *p, *found_pack = NULL;
6777a59f 697 off_t found_offset = 0;
29b734e4 698 int ix;
bc32fed5 699 unsigned hash = name_hash(name);
29b734e4
NP
700
701 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
702 if (ix >= 0) {
703 if (exclude) {
704 entry = objects + object_ix[ix] - 1;
81a216a5
NP
705 if (!entry->preferred_base)
706 nr_result--;
29b734e4
NP
707 entry->preferred_base = 1;
708 }
709 return 0;
710 }
c323ac7d 711
daae0625
BC
712 if (!exclude && local && has_loose_object_nonlocal(sha1))
713 return 0;
714
5c49c116
NP
715 for (p = packed_git; p; p = p->next) {
716 off_t offset = find_pack_entry_one(sha1, p);
717 if (offset) {
718 if (!found_pack) {
719 found_offset = offset;
720 found_pack = p;
64560374 721 }
5c49c116
NP
722 if (exclude)
723 break;
724 if (incremental)
725 return 0;
726 if (local && !p->pack_local)
727 return 0;
e96fb9b8
BC
728 if (ignore_packed_keep && p->pack_local && p->pack_keep)
729 return 0;
64560374
LT
730 }
731 }
eb019375 732
29b734e4
NP
733 if (nr_objects >= nr_alloc) {
734 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 735 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 736 }
29b734e4
NP
737
738 entry = objects + nr_objects++;
c323ac7d 739 memset(entry, 0, sizeof(*entry));
aa7e44bf 740 hashcpy(entry->idx.sha1, sha1);
27225f2e 741 entry->hash = hash;
5c49c116
NP
742 if (type)
743 entry->type = type;
29b734e4
NP
744 if (exclude)
745 entry->preferred_base = 1;
81a216a5
NP
746 else
747 nr_result++;
29b734e4
NP
748 if (found_pack) {
749 entry->in_pack = found_pack;
750 entry->in_pack_offset = found_offset;
751 }
7a979d99
JH
752
753 if (object_ix_hashsz * 3 <= nr_objects * 4)
754 rehash_objects();
29b734e4
NP
755 else
756 object_ix[-1 - ix] = nr_objects;
7a979d99 757
4d4fcc54 758 display_progress(progress_state, nr_objects);
29b734e4 759
a74db82e
JH
760 if (name && no_try_delta(name))
761 entry->no_try_delta = 1;
762
29b734e4 763 return 1;
c323ac7d
LT
764}
765
5379a5c5
JH
766struct pbase_tree_cache {
767 unsigned char sha1[20];
768 int ref;
769 int temporary;
770 void *tree_data;
771 unsigned long tree_size;
772};
773
774static struct pbase_tree_cache *(pbase_tree_cache[256]);
775static int pbase_tree_cache_ix(const unsigned char *sha1)
776{
777 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
778}
779static int pbase_tree_cache_ix_incr(int ix)
780{
781 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
782}
783
784static struct pbase_tree {
785 struct pbase_tree *next;
786 /* This is a phony "cache" entry; we are not
787 * going to evict it nor find it through _get()
788 * mechanism -- this is for the toplevel node that
789 * would almost always change with any commit.
790 */
791 struct pbase_tree_cache pcache;
792} *pbase_tree;
793
794static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
795{
796 struct pbase_tree_cache *ent, *nent;
797 void *data;
798 unsigned long size;
21666f1a 799 enum object_type type;
5379a5c5
JH
800 int neigh;
801 int my_ix = pbase_tree_cache_ix(sha1);
802 int available_ix = -1;
803
804 /* pbase-tree-cache acts as a limited hashtable.
805 * your object will be found at your index or within a few
806 * slots after that slot if it is cached.
807 */
808 for (neigh = 0; neigh < 8; neigh++) {
809 ent = pbase_tree_cache[my_ix];
a89fccd2 810 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
811 ent->ref++;
812 return ent;
813 }
814 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
815 ((0 <= available_ix) &&
816 (!ent && pbase_tree_cache[available_ix])))
817 available_ix = my_ix;
818 if (!ent)
819 break;
820 my_ix = pbase_tree_cache_ix_incr(my_ix);
821 }
822
823 /* Did not find one. Either we got a bogus request or
824 * we need to read and perhaps cache.
825 */
21666f1a 826 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
827 if (!data)
828 return NULL;
21666f1a 829 if (type != OBJ_TREE) {
5379a5c5
JH
830 free(data);
831 return NULL;
832 }
833
834 /* We need to either cache or return a throwaway copy */
835
836 if (available_ix < 0)
837 ent = NULL;
838 else {
839 ent = pbase_tree_cache[available_ix];
840 my_ix = available_ix;
841 }
842
843 if (!ent) {
844 nent = xmalloc(sizeof(*nent));
845 nent->temporary = (available_ix < 0);
846 }
847 else {
848 /* evict and reuse */
849 free(ent->tree_data);
850 nent = ent;
851 }
e702496e 852 hashcpy(nent->sha1, sha1);
5379a5c5
JH
853 nent->tree_data = data;
854 nent->tree_size = size;
855 nent->ref = 1;
856 if (!nent->temporary)
857 pbase_tree_cache[my_ix] = nent;
858 return nent;
859}
860
861static void pbase_tree_put(struct pbase_tree_cache *cache)
862{
863 if (!cache->temporary) {
864 cache->ref--;
865 return;
866 }
867 free(cache->tree_data);
868 free(cache);
869}
870
871static int name_cmp_len(const char *name)
872{
873 int i;
874 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
875 ;
876 return i;
877}
878
879static void add_pbase_object(struct tree_desc *tree,
5379a5c5 880 const char *name,
ce0bd642
LT
881 int cmplen,
882 const char *fullname)
3f9ac8d2 883{
4c068a98 884 struct name_entry entry;
8a5a8d6c 885 int cmp;
4c068a98
LT
886
887 while (tree_entry(tree,&entry)) {
1211be6b
LT
888 if (S_ISGITLINK(entry.mode))
889 continue;
8a5a8d6c
NP
890 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
891 memcmp(name, entry.path, cmplen);
892 if (cmp > 0)
7a979d99 893 continue;
8a5a8d6c
NP
894 if (cmp < 0)
895 return;
5379a5c5 896 if (name[cmplen] != '/') {
5c49c116 897 add_object_entry(entry.sha1,
4d1012c3 898 object_type(entry.mode),
bc32fed5 899 fullname, 1);
5379a5c5
JH
900 return;
901 }
8a5a8d6c 902 if (S_ISDIR(entry.mode)) {
7a979d99 903 struct tree_desc sub;
5379a5c5
JH
904 struct pbase_tree_cache *tree;
905 const char *down = name+cmplen+1;
906 int downlen = name_cmp_len(down);
907
4c068a98 908 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
909 if (!tree)
910 return;
6fda5e51 911 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 912
ce0bd642 913 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
914 pbase_tree_put(tree);
915 }
916 }
917}
1d6b38cc 918
5379a5c5
JH
919static unsigned *done_pbase_paths;
920static int done_pbase_paths_num;
921static int done_pbase_paths_alloc;
922static int done_pbase_path_pos(unsigned hash)
923{
924 int lo = 0;
925 int hi = done_pbase_paths_num;
926 while (lo < hi) {
927 int mi = (hi + lo) / 2;
928 if (done_pbase_paths[mi] == hash)
929 return mi;
930 if (done_pbase_paths[mi] < hash)
931 hi = mi;
932 else
933 lo = mi + 1;
934 }
935 return -lo-1;
936}
937
938static int check_pbase_path(unsigned hash)
939{
940 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
941 if (0 <= pos)
942 return 1;
943 pos = -pos - 1;
944 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
945 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
946 done_pbase_paths = xrealloc(done_pbase_paths,
947 done_pbase_paths_alloc *
948 sizeof(unsigned));
949 }
950 done_pbase_paths_num++;
951 if (pos < done_pbase_paths_num)
952 memmove(done_pbase_paths + pos + 1,
953 done_pbase_paths + pos,
954 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
955 done_pbase_paths[pos] = hash;
956 return 0;
957}
958
bc32fed5 959static void add_preferred_base_object(const char *name)
5379a5c5
JH
960{
961 struct pbase_tree *it;
8a5a8d6c 962 int cmplen;
bc32fed5 963 unsigned hash = name_hash(name);
5379a5c5 964
8a5a8d6c 965 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
966 return;
967
8a5a8d6c 968 cmplen = name_cmp_len(name);
5379a5c5
JH
969 for (it = pbase_tree; it; it = it->next) {
970 if (cmplen == 0) {
bc32fed5 971 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
972 }
973 else {
974 struct tree_desc tree;
6fda5e51 975 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 976 add_pbase_object(&tree, name, cmplen, name);
7a979d99 977 }
3f9ac8d2 978 }
3f9ac8d2
JH
979}
980
7a979d99 981static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 982{
5379a5c5
JH
983 struct pbase_tree *it;
984 void *data;
985 unsigned long size;
986 unsigned char tree_sha1[20];
1d6b38cc 987
8d1d8f83
JH
988 if (window <= num_preferred_base++)
989 return;
990
5379a5c5
JH
991 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
992 if (!data)
7a979d99 993 return;
5379a5c5
JH
994
995 for (it = pbase_tree; it; it = it->next) {
a89fccd2 996 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
997 free(data);
998 return;
999 }
1000 }
1001
1002 it = xcalloc(1, sizeof(*it));
1003 it->next = pbase_tree;
1004 pbase_tree = it;
1005
e702496e 1006 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1007 it->pcache.tree_data = data;
1008 it->pcache.tree_size = size;
3f9ac8d2
JH
1009}
1010
0ef95f72
NP
1011static void cleanup_preferred_base(void)
1012{
1013 struct pbase_tree *it;
1014 unsigned i;
1015
1016 it = pbase_tree;
1017 pbase_tree = NULL;
1018 while (it) {
1019 struct pbase_tree *this = it;
1020 it = this->next;
1021 free(this->pcache.tree_data);
1022 free(this);
1023 }
1024
1025 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1026 if (!pbase_tree_cache[i])
1027 continue;
1028 free(pbase_tree_cache[i]->tree_data);
1029 free(pbase_tree_cache[i]);
1030 pbase_tree_cache[i] = NULL;
1031 }
1032
1033 free(done_pbase_paths);
1034 done_pbase_paths = NULL;
1035 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1036}
1037
c323ac7d
LT
1038static void check_object(struct object_entry *entry)
1039{
5c49c116 1040 if (entry->in_pack) {
780e6e73 1041 struct packed_git *p = entry->in_pack;
03e79c88 1042 struct pack_window *w_curs = NULL;
5c49c116
NP
1043 const unsigned char *base_ref = NULL;
1044 struct object_entry *base_entry;
1045 unsigned long used, used_0;
6777a59f 1046 unsigned int avail;
5c49c116
NP
1047 off_t ofs;
1048 unsigned char *buf, c;
780e6e73 1049
6777a59f 1050 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1051
5c49c116 1052 /*
fa736f72
NP
1053 * We want in_pack_type even if we do not reuse delta
1054 * since non-delta representations could still be reused.
ab7cd7bb 1055 */
09ded04b 1056 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1057 &entry->in_pack_type,
1058 &entry->size);
03d66015
NP
1059 if (used == 0)
1060 goto give_up;
ab7cd7bb 1061
5c49c116
NP
1062 /*
1063 * Determine if this is a delta and if so whether we can
1064 * reuse it or not. Otherwise let's find out as cheaply as
1065 * possible what the actual type and size for this object is.
3f9ac8d2 1066 */
5c49c116
NP
1067 switch (entry->in_pack_type) {
1068 default:
1069 /* Not a delta hence we've already got all we need. */
1070 entry->type = entry->in_pack_type;
1071 entry->in_pack_header_size = used;
03d66015
NP
1072 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1073 goto give_up;
5c49c116
NP
1074 unuse_pack(&w_curs);
1075 return;
1076 case OBJ_REF_DELTA:
a7de7130 1077 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1078 base_ref = use_pack(p, &w_curs,
1079 entry->in_pack_offset + used, NULL);
1080 entry->in_pack_header_size = used + 20;
1081 break;
1082 case OBJ_OFS_DELTA:
1083 buf = use_pack(p, &w_curs,
1084 entry->in_pack_offset + used, NULL);
1085 used_0 = 0;
1086 c = buf[used_0++];
1087 ofs = c & 127;
1088 while (c & 128) {
1089 ofs += 1;
03d66015
NP
1090 if (!ofs || MSB(ofs, 7)) {
1091 error("delta base offset overflow in pack for %s",
1092 sha1_to_hex(entry->idx.sha1));
1093 goto give_up;
1094 }
5c49c116
NP
1095 c = buf[used_0++];
1096 ofs = (ofs << 7) + (c & 127);
780e6e73 1097 }
5c49c116 1098 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1099 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1100 error("delta base offset out of bound for %s",
1101 sha1_to_hex(entry->idx.sha1));
1102 goto give_up;
1103 }
a7de7130 1104 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1105 struct revindex_entry *revidx;
1106 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1107 if (!revidx)
1108 goto give_up;
3449f8c4
NP
1109 base_ref = nth_packed_object_sha1(p, revidx->nr);
1110 }
5c49c116
NP
1111 entry->in_pack_header_size = used + used_0;
1112 break;
780e6e73 1113 }
780e6e73 1114
5c49c116
NP
1115 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1116 /*
1117 * If base_ref was set above that means we wish to
1118 * reuse delta data, and we even found that base
1119 * in the list of objects we want to pack. Goodie!
1120 *
1121 * Depth value does not matter - find_deltas() will
1122 * never consider reused delta as the base object to
1123 * deltify other objects against, in order to avoid
1124 * circular deltas.
3f9ac8d2 1125 */
780e6e73 1126 entry->type = entry->in_pack_type;
5c49c116 1127 entry->delta = base_entry;
64bd76b1 1128 entry->delta_size = entry->size;
15b4d577
JH
1129 entry->delta_sibling = base_entry->delta_child;
1130 base_entry->delta_child = entry;
5c49c116
NP
1131 unuse_pack(&w_curs);
1132 return;
1133 }
ab7cd7bb 1134
5c49c116
NP
1135 if (entry->type) {
1136 /*
1137 * This must be a delta and we already know what the
1138 * final object type is. Let's extract the actual
1139 * object size from the delta header.
1140 */
1141 entry->size = get_size_from_delta(p, &w_curs,
1142 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1143 if (entry->size == 0)
1144 goto give_up;
5c49c116 1145 unuse_pack(&w_curs);
3f9ac8d2
JH
1146 return;
1147 }
5c49c116
NP
1148
1149 /*
1150 * No choice but to fall back to the recursive delta walk
1151 * with sha1_object_info() to find about the object type
1152 * at this point...
1153 */
03d66015 1154 give_up:
5c49c116 1155 unuse_pack(&w_curs);
36e4d74a 1156 }
3f9ac8d2 1157
aa7e44bf 1158 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1159 /*
1160 * The error condition is checked in prepare_pack(). This is
1161 * to permit a missing preferred base object to be ignored
1162 * as a preferred base. Doing so can result in a larger
1163 * pack file, but the transfer will still take place.
1164 */
3f9ac8d2
JH
1165}
1166
5c49c116
NP
1167static int pack_offset_sort(const void *_a, const void *_b)
1168{
1169 const struct object_entry *a = *(struct object_entry **)_a;
1170 const struct object_entry *b = *(struct object_entry **)_b;
1171
1172 /* avoid filesystem trashing with loose objects */
1173 if (!a->in_pack && !b->in_pack)
aa7e44bf 1174 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1175
1176 if (a->in_pack < b->in_pack)
1177 return -1;
1178 if (a->in_pack > b->in_pack)
1179 return 1;
1180 return a->in_pack_offset < b->in_pack_offset ? -1 :
1181 (a->in_pack_offset > b->in_pack_offset);
1182}
1183
c323ac7d
LT
1184static void get_object_details(void)
1185{
7cadf491 1186 uint32_t i;
5c49c116
NP
1187 struct object_entry **sorted_by_offset;
1188
1189 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1190 for (i = 0; i < nr_objects; i++)
1191 sorted_by_offset[i] = objects + i;
1192 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1193
5c49c116
NP
1194 for (i = 0; i < nr_objects; i++)
1195 check_object(sorted_by_offset[i]);
3449f8c4 1196
5c49c116 1197 free(sorted_by_offset);
c323ac7d
LT
1198}
1199
b904166c
NP
1200/*
1201 * We search for deltas in a list sorted by type, by filename hash, and then
1202 * by size, so that we see progressively smaller and smaller files.
1203 * That's because we prefer deltas to be from the bigger file
1204 * to the smaller -- deletes are potentially cheaper, but perhaps
1205 * more importantly, the bigger file is likely the more recent
1206 * one. The deepest deltas are therefore the oldest objects which are
1207 * less susceptible to be accessed often.
1208 */
9668cf59 1209static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1210{
9668cf59
NP
1211 const struct object_entry *a = *(struct object_entry **)_a;
1212 const struct object_entry *b = *(struct object_entry **)_b;
1213
c323ac7d 1214 if (a->type > b->type)
27225f2e 1215 return -1;
b904166c 1216 if (a->type < b->type)
27225f2e 1217 return 1;
b904166c 1218 if (a->hash > b->hash)
7a979d99 1219 return -1;
b904166c 1220 if (a->hash < b->hash)
7a979d99 1221 return 1;
b904166c 1222 if (a->preferred_base > b->preferred_base)
c323ac7d 1223 return -1;
b904166c
NP
1224 if (a->preferred_base < b->preferred_base)
1225 return 1;
c323ac7d 1226 if (a->size > b->size)
b904166c
NP
1227 return -1;
1228 if (a->size < b->size)
c323ac7d 1229 return 1;
b904166c 1230 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1231}
1232
1233struct unpacked {
1234 struct object_entry *entry;
1235 void *data;
f6c7081a 1236 struct delta_index *index;
5a235b5e 1237 unsigned depth;
c323ac7d
LT
1238};
1239
d250626c
NP
1240static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1241 unsigned long delta_size)
074b2eea
MK
1242{
1243 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1244 return 0;
1245
e3dfddb3
MK
1246 if (delta_size < cache_max_small_delta_size)
1247 return 1;
1248
074b2eea
MK
1249 /* cache delta, if objects are large enough compared to delta size */
1250 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1251 return 1;
1252
1253 return 0;
1254}
1255
8ecce684
NP
1256#ifdef THREADED_DELTA_SEARCH
1257
1258static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1259#define read_lock() pthread_mutex_lock(&read_mutex)
1260#define read_unlock() pthread_mutex_unlock(&read_mutex)
1261
3c701839
NP
1262static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1263#define cache_lock() pthread_mutex_lock(&cache_mutex)
1264#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1265
8ecce684
NP
1266static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1267#define progress_lock() pthread_mutex_lock(&progress_mutex)
1268#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1269
1270#else
1271
e1ef8673
JH
1272#define read_lock() (void)0
1273#define read_unlock() (void)0
1274#define cache_lock() (void)0
1275#define cache_unlock() (void)0
1276#define progress_lock() (void)0
1277#define progress_unlock() (void)0
8ecce684
NP
1278
1279#endif
1280
f6c7081a 1281static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1282 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1283{
f6c7081a
NP
1284 struct object_entry *trg_entry = trg->entry;
1285 struct object_entry *src_entry = src->entry;
560b25a8 1286 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1287 unsigned ref_depth;
21666f1a 1288 enum object_type type;
c323ac7d
LT
1289 void *delta_buf;
1290
1291 /* Don't bother doing diffs between different types */
f6c7081a 1292 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1293 return -1;
1294
51d1e83f
LT
1295 /*
1296 * We do not bother to try a delta that we discarded
8dbbd14e 1297 * on an earlier try, but only when reusing delta data.
51d1e83f 1298 */
a7de7130 1299 if (reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1300 trg_entry->in_pack == src_entry->in_pack &&
1301 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1302 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1303 return 0;
1304
898b14ce 1305 /* Let's not bust the allowed depth. */
5a235b5e 1306 if (src->depth >= max_depth)
d116a45a 1307 return 0;
c323ac7d 1308
c3b06a69 1309 /* Now some size filtering heuristics. */
560b25a8 1310 trg_size = trg_entry->size;
c83f032e
NP
1311 if (!trg_entry->delta) {
1312 max_size = trg_size/2 - 20;
1313 ref_depth = 1;
1314 } else {
1315 max_size = trg_entry->delta_size;
5a235b5e 1316 ref_depth = trg->depth;
c83f032e 1317 }
720fe22d 1318 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1319 (max_depth - ref_depth + 1);
c3b06a69
NP
1320 if (max_size == 0)
1321 return 0;
f6c7081a 1322 src_size = src_entry->size;
560b25a8 1323 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1324 if (sizediff >= max_size)
f527cb8c 1325 return 0;
a1dab41a
BD
1326 if (trg_size < src_size / 32)
1327 return 0;
f6c7081a 1328
560b25a8
NP
1329 /* Load data if not already done */
1330 if (!trg->data) {
8ecce684 1331 read_lock();
aa7e44bf 1332 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1333 read_unlock();
2e3404c3
JH
1334 if (!trg->data)
1335 die("object %s cannot be read",
1336 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1337 if (sz != trg_size)
1338 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1339 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1340 *mem_usage += sz;
560b25a8
NP
1341 }
1342 if (!src->data) {
8ecce684 1343 read_lock();
aa7e44bf 1344 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1345 read_unlock();
2e3404c3
JH
1346 if (!src->data)
1347 die("object %s cannot be read",
1348 sha1_to_hex(src_entry->idx.sha1));
560b25a8
NP
1349 if (sz != src_size)
1350 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1351 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1352 *mem_usage += sz;
560b25a8
NP
1353 }
1354 if (!src->index) {
1355 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1356 if (!src->index) {
1357 static int warned = 0;
1358 if (!warned++)
1359 warning("suboptimal pack - out of memory");
1360 return 0;
1361 }
ef0316fc 1362 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1363 }
1364
1365 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1366 if (!delta_buf)
75c42d8c 1367 return 0;
f6c7081a 1368
9e2d57a0 1369 if (trg_entry->delta) {
848d732c
BD
1370 /* Prefer only shallower same-sized deltas. */
1371 if (delta_size == trg_entry->delta_size &&
5a235b5e 1372 src->depth + 1 >= trg->depth) {
848d732c
BD
1373 free(delta_buf);
1374 return 0;
1375 }
074b2eea 1376 }
9e2d57a0 1377
3c701839
NP
1378 /*
1379 * Handle memory allocation outside of the cache
1380 * accounting lock. Compiler will optimize the strangeness
1381 * away when THREADED_DELTA_SEARCH is not defined.
1382 */
8e0f7003 1383 free(trg_entry->delta_data);
3c701839 1384 cache_lock();
9e2d57a0
NP
1385 if (trg_entry->delta_data) {
1386 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1387 trg_entry->delta_data = NULL;
1388 }
d250626c 1389 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1390 delta_cache_size += delta_size;
3c701839
NP
1391 cache_unlock();
1392 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1393 } else {
1394 cache_unlock();
074b2eea 1395 free(delta_buf);
3c701839
NP
1396 }
1397
b7a28f78
NP
1398 trg_entry->delta = src_entry;
1399 trg_entry->delta_size = delta_size;
1400 trg->depth = src->depth + 1;
1401
f6c7081a 1402 return 1;
c323ac7d
LT
1403}
1404
898b14ce 1405static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1406{
898b14ce
NP
1407 struct object_entry *child = me->delta_child;
1408 unsigned int m = n;
1409 while (child) {
1410 unsigned int c = check_delta_limit(child, n + 1);
1411 if (m < c)
1412 m = c;
1413 child = child->delta_sibling;
1414 }
1415 return m;
b2504a0d
NP
1416}
1417
75ad235c 1418static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1419{
ef0316fc 1420 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1421 free_delta_index(n->index);
1422 n->index = NULL;
1423 if (n->data) {
ef0316fc 1424 freed_mem += n->entry->size;
a97773ce
BD
1425 free(n->data);
1426 n->data = NULL;
a97773ce
BD
1427 }
1428 n->entry = NULL;
7d7baa5e 1429 n->depth = 0;
ef0316fc 1430 return freed_mem;
a97773ce
BD
1431}
1432
384b32c0 1433static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1434 int window, int depth, unsigned *processed)
c323ac7d 1435{
384b32c0 1436 uint32_t i, idx = 0, count = 0;
7cadf491 1437 struct unpacked *array;
ef0316fc 1438 unsigned long mem_usage = 0;
c323ac7d 1439
19d4b416 1440 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1441
384b32c0 1442 for (;;) {
421b488a 1443 struct object_entry *entry;
c323ac7d 1444 struct unpacked *n = array + idx;
ef0316fc 1445 int j, max_depth, best_base = -1;
c323ac7d 1446
384b32c0
NP
1447 progress_lock();
1448 if (!*list_size) {
1449 progress_unlock();
1450 break;
1451 }
421b488a 1452 entry = *list++;
384b32c0
NP
1453 (*list_size)--;
1454 if (!entry->preferred_base) {
1455 (*processed)++;
1456 display_progress(progress_state, *processed);
1457 }
1458 progress_unlock();
1459
ef0316fc 1460 mem_usage -= free_unpacked(n);
c323ac7d 1461 n->entry = entry;
ab7cd7bb 1462
a97773ce 1463 while (window_memory_limit &&
ef0316fc 1464 mem_usage > window_memory_limit &&
a97773ce
BD
1465 count > 1) {
1466 uint32_t tail = (idx + window - count) % window;
75ad235c 1467 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1468 count--;
1469 }
1470
75d39853
NP
1471 /* We do not compute delta to *create* objects we are not
1472 * going to pack.
1473 */
1474 if (entry->preferred_base)
1475 goto next;
1476
898b14ce
NP
1477 /*
1478 * If the current object is at pack edge, take the depth the
1479 * objects that depend on the current object into account
1480 * otherwise they would become too deep.
1481 */
1482 max_depth = depth;
1483 if (entry->delta_child) {
1484 max_depth -= check_delta_limit(entry, 0);
1485 if (max_depth <= 0)
1486 goto next;
1487 }
1488
78817c15
LT
1489 j = window;
1490 while (--j > 0) {
77639870 1491 int ret;
7cadf491 1492 uint32_t other_idx = idx + j;
c323ac7d 1493 struct unpacked *m;
78817c15
LT
1494 if (other_idx >= window)
1495 other_idx -= window;
c323ac7d
LT
1496 m = array + other_idx;
1497 if (!m->entry)
1498 break;
ef0316fc 1499 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1500 if (ret < 0)
c323ac7d 1501 break;
77639870
JH
1502 else if (ret > 0)
1503 best_base = other_idx;
c323ac7d 1504 }
898b14ce 1505
ed4a9031
NP
1506 /*
1507 * If we decided to cache the delta data, then it is best
1508 * to compress it right away. First because we have to do
1509 * it anyway, and doing it here while we're threaded will
1510 * save a lot of time in the non threaded write phase,
1511 * as well as allow for caching more deltas within
1512 * the same cache size limit.
1513 * ...
1514 * But only if not writing to stdout, since in that case
1515 * the network is most likely throttling writes anyway,
1516 * and therefore it is best to go to the write phase ASAP
1517 * instead, as we can afford spending more time compressing
1518 * between writes at that moment.
1519 */
1520 if (entry->delta_data && !pack_to_stdout) {
1521 entry->z_delta_size = do_compress(&entry->delta_data,
1522 entry->delta_size);
1523 cache_lock();
1524 delta_cache_size -= entry->delta_size;
1525 delta_cache_size += entry->z_delta_size;
1526 cache_unlock();
1527 }
1528
70ca1a3f
JH
1529 /* if we made n a delta, and if n is already at max
1530 * depth, leaving it in the window is pointless. we
1531 * should evict it first.
70ca1a3f 1532 */
70baf5d4 1533 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1534 continue;
ff45715c 1535
77639870
JH
1536 /*
1537 * Move the best delta base up in the window, after the
1538 * currently deltified object, to keep it longer. It will
1539 * be the first base object to be attempted next.
1540 */
1541 if (entry->delta) {
1542 struct unpacked swap = array[best_base];
1543 int dist = (window + idx - best_base) % window;
1544 int dst = best_base;
1545 while (dist--) {
1546 int src = (dst + 1) % window;
1547 array[dst] = array[src];
1548 dst = src;
1549 }
1550 array[dst] = swap;
1551 }
1552
898b14ce 1553 next:
521a4f4c 1554 idx++;
a97773ce
BD
1555 if (count + 1 < window)
1556 count++;
521a4f4c
LT
1557 if (idx >= window)
1558 idx = 0;
384b32c0 1559 }
adee7bdf 1560
f6c7081a 1561 for (i = 0; i < window; ++i) {
ff45715c 1562 free_delta_index(array[i].index);
adee7bdf 1563 free(array[i].data);
f6c7081a 1564 }
adee7bdf 1565 free(array);
c323ac7d
LT
1566}
1567
8ecce684
NP
1568#ifdef THREADED_DELTA_SEARCH
1569
50f22ada
JS
1570/*
1571 * The main thread waits on the condition that (at least) one of the workers
1572 * has stopped working (which is indicated in the .working member of
1573 * struct thread_params).
1574 * When a work thread has completed its work, it sets .working to 0 and
1575 * signals the main thread and waits on the condition that .data_ready
1576 * becomes 1.
1577 */
1578
8ecce684
NP
1579struct thread_params {
1580 pthread_t thread;
1581 struct object_entry **list;
1582 unsigned list_size;
384b32c0 1583 unsigned remaining;
8ecce684
NP
1584 int window;
1585 int depth;
50f22ada
JS
1586 int working;
1587 int data_ready;
1588 pthread_mutex_t mutex;
1589 pthread_cond_t cond;
8ecce684
NP
1590 unsigned *processed;
1591};
1592
50f22ada 1593static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
c2a33679 1594
8ecce684
NP
1595static void *threaded_find_deltas(void *arg)
1596{
c2a33679
NP
1597 struct thread_params *me = arg;
1598
50f22ada 1599 while (me->remaining) {
384b32c0 1600 find_deltas(me->list, &me->remaining,
c2a33679 1601 me->window, me->depth, me->processed);
50f22ada
JS
1602
1603 progress_lock();
1604 me->working = 0;
1605 pthread_cond_signal(&progress_cond);
1606 progress_unlock();
1607
1608 /*
1609 * We must not set ->data_ready before we wait on the
1610 * condition because the main thread may have set it to 1
1611 * before we get here. In order to be sure that new
1612 * work is available if we see 1 in ->data_ready, it
1613 * was initialized to 0 before this thread was spawned
1614 * and we reset it to 0 right away.
1615 */
1616 pthread_mutex_lock(&me->mutex);
1617 while (!me->data_ready)
1618 pthread_cond_wait(&me->cond, &me->mutex);
1619 me->data_ready = 0;
1620 pthread_mutex_unlock(&me->mutex);
c2a33679 1621 }
50f22ada
JS
1622 /* leave ->working 1 so that this doesn't get more work assigned */
1623 return NULL;
8ecce684
NP
1624}
1625
8ecce684
NP
1626static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1627 int window, int depth, unsigned *processed)
1628{
dcda3614 1629 struct thread_params *p;
384b32c0 1630 int i, ret, active_threads = 0;
c2a33679 1631
ef055571
NP
1632 if (!delta_search_threads) /* --threads=0 means autodetect */
1633 delta_search_threads = online_cpus();
367f4a43 1634 if (delta_search_threads <= 1) {
384b32c0 1635 find_deltas(list, &list_size, window, depth, processed);
367f4a43
NP
1636 return;
1637 }
43cc2b42 1638 if (progress > pack_to_stdout)
b6c29915 1639 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1640 delta_search_threads);
dcda3614 1641 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 1642
50f22ada 1643 /* Partition the work amongst work threads. */
367f4a43 1644 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1645 unsigned sub_size = list_size / (delta_search_threads - i);
1646
bf874896
NP
1647 /* don't use too small segments or no deltas will be found */
1648 if (sub_size < 2*window && i+1 < delta_search_threads)
1649 sub_size = 0;
1650
8ecce684
NP
1651 p[i].window = window;
1652 p[i].depth = depth;
1653 p[i].processed = processed;
50f22ada
JS
1654 p[i].working = 1;
1655 p[i].data_ready = 0;
c2a33679 1656
59921b4b 1657 /* try to split chunks on "path" boundaries */
6fc74703
NP
1658 while (sub_size && sub_size < list_size &&
1659 list[sub_size]->hash &&
384b32c0
NP
1660 list[sub_size]->hash == list[sub_size-1]->hash)
1661 sub_size++;
1662
50f22ada
JS
1663 p[i].list = list;
1664 p[i].list_size = sub_size;
1665 p[i].remaining = sub_size;
59921b4b 1666
384b32c0
NP
1667 list += sub_size;
1668 list_size -= sub_size;
1669 }
1670
50f22ada
JS
1671 /* Start work threads. */
1672 for (i = 0; i < delta_search_threads; i++) {
1673 if (!p[i].list_size)
1674 continue;
68e6a4f8
JS
1675 pthread_mutex_init(&p[i].mutex, NULL);
1676 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1677 ret = pthread_create(&p[i].thread, NULL,
1678 threaded_find_deltas, &p[i]);
1679 if (ret)
1680 die("unable to create thread: %s", strerror(ret));
1681 active_threads++;
1682 }
1683
384b32c0
NP
1684 /*
1685 * Now let's wait for work completion. Each time a thread is done
1686 * with its work, we steal half of the remaining work from the
1687 * thread with the largest number of unprocessed objects and give
1688 * it to that newly idle thread. This ensure good load balancing
1689 * until the remaining object list segments are simply too short
1690 * to be worth splitting anymore.
1691 */
50f22ada
JS
1692 while (active_threads) {
1693 struct thread_params *target = NULL;
384b32c0
NP
1694 struct thread_params *victim = NULL;
1695 unsigned sub_size = 0;
384b32c0
NP
1696
1697 progress_lock();
50f22ada
JS
1698 for (;;) {
1699 for (i = 0; !target && i < delta_search_threads; i++)
1700 if (!p[i].working)
1701 target = &p[i];
1702 if (target)
1703 break;
1704 pthread_cond_wait(&progress_cond, &progress_mutex);
1705 }
1706
384b32c0
NP
1707 for (i = 0; i < delta_search_threads; i++)
1708 if (p[i].remaining > 2*window &&
1709 (!victim || victim->remaining < p[i].remaining))
1710 victim = &p[i];
1711 if (victim) {
1712 sub_size = victim->remaining / 2;
1713 list = victim->list + victim->list_size - sub_size;
1714 while (sub_size && list[0]->hash &&
1715 list[0]->hash == list[-1]->hash) {
1716 list++;
1717 sub_size--;
1718 }
eb9688ff
NP
1719 if (!sub_size) {
1720 /*
1721 * It is possible for some "paths" to have
1722 * so many objects that no hash boundary
1723 * might be found. Let's just steal the
1724 * exact half in that case.
1725 */
1726 sub_size = victim->remaining / 2;
1727 list -= sub_size;
1728 }
384b32c0
NP
1729 target->list = list;
1730 victim->list_size -= sub_size;
1731 victim->remaining -= sub_size;
1732 }
384b32c0
NP
1733 target->list_size = sub_size;
1734 target->remaining = sub_size;
50f22ada
JS
1735 target->working = 1;
1736 progress_unlock();
1737
1738 pthread_mutex_lock(&target->mutex);
1739 target->data_ready = 1;
1740 pthread_cond_signal(&target->cond);
1741 pthread_mutex_unlock(&target->mutex);
c2a33679 1742
384b32c0 1743 if (!sub_size) {
b81d9af7 1744 pthread_join(target->thread, NULL);
50f22ada
JS
1745 pthread_cond_destroy(&target->cond);
1746 pthread_mutex_destroy(&target->mutex);
384b32c0 1747 active_threads--;
c2a33679 1748 }
50f22ada 1749 }
dcda3614 1750 free(p);
8ecce684
NP
1751}
1752
1753#else
384b32c0 1754#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
1755#endif
1756
f0a24aa5
SP
1757static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1758{
1759 unsigned char peeled[20];
1760
1761 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1762 !peel_ref(path, peeled) && /* peelable? */
1763 !is_null_sha1(peeled) && /* annotated tag? */
1764 locate_object_entry(peeled)) /* object packed? */
1765 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1766 return 0;
1767}
1768
f3123c4a
JH
1769static void prepare_pack(int window, int depth)
1770{
9668cf59 1771 struct object_entry **delta_list;
6e1c2344
RJ
1772 uint32_t i, nr_deltas;
1773 unsigned n;
9668cf59 1774
3f9ac8d2 1775 get_object_details();
9668cf59 1776
0e8189e2
NP
1777 /*
1778 * If we're locally repacking then we need to be doubly careful
1779 * from now on in order to make sure no stealth corruption gets
1780 * propagated to the new pack. Clients receiving streamed packs
1781 * should validate everything they get anyway so no need to incur
1782 * the additional cost here in that case.
1783 */
1784 if (!pack_to_stdout)
1785 do_check_packed_object_crc = 1;
1786
75d39853 1787 if (!nr_objects || !window || !depth)
9668cf59
NP
1788 return;
1789
1790 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1791 nr_deltas = n = 0;
1792
1793 for (i = 0; i < nr_objects; i++) {
1794 struct object_entry *entry = objects + i;
1795
1796 if (entry->delta)
1797 /* This happens if we decided to reuse existing
a7de7130 1798 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
1799 */
1800 continue;
1801
1802 if (entry->size < 50)
1803 continue;
1804
1805 if (entry->no_try_delta)
1806 continue;
1807
6d6f9cdd 1808 if (!entry->preferred_base) {
75d39853 1809 nr_deltas++;
6d6f9cdd
SP
1810 if (entry->type < 0)
1811 die("unable to get type of object %s",
1812 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
1813 } else {
1814 if (entry->type < 0) {
1815 /*
1816 * This object is not found, but we
1817 * don't have to include it anyway.
1818 */
1819 continue;
1820 }
6d6f9cdd 1821 }
75d39853
NP
1822
1823 delta_list[n++] = entry;
1824 }
1825
2f8b8947 1826 if (nr_deltas && n > 1) {
e334977d
NP
1827 unsigned nr_done = 0;
1828 if (progress)
dc6a0757
NP
1829 progress_state = start_progress("Compressing objects",
1830 nr_deltas);
75d39853 1831 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1832 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1833 stop_progress(&progress_state);
e334977d
NP
1834 if (nr_done != nr_deltas)
1835 die("inconsistency with delta count");
75d39853 1836 }
9668cf59 1837 free(delta_list);
f3123c4a
JH
1838}
1839
ef90d6d4 1840static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 1841{
eeefa7c9 1842 if (!strcmp(k, "pack.window")) {
4812a93a
JK
1843 window = git_config_int(k, v);
1844 return 0;
1845 }
a97773ce
BD
1846 if (!strcmp(k, "pack.windowmemory")) {
1847 window_memory_limit = git_config_ulong(k, v);
1848 return 0;
1849 }
1850 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1851 depth = git_config_int(k, v);
1852 return 0;
1853 }
960ccca6
DH
1854 if (!strcmp(k, "pack.compression")) {
1855 int level = git_config_int(k, v);
1856 if (level == -1)
1857 level = Z_DEFAULT_COMPRESSION;
1858 else if (level < 0 || level > Z_BEST_COMPRESSION)
1859 die("bad pack compression level %d", level);
1860 pack_compression_level = level;
1861 pack_compression_seen = 1;
1862 return 0;
1863 }
074b2eea
MK
1864 if (!strcmp(k, "pack.deltacachesize")) {
1865 max_delta_cache_size = git_config_int(k, v);
1866 return 0;
1867 }
e3dfddb3
MK
1868 if (!strcmp(k, "pack.deltacachelimit")) {
1869 cache_max_small_delta_size = git_config_int(k, v);
1870 return 0;
1871 }
693b86ff
NP
1872 if (!strcmp(k, "pack.threads")) {
1873 delta_search_threads = git_config_int(k, v);
833e3df1 1874 if (delta_search_threads < 0)
693b86ff
NP
1875 die("invalid number of threads specified (%d)",
1876 delta_search_threads);
1877#ifndef THREADED_DELTA_SEARCH
833e3df1 1878 if (delta_search_threads != 1)
693b86ff
NP
1879 warning("no threads support, ignoring %s", k);
1880#endif
1881 return 0;
1882 }
4d00bda2
NP
1883 if (!strcmp(k, "pack.indexversion")) {
1884 pack_idx_default_version = git_config_int(k, v);
1885 if (pack_idx_default_version > 2)
6e1c2344
RJ
1886 die("bad pack.indexversion=%"PRIu32,
1887 pack_idx_default_version);
4d00bda2
NP
1888 return 0;
1889 }
2b84b5a8
JS
1890 if (!strcmp(k, "pack.packsizelimit")) {
1891 pack_size_limit_cfg = git_config_ulong(k, v);
1892 return 0;
1893 }
ef90d6d4 1894 return git_default_config(k, v, cb);
4812a93a
JK
1895}
1896
b5d97e6b 1897static void read_object_list_from_stdin(void)
c323ac7d 1898{
b5d97e6b
JH
1899 char line[40 + 1 + PATH_MAX + 2];
1900 unsigned char sha1[20];
b2504a0d 1901
da93d12b 1902 for (;;) {
da93d12b
LT
1903 if (!fgets(line, sizeof(line), stdin)) {
1904 if (feof(stdin))
1905 break;
1906 if (!ferror(stdin))
1907 die("fgets returned NULL, not EOF, not error!");
687dd75c 1908 if (errno != EINTR)
d824cbba 1909 die_errno("fgets");
687dd75c
JH
1910 clearerr(stdin);
1911 continue;
da93d12b 1912 }
7a979d99
JH
1913 if (line[0] == '-') {
1914 if (get_sha1_hex(line+1, sha1))
1915 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1916 line);
8d1d8f83 1917 add_preferred_base(sha1);
7a979d99 1918 continue;
21fcd1bd 1919 }
c323ac7d 1920 if (get_sha1_hex(line, sha1))
ef07618f 1921 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1922
bc32fed5
JH
1923 add_preferred_base_object(line+41);
1924 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1925 }
b5d97e6b
JH
1926}
1927
08cdfb13
JH
1928#define OBJECT_ADDED (1u<<20)
1929
11c211fa 1930static void show_commit(struct commit *commit, void *data)
b5d97e6b 1931{
bc32fed5 1932 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1933 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1934}
1935
cf2ab916 1936static void show_object(struct object *obj, const struct name_path *path, const char *last)
b5d97e6b 1937{
cf2ab916
LT
1938 char *name = path_name(path, last);
1939
8d2dfc49
LT
1940 add_preferred_base_object(name);
1941 add_object_entry(obj->sha1, obj->type, name, 0);
1942 obj->flags |= OBJECT_ADDED;
1943
1944 /*
1945 * We will have generated the hash from the name,
1946 * but not saved a pointer to it - we can free it
1947 */
1948 free((char *)name);
b5d97e6b
JH
1949}
1950
8d1d8f83
JH
1951static void show_edge(struct commit *commit)
1952{
1953 add_preferred_base(commit->object.sha1);
1954}
1955
08cdfb13
JH
1956struct in_pack_object {
1957 off_t offset;
1958 struct object *object;
1959};
1960
1961struct in_pack {
1962 int alloc;
1963 int nr;
1964 struct in_pack_object *array;
1965};
1966
1967static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1968{
1969 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1970 in_pack->array[in_pack->nr].object = object;
1971 in_pack->nr++;
1972}
1973
1974/*
1975 * Compare the objects in the offset order, in order to emulate the
f18d244a 1976 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
1977 */
1978static int ofscmp(const void *a_, const void *b_)
1979{
1980 struct in_pack_object *a = (struct in_pack_object *)a_;
1981 struct in_pack_object *b = (struct in_pack_object *)b_;
1982
1983 if (a->offset < b->offset)
1984 return -1;
1985 else if (a->offset > b->offset)
1986 return 1;
1987 else
1988 return hashcmp(a->object->sha1, b->object->sha1);
1989}
1990
1991static void add_objects_in_unpacked_packs(struct rev_info *revs)
1992{
1993 struct packed_git *p;
1994 struct in_pack in_pack;
1995 uint32_t i;
1996
1997 memset(&in_pack, 0, sizeof(in_pack));
1998
1999 for (p = packed_git; p; p = p->next) {
2000 const unsigned char *sha1;
2001 struct object *o;
2002
79bc4c71 2003 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2004 continue;
2005 if (open_pack_index(p))
2006 die("cannot open pack index");
2007
2008 ALLOC_GROW(in_pack.array,
2009 in_pack.nr + p->num_objects,
2010 in_pack.alloc);
2011
2012 for (i = 0; i < p->num_objects; i++) {
2013 sha1 = nth_packed_object_sha1(p, i);
2014 o = lookup_unknown_object(sha1);
2015 if (!(o->flags & OBJECT_ADDED))
2016 mark_in_pack_object(o, p, &in_pack);
2017 o->flags |= OBJECT_ADDED;
2018 }
2019 }
2020
2021 if (in_pack.nr) {
2022 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2023 ofscmp);
2024 for (i = 0; i < in_pack.nr; i++) {
2025 struct object *o = in_pack.array[i].object;
2026 add_object_entry(o->sha1, o->type, "", 0);
2027 }
2028 }
2029 free(in_pack.array);
2030}
2031
094085e3
BC
2032static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2033{
2034 static struct packed_git *last_found = (void *)1;
2035 struct packed_git *p;
2036
2037 p = (last_found != (void *)1) ? last_found : packed_git;
2038
2039 while (p) {
2040 if ((!p->pack_local || p->pack_keep) &&
2041 find_pack_entry_one(sha1, p)) {
2042 last_found = p;
2043 return 1;
2044 }
2045 if (p == last_found)
2046 p = packed_git;
2047 else
2048 p = p->next;
2049 if (p == last_found)
2050 p = p->next;
2051 }
2052 return 0;
2053}
2054
ca11b212
NP
2055static void loosen_unused_packed_objects(struct rev_info *revs)
2056{
2057 struct packed_git *p;
2058 uint32_t i;
2059 const unsigned char *sha1;
2060
2061 for (p = packed_git; p; p = p->next) {
79bc4c71 2062 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2063 continue;
2064
2065 if (open_pack_index(p))
2066 die("cannot open pack index");
2067
2068 for (i = 0; i < p->num_objects; i++) {
2069 sha1 = nth_packed_object_sha1(p, i);
094085e3
BC
2070 if (!locate_object_entry(sha1) &&
2071 !has_sha1_pack_kept_or_nonlocal(sha1))
ca11b212
NP
2072 if (force_object_loose(sha1, p->mtime))
2073 die("unable to force loose object");
2074 }
2075 }
2076}
2077
8d1d8f83 2078static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2079{
2080 struct rev_info revs;
2081 char line[1000];
b5d97e6b
JH
2082 int flags = 0;
2083
b5d97e6b
JH
2084 init_revisions(&revs, NULL);
2085 save_commit_buffer = 0;
b5d97e6b
JH
2086 setup_revisions(ac, av, &revs, NULL);
2087
b5d97e6b
JH
2088 while (fgets(line, sizeof(line), stdin) != NULL) {
2089 int len = strlen(line);
872c930d 2090 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2091 line[--len] = 0;
2092 if (!len)
2093 break;
2094 if (*line == '-') {
2095 if (!strcmp(line, "--not")) {
2096 flags ^= UNINTERESTING;
2097 continue;
2098 }
2099 die("not a rev '%s'", line);
2100 }
2101 if (handle_revision_arg(line, &revs, flags, 1))
2102 die("bad revision '%s'", line);
2103 }
2104
3d51e1b5
MK
2105 if (prepare_revision_walk(&revs))
2106 die("revision walk setup failed");
8d1d8f83 2107 mark_edges_uninteresting(revs.commits, &revs, show_edge);
11c211fa 2108 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13
JH
2109
2110 if (keep_unreachable)
2111 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2112 if (unpack_unreachable)
2113 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2114}
2115
b6b32ccb
JH
2116static int adjust_perm(const char *path, mode_t mode)
2117{
2118 if (chmod(path, mode))
2119 return -1;
2120 return adjust_shared_perm(path);
2121}
2122
b5d97e6b
JH
2123int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2124{
b5d97e6b 2125 int use_internal_rev_list = 0;
8d1d8f83 2126 int thin = 0;
7cadf491 2127 uint32_t i;
ffa84ffb
RD
2128 const char **rp_av;
2129 int rp_ac_alloc = 64;
8d1d8f83
JH
2130 int rp_ac;
2131
dae556bd
CC
2132 read_replace_refs = 0;
2133
ffa84ffb
RD
2134 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2135
8d1d8f83
JH
2136 rp_av[0] = "pack-objects";
2137 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2138 rp_ac = 2;
b5d97e6b 2139
ef90d6d4 2140 git_config(git_pack_config, NULL);
960ccca6
DH
2141 if (!pack_compression_seen && core_compression_seen)
2142 pack_compression_level = core_compression_level;
b5d97e6b
JH
2143
2144 progress = isatty(2);
2145 for (i = 1; i < argc; i++) {
2146 const char *arg = argv[i];
2147
2148 if (*arg != '-')
2149 break;
2150
2151 if (!strcmp("--non-empty", arg)) {
2152 non_empty = 1;
2153 continue;
2154 }
2155 if (!strcmp("--local", arg)) {
2156 local = 1;
2157 continue;
2158 }
b5d97e6b
JH
2159 if (!strcmp("--incremental", arg)) {
2160 incremental = 1;
2161 continue;
2162 }
e96fb9b8
BC
2163 if (!strcmp("--honor-pack-keep", arg)) {
2164 ignore_packed_keep = 1;
2165 continue;
2166 }
960ccca6
DH
2167 if (!prefixcmp(arg, "--compression=")) {
2168 char *end;
2169 int level = strtoul(arg+14, &end, 0);
2170 if (!arg[14] || *end)
2171 usage(pack_usage);
2172 if (level == -1)
2173 level = Z_DEFAULT_COMPRESSION;
2174 else if (level < 0 || level > Z_BEST_COMPRESSION)
2175 die("bad pack compression level %d", level);
2176 pack_compression_level = level;
2177 continue;
2178 }
6b94b1a0
DH
2179 if (!prefixcmp(arg, "--max-pack-size=")) {
2180 char *end;
2b84b5a8 2181 pack_size_limit_cfg = 0;
6b94b1a0
DH
2182 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2183 if (!arg[16] || *end)
2184 usage(pack_usage);
2185 continue;
2186 }
599065a3 2187 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2188 char *end;
2189 window = strtoul(arg+9, &end, 0);
2190 if (!arg[9] || *end)
2191 usage(pack_usage);
2192 continue;
2193 }
a97773ce
BD
2194 if (!prefixcmp(arg, "--window-memory=")) {
2195 if (!git_parse_ulong(arg+16, &window_memory_limit))
2196 usage(pack_usage);
2197 continue;
2198 }
367f4a43
NP
2199 if (!prefixcmp(arg, "--threads=")) {
2200 char *end;
2201 delta_search_threads = strtoul(arg+10, &end, 0);
833e3df1 2202 if (!arg[10] || *end || delta_search_threads < 0)
367f4a43
NP
2203 usage(pack_usage);
2204#ifndef THREADED_DELTA_SEARCH
833e3df1 2205 if (delta_search_threads != 1)
367f4a43
NP
2206 warning("no threads support, "
2207 "ignoring %s", arg);
2208#endif
2209 continue;
2210 }
599065a3 2211 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2212 char *end;
2213 depth = strtoul(arg+8, &end, 0);
2214 if (!arg[8] || *end)
2215 usage(pack_usage);
2216 continue;
2217 }
2218 if (!strcmp("--progress", arg)) {
2219 progress = 1;
2220 continue;
2221 }
231f240b
NP
2222 if (!strcmp("--all-progress", arg)) {
2223 progress = 2;
2224 continue;
2225 }
b5d97e6b
JH
2226 if (!strcmp("-q", arg)) {
2227 progress = 0;
2228 continue;
2229 }
2230 if (!strcmp("--no-reuse-delta", arg)) {
a7de7130 2231 reuse_delta = 0;
b5d97e6b
JH
2232 continue;
2233 }
fa736f72 2234 if (!strcmp("--no-reuse-object", arg)) {
a7de7130 2235 reuse_object = reuse_delta = 0;
fa736f72
NP
2236 continue;
2237 }
be6b1914 2238 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2239 allow_ofs_delta = 1;
be6b1914
NP
2240 continue;
2241 }
b5d97e6b
JH
2242 if (!strcmp("--stdout", arg)) {
2243 pack_to_stdout = 1;
2244 continue;
2245 }
2246 if (!strcmp("--revs", arg)) {
2247 use_internal_rev_list = 1;
2248 continue;
2249 }
08cdfb13
JH
2250 if (!strcmp("--keep-unreachable", arg)) {
2251 keep_unreachable = 1;
2252 continue;
2253 }
ca11b212
NP
2254 if (!strcmp("--unpack-unreachable", arg)) {
2255 unpack_unreachable = 1;
2256 continue;
2257 }
f0a24aa5
SP
2258 if (!strcmp("--include-tag", arg)) {
2259 include_tag = 1;
2260 continue;
2261 }
8d1d8f83 2262 if (!strcmp("--unpacked", arg) ||
63049292 2263 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2264 !strcmp("--all", arg)) {
2265 use_internal_rev_list = 1;
ffa84ffb
RD
2266 if (rp_ac >= rp_ac_alloc - 1) {
2267 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2268 rp_av = xrealloc(rp_av,
2269 rp_ac_alloc * sizeof(*rp_av));
2270 }
8d1d8f83 2271 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2272 continue;
2273 }
8d1d8f83
JH
2274 if (!strcmp("--thin", arg)) {
2275 use_internal_rev_list = 1;
2276 thin = 1;
2277 rp_av[1] = "--objects-edge";
b5d97e6b 2278 continue;
4ba7d711
NP
2279 }
2280 if (!prefixcmp(arg, "--index-version=")) {
2281 char *c;
aa7e44bf
GB
2282 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2283 if (pack_idx_default_version > 2)
4ba7d711
NP
2284 die("bad %s", arg);
2285 if (*c == ',')
aa7e44bf
GB
2286 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2287 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2288 die("bad %s", arg);
2289 continue;
b5d97e6b 2290 }
7f3140cd
JS
2291 if (!strcmp(arg, "--keep-true-parents")) {
2292 grafts_replace_parents = 0;
2293 continue;
2294 }
b5d97e6b
JH
2295 usage(pack_usage);
2296 }
2297
2298 /* Traditionally "pack-objects [options] base extra" failed;
2299 * we would however want to take refs parameter that would
2300 * have been given to upstream rev-list ourselves, which means
2301 * we somehow want to say what the base name is. So the
2302 * syntax would be:
2303 *
2304 * pack-objects [options] base <refs...>
2305 *
2306 * in other words, we would treat the first non-option as the
2307 * base_name and send everything else to the internal revision
2308 * walker.
2309 */
2310
2311 if (!pack_to_stdout)
2312 base_name = argv[i++];
2313
2314 if (pack_to_stdout != !base_name)
2315 usage(pack_usage);
2316
2b84b5a8
JS
2317 if (!pack_to_stdout && !pack_size_limit)
2318 pack_size_limit = pack_size_limit_cfg;
2319
01c12a23
DH
2320 if (pack_to_stdout && pack_size_limit)
2321 die("--max-pack-size cannot be used to build a pack for transfer.");
2322
8d1d8f83
JH
2323 if (!pack_to_stdout && thin)
2324 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2325
ca11b212
NP
2326 if (keep_unreachable && unpack_unreachable)
2327 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2328
b5d97e6b
JH
2329 prepare_packed_git();
2330
13aaf148 2331 if (progress)
dc6a0757 2332 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2333 if (!use_internal_rev_list)
2334 read_object_list_from_stdin();
8d1d8f83
JH
2335 else {
2336 rp_av[rp_ac] = NULL;
2337 get_object_list(rp_ac, rp_av);
2338 }
0ef95f72 2339 cleanup_preferred_base();
f0a24aa5
SP
2340 if (include_tag && nr_result)
2341 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2342 stop_progress(&progress_state);
96a02f8f 2343
f0b0af1b 2344 if (non_empty && !nr_result)
1c4a2912 2345 return 0;
f7ae6a93
NP
2346 if (nr_result)
2347 prepare_pack(window, depth);
d01fb92f 2348 write_pack_file();
ab7cd7bb 2349 if (progress)
6e1c2344
RJ
2350 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2351 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2352 written, written_delta, reused, reused_delta);
c323ac7d
LT
2353 return 0;
2354}