make pack-objects a bit more resilient to repo corruption
[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
8ecce684 21#include <pthread.h>
93749194 22#include "thread-utils.h"
8ecce684
NP
23#endif
24
812bdbc3 25static const char pack_usage[] =
0adda936 26 "git pack-objects [ -q | --progress | --all-progress ]\n"
4f366275 27 " [--all-progress-implied]\n"
62b4698e
ŠN
28 " [--max-pack-size=<n>] [--local] [--incremental]\n"
29 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
812bdbc3 30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
62b4698e 31 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
812bdbc3 32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
86953531 33 " [--keep-unreachable | --unpack-unreachable]\n"
88422065 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
780e6e73
NP
157/*
158 * we are going to reuse the existing object data as is. make
159 * sure it is not corrupt.
160 */
079afb18
SP
161static int check_pack_inflate(struct packed_git *p,
162 struct pack_window **w_curs,
6777a59f
SP
163 off_t offset,
164 off_t len,
079afb18
SP
165 unsigned long expect)
166{
167 z_stream stream;
168 unsigned char fakebuf[4096], *in;
169 int st;
170
171 memset(&stream, 0, sizeof(stream));
39c68542 172 git_inflate_init(&stream);
079afb18
SP
173 do {
174 in = use_pack(p, w_curs, offset, &stream.avail_in);
175 stream.next_in = in;
176 stream.next_out = fakebuf;
177 stream.avail_out = sizeof(fakebuf);
39c68542 178 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
179 offset += stream.next_in - in;
180 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 181 git_inflate_end(&stream);
079afb18
SP
182 return (st == Z_STREAM_END &&
183 stream.total_out == expect &&
184 stream.total_in == len) ? 0 : -1;
185}
186
187static void copy_pack_data(struct sha1file *f,
188 struct packed_git *p,
189 struct pack_window **w_curs,
6777a59f
SP
190 off_t offset,
191 off_t len)
079afb18
SP
192{
193 unsigned char *in;
194 unsigned int avail;
195
196 while (len) {
197 in = use_pack(p, w_curs, offset, &avail);
198 if (avail > len)
6777a59f 199 avail = (unsigned int)len;
079afb18
SP
200 sha1write(f, in, avail);
201 offset += avail;
202 len -= avail;
203 }
204}
205
d7dd0223 206static unsigned long write_object(struct sha1file *f,
17b08f2c
DH
207 struct object_entry *entry,
208 off_t write_offset)
c323ac7d 209{
30ebb40a 210 unsigned long size, limit, datalen;
3f9ac8d2 211 void *buf;
2c5ef824 212 unsigned char header[10], dheader[10];
6777a59f 213 unsigned hdrlen;
2c5ef824
NP
214 enum object_type type;
215 int usable_delta, to_reuse;
c323ac7d 216
78d1e84f
NP
217 if (!pack_to_stdout)
218 crc32_begin(f);
219
2c5ef824
NP
220 type = entry->type;
221
a2430dde 222 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
223 if (!pack_size_limit || !nr_written)
224 limit = 0;
225 else if (pack_size_limit <= write_offset)
226 /*
227 * the earlier object did not fit the limit; avoid
228 * mistaking this with unlimited (i.e. limit = 0).
229 */
230 limit = 1;
231 else
232 limit = pack_size_limit - write_offset;
2c5ef824
NP
233
234 if (!entry->delta)
235 usable_delta = 0; /* no delta */
236 else if (!pack_size_limit)
237 usable_delta = 1; /* unlimited packfile */
238 else if (entry->delta->idx.offset == (off_t)-1)
239 usable_delta = 0; /* base was written to another pack */
240 else if (entry->delta->idx.offset)
241 usable_delta = 1; /* base already exists in this pack */
242 else
243 usable_delta = 0; /* base could end up in another pack */
244
a7de7130 245 if (!reuse_object)
fa736f72
NP
246 to_reuse = 0; /* explicit */
247 else if (!entry->in_pack)
ab7cd7bb 248 to_reuse = 0; /* can't reuse what we don't have */
2c5ef824 249 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
17b08f2c
DH
250 /* check_object() decided it for us ... */
251 to_reuse = usable_delta;
252 /* ... but pack split may override that */
2c5ef824 253 else if (type != entry->in_pack_type)
ab7cd7bb
JH
254 to_reuse = 0; /* pack has delta which is unusable */
255 else if (entry->delta)
256 to_reuse = 0; /* we want to pack afresh */
257 else
258 to_reuse = 1; /* we have it in-pack undeltified,
259 * and we do not need to deltify it.
260 */
261
df6d6101 262 if (!to_reuse) {
64bd76b1 263 no_reuse:
5476a8ad 264 if (!usable_delta) {
2c5ef824 265 buf = read_sha1_file(entry->idx.sha1, &type, &size);
5476a8ad 266 if (!buf)
aa7e44bf 267 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
ed4a9031
NP
268 /*
269 * make sure no cached delta data remains from a
3ea3c215 270 * previous attempt before a pack split occurred.
ed4a9031
NP
271 */
272 free(entry->delta_data);
273 entry->delta_data = NULL;
274 entry->z_delta_size = 0;
5476a8ad 275 } else if (entry->delta_data) {
3f9ac8d2 276 size = entry->delta_size;
5476a8ad
NP
277 buf = entry->delta_data;
278 entry->delta_data = NULL;
2c5ef824 279 type = (allow_ofs_delta && entry->delta->idx.offset) ?
be6b1914 280 OBJ_OFS_DELTA : OBJ_REF_DELTA;
17b08f2c 281 } else {
3613f9b4 282 buf = get_delta(entry);
5476a8ad 283 size = entry->delta_size;
2c5ef824 284 type = (allow_ofs_delta && entry->delta->idx.offset) ?
5476a8ad 285 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2 286 }
ed4a9031
NP
287
288 if (entry->z_delta_size)
289 datalen = entry->z_delta_size;
290 else
291 datalen = do_compress(&buf, size);
8c3c7b2a 292
3f9ac8d2
JH
293 /*
294 * The object header is a byte of 'type' followed by zero or
be6b1914 295 * more bytes of length.
3f9ac8d2 296 */
1b22b6c8 297 hdrlen = encode_in_pack_object_header(type, size, header);
3f9ac8d2 298
2c5ef824 299 if (type == OBJ_OFS_DELTA) {
be6b1914
NP
300 /*
301 * Deltas with relative base contain an additional
302 * encoding of the relative offset for the delta
303 * base from this object's position in the pack.
304 */
aa7e44bf 305 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
306 unsigned pos = sizeof(dheader) - 1;
307 dheader[pos] = ofs & 127;
be6b1914 308 while (ofs >>= 7)
17b08f2c
DH
309 dheader[--pos] = 128 | (--ofs & 127);
310 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
17b08f2c
DH
311 free(buf);
312 return 0;
313 }
314 sha1write(f, header, hdrlen);
315 sha1write(f, dheader + pos, sizeof(dheader) - pos);
316 hdrlen += sizeof(dheader) - pos;
2c5ef824 317 } else if (type == OBJ_REF_DELTA) {
be6b1914
NP
318 /*
319 * Deltas with a base reference contain
320 * an additional 20 bytes for the base sha1.
321 */
17b08f2c 322 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
17b08f2c
DH
323 free(buf);
324 return 0;
325 }
326 sha1write(f, header, hdrlen);
aa7e44bf 327 sha1write(f, entry->delta->idx.sha1, 20);
3f9ac8d2 328 hdrlen += 20;
17b08f2c
DH
329 } else {
330 if (limit && hdrlen + datalen + 20 >= limit) {
17b08f2c
DH
331 free(buf);
332 return 0;
333 }
334 sha1write(f, header, hdrlen);
3f9ac8d2 335 }
30ebb40a 336 sha1write(f, buf, datalen);
3f9ac8d2 337 free(buf);
c323ac7d 338 }
3f9ac8d2
JH
339 else {
340 struct packed_git *p = entry->in_pack;
03e79c88 341 struct pack_window *w_curs = NULL;
91ecbeca 342 struct revindex_entry *revidx;
6777a59f 343 off_t offset;
3f9ac8d2 344
64bd76b1 345 if (entry->delta)
2c5ef824 346 type = (allow_ofs_delta && entry->delta->idx.offset) ?
780e6e73 347 OBJ_OFS_DELTA : OBJ_REF_DELTA;
1b22b6c8 348 hdrlen = encode_in_pack_object_header(type, entry->size, header);
64bd76b1 349
91ecbeca 350 offset = entry->in_pack_offset;
3449f8c4 351 revidx = find_pack_revindex(p, offset);
91ecbeca
NP
352 datalen = revidx[1].offset - offset;
353 if (!pack_to_stdout && p->index_version > 1 &&
64bd76b1
NP
354 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
355 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
356 unuse_pack(&w_curs);
357 goto no_reuse;
358 }
359
91ecbeca
NP
360 offset += entry->in_pack_header_size;
361 datalen -= entry->in_pack_header_size;
64bd76b1
NP
362 if (!pack_to_stdout && p->index_version == 1 &&
363 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
364 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
365 unuse_pack(&w_curs);
366 goto no_reuse;
367 }
368
2c5ef824 369 if (type == OBJ_OFS_DELTA) {
aa7e44bf 370 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
371 unsigned pos = sizeof(dheader) - 1;
372 dheader[pos] = ofs & 127;
373 while (ofs >>= 7)
374 dheader[--pos] = 128 | (--ofs & 127);
59dd9ed1
NP
375 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
376 unuse_pack(&w_curs);
17b08f2c 377 return 0;
59dd9ed1 378 }
17b08f2c
DH
379 sha1write(f, header, hdrlen);
380 sha1write(f, dheader + pos, sizeof(dheader) - pos);
381 hdrlen += sizeof(dheader) - pos;
64bd76b1 382 reused_delta++;
2c5ef824 383 } else if (type == OBJ_REF_DELTA) {
59dd9ed1
NP
384 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
385 unuse_pack(&w_curs);
17b08f2c 386 return 0;
59dd9ed1 387 }
17b08f2c 388 sha1write(f, header, hdrlen);
aa7e44bf 389 sha1write(f, entry->delta->idx.sha1, 20);
17b08f2c 390 hdrlen += 20;
64bd76b1 391 reused_delta++;
17b08f2c 392 } else {
59dd9ed1
NP
393 if (limit && hdrlen + datalen + 20 >= limit) {
394 unuse_pack(&w_curs);
17b08f2c 395 return 0;
59dd9ed1 396 }
17b08f2c
DH
397 sha1write(f, header, hdrlen);
398 }
079afb18 399 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 400 unuse_pack(&w_curs);
3f9ac8d2 401 reused++;
a733cb60 402 }
17b08f2c 403 if (usable_delta)
ab7cd7bb 404 written_delta++;
3f9ac8d2 405 written++;
78d1e84f 406 if (!pack_to_stdout)
aa7e44bf 407 entry->idx.crc32 = crc32_end(f);
c323ac7d
LT
408 return hdrlen + datalen;
409}
410
6ed7f25e 411static int write_one(struct sha1file *f,
9d5ab962 412 struct object_entry *e,
6ed7f25e 413 off_t *offset)
9d5ab962 414{
d7dd0223
NP
415 unsigned long size;
416
417 /* offset is non zero if object is written already. */
aa7e44bf 418 if (e->idx.offset || e->preferred_base)
a2430dde 419 return -1;
d7dd0223 420
720c9f7b
NP
421 /* if we are deltified, write out base object first. */
422 if (e->delta && !write_one(f, e->delta, offset))
423 return 0;
d7dd0223 424
6ed7f25e
NP
425 e->idx.offset = *offset;
426 size = write_object(f, e, *offset);
17b08f2c 427 if (!size) {
aa7e44bf 428 e->idx.offset = 0;
17b08f2c
DH
429 return 0;
430 }
79814f42 431 written_list[nr_written++] = &e->idx;
d7dd0223
NP
432
433 /* make sure off_t is sufficiently large not to wrap */
c03c8315 434 if (signed_add_overflows(*offset, size))
d7dd0223 435 die("pack too large for current definition of off_t");
6ed7f25e
NP
436 *offset += size;
437 return 1;
9d5ab962
JH
438}
439
d01fb92f 440static void write_pack_file(void)
c323ac7d 441{
ebe27b13 442 uint32_t i = 0, j;
d22b9290 443 struct sha1file *f;
6ed7f25e 444 off_t offset;
a733cb60 445 struct pack_header hdr;
ebe27b13 446 uint32_t nr_remaining = nr_result;
f746bae8 447 time_t last_mtime = 0;
81a216a5 448
bcd7954e 449 if (progress > pack_to_stdout)
dc6a0757 450 progress_state = start_progress("Writing objects", nr_result);
79814f42 451 written_list = xmalloc(nr_objects * sizeof(*written_list));
183bdb2c 452
ebe27b13 453 do {
aa7e44bf 454 unsigned char sha1[20];
7ba502c4 455 char *pack_tmp_name = NULL;
aa7e44bf 456
ebe27b13 457 if (pack_to_stdout) {
2a128d63 458 f = sha1fd_throughput(1, "<stdout>", progress_state);
ebe27b13 459 } else {
7ba502c4
NP
460 char tmpname[PATH_MAX];
461 int fd;
6e180cdc
JH
462 fd = odb_mkstemp(tmpname, sizeof(tmpname),
463 "pack/tmp_pack_XXXXXX");
ebe27b13
DH
464 pack_tmp_name = xstrdup(tmpname);
465 f = sha1fd(fd, pack_tmp_name);
466 }
467
468 hdr.hdr_signature = htonl(PACK_SIGNATURE);
469 hdr.hdr_version = htonl(PACK_VERSION);
470 hdr.hdr_entries = htonl(nr_remaining);
471 sha1write(f, &hdr, sizeof(hdr));
472 offset = sizeof(hdr);
473 nr_written = 0;
720c9f7b
NP
474 for (; i < nr_objects; i++) {
475 if (!write_one(f, objects + i, &offset))
476 break;
477 display_progress(progress_state, written);
478 }
c553ca25 479
ebe27b13
DH
480 /*
481 * Did we write the wrong # entries in the header?
482 * If so, rewrite it like in fast-import
483 */
54352bb2
LT
484 if (pack_to_stdout) {
485 sha1close(f, sha1, CSUM_CLOSE);
486 } else if (nr_written == nr_remaining) {
487 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 488 } else {
ac0463ed 489 int fd = sha1close(f, sha1, 0);
abeb40e5 490 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 491 nr_written, sha1, offset);
7ba502c4 492 close(fd);
ebe27b13
DH
493 }
494
495 if (!pack_to_stdout) {
f746bae8 496 struct stat st;
3bb72562
LT
497 const char *idx_tmp_name;
498 char tmpname[PATH_MAX];
d01fb92f 499
79814f42
NP
500 idx_tmp_name = write_idx_file(NULL, written_list,
501 nr_written, sha1);
f746bae8 502
d01fb92f 503 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
aa7e44bf 504 base_name, sha1_to_hex(sha1));
c74faea1 505 free_pack_by_name(tmpname);
f80c7ae8 506 if (adjust_shared_perm(pack_tmp_name))
d824cbba 507 die_errno("unable to make temporary pack file readable");
d01fb92f 508 if (rename(pack_tmp_name, tmpname))
d824cbba 509 die_errno("unable to rename temporary pack file");
f746bae8
NP
510
511 /*
512 * Packs are runtime accessed in their mtime
513 * order since newer packs are more likely to contain
514 * younger objects. So if we are creating multiple
515 * packs then we should modify the mtime of later ones
516 * to preserve this property.
517 */
518 if (stat(tmpname, &st) < 0) {
519 warning("failed to stat %s: %s",
520 tmpname, strerror(errno));
521 } else if (!last_mtime) {
522 last_mtime = st.st_mtime;
523 } else {
524 struct utimbuf utb;
525 utb.actime = st.st_atime;
526 utb.modtime = --last_mtime;
527 if (utime(tmpname, &utb) < 0)
528 warning("failed utime() on %s: %s",
529 tmpname, strerror(errno));
530 }
531
d01fb92f 532 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 533 base_name, sha1_to_hex(sha1));
f80c7ae8 534 if (adjust_shared_perm(idx_tmp_name))
d824cbba 535 die_errno("unable to make temporary index file readable");
d01fb92f 536 if (rename(idx_tmp_name, tmpname))
d824cbba 537 die_errno("unable to rename temporary index file");
f746bae8 538
3bb72562 539 free((void *) idx_tmp_name);
7ba502c4 540 free(pack_tmp_name);
aa7e44bf 541 puts(sha1_to_hex(sha1));
ebe27b13
DH
542 }
543
544 /* mark written objects as written to previous pack */
545 for (j = 0; j < nr_written; j++) {
79814f42 546 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
547 }
548 nr_remaining -= nr_written;
720c9f7b 549 } while (nr_remaining && i < nr_objects);
ebe27b13
DH
550
551 free(written_list);
4d4fcc54 552 stop_progress(&progress_state);
67c08ce1 553 if (written != nr_result)
6e1c2344
RJ
554 die("wrote %"PRIu32" objects while expecting %"PRIu32,
555 written, nr_result);
c323ac7d
LT
556}
557
7a979d99
JH
558static int locate_object_entry_hash(const unsigned char *sha1)
559{
560 int i;
561 unsigned int ui;
562 memcpy(&ui, sha1, sizeof(unsigned int));
563 i = ui % object_ix_hashsz;
564 while (0 < object_ix[i]) {
aa7e44bf 565 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
566 return i;
567 if (++i == object_ix_hashsz)
568 i = 0;
569 }
570 return -1 - i;
571}
572
573static struct object_entry *locate_object_entry(const unsigned char *sha1)
574{
575 int i;
576
577 if (!object_ix_hashsz)
578 return NULL;
579
580 i = locate_object_entry_hash(sha1);
581 if (0 <= i)
582 return &objects[object_ix[i]-1];
583 return NULL;
584}
585
586static void rehash_objects(void)
c323ac7d 587{
7cadf491 588 uint32_t i;
7a979d99
JH
589 struct object_entry *oe;
590
591 object_ix_hashsz = nr_objects * 3;
592 if (object_ix_hashsz < 1024)
593 object_ix_hashsz = 1024;
594 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 595 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 596 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 597 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
598 if (0 <= ix)
599 continue;
600 ix = -1 - ix;
601 object_ix[ix] = i + 1;
602 }
603}
604
ce0bd642 605static unsigned name_hash(const char *name)
1d6b38cc 606{
48fb7deb 607 unsigned c, hash = 0;
ce0bd642 608
bc32fed5
JH
609 if (!name)
610 return 0;
611
eeef7135 612 /*
ce0bd642
LT
613 * This effectively just creates a sortable number from the
614 * last sixteen non-whitespace characters. Last characters
615 * count "most", so things that end in ".c" sort together.
eeef7135 616 */
ce0bd642
LT
617 while ((c = *name++) != 0) {
618 if (isspace(c))
619 continue;
620 hash = (hash >> 2) + (c << 24);
621 }
1d6b38cc
JH
622 return hash;
623}
624
a74db82e
JH
625static void setup_delta_attr_check(struct git_attr_check *check)
626{
627 static struct git_attr *attr_delta;
628
629 if (!attr_delta)
7fb0eaa2 630 attr_delta = git_attr("delta");
a74db82e
JH
631
632 check[0].attr = attr_delta;
633}
634
635static int no_try_delta(const char *path)
636{
637 struct git_attr_check check[1];
638
639 setup_delta_attr_check(check);
640 if (git_checkattr(path, ARRAY_SIZE(check), check))
641 return 0;
642 if (ATTR_FALSE(check->value))
643 return 1;
644 return 0;
645}
646
5c49c116 647static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 648 const char *name, int exclude)
c323ac7d 649{
c323ac7d 650 struct object_entry *entry;
29b734e4 651 struct packed_git *p, *found_pack = NULL;
6777a59f 652 off_t found_offset = 0;
29b734e4 653 int ix;
bc32fed5 654 unsigned hash = name_hash(name);
29b734e4
NP
655
656 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
657 if (ix >= 0) {
658 if (exclude) {
659 entry = objects + object_ix[ix] - 1;
81a216a5
NP
660 if (!entry->preferred_base)
661 nr_result--;
29b734e4
NP
662 entry->preferred_base = 1;
663 }
664 return 0;
665 }
c323ac7d 666
daae0625
BC
667 if (!exclude && local && has_loose_object_nonlocal(sha1))
668 return 0;
669
5c49c116
NP
670 for (p = packed_git; p; p = p->next) {
671 off_t offset = find_pack_entry_one(sha1, p);
672 if (offset) {
673 if (!found_pack) {
674 found_offset = offset;
675 found_pack = p;
64560374 676 }
5c49c116
NP
677 if (exclude)
678 break;
679 if (incremental)
680 return 0;
681 if (local && !p->pack_local)
682 return 0;
e96fb9b8
BC
683 if (ignore_packed_keep && p->pack_local && p->pack_keep)
684 return 0;
64560374
LT
685 }
686 }
eb019375 687
29b734e4
NP
688 if (nr_objects >= nr_alloc) {
689 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 690 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 691 }
29b734e4
NP
692
693 entry = objects + nr_objects++;
c323ac7d 694 memset(entry, 0, sizeof(*entry));
aa7e44bf 695 hashcpy(entry->idx.sha1, sha1);
27225f2e 696 entry->hash = hash;
5c49c116
NP
697 if (type)
698 entry->type = type;
29b734e4
NP
699 if (exclude)
700 entry->preferred_base = 1;
81a216a5
NP
701 else
702 nr_result++;
29b734e4
NP
703 if (found_pack) {
704 entry->in_pack = found_pack;
705 entry->in_pack_offset = found_offset;
706 }
7a979d99
JH
707
708 if (object_ix_hashsz * 3 <= nr_objects * 4)
709 rehash_objects();
29b734e4
NP
710 else
711 object_ix[-1 - ix] = nr_objects;
7a979d99 712
4d4fcc54 713 display_progress(progress_state, nr_objects);
29b734e4 714
a74db82e
JH
715 if (name && no_try_delta(name))
716 entry->no_try_delta = 1;
717
29b734e4 718 return 1;
c323ac7d
LT
719}
720
5379a5c5
JH
721struct pbase_tree_cache {
722 unsigned char sha1[20];
723 int ref;
724 int temporary;
725 void *tree_data;
726 unsigned long tree_size;
727};
728
729static struct pbase_tree_cache *(pbase_tree_cache[256]);
730static int pbase_tree_cache_ix(const unsigned char *sha1)
731{
732 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
733}
734static int pbase_tree_cache_ix_incr(int ix)
735{
736 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
737}
738
739static struct pbase_tree {
740 struct pbase_tree *next;
741 /* This is a phony "cache" entry; we are not
742 * going to evict it nor find it through _get()
743 * mechanism -- this is for the toplevel node that
744 * would almost always change with any commit.
745 */
746 struct pbase_tree_cache pcache;
747} *pbase_tree;
748
749static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
750{
751 struct pbase_tree_cache *ent, *nent;
752 void *data;
753 unsigned long size;
21666f1a 754 enum object_type type;
5379a5c5
JH
755 int neigh;
756 int my_ix = pbase_tree_cache_ix(sha1);
757 int available_ix = -1;
758
759 /* pbase-tree-cache acts as a limited hashtable.
760 * your object will be found at your index or within a few
761 * slots after that slot if it is cached.
762 */
763 for (neigh = 0; neigh < 8; neigh++) {
764 ent = pbase_tree_cache[my_ix];
a89fccd2 765 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
766 ent->ref++;
767 return ent;
768 }
769 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
770 ((0 <= available_ix) &&
771 (!ent && pbase_tree_cache[available_ix])))
772 available_ix = my_ix;
773 if (!ent)
774 break;
775 my_ix = pbase_tree_cache_ix_incr(my_ix);
776 }
777
778 /* Did not find one. Either we got a bogus request or
779 * we need to read and perhaps cache.
780 */
21666f1a 781 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
782 if (!data)
783 return NULL;
21666f1a 784 if (type != OBJ_TREE) {
5379a5c5
JH
785 free(data);
786 return NULL;
787 }
788
789 /* We need to either cache or return a throwaway copy */
790
791 if (available_ix < 0)
792 ent = NULL;
793 else {
794 ent = pbase_tree_cache[available_ix];
795 my_ix = available_ix;
796 }
797
798 if (!ent) {
799 nent = xmalloc(sizeof(*nent));
800 nent->temporary = (available_ix < 0);
801 }
802 else {
803 /* evict and reuse */
804 free(ent->tree_data);
805 nent = ent;
806 }
e702496e 807 hashcpy(nent->sha1, sha1);
5379a5c5
JH
808 nent->tree_data = data;
809 nent->tree_size = size;
810 nent->ref = 1;
811 if (!nent->temporary)
812 pbase_tree_cache[my_ix] = nent;
813 return nent;
814}
815
816static void pbase_tree_put(struct pbase_tree_cache *cache)
817{
818 if (!cache->temporary) {
819 cache->ref--;
820 return;
821 }
822 free(cache->tree_data);
823 free(cache);
824}
825
826static int name_cmp_len(const char *name)
827{
828 int i;
829 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
830 ;
831 return i;
832}
833
834static void add_pbase_object(struct tree_desc *tree,
5379a5c5 835 const char *name,
ce0bd642
LT
836 int cmplen,
837 const char *fullname)
3f9ac8d2 838{
4c068a98 839 struct name_entry entry;
8a5a8d6c 840 int cmp;
4c068a98
LT
841
842 while (tree_entry(tree,&entry)) {
1211be6b
LT
843 if (S_ISGITLINK(entry.mode))
844 continue;
8a5a8d6c
NP
845 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
846 memcmp(name, entry.path, cmplen);
847 if (cmp > 0)
7a979d99 848 continue;
8a5a8d6c
NP
849 if (cmp < 0)
850 return;
5379a5c5 851 if (name[cmplen] != '/') {
5c49c116 852 add_object_entry(entry.sha1,
4d1012c3 853 object_type(entry.mode),
bc32fed5 854 fullname, 1);
5379a5c5
JH
855 return;
856 }
8a5a8d6c 857 if (S_ISDIR(entry.mode)) {
7a979d99 858 struct tree_desc sub;
5379a5c5
JH
859 struct pbase_tree_cache *tree;
860 const char *down = name+cmplen+1;
861 int downlen = name_cmp_len(down);
862
4c068a98 863 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
864 if (!tree)
865 return;
6fda5e51 866 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 867
ce0bd642 868 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
869 pbase_tree_put(tree);
870 }
871 }
872}
1d6b38cc 873
5379a5c5
JH
874static unsigned *done_pbase_paths;
875static int done_pbase_paths_num;
876static int done_pbase_paths_alloc;
877static int done_pbase_path_pos(unsigned hash)
878{
879 int lo = 0;
880 int hi = done_pbase_paths_num;
881 while (lo < hi) {
882 int mi = (hi + lo) / 2;
883 if (done_pbase_paths[mi] == hash)
884 return mi;
885 if (done_pbase_paths[mi] < hash)
886 hi = mi;
887 else
888 lo = mi + 1;
889 }
890 return -lo-1;
891}
892
893static int check_pbase_path(unsigned hash)
894{
895 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
896 if (0 <= pos)
897 return 1;
898 pos = -pos - 1;
899 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
900 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
901 done_pbase_paths = xrealloc(done_pbase_paths,
902 done_pbase_paths_alloc *
903 sizeof(unsigned));
904 }
905 done_pbase_paths_num++;
906 if (pos < done_pbase_paths_num)
907 memmove(done_pbase_paths + pos + 1,
908 done_pbase_paths + pos,
909 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
910 done_pbase_paths[pos] = hash;
911 return 0;
912}
913
bc32fed5 914static void add_preferred_base_object(const char *name)
5379a5c5
JH
915{
916 struct pbase_tree *it;
8a5a8d6c 917 int cmplen;
bc32fed5 918 unsigned hash = name_hash(name);
5379a5c5 919
8a5a8d6c 920 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
921 return;
922
8a5a8d6c 923 cmplen = name_cmp_len(name);
5379a5c5
JH
924 for (it = pbase_tree; it; it = it->next) {
925 if (cmplen == 0) {
bc32fed5 926 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
927 }
928 else {
929 struct tree_desc tree;
6fda5e51 930 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 931 add_pbase_object(&tree, name, cmplen, name);
7a979d99 932 }
3f9ac8d2 933 }
3f9ac8d2
JH
934}
935
7a979d99 936static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 937{
5379a5c5
JH
938 struct pbase_tree *it;
939 void *data;
940 unsigned long size;
941 unsigned char tree_sha1[20];
1d6b38cc 942
8d1d8f83
JH
943 if (window <= num_preferred_base++)
944 return;
945
5379a5c5
JH
946 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
947 if (!data)
7a979d99 948 return;
5379a5c5
JH
949
950 for (it = pbase_tree; it; it = it->next) {
a89fccd2 951 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
952 free(data);
953 return;
954 }
955 }
956
957 it = xcalloc(1, sizeof(*it));
958 it->next = pbase_tree;
959 pbase_tree = it;
960
e702496e 961 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
962 it->pcache.tree_data = data;
963 it->pcache.tree_size = size;
3f9ac8d2
JH
964}
965
0ef95f72
NP
966static void cleanup_preferred_base(void)
967{
968 struct pbase_tree *it;
969 unsigned i;
970
971 it = pbase_tree;
972 pbase_tree = NULL;
973 while (it) {
974 struct pbase_tree *this = it;
975 it = this->next;
976 free(this->pcache.tree_data);
977 free(this);
978 }
979
980 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
981 if (!pbase_tree_cache[i])
982 continue;
983 free(pbase_tree_cache[i]->tree_data);
984 free(pbase_tree_cache[i]);
985 pbase_tree_cache[i] = NULL;
986 }
987
988 free(done_pbase_paths);
989 done_pbase_paths = NULL;
990 done_pbase_paths_num = done_pbase_paths_alloc = 0;
991}
992
c323ac7d
LT
993static void check_object(struct object_entry *entry)
994{
5c49c116 995 if (entry->in_pack) {
780e6e73 996 struct packed_git *p = entry->in_pack;
03e79c88 997 struct pack_window *w_curs = NULL;
5c49c116
NP
998 const unsigned char *base_ref = NULL;
999 struct object_entry *base_entry;
1000 unsigned long used, used_0;
6777a59f 1001 unsigned int avail;
5c49c116
NP
1002 off_t ofs;
1003 unsigned char *buf, c;
780e6e73 1004
6777a59f 1005 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1006
5c49c116 1007 /*
fa736f72
NP
1008 * We want in_pack_type even if we do not reuse delta
1009 * since non-delta representations could still be reused.
ab7cd7bb 1010 */
09ded04b 1011 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1012 &entry->in_pack_type,
1013 &entry->size);
03d66015
NP
1014 if (used == 0)
1015 goto give_up;
ab7cd7bb 1016
5c49c116
NP
1017 /*
1018 * Determine if this is a delta and if so whether we can
1019 * reuse it or not. Otherwise let's find out as cheaply as
1020 * possible what the actual type and size for this object is.
3f9ac8d2 1021 */
5c49c116
NP
1022 switch (entry->in_pack_type) {
1023 default:
1024 /* Not a delta hence we've already got all we need. */
1025 entry->type = entry->in_pack_type;
1026 entry->in_pack_header_size = used;
03d66015
NP
1027 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1028 goto give_up;
5c49c116
NP
1029 unuse_pack(&w_curs);
1030 return;
1031 case OBJ_REF_DELTA:
a7de7130 1032 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1033 base_ref = use_pack(p, &w_curs,
1034 entry->in_pack_offset + used, NULL);
1035 entry->in_pack_header_size = used + 20;
1036 break;
1037 case OBJ_OFS_DELTA:
1038 buf = use_pack(p, &w_curs,
1039 entry->in_pack_offset + used, NULL);
1040 used_0 = 0;
1041 c = buf[used_0++];
1042 ofs = c & 127;
1043 while (c & 128) {
1044 ofs += 1;
03d66015
NP
1045 if (!ofs || MSB(ofs, 7)) {
1046 error("delta base offset overflow in pack for %s",
1047 sha1_to_hex(entry->idx.sha1));
1048 goto give_up;
1049 }
5c49c116
NP
1050 c = buf[used_0++];
1051 ofs = (ofs << 7) + (c & 127);
780e6e73 1052 }
5c49c116 1053 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1054 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1055 error("delta base offset out of bound for %s",
1056 sha1_to_hex(entry->idx.sha1));
1057 goto give_up;
1058 }
a7de7130 1059 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1060 struct revindex_entry *revidx;
1061 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1062 if (!revidx)
1063 goto give_up;
3449f8c4
NP
1064 base_ref = nth_packed_object_sha1(p, revidx->nr);
1065 }
5c49c116
NP
1066 entry->in_pack_header_size = used + used_0;
1067 break;
780e6e73 1068 }
780e6e73 1069
5c49c116
NP
1070 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1071 /*
1072 * If base_ref was set above that means we wish to
1073 * reuse delta data, and we even found that base
1074 * in the list of objects we want to pack. Goodie!
1075 *
1076 * Depth value does not matter - find_deltas() will
1077 * never consider reused delta as the base object to
1078 * deltify other objects against, in order to avoid
1079 * circular deltas.
3f9ac8d2 1080 */
780e6e73 1081 entry->type = entry->in_pack_type;
5c49c116 1082 entry->delta = base_entry;
64bd76b1 1083 entry->delta_size = entry->size;
15b4d577
JH
1084 entry->delta_sibling = base_entry->delta_child;
1085 base_entry->delta_child = entry;
5c49c116
NP
1086 unuse_pack(&w_curs);
1087 return;
1088 }
ab7cd7bb 1089
5c49c116
NP
1090 if (entry->type) {
1091 /*
1092 * This must be a delta and we already know what the
1093 * final object type is. Let's extract the actual
1094 * object size from the delta header.
1095 */
1096 entry->size = get_size_from_delta(p, &w_curs,
1097 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1098 if (entry->size == 0)
1099 goto give_up;
5c49c116 1100 unuse_pack(&w_curs);
3f9ac8d2
JH
1101 return;
1102 }
5c49c116
NP
1103
1104 /*
1105 * No choice but to fall back to the recursive delta walk
1106 * with sha1_object_info() to find about the object type
1107 * at this point...
1108 */
03d66015 1109 give_up:
5c49c116 1110 unuse_pack(&w_curs);
36e4d74a 1111 }
3f9ac8d2 1112
aa7e44bf 1113 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1114 /*
1115 * The error condition is checked in prepare_pack(). This is
1116 * to permit a missing preferred base object to be ignored
1117 * as a preferred base. Doing so can result in a larger
1118 * pack file, but the transfer will still take place.
1119 */
3f9ac8d2
JH
1120}
1121
5c49c116
NP
1122static int pack_offset_sort(const void *_a, const void *_b)
1123{
1124 const struct object_entry *a = *(struct object_entry **)_a;
1125 const struct object_entry *b = *(struct object_entry **)_b;
1126
1127 /* avoid filesystem trashing with loose objects */
1128 if (!a->in_pack && !b->in_pack)
aa7e44bf 1129 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1130
1131 if (a->in_pack < b->in_pack)
1132 return -1;
1133 if (a->in_pack > b->in_pack)
1134 return 1;
1135 return a->in_pack_offset < b->in_pack_offset ? -1 :
1136 (a->in_pack_offset > b->in_pack_offset);
1137}
1138
c323ac7d
LT
1139static void get_object_details(void)
1140{
7cadf491 1141 uint32_t i;
5c49c116
NP
1142 struct object_entry **sorted_by_offset;
1143
1144 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1145 for (i = 0; i < nr_objects; i++)
1146 sorted_by_offset[i] = objects + i;
1147 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1148
5c49c116
NP
1149 for (i = 0; i < nr_objects; i++)
1150 check_object(sorted_by_offset[i]);
3449f8c4 1151
5c49c116 1152 free(sorted_by_offset);
c323ac7d
LT
1153}
1154
b904166c
NP
1155/*
1156 * We search for deltas in a list sorted by type, by filename hash, and then
1157 * by size, so that we see progressively smaller and smaller files.
1158 * That's because we prefer deltas to be from the bigger file
1159 * to the smaller -- deletes are potentially cheaper, but perhaps
1160 * more importantly, the bigger file is likely the more recent
1161 * one. The deepest deltas are therefore the oldest objects which are
1162 * less susceptible to be accessed often.
1163 */
9668cf59 1164static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1165{
9668cf59
NP
1166 const struct object_entry *a = *(struct object_entry **)_a;
1167 const struct object_entry *b = *(struct object_entry **)_b;
1168
c323ac7d 1169 if (a->type > b->type)
27225f2e 1170 return -1;
b904166c 1171 if (a->type < b->type)
27225f2e 1172 return 1;
b904166c 1173 if (a->hash > b->hash)
7a979d99 1174 return -1;
b904166c 1175 if (a->hash < b->hash)
7a979d99 1176 return 1;
b904166c 1177 if (a->preferred_base > b->preferred_base)
c323ac7d 1178 return -1;
b904166c
NP
1179 if (a->preferred_base < b->preferred_base)
1180 return 1;
c323ac7d 1181 if (a->size > b->size)
b904166c
NP
1182 return -1;
1183 if (a->size < b->size)
c323ac7d 1184 return 1;
b904166c 1185 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1186}
1187
1188struct unpacked {
1189 struct object_entry *entry;
1190 void *data;
f6c7081a 1191 struct delta_index *index;
5a235b5e 1192 unsigned depth;
c323ac7d
LT
1193};
1194
d250626c
NP
1195static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1196 unsigned long delta_size)
074b2eea
MK
1197{
1198 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1199 return 0;
1200
e3dfddb3
MK
1201 if (delta_size < cache_max_small_delta_size)
1202 return 1;
1203
074b2eea
MK
1204 /* cache delta, if objects are large enough compared to delta size */
1205 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1206 return 1;
1207
1208 return 0;
1209}
1210
7eb151d6 1211#ifndef NO_PTHREADS
8ecce684 1212
44626dc7 1213static pthread_mutex_t read_mutex;
8ecce684
NP
1214#define read_lock() pthread_mutex_lock(&read_mutex)
1215#define read_unlock() pthread_mutex_unlock(&read_mutex)
1216
44626dc7 1217static pthread_mutex_t cache_mutex;
3c701839
NP
1218#define cache_lock() pthread_mutex_lock(&cache_mutex)
1219#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1220
44626dc7 1221static pthread_mutex_t progress_mutex;
8ecce684
NP
1222#define progress_lock() pthread_mutex_lock(&progress_mutex)
1223#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1224
1225#else
1226
e1ef8673
JH
1227#define read_lock() (void)0
1228#define read_unlock() (void)0
1229#define cache_lock() (void)0
1230#define cache_unlock() (void)0
1231#define progress_lock() (void)0
1232#define progress_unlock() (void)0
8ecce684
NP
1233
1234#endif
1235
f6c7081a 1236static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1237 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1238{
f6c7081a
NP
1239 struct object_entry *trg_entry = trg->entry;
1240 struct object_entry *src_entry = src->entry;
560b25a8 1241 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1242 unsigned ref_depth;
21666f1a 1243 enum object_type type;
c323ac7d
LT
1244 void *delta_buf;
1245
1246 /* Don't bother doing diffs between different types */
f6c7081a 1247 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1248 return -1;
1249
51d1e83f
LT
1250 /*
1251 * We do not bother to try a delta that we discarded
8dbbd14e 1252 * on an earlier try, but only when reusing delta data.
51d1e83f 1253 */
a7de7130 1254 if (reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1255 trg_entry->in_pack == src_entry->in_pack &&
1256 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1257 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1258 return 0;
1259
898b14ce 1260 /* Let's not bust the allowed depth. */
5a235b5e 1261 if (src->depth >= max_depth)
d116a45a 1262 return 0;
c323ac7d 1263
c3b06a69 1264 /* Now some size filtering heuristics. */
560b25a8 1265 trg_size = trg_entry->size;
c83f032e
NP
1266 if (!trg_entry->delta) {
1267 max_size = trg_size/2 - 20;
1268 ref_depth = 1;
1269 } else {
1270 max_size = trg_entry->delta_size;
5a235b5e 1271 ref_depth = trg->depth;
c83f032e 1272 }
720fe22d 1273 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1274 (max_depth - ref_depth + 1);
c3b06a69
NP
1275 if (max_size == 0)
1276 return 0;
f6c7081a 1277 src_size = src_entry->size;
560b25a8 1278 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1279 if (sizediff >= max_size)
f527cb8c 1280 return 0;
a1dab41a
BD
1281 if (trg_size < src_size / 32)
1282 return 0;
f6c7081a 1283
560b25a8
NP
1284 /* Load data if not already done */
1285 if (!trg->data) {
8ecce684 1286 read_lock();
aa7e44bf 1287 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1288 read_unlock();
2e3404c3
JH
1289 if (!trg->data)
1290 die("object %s cannot be read",
1291 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1292 if (sz != trg_size)
1293 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1294 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1295 *mem_usage += sz;
560b25a8
NP
1296 }
1297 if (!src->data) {
8ecce684 1298 read_lock();
aa7e44bf 1299 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1300 read_unlock();
71064a95
NP
1301 if (!src->data) {
1302 if (src_entry->preferred_base) {
1303 static int warned = 0;
1304 if (!warned++)
1305 warning("object %s cannot be read",
1306 sha1_to_hex(src_entry->idx.sha1));
1307 /*
1308 * Those objects are not included in the
1309 * resulting pack. Be resilient and ignore
1310 * them if they can't be read, in case the
1311 * pack could be created nevertheless.
1312 */
1313 return 0;
1314 }
2e3404c3
JH
1315 die("object %s cannot be read",
1316 sha1_to_hex(src_entry->idx.sha1));
71064a95 1317 }
560b25a8
NP
1318 if (sz != src_size)
1319 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1320 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1321 *mem_usage += sz;
560b25a8
NP
1322 }
1323 if (!src->index) {
1324 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1325 if (!src->index) {
1326 static int warned = 0;
1327 if (!warned++)
1328 warning("suboptimal pack - out of memory");
1329 return 0;
1330 }
ef0316fc 1331 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1332 }
1333
1334 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1335 if (!delta_buf)
75c42d8c 1336 return 0;
f6c7081a 1337
9e2d57a0 1338 if (trg_entry->delta) {
848d732c
BD
1339 /* Prefer only shallower same-sized deltas. */
1340 if (delta_size == trg_entry->delta_size &&
5a235b5e 1341 src->depth + 1 >= trg->depth) {
848d732c
BD
1342 free(delta_buf);
1343 return 0;
1344 }
074b2eea 1345 }
9e2d57a0 1346
3c701839
NP
1347 /*
1348 * Handle memory allocation outside of the cache
1349 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1350 * away when NO_PTHREADS is defined.
3c701839 1351 */
8e0f7003 1352 free(trg_entry->delta_data);
3c701839 1353 cache_lock();
9e2d57a0
NP
1354 if (trg_entry->delta_data) {
1355 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1356 trg_entry->delta_data = NULL;
1357 }
d250626c 1358 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1359 delta_cache_size += delta_size;
3c701839
NP
1360 cache_unlock();
1361 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1362 } else {
1363 cache_unlock();
074b2eea 1364 free(delta_buf);
3c701839
NP
1365 }
1366
b7a28f78
NP
1367 trg_entry->delta = src_entry;
1368 trg_entry->delta_size = delta_size;
1369 trg->depth = src->depth + 1;
1370
f6c7081a 1371 return 1;
c323ac7d
LT
1372}
1373
898b14ce 1374static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1375{
898b14ce
NP
1376 struct object_entry *child = me->delta_child;
1377 unsigned int m = n;
1378 while (child) {
1379 unsigned int c = check_delta_limit(child, n + 1);
1380 if (m < c)
1381 m = c;
1382 child = child->delta_sibling;
1383 }
1384 return m;
b2504a0d
NP
1385}
1386
75ad235c 1387static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1388{
ef0316fc 1389 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1390 free_delta_index(n->index);
1391 n->index = NULL;
1392 if (n->data) {
ef0316fc 1393 freed_mem += n->entry->size;
a97773ce
BD
1394 free(n->data);
1395 n->data = NULL;
a97773ce
BD
1396 }
1397 n->entry = NULL;
7d7baa5e 1398 n->depth = 0;
ef0316fc 1399 return freed_mem;
a97773ce
BD
1400}
1401
384b32c0 1402static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1403 int window, int depth, unsigned *processed)
c323ac7d 1404{
384b32c0 1405 uint32_t i, idx = 0, count = 0;
7cadf491 1406 struct unpacked *array;
ef0316fc 1407 unsigned long mem_usage = 0;
c323ac7d 1408
19d4b416 1409 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1410
384b32c0 1411 for (;;) {
421b488a 1412 struct object_entry *entry;
c323ac7d 1413 struct unpacked *n = array + idx;
ef0316fc 1414 int j, max_depth, best_base = -1;
c323ac7d 1415
384b32c0
NP
1416 progress_lock();
1417 if (!*list_size) {
1418 progress_unlock();
1419 break;
1420 }
421b488a 1421 entry = *list++;
384b32c0
NP
1422 (*list_size)--;
1423 if (!entry->preferred_base) {
1424 (*processed)++;
1425 display_progress(progress_state, *processed);
1426 }
1427 progress_unlock();
1428
ef0316fc 1429 mem_usage -= free_unpacked(n);
c323ac7d 1430 n->entry = entry;
ab7cd7bb 1431
a97773ce 1432 while (window_memory_limit &&
ef0316fc 1433 mem_usage > window_memory_limit &&
a97773ce
BD
1434 count > 1) {
1435 uint32_t tail = (idx + window - count) % window;
75ad235c 1436 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1437 count--;
1438 }
1439
75d39853
NP
1440 /* We do not compute delta to *create* objects we are not
1441 * going to pack.
1442 */
1443 if (entry->preferred_base)
1444 goto next;
1445
898b14ce
NP
1446 /*
1447 * If the current object is at pack edge, take the depth the
1448 * objects that depend on the current object into account
1449 * otherwise they would become too deep.
1450 */
1451 max_depth = depth;
1452 if (entry->delta_child) {
1453 max_depth -= check_delta_limit(entry, 0);
1454 if (max_depth <= 0)
1455 goto next;
1456 }
1457
78817c15
LT
1458 j = window;
1459 while (--j > 0) {
77639870 1460 int ret;
7cadf491 1461 uint32_t other_idx = idx + j;
c323ac7d 1462 struct unpacked *m;
78817c15
LT
1463 if (other_idx >= window)
1464 other_idx -= window;
c323ac7d
LT
1465 m = array + other_idx;
1466 if (!m->entry)
1467 break;
ef0316fc 1468 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1469 if (ret < 0)
c323ac7d 1470 break;
77639870
JH
1471 else if (ret > 0)
1472 best_base = other_idx;
c323ac7d 1473 }
898b14ce 1474
ed4a9031
NP
1475 /*
1476 * If we decided to cache the delta data, then it is best
1477 * to compress it right away. First because we have to do
1478 * it anyway, and doing it here while we're threaded will
1479 * save a lot of time in the non threaded write phase,
1480 * as well as allow for caching more deltas within
1481 * the same cache size limit.
1482 * ...
1483 * But only if not writing to stdout, since in that case
1484 * the network is most likely throttling writes anyway,
1485 * and therefore it is best to go to the write phase ASAP
1486 * instead, as we can afford spending more time compressing
1487 * between writes at that moment.
1488 */
1489 if (entry->delta_data && !pack_to_stdout) {
1490 entry->z_delta_size = do_compress(&entry->delta_data,
1491 entry->delta_size);
1492 cache_lock();
1493 delta_cache_size -= entry->delta_size;
1494 delta_cache_size += entry->z_delta_size;
1495 cache_unlock();
1496 }
1497
70ca1a3f
JH
1498 /* if we made n a delta, and if n is already at max
1499 * depth, leaving it in the window is pointless. we
1500 * should evict it first.
70ca1a3f 1501 */
70baf5d4 1502 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1503 continue;
ff45715c 1504
77639870
JH
1505 /*
1506 * Move the best delta base up in the window, after the
1507 * currently deltified object, to keep it longer. It will
1508 * be the first base object to be attempted next.
1509 */
1510 if (entry->delta) {
1511 struct unpacked swap = array[best_base];
1512 int dist = (window + idx - best_base) % window;
1513 int dst = best_base;
1514 while (dist--) {
1515 int src = (dst + 1) % window;
1516 array[dst] = array[src];
1517 dst = src;
1518 }
1519 array[dst] = swap;
1520 }
1521
898b14ce 1522 next:
521a4f4c 1523 idx++;
a97773ce
BD
1524 if (count + 1 < window)
1525 count++;
521a4f4c
LT
1526 if (idx >= window)
1527 idx = 0;
384b32c0 1528 }
adee7bdf 1529
f6c7081a 1530 for (i = 0; i < window; ++i) {
ff45715c 1531 free_delta_index(array[i].index);
adee7bdf 1532 free(array[i].data);
f6c7081a 1533 }
adee7bdf 1534 free(array);
c323ac7d
LT
1535}
1536
7eb151d6 1537#ifndef NO_PTHREADS
8ecce684 1538
a9a74636
NP
1539static void try_to_free_from_threads(size_t size)
1540{
1541 read_lock();
1542 release_pack_memory(size, -1);
1543 read_unlock();
1544}
1545
851c34b0
JS
1546try_to_free_t old_try_to_free_routine;
1547
50f22ada
JS
1548/*
1549 * The main thread waits on the condition that (at least) one of the workers
1550 * has stopped working (which is indicated in the .working member of
1551 * struct thread_params).
1552 * When a work thread has completed its work, it sets .working to 0 and
1553 * signals the main thread and waits on the condition that .data_ready
1554 * becomes 1.
1555 */
1556
8ecce684
NP
1557struct thread_params {
1558 pthread_t thread;
1559 struct object_entry **list;
1560 unsigned list_size;
384b32c0 1561 unsigned remaining;
8ecce684
NP
1562 int window;
1563 int depth;
50f22ada
JS
1564 int working;
1565 int data_ready;
1566 pthread_mutex_t mutex;
1567 pthread_cond_t cond;
8ecce684
NP
1568 unsigned *processed;
1569};
1570
44626dc7
AH
1571static pthread_cond_t progress_cond;
1572
1573/*
1574 * Mutex and conditional variable can't be statically-initialized on Windows.
1575 */
1576static void init_threaded_search(void)
1577{
93749194 1578 init_recursive_mutex(&read_mutex);
44626dc7
AH
1579 pthread_mutex_init(&cache_mutex, NULL);
1580 pthread_mutex_init(&progress_mutex, NULL);
1581 pthread_cond_init(&progress_cond, NULL);
851c34b0 1582 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
1583}
1584
1585static void cleanup_threaded_search(void)
1586{
851c34b0 1587 set_try_to_free_routine(old_try_to_free_routine);
44626dc7
AH
1588 pthread_cond_destroy(&progress_cond);
1589 pthread_mutex_destroy(&read_mutex);
1590 pthread_mutex_destroy(&cache_mutex);
1591 pthread_mutex_destroy(&progress_mutex);
1592}
c2a33679 1593
8ecce684
NP
1594static void *threaded_find_deltas(void *arg)
1595{
c2a33679
NP
1596 struct thread_params *me = arg;
1597
50f22ada 1598 while (me->remaining) {
384b32c0 1599 find_deltas(me->list, &me->remaining,
c2a33679 1600 me->window, me->depth, me->processed);
50f22ada
JS
1601
1602 progress_lock();
1603 me->working = 0;
1604 pthread_cond_signal(&progress_cond);
1605 progress_unlock();
1606
1607 /*
1608 * We must not set ->data_ready before we wait on the
1609 * condition because the main thread may have set it to 1
1610 * before we get here. In order to be sure that new
1611 * work is available if we see 1 in ->data_ready, it
1612 * was initialized to 0 before this thread was spawned
1613 * and we reset it to 0 right away.
1614 */
1615 pthread_mutex_lock(&me->mutex);
1616 while (!me->data_ready)
1617 pthread_cond_wait(&me->cond, &me->mutex);
1618 me->data_ready = 0;
1619 pthread_mutex_unlock(&me->mutex);
c2a33679 1620 }
50f22ada
JS
1621 /* leave ->working 1 so that this doesn't get more work assigned */
1622 return NULL;
8ecce684
NP
1623}
1624
8ecce684
NP
1625static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1626 int window, int depth, unsigned *processed)
1627{
dcda3614 1628 struct thread_params *p;
384b32c0 1629 int i, ret, active_threads = 0;
c2a33679 1630
44626dc7
AH
1631 init_threaded_search();
1632
ef055571
NP
1633 if (!delta_search_threads) /* --threads=0 means autodetect */
1634 delta_search_threads = online_cpus();
367f4a43 1635 if (delta_search_threads <= 1) {
384b32c0 1636 find_deltas(list, &list_size, window, depth, processed);
44626dc7 1637 cleanup_threaded_search();
367f4a43
NP
1638 return;
1639 }
43cc2b42 1640 if (progress > pack_to_stdout)
b6c29915 1641 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1642 delta_search_threads);
dcda3614 1643 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 1644
50f22ada 1645 /* Partition the work amongst work threads. */
367f4a43 1646 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1647 unsigned sub_size = list_size / (delta_search_threads - i);
1648
bf874896
NP
1649 /* don't use too small segments or no deltas will be found */
1650 if (sub_size < 2*window && i+1 < delta_search_threads)
1651 sub_size = 0;
1652
8ecce684
NP
1653 p[i].window = window;
1654 p[i].depth = depth;
1655 p[i].processed = processed;
50f22ada
JS
1656 p[i].working = 1;
1657 p[i].data_ready = 0;
c2a33679 1658
59921b4b 1659 /* try to split chunks on "path" boundaries */
6fc74703
NP
1660 while (sub_size && sub_size < list_size &&
1661 list[sub_size]->hash &&
384b32c0
NP
1662 list[sub_size]->hash == list[sub_size-1]->hash)
1663 sub_size++;
1664
50f22ada
JS
1665 p[i].list = list;
1666 p[i].list_size = sub_size;
1667 p[i].remaining = sub_size;
59921b4b 1668
384b32c0
NP
1669 list += sub_size;
1670 list_size -= sub_size;
1671 }
1672
50f22ada
JS
1673 /* Start work threads. */
1674 for (i = 0; i < delta_search_threads; i++) {
1675 if (!p[i].list_size)
1676 continue;
68e6a4f8
JS
1677 pthread_mutex_init(&p[i].mutex, NULL);
1678 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1679 ret = pthread_create(&p[i].thread, NULL,
1680 threaded_find_deltas, &p[i]);
1681 if (ret)
1682 die("unable to create thread: %s", strerror(ret));
1683 active_threads++;
1684 }
1685
384b32c0
NP
1686 /*
1687 * Now let's wait for work completion. Each time a thread is done
1688 * with its work, we steal half of the remaining work from the
1689 * thread with the largest number of unprocessed objects and give
1690 * it to that newly idle thread. This ensure good load balancing
1691 * until the remaining object list segments are simply too short
1692 * to be worth splitting anymore.
1693 */
50f22ada
JS
1694 while (active_threads) {
1695 struct thread_params *target = NULL;
384b32c0
NP
1696 struct thread_params *victim = NULL;
1697 unsigned sub_size = 0;
384b32c0
NP
1698
1699 progress_lock();
50f22ada
JS
1700 for (;;) {
1701 for (i = 0; !target && i < delta_search_threads; i++)
1702 if (!p[i].working)
1703 target = &p[i];
1704 if (target)
1705 break;
1706 pthread_cond_wait(&progress_cond, &progress_mutex);
1707 }
1708
384b32c0
NP
1709 for (i = 0; i < delta_search_threads; i++)
1710 if (p[i].remaining > 2*window &&
1711 (!victim || victim->remaining < p[i].remaining))
1712 victim = &p[i];
1713 if (victim) {
1714 sub_size = victim->remaining / 2;
1715 list = victim->list + victim->list_size - sub_size;
1716 while (sub_size && list[0]->hash &&
1717 list[0]->hash == list[-1]->hash) {
1718 list++;
1719 sub_size--;
1720 }
eb9688ff
NP
1721 if (!sub_size) {
1722 /*
1723 * It is possible for some "paths" to have
1724 * so many objects that no hash boundary
1725 * might be found. Let's just steal the
1726 * exact half in that case.
1727 */
1728 sub_size = victim->remaining / 2;
1729 list -= sub_size;
1730 }
384b32c0
NP
1731 target->list = list;
1732 victim->list_size -= sub_size;
1733 victim->remaining -= sub_size;
1734 }
384b32c0
NP
1735 target->list_size = sub_size;
1736 target->remaining = sub_size;
50f22ada
JS
1737 target->working = 1;
1738 progress_unlock();
1739
1740 pthread_mutex_lock(&target->mutex);
1741 target->data_ready = 1;
1742 pthread_cond_signal(&target->cond);
1743 pthread_mutex_unlock(&target->mutex);
c2a33679 1744
384b32c0 1745 if (!sub_size) {
b81d9af7 1746 pthread_join(target->thread, NULL);
50f22ada
JS
1747 pthread_cond_destroy(&target->cond);
1748 pthread_mutex_destroy(&target->mutex);
384b32c0 1749 active_threads--;
c2a33679 1750 }
50f22ada 1751 }
44626dc7 1752 cleanup_threaded_search();
dcda3614 1753 free(p);
8ecce684
NP
1754}
1755
1756#else
384b32c0 1757#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
1758#endif
1759
f0a24aa5
SP
1760static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1761{
1762 unsigned char peeled[20];
1763
1764 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1765 !peel_ref(path, peeled) && /* peelable? */
1766 !is_null_sha1(peeled) && /* annotated tag? */
1767 locate_object_entry(peeled)) /* object packed? */
1768 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1769 return 0;
1770}
1771
f3123c4a
JH
1772static void prepare_pack(int window, int depth)
1773{
9668cf59 1774 struct object_entry **delta_list;
6e1c2344
RJ
1775 uint32_t i, nr_deltas;
1776 unsigned n;
9668cf59 1777
3f9ac8d2 1778 get_object_details();
9668cf59 1779
0e8189e2
NP
1780 /*
1781 * If we're locally repacking then we need to be doubly careful
1782 * from now on in order to make sure no stealth corruption gets
1783 * propagated to the new pack. Clients receiving streamed packs
1784 * should validate everything they get anyway so no need to incur
1785 * the additional cost here in that case.
1786 */
1787 if (!pack_to_stdout)
1788 do_check_packed_object_crc = 1;
1789
75d39853 1790 if (!nr_objects || !window || !depth)
9668cf59
NP
1791 return;
1792
1793 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1794 nr_deltas = n = 0;
1795
1796 for (i = 0; i < nr_objects; i++) {
1797 struct object_entry *entry = objects + i;
1798
1799 if (entry->delta)
1800 /* This happens if we decided to reuse existing
a7de7130 1801 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
1802 */
1803 continue;
1804
1805 if (entry->size < 50)
1806 continue;
1807
1808 if (entry->no_try_delta)
1809 continue;
1810
6d6f9cdd 1811 if (!entry->preferred_base) {
75d39853 1812 nr_deltas++;
6d6f9cdd
SP
1813 if (entry->type < 0)
1814 die("unable to get type of object %s",
1815 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
1816 } else {
1817 if (entry->type < 0) {
1818 /*
1819 * This object is not found, but we
1820 * don't have to include it anyway.
1821 */
1822 continue;
1823 }
6d6f9cdd 1824 }
75d39853
NP
1825
1826 delta_list[n++] = entry;
1827 }
1828
2f8b8947 1829 if (nr_deltas && n > 1) {
e334977d
NP
1830 unsigned nr_done = 0;
1831 if (progress)
dc6a0757
NP
1832 progress_state = start_progress("Compressing objects",
1833 nr_deltas);
75d39853 1834 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1835 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1836 stop_progress(&progress_state);
e334977d
NP
1837 if (nr_done != nr_deltas)
1838 die("inconsistency with delta count");
75d39853 1839 }
9668cf59 1840 free(delta_list);
f3123c4a
JH
1841}
1842
ef90d6d4 1843static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 1844{
eeefa7c9 1845 if (!strcmp(k, "pack.window")) {
4812a93a
JK
1846 window = git_config_int(k, v);
1847 return 0;
1848 }
a97773ce
BD
1849 if (!strcmp(k, "pack.windowmemory")) {
1850 window_memory_limit = git_config_ulong(k, v);
1851 return 0;
1852 }
1853 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1854 depth = git_config_int(k, v);
1855 return 0;
1856 }
960ccca6
DH
1857 if (!strcmp(k, "pack.compression")) {
1858 int level = git_config_int(k, v);
1859 if (level == -1)
1860 level = Z_DEFAULT_COMPRESSION;
1861 else if (level < 0 || level > Z_BEST_COMPRESSION)
1862 die("bad pack compression level %d", level);
1863 pack_compression_level = level;
1864 pack_compression_seen = 1;
1865 return 0;
1866 }
074b2eea
MK
1867 if (!strcmp(k, "pack.deltacachesize")) {
1868 max_delta_cache_size = git_config_int(k, v);
1869 return 0;
1870 }
e3dfddb3
MK
1871 if (!strcmp(k, "pack.deltacachelimit")) {
1872 cache_max_small_delta_size = git_config_int(k, v);
1873 return 0;
1874 }
693b86ff
NP
1875 if (!strcmp(k, "pack.threads")) {
1876 delta_search_threads = git_config_int(k, v);
833e3df1 1877 if (delta_search_threads < 0)
693b86ff
NP
1878 die("invalid number of threads specified (%d)",
1879 delta_search_threads);
7eb151d6 1880#ifdef NO_PTHREADS
833e3df1 1881 if (delta_search_threads != 1)
693b86ff
NP
1882 warning("no threads support, ignoring %s", k);
1883#endif
1884 return 0;
1885 }
4d00bda2
NP
1886 if (!strcmp(k, "pack.indexversion")) {
1887 pack_idx_default_version = git_config_int(k, v);
1888 if (pack_idx_default_version > 2)
6e1c2344
RJ
1889 die("bad pack.indexversion=%"PRIu32,
1890 pack_idx_default_version);
4d00bda2
NP
1891 return 0;
1892 }
2b84b5a8
JS
1893 if (!strcmp(k, "pack.packsizelimit")) {
1894 pack_size_limit_cfg = git_config_ulong(k, v);
1895 return 0;
1896 }
ef90d6d4 1897 return git_default_config(k, v, cb);
4812a93a
JK
1898}
1899
b5d97e6b 1900static void read_object_list_from_stdin(void)
c323ac7d 1901{
b5d97e6b
JH
1902 char line[40 + 1 + PATH_MAX + 2];
1903 unsigned char sha1[20];
b2504a0d 1904
da93d12b 1905 for (;;) {
da93d12b
LT
1906 if (!fgets(line, sizeof(line), stdin)) {
1907 if (feof(stdin))
1908 break;
1909 if (!ferror(stdin))
1910 die("fgets returned NULL, not EOF, not error!");
687dd75c 1911 if (errno != EINTR)
d824cbba 1912 die_errno("fgets");
687dd75c
JH
1913 clearerr(stdin);
1914 continue;
da93d12b 1915 }
7a979d99
JH
1916 if (line[0] == '-') {
1917 if (get_sha1_hex(line+1, sha1))
1918 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1919 line);
8d1d8f83 1920 add_preferred_base(sha1);
7a979d99 1921 continue;
21fcd1bd 1922 }
c323ac7d 1923 if (get_sha1_hex(line, sha1))
ef07618f 1924 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1925
bc32fed5
JH
1926 add_preferred_base_object(line+41);
1927 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1928 }
b5d97e6b
JH
1929}
1930
08cdfb13
JH
1931#define OBJECT_ADDED (1u<<20)
1932
11c211fa 1933static void show_commit(struct commit *commit, void *data)
b5d97e6b 1934{
bc32fed5 1935 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1936 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1937}
1938
cf2ab916 1939static void show_object(struct object *obj, const struct name_path *path, const char *last)
b5d97e6b 1940{
cf2ab916
LT
1941 char *name = path_name(path, last);
1942
8d2dfc49
LT
1943 add_preferred_base_object(name);
1944 add_object_entry(obj->sha1, obj->type, name, 0);
1945 obj->flags |= OBJECT_ADDED;
1946
1947 /*
1948 * We will have generated the hash from the name,
1949 * but not saved a pointer to it - we can free it
1950 */
1951 free((char *)name);
b5d97e6b
JH
1952}
1953
8d1d8f83
JH
1954static void show_edge(struct commit *commit)
1955{
1956 add_preferred_base(commit->object.sha1);
1957}
1958
08cdfb13
JH
1959struct in_pack_object {
1960 off_t offset;
1961 struct object *object;
1962};
1963
1964struct in_pack {
1965 int alloc;
1966 int nr;
1967 struct in_pack_object *array;
1968};
1969
1970static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1971{
1972 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1973 in_pack->array[in_pack->nr].object = object;
1974 in_pack->nr++;
1975}
1976
1977/*
1978 * Compare the objects in the offset order, in order to emulate the
f18d244a 1979 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
1980 */
1981static int ofscmp(const void *a_, const void *b_)
1982{
1983 struct in_pack_object *a = (struct in_pack_object *)a_;
1984 struct in_pack_object *b = (struct in_pack_object *)b_;
1985
1986 if (a->offset < b->offset)
1987 return -1;
1988 else if (a->offset > b->offset)
1989 return 1;
1990 else
1991 return hashcmp(a->object->sha1, b->object->sha1);
1992}
1993
1994static void add_objects_in_unpacked_packs(struct rev_info *revs)
1995{
1996 struct packed_git *p;
1997 struct in_pack in_pack;
1998 uint32_t i;
1999
2000 memset(&in_pack, 0, sizeof(in_pack));
2001
2002 for (p = packed_git; p; p = p->next) {
2003 const unsigned char *sha1;
2004 struct object *o;
2005
79bc4c71 2006 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2007 continue;
2008 if (open_pack_index(p))
2009 die("cannot open pack index");
2010
2011 ALLOC_GROW(in_pack.array,
2012 in_pack.nr + p->num_objects,
2013 in_pack.alloc);
2014
2015 for (i = 0; i < p->num_objects; i++) {
2016 sha1 = nth_packed_object_sha1(p, i);
2017 o = lookup_unknown_object(sha1);
2018 if (!(o->flags & OBJECT_ADDED))
2019 mark_in_pack_object(o, p, &in_pack);
2020 o->flags |= OBJECT_ADDED;
2021 }
2022 }
2023
2024 if (in_pack.nr) {
2025 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2026 ofscmp);
2027 for (i = 0; i < in_pack.nr; i++) {
2028 struct object *o = in_pack.array[i].object;
2029 add_object_entry(o->sha1, o->type, "", 0);
2030 }
2031 }
2032 free(in_pack.array);
2033}
2034
094085e3
BC
2035static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2036{
2037 static struct packed_git *last_found = (void *)1;
2038 struct packed_git *p;
2039
2040 p = (last_found != (void *)1) ? last_found : packed_git;
2041
2042 while (p) {
2043 if ((!p->pack_local || p->pack_keep) &&
2044 find_pack_entry_one(sha1, p)) {
2045 last_found = p;
2046 return 1;
2047 }
2048 if (p == last_found)
2049 p = packed_git;
2050 else
2051 p = p->next;
2052 if (p == last_found)
2053 p = p->next;
2054 }
2055 return 0;
2056}
2057
ca11b212
NP
2058static void loosen_unused_packed_objects(struct rev_info *revs)
2059{
2060 struct packed_git *p;
2061 uint32_t i;
2062 const unsigned char *sha1;
2063
2064 for (p = packed_git; p; p = p->next) {
79bc4c71 2065 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2066 continue;
2067
2068 if (open_pack_index(p))
2069 die("cannot open pack index");
2070
2071 for (i = 0; i < p->num_objects; i++) {
2072 sha1 = nth_packed_object_sha1(p, i);
094085e3
BC
2073 if (!locate_object_entry(sha1) &&
2074 !has_sha1_pack_kept_or_nonlocal(sha1))
ca11b212
NP
2075 if (force_object_loose(sha1, p->mtime))
2076 die("unable to force loose object");
2077 }
2078 }
2079}
2080
8d1d8f83 2081static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2082{
2083 struct rev_info revs;
2084 char line[1000];
b5d97e6b
JH
2085 int flags = 0;
2086
b5d97e6b
JH
2087 init_revisions(&revs, NULL);
2088 save_commit_buffer = 0;
b5d97e6b
JH
2089 setup_revisions(ac, av, &revs, NULL);
2090
b5d97e6b
JH
2091 while (fgets(line, sizeof(line), stdin) != NULL) {
2092 int len = strlen(line);
872c930d 2093 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2094 line[--len] = 0;
2095 if (!len)
2096 break;
2097 if (*line == '-') {
2098 if (!strcmp(line, "--not")) {
2099 flags ^= UNINTERESTING;
2100 continue;
2101 }
2102 die("not a rev '%s'", line);
2103 }
2104 if (handle_revision_arg(line, &revs, flags, 1))
2105 die("bad revision '%s'", line);
2106 }
2107
3d51e1b5
MK
2108 if (prepare_revision_walk(&revs))
2109 die("revision walk setup failed");
8d1d8f83 2110 mark_edges_uninteresting(revs.commits, &revs, show_edge);
11c211fa 2111 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13
JH
2112
2113 if (keep_unreachable)
2114 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2115 if (unpack_unreachable)
2116 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2117}
2118
2119int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2120{
b5d97e6b 2121 int use_internal_rev_list = 0;
8d1d8f83 2122 int thin = 0;
4f366275 2123 int all_progress_implied = 0;
7cadf491 2124 uint32_t i;
ffa84ffb
RD
2125 const char **rp_av;
2126 int rp_ac_alloc = 64;
8d1d8f83
JH
2127 int rp_ac;
2128
dae556bd
CC
2129 read_replace_refs = 0;
2130
ffa84ffb
RD
2131 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2132
8d1d8f83
JH
2133 rp_av[0] = "pack-objects";
2134 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2135 rp_ac = 2;
b5d97e6b 2136
ef90d6d4 2137 git_config(git_pack_config, NULL);
960ccca6
DH
2138 if (!pack_compression_seen && core_compression_seen)
2139 pack_compression_level = core_compression_level;
b5d97e6b
JH
2140
2141 progress = isatty(2);
2142 for (i = 1; i < argc; i++) {
2143 const char *arg = argv[i];
2144
2145 if (*arg != '-')
2146 break;
2147
2148 if (!strcmp("--non-empty", arg)) {
2149 non_empty = 1;
2150 continue;
2151 }
2152 if (!strcmp("--local", arg)) {
2153 local = 1;
2154 continue;
2155 }
b5d97e6b
JH
2156 if (!strcmp("--incremental", arg)) {
2157 incremental = 1;
2158 continue;
2159 }
e96fb9b8
BC
2160 if (!strcmp("--honor-pack-keep", arg)) {
2161 ignore_packed_keep = 1;
2162 continue;
2163 }
960ccca6
DH
2164 if (!prefixcmp(arg, "--compression=")) {
2165 char *end;
2166 int level = strtoul(arg+14, &end, 0);
2167 if (!arg[14] || *end)
2168 usage(pack_usage);
2169 if (level == -1)
2170 level = Z_DEFAULT_COMPRESSION;
2171 else if (level < 0 || level > Z_BEST_COMPRESSION)
2172 die("bad pack compression level %d", level);
2173 pack_compression_level = level;
2174 continue;
2175 }
6b94b1a0 2176 if (!prefixcmp(arg, "--max-pack-size=")) {
2b84b5a8 2177 pack_size_limit_cfg = 0;
07cf0f24 2178 if (!git_parse_ulong(arg+16, &pack_size_limit))
6b94b1a0
DH
2179 usage(pack_usage);
2180 continue;
2181 }
599065a3 2182 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2183 char *end;
2184 window = strtoul(arg+9, &end, 0);
2185 if (!arg[9] || *end)
2186 usage(pack_usage);
2187 continue;
2188 }
a97773ce
BD
2189 if (!prefixcmp(arg, "--window-memory=")) {
2190 if (!git_parse_ulong(arg+16, &window_memory_limit))
2191 usage(pack_usage);
2192 continue;
2193 }
367f4a43
NP
2194 if (!prefixcmp(arg, "--threads=")) {
2195 char *end;
2196 delta_search_threads = strtoul(arg+10, &end, 0);
833e3df1 2197 if (!arg[10] || *end || delta_search_threads < 0)
367f4a43 2198 usage(pack_usage);
7eb151d6 2199#ifdef NO_PTHREADS
833e3df1 2200 if (delta_search_threads != 1)
367f4a43
NP
2201 warning("no threads support, "
2202 "ignoring %s", arg);
2203#endif
2204 continue;
2205 }
599065a3 2206 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2207 char *end;
2208 depth = strtoul(arg+8, &end, 0);
2209 if (!arg[8] || *end)
2210 usage(pack_usage);
2211 continue;
2212 }
2213 if (!strcmp("--progress", arg)) {
2214 progress = 1;
2215 continue;
2216 }
231f240b
NP
2217 if (!strcmp("--all-progress", arg)) {
2218 progress = 2;
2219 continue;
2220 }
4f366275
NP
2221 if (!strcmp("--all-progress-implied", arg)) {
2222 all_progress_implied = 1;
2223 continue;
2224 }
b5d97e6b
JH
2225 if (!strcmp("-q", arg)) {
2226 progress = 0;
2227 continue;
2228 }
2229 if (!strcmp("--no-reuse-delta", arg)) {
a7de7130 2230 reuse_delta = 0;
b5d97e6b
JH
2231 continue;
2232 }
fa736f72 2233 if (!strcmp("--no-reuse-object", arg)) {
a7de7130 2234 reuse_object = reuse_delta = 0;
fa736f72
NP
2235 continue;
2236 }
be6b1914 2237 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2238 allow_ofs_delta = 1;
be6b1914
NP
2239 continue;
2240 }
b5d97e6b
JH
2241 if (!strcmp("--stdout", arg)) {
2242 pack_to_stdout = 1;
2243 continue;
2244 }
2245 if (!strcmp("--revs", arg)) {
2246 use_internal_rev_list = 1;
2247 continue;
2248 }
08cdfb13
JH
2249 if (!strcmp("--keep-unreachable", arg)) {
2250 keep_unreachable = 1;
2251 continue;
2252 }
ca11b212
NP
2253 if (!strcmp("--unpack-unreachable", arg)) {
2254 unpack_unreachable = 1;
2255 continue;
2256 }
f0a24aa5
SP
2257 if (!strcmp("--include-tag", arg)) {
2258 include_tag = 1;
2259 continue;
2260 }
8d1d8f83 2261 if (!strcmp("--unpacked", arg) ||
63049292 2262 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2263 !strcmp("--all", arg)) {
2264 use_internal_rev_list = 1;
ffa84ffb
RD
2265 if (rp_ac >= rp_ac_alloc - 1) {
2266 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2267 rp_av = xrealloc(rp_av,
2268 rp_ac_alloc * sizeof(*rp_av));
2269 }
8d1d8f83 2270 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2271 continue;
2272 }
8d1d8f83
JH
2273 if (!strcmp("--thin", arg)) {
2274 use_internal_rev_list = 1;
2275 thin = 1;
2276 rp_av[1] = "--objects-edge";
b5d97e6b 2277 continue;
4ba7d711
NP
2278 }
2279 if (!prefixcmp(arg, "--index-version=")) {
2280 char *c;
aa7e44bf
GB
2281 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2282 if (pack_idx_default_version > 2)
4ba7d711
NP
2283 die("bad %s", arg);
2284 if (*c == ',')
aa7e44bf
GB
2285 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2286 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2287 die("bad %s", arg);
2288 continue;
b5d97e6b 2289 }
7f3140cd
JS
2290 if (!strcmp(arg, "--keep-true-parents")) {
2291 grafts_replace_parents = 0;
2292 continue;
2293 }
b5d97e6b
JH
2294 usage(pack_usage);
2295 }
2296
2297 /* Traditionally "pack-objects [options] base extra" failed;
2298 * we would however want to take refs parameter that would
2299 * have been given to upstream rev-list ourselves, which means
2300 * we somehow want to say what the base name is. So the
2301 * syntax would be:
2302 *
2303 * pack-objects [options] base <refs...>
2304 *
2305 * in other words, we would treat the first non-option as the
2306 * base_name and send everything else to the internal revision
2307 * walker.
2308 */
2309
2310 if (!pack_to_stdout)
2311 base_name = argv[i++];
2312
2313 if (pack_to_stdout != !base_name)
2314 usage(pack_usage);
2315
2b84b5a8
JS
2316 if (!pack_to_stdout && !pack_size_limit)
2317 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
2318 if (pack_to_stdout && pack_size_limit)
2319 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
2320 if (pack_size_limit && pack_size_limit < 1024*1024) {
2321 warning("minimum pack size limit is 1 MiB");
2322 pack_size_limit = 1024*1024;
2323 }
01c12a23 2324
8d1d8f83
JH
2325 if (!pack_to_stdout && thin)
2326 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2327
ca11b212
NP
2328 if (keep_unreachable && unpack_unreachable)
2329 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2330
4f366275
NP
2331 if (progress && all_progress_implied)
2332 progress = 2;
2333
b5d97e6b
JH
2334 prepare_packed_git();
2335
13aaf148 2336 if (progress)
dc6a0757 2337 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2338 if (!use_internal_rev_list)
2339 read_object_list_from_stdin();
8d1d8f83
JH
2340 else {
2341 rp_av[rp_ac] = NULL;
2342 get_object_list(rp_ac, rp_av);
2343 }
0ef95f72 2344 cleanup_preferred_base();
f0a24aa5
SP
2345 if (include_tag && nr_result)
2346 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2347 stop_progress(&progress_state);
96a02f8f 2348
f0b0af1b 2349 if (non_empty && !nr_result)
1c4a2912 2350 return 0;
f7ae6a93
NP
2351 if (nr_result)
2352 prepare_pack(window, depth);
d01fb92f 2353 write_pack_file();
ab7cd7bb 2354 if (progress)
6e1c2344
RJ
2355 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2356 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2357 written, written_delta, reused, reused_delta);
c323ac7d
LT
2358 return 0;
2359}