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