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