i18n: notes: mark parseopt strings for translation
[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"
cf2ba13a 19#include "streaming.h"
93749194 20#include "thread-utils.h"
8ecce684 21
99fb6e04
NTND
22static const char *pack_usage[] = {
23 "git pack-objects --stdout [options...] [< ref-list | < object-list]",
24 "git pack-objects [options...] base-name [< ref-list | < object-list]",
25 NULL
26};
c323ac7d 27
c323ac7d 28struct object_entry {
aa7e44bf 29 struct pack_idx_entry idx;
3f9ac8d2 30 unsigned long size; /* uncompressed size */
3f9ac8d2 31 struct packed_git *in_pack; /* already in pack */
6777a59f 32 off_t in_pack_offset;
a3fbf4df 33 struct object_entry *delta; /* delta base object */
82e5a82f 34 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
35 struct object_entry *delta_sibling; /* other deltified objects who
36 * uses the same base as me
37 */
074b2eea 38 void *delta_data; /* cached delta (uncompressed) */
a3fbf4df 39 unsigned long delta_size; /* delta data size (uncompressed) */
ed4a9031 40 unsigned long z_delta_size; /* delta data size (compressed) */
5a235b5e 41 unsigned int hash; /* name hint hash */
a3fbf4df
NP
42 enum object_type type;
43 enum object_type in_pack_type; /* could be delta */
44 unsigned char in_pack_header_size;
45 unsigned char preferred_base; /* we do not pack this, but is available
a74db82e 46 * to be used as the base object to delta
a3fbf4df
NP
47 * objects against.
48 */
a74db82e 49 unsigned char no_try_delta;
1b4bb16b
JH
50 unsigned char tagged; /* near the very tip of refs */
51 unsigned char filled; /* assigned write-order */
c323ac7d
LT
52};
53
3f9ac8d2 54/*
82e5a82f 55 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
56 * expanded). nr_objects & nr_alloc controls this array. They are stored
57 * in the order we see -- typically rev-list --objects order that gives us
58 * nice "minimum seek" order.
3f9ac8d2 59 */
e4d58311 60static struct object_entry *objects;
79814f42 61static struct pack_idx_entry **written_list;
d01fb92f 62static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
3f9ac8d2 63
96f1e58f 64static int non_empty;
a7de7130 65static int reuse_delta = 1, reuse_object = 1;
e5e9714a 66static int keep_unreachable, unpack_unreachable, include_tag;
7e52f566 67static unsigned long unpack_unreachable_expiration;
96f1e58f
DR
68static int local;
69static int incremental;
e96fb9b8 70static int ignore_packed_keep;
be6b1914 71static int allow_ofs_delta;
ebcfb379 72static struct pack_idx_option pack_idx_opts;
d01fb92f 73static const char *base_name;
024701f1 74static int progress = 1;
4812a93a 75static int window = 10;
568508e7 76static unsigned long pack_size_limit;
618e613a 77static int depth = 50;
43cc2b42 78static int delta_search_threads;
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 85static unsigned long delta_cache_size = 0;
5749b0b2 86static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
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;
1b4bb16b 98static struct object_entry *locate_object_entry(const unsigned char *sha1);
3f9ac8d2 99
3f9ac8d2
JH
100/*
101 * stats
102 */
7cadf491
SP
103static uint32_t written, written_delta;
104static uint32_t reused, reused_delta;
3f9ac8d2 105
780e6e73 106
3613f9b4 107static void *get_delta(struct object_entry *entry)
c323ac7d 108{
3613f9b4
NP
109 unsigned long size, base_size, delta_size;
110 void *buf, *base_buf, *delta_buf;
21666f1a 111 enum object_type type;
c323ac7d 112
3613f9b4
NP
113 buf = read_sha1_file(entry->idx.sha1, &type, &size);
114 if (!buf)
115 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
116 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
117 if (!base_buf)
aa7e44bf 118 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
3613f9b4 119 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 120 buf, size, &delta_size, 0);
3613f9b4 121 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 122 die("delta size changed");
3613f9b4
NP
123 free(buf);
124 free(base_buf);
c323ac7d
LT
125 return delta_buf;
126}
127
30ebb40a
NP
128static unsigned long do_compress(void **pptr, unsigned long size)
129{
ef49a7a0 130 git_zstream stream;
30ebb40a
NP
131 void *in, *out;
132 unsigned long maxsize;
133
134 memset(&stream, 0, sizeof(stream));
55bb5c91 135 git_deflate_init(&stream, pack_compression_level);
225a6f10 136 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
137
138 in = *pptr;
139 out = xmalloc(maxsize);
140 *pptr = out;
141
142 stream.next_in = in;
143 stream.avail_in = size;
144 stream.next_out = out;
145 stream.avail_out = maxsize;
55bb5c91 146 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 147 ; /* nothing */
55bb5c91 148 git_deflate_end(&stream);
30ebb40a
NP
149
150 free(in);
151 return stream.total_out;
152}
153
cf2ba13a
NTND
154static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
155 const unsigned char *sha1)
156{
157 git_zstream stream;
158 unsigned char ibuf[1024 * 16];
159 unsigned char obuf[1024 * 16];
160 unsigned long olen = 0;
161
162 memset(&stream, 0, sizeof(stream));
163 git_deflate_init(&stream, pack_compression_level);
164
165 for (;;) {
166 ssize_t readlen;
167 int zret = Z_OK;
168 readlen = read_istream(st, ibuf, sizeof(ibuf));
169 if (readlen == -1)
170 die(_("unable to read %s"), sha1_to_hex(sha1));
171
172 stream.next_in = ibuf;
173 stream.avail_in = readlen;
174 while ((stream.avail_in || readlen == 0) &&
175 (zret == Z_OK || zret == Z_BUF_ERROR)) {
176 stream.next_out = obuf;
177 stream.avail_out = sizeof(obuf);
178 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
179 sha1write(f, obuf, stream.next_out - obuf);
180 olen += stream.next_out - obuf;
181 }
182 if (stream.avail_in)
183 die(_("deflate error (%d)"), zret);
184 if (readlen == 0) {
185 if (zret != Z_STREAM_END)
186 die(_("deflate error (%d)"), zret);
187 break;
188 }
189 }
190 git_deflate_end(&stream);
191 return olen;
192}
193
780e6e73
NP
194/*
195 * we are going to reuse the existing object data as is. make
196 * sure it is not corrupt.
197 */
079afb18
SP
198static int check_pack_inflate(struct packed_git *p,
199 struct pack_window **w_curs,
6777a59f
SP
200 off_t offset,
201 off_t len,
079afb18
SP
202 unsigned long expect)
203{
ef49a7a0 204 git_zstream stream;
079afb18
SP
205 unsigned char fakebuf[4096], *in;
206 int st;
207
208 memset(&stream, 0, sizeof(stream));
39c68542 209 git_inflate_init(&stream);
079afb18
SP
210 do {
211 in = use_pack(p, w_curs, offset, &stream.avail_in);
212 stream.next_in = in;
213 stream.next_out = fakebuf;
214 stream.avail_out = sizeof(fakebuf);
39c68542 215 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
216 offset += stream.next_in - in;
217 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 218 git_inflate_end(&stream);
079afb18
SP
219 return (st == Z_STREAM_END &&
220 stream.total_out == expect &&
221 stream.total_in == len) ? 0 : -1;
222}
223
224static void copy_pack_data(struct sha1file *f,
225 struct packed_git *p,
226 struct pack_window **w_curs,
6777a59f
SP
227 off_t offset,
228 off_t len)
079afb18
SP
229{
230 unsigned char *in;
ef49a7a0 231 unsigned long avail;
079afb18
SP
232
233 while (len) {
234 in = use_pack(p, w_curs, offset, &avail);
235 if (avail > len)
ef49a7a0 236 avail = (unsigned long)len;
079afb18
SP
237 sha1write(f, in, avail);
238 offset += avail;
239 len -= avail;
240 }
241}
242
1b4bb16b 243/* Return 0 if we will bust the pack-size limit */
c9018b03
NTND
244static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
245 unsigned long limit, int usable_delta)
c323ac7d 246{
c9018b03 247 unsigned long size, datalen;
2c5ef824 248 unsigned char header[10], dheader[10];
6777a59f 249 unsigned hdrlen;
2c5ef824 250 enum object_type type;
c9018b03 251 void *buf;
cf2ba13a 252 struct git_istream *st = NULL;
c9018b03
NTND
253
254 if (!usable_delta) {
cf2ba13a
NTND
255 if (entry->type == OBJ_BLOB &&
256 entry->size > big_file_threshold &&
257 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
258 buf = NULL;
259 else {
260 buf = read_sha1_file(entry->idx.sha1, &type, &size);
261 if (!buf)
262 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
263 }
c9018b03
NTND
264 /*
265 * make sure no cached delta data remains from a
266 * previous attempt before a pack split occurred.
267 */
268 free(entry->delta_data);
269 entry->delta_data = NULL;
270 entry->z_delta_size = 0;
271 } else if (entry->delta_data) {
272 size = entry->delta_size;
273 buf = entry->delta_data;
274 entry->delta_data = NULL;
275 type = (allow_ofs_delta && entry->delta->idx.offset) ?
276 OBJ_OFS_DELTA : OBJ_REF_DELTA;
277 } else {
278 buf = get_delta(entry);
279 size = entry->delta_size;
280 type = (allow_ofs_delta && entry->delta->idx.offset) ?
281 OBJ_OFS_DELTA : OBJ_REF_DELTA;
282 }
283
cf2ba13a
NTND
284 if (st) /* large blob case, just assume we don't compress well */
285 datalen = size;
286 else if (entry->z_delta_size)
c9018b03
NTND
287 datalen = entry->z_delta_size;
288 else
289 datalen = do_compress(&buf, size);
290
291 /*
292 * The object header is a byte of 'type' followed by zero or
293 * more bytes of length.
294 */
295 hdrlen = encode_in_pack_object_header(type, size, header);
296
297 if (type == OBJ_OFS_DELTA) {
298 /*
299 * Deltas with relative base contain an additional
300 * encoding of the relative offset for the delta
301 * base from this object's position in the pack.
302 */
303 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
304 unsigned pos = sizeof(dheader) - 1;
305 dheader[pos] = ofs & 127;
306 while (ofs >>= 7)
307 dheader[--pos] = 128 | (--ofs & 127);
308 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
cf2ba13a
NTND
309 if (st)
310 close_istream(st);
c9018b03
NTND
311 free(buf);
312 return 0;
313 }
314 sha1write(f, header, hdrlen);
315 sha1write(f, dheader + pos, sizeof(dheader) - pos);
316 hdrlen += sizeof(dheader) - pos;
317 } else if (type == OBJ_REF_DELTA) {
318 /*
319 * Deltas with a base reference contain
320 * an additional 20 bytes for the base sha1.
321 */
322 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
cf2ba13a
NTND
323 if (st)
324 close_istream(st);
c9018b03
NTND
325 free(buf);
326 return 0;
327 }
328 sha1write(f, header, hdrlen);
329 sha1write(f, entry->delta->idx.sha1, 20);
330 hdrlen += 20;
331 } else {
332 if (limit && hdrlen + datalen + 20 >= limit) {
cf2ba13a
NTND
333 if (st)
334 close_istream(st);
c9018b03
NTND
335 free(buf);
336 return 0;
337 }
338 sha1write(f, header, hdrlen);
339 }
cf2ba13a
NTND
340 if (st) {
341 datalen = write_large_blob_data(st, f, entry->idx.sha1);
342 close_istream(st);
343 } else {
344 sha1write(f, buf, datalen);
345 free(buf);
346 }
c9018b03
NTND
347
348 return hdrlen + datalen;
349}
350
351/* Return 0 if we will bust the pack-size limit */
352static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
353 unsigned long limit, int usable_delta)
354{
355 struct packed_git *p = entry->in_pack;
356 struct pack_window *w_curs = NULL;
357 struct revindex_entry *revidx;
358 off_t offset;
359 enum object_type type = entry->type;
360 unsigned long datalen;
361 unsigned char header[10], dheader[10];
362 unsigned hdrlen;
363
364 if (entry->delta)
365 type = (allow_ofs_delta && entry->delta->idx.offset) ?
366 OBJ_OFS_DELTA : OBJ_REF_DELTA;
367 hdrlen = encode_in_pack_object_header(type, entry->size, header);
368
369 offset = entry->in_pack_offset;
370 revidx = find_pack_revindex(p, offset);
371 datalen = revidx[1].offset - offset;
372 if (!pack_to_stdout && p->index_version > 1 &&
373 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
374 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
375 unuse_pack(&w_curs);
376 return write_no_reuse_object(f, entry, limit, usable_delta);
377 }
378
379 offset += entry->in_pack_header_size;
380 datalen -= entry->in_pack_header_size;
381
382 if (!pack_to_stdout && p->index_version == 1 &&
383 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
384 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
385 unuse_pack(&w_curs);
386 return write_no_reuse_object(f, entry, limit, usable_delta);
387 }
388
389 if (type == OBJ_OFS_DELTA) {
390 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
391 unsigned pos = sizeof(dheader) - 1;
392 dheader[pos] = ofs & 127;
393 while (ofs >>= 7)
394 dheader[--pos] = 128 | (--ofs & 127);
395 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
396 unuse_pack(&w_curs);
397 return 0;
398 }
399 sha1write(f, header, hdrlen);
400 sha1write(f, dheader + pos, sizeof(dheader) - pos);
401 hdrlen += sizeof(dheader) - pos;
402 reused_delta++;
403 } else if (type == OBJ_REF_DELTA) {
404 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
405 unuse_pack(&w_curs);
406 return 0;
407 }
408 sha1write(f, header, hdrlen);
409 sha1write(f, entry->delta->idx.sha1, 20);
410 hdrlen += 20;
411 reused_delta++;
412 } else {
413 if (limit && hdrlen + datalen + 20 >= limit) {
414 unuse_pack(&w_curs);
415 return 0;
416 }
417 sha1write(f, header, hdrlen);
418 }
419 copy_pack_data(f, p, &w_curs, offset, datalen);
420 unuse_pack(&w_curs);
421 reused++;
422 return hdrlen + datalen;
423}
424
425/* Return 0 if we will bust the pack-size limit */
426static unsigned long write_object(struct sha1file *f,
427 struct object_entry *entry,
428 off_t write_offset)
429{
430 unsigned long limit, len;
2c5ef824 431 int usable_delta, to_reuse;
c323ac7d 432
78d1e84f
NP
433 if (!pack_to_stdout)
434 crc32_begin(f);
435
a2430dde 436 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
437 if (!pack_size_limit || !nr_written)
438 limit = 0;
439 else if (pack_size_limit <= write_offset)
440 /*
441 * the earlier object did not fit the limit; avoid
442 * mistaking this with unlimited (i.e. limit = 0).
443 */
444 limit = 1;
445 else
446 limit = pack_size_limit - write_offset;
2c5ef824
NP
447
448 if (!entry->delta)
449 usable_delta = 0; /* no delta */
450 else if (!pack_size_limit)
451 usable_delta = 1; /* unlimited packfile */
452 else if (entry->delta->idx.offset == (off_t)-1)
453 usable_delta = 0; /* base was written to another pack */
454 else if (entry->delta->idx.offset)
455 usable_delta = 1; /* base already exists in this pack */
456 else
457 usable_delta = 0; /* base could end up in another pack */
458
a7de7130 459 if (!reuse_object)
fa736f72
NP
460 to_reuse = 0; /* explicit */
461 else if (!entry->in_pack)
ab7cd7bb 462 to_reuse = 0; /* can't reuse what we don't have */
c9018b03 463 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
17b08f2c
DH
464 /* check_object() decided it for us ... */
465 to_reuse = usable_delta;
466 /* ... but pack split may override that */
c9018b03 467 else if (entry->type != entry->in_pack_type)
ab7cd7bb
JH
468 to_reuse = 0; /* pack has delta which is unusable */
469 else if (entry->delta)
470 to_reuse = 0; /* we want to pack afresh */
471 else
472 to_reuse = 1; /* we have it in-pack undeltified,
473 * and we do not need to deltify it.
474 */
475
c9018b03
NTND
476 if (!to_reuse)
477 len = write_no_reuse_object(f, entry, limit, usable_delta);
478 else
479 len = write_reuse_object(f, entry, limit, usable_delta);
480 if (!len)
481 return 0;
64bd76b1 482
17b08f2c 483 if (usable_delta)
ab7cd7bb 484 written_delta++;
3f9ac8d2 485 written++;
78d1e84f 486 if (!pack_to_stdout)
aa7e44bf 487 entry->idx.crc32 = crc32_end(f);
c9018b03 488 return len;
c323ac7d
LT
489}
490
f63c79db
JH
491enum write_one_status {
492 WRITE_ONE_SKIP = -1, /* already written */
493 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
494 WRITE_ONE_WRITTEN = 1, /* normal */
495 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
496};
497
498static enum write_one_status write_one(struct sha1file *f,
499 struct object_entry *e,
500 off_t *offset)
9d5ab962 501{
d7dd0223 502 unsigned long size;
f63c79db 503 int recursing;
d7dd0223 504
f63c79db
JH
505 /*
506 * we set offset to 1 (which is an impossible value) to mark
507 * the fact that this object is involved in "write its base
508 * first before writing a deltified object" recursion.
509 */
510 recursing = (e->idx.offset == 1);
511 if (recursing) {
512 warning("recursive delta detected for object %s",
513 sha1_to_hex(e->idx.sha1));
514 return WRITE_ONE_RECURSIVE;
515 } else if (e->idx.offset || e->preferred_base) {
516 /* offset is non zero if object is written already. */
517 return WRITE_ONE_SKIP;
518 }
d7dd0223 519
720c9f7b 520 /* if we are deltified, write out base object first. */
f63c79db
JH
521 if (e->delta) {
522 e->idx.offset = 1; /* now recurse */
523 switch (write_one(f, e->delta, offset)) {
524 case WRITE_ONE_RECURSIVE:
525 /* we cannot depend on this one */
526 e->delta = NULL;
527 break;
528 default:
529 break;
530 case WRITE_ONE_BREAK:
531 e->idx.offset = recursing;
532 return WRITE_ONE_BREAK;
533 }
534 }
d7dd0223 535
6ed7f25e
NP
536 e->idx.offset = *offset;
537 size = write_object(f, e, *offset);
17b08f2c 538 if (!size) {
f63c79db
JH
539 e->idx.offset = recursing;
540 return WRITE_ONE_BREAK;
17b08f2c 541 }
79814f42 542 written_list[nr_written++] = &e->idx;
d7dd0223
NP
543
544 /* make sure off_t is sufficiently large not to wrap */
c03c8315 545 if (signed_add_overflows(*offset, size))
d7dd0223 546 die("pack too large for current definition of off_t");
6ed7f25e 547 *offset += size;
f63c79db 548 return WRITE_ONE_WRITTEN;
9d5ab962
JH
549}
550
1b4bb16b
JH
551static int mark_tagged(const char *path, const unsigned char *sha1, int flag,
552 void *cb_data)
553{
554 unsigned char peeled[20];
555 struct object_entry *entry = locate_object_entry(sha1);
556
557 if (entry)
558 entry->tagged = 1;
559 if (!peel_ref(path, peeled)) {
560 entry = locate_object_entry(peeled);
561 if (entry)
562 entry->tagged = 1;
563 }
564 return 0;
565}
566
be126818 567static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 568 unsigned int *endp,
1b4bb16b
JH
569 struct object_entry *e)
570{
571 if (e->filled)
572 return;
573 wo[(*endp)++] = e;
574 e->filled = 1;
575}
576
577static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 578 unsigned int *endp,
1b4bb16b
JH
579 struct object_entry *e)
580{
f380872f
DM
581 int add_to_order = 1;
582 while (e) {
583 if (add_to_order) {
584 struct object_entry *s;
585 /* add this node... */
586 add_to_write_order(wo, endp, e);
587 /* all its siblings... */
588 for (s = e->delta_sibling; s; s = s->delta_sibling) {
589 add_to_write_order(wo, endp, s);
590 }
591 }
592 /* drop down a level to add left subtree nodes if possible */
593 if (e->delta_child) {
594 add_to_order = 1;
595 e = e->delta_child;
596 } else {
597 add_to_order = 0;
598 /* our sibling might have some children, it is next */
599 if (e->delta_sibling) {
600 e = e->delta_sibling;
601 continue;
602 }
603 /* go back to our parent node */
604 e = e->delta;
605 while (e && !e->delta_sibling) {
606 /* we're on the right side of a subtree, keep
607 * going up until we can go right again */
608 e = e->delta;
609 }
610 if (!e) {
611 /* done- we hit our original root node */
612 return;
613 }
614 /* pass it off to sibling at this level */
615 e = e->delta_sibling;
616 }
617 };
1b4bb16b
JH
618}
619
620static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 621 unsigned int *endp,
1b4bb16b
JH
622 struct object_entry *e)
623{
624 struct object_entry *root;
625
626 for (root = e; root->delta; root = root->delta)
627 ; /* nothing */
1b4bb16b
JH
628 add_descendants_to_write_order(wo, endp, root);
629}
630
631static struct object_entry **compute_write_order(void)
632{
38d4debb 633 unsigned int i, wo_end, last_untagged;
1b4bb16b
JH
634
635 struct object_entry **wo = xmalloc(nr_objects * sizeof(*wo));
636
637 for (i = 0; i < nr_objects; i++) {
638 objects[i].tagged = 0;
639 objects[i].filled = 0;
640 objects[i].delta_child = NULL;
641 objects[i].delta_sibling = NULL;
642 }
643
644 /*
645 * Fully connect delta_child/delta_sibling network.
646 * Make sure delta_sibling is sorted in the original
647 * recency order.
648 */
92bef1a1
DM
649 for (i = nr_objects; i > 0;) {
650 struct object_entry *e = &objects[--i];
1b4bb16b
JH
651 if (!e->delta)
652 continue;
653 /* Mark me as the first child */
654 e->delta_sibling = e->delta->delta_child;
655 e->delta->delta_child = e;
656 }
657
658 /*
659 * Mark objects that are at the tip of tags.
660 */
661 for_each_tag_ref(mark_tagged, NULL);
662
663 /*
38d4debb 664 * Give the objects in the original recency order until
1b4bb16b
JH
665 * we see a tagged tip.
666 */
667 for (i = wo_end = 0; i < nr_objects; i++) {
668 if (objects[i].tagged)
669 break;
670 add_to_write_order(wo, &wo_end, &objects[i]);
671 }
38d4debb 672 last_untagged = i;
1b4bb16b
JH
673
674 /*
675 * Then fill all the tagged tips.
676 */
677 for (; i < nr_objects; i++) {
678 if (objects[i].tagged)
679 add_to_write_order(wo, &wo_end, &objects[i]);
680 }
681
682 /*
683 * And then all remaining commits and tags.
684 */
38d4debb 685 for (i = last_untagged; i < nr_objects; i++) {
1b4bb16b
JH
686 if (objects[i].type != OBJ_COMMIT &&
687 objects[i].type != OBJ_TAG)
688 continue;
689 add_to_write_order(wo, &wo_end, &objects[i]);
690 }
691
692 /*
693 * And then all the trees.
694 */
38d4debb 695 for (i = last_untagged; i < nr_objects; i++) {
1b4bb16b
JH
696 if (objects[i].type != OBJ_TREE)
697 continue;
698 add_to_write_order(wo, &wo_end, &objects[i]);
699 }
700
701 /*
702 * Finally all the rest in really tight order
703 */
38d4debb
DM
704 for (i = last_untagged; i < nr_objects; i++) {
705 if (!objects[i].filled)
706 add_family_to_write_order(wo, &wo_end, &objects[i]);
707 }
708
709 if (wo_end != nr_objects)
710 die("ordered %u objects, expected %"PRIu32, wo_end, nr_objects);
1b4bb16b
JH
711
712 return wo;
713}
714
d01fb92f 715static void write_pack_file(void)
c323ac7d 716{
ebe27b13 717 uint32_t i = 0, j;
d22b9290 718 struct sha1file *f;
6ed7f25e 719 off_t offset;
ebe27b13 720 uint32_t nr_remaining = nr_result;
f746bae8 721 time_t last_mtime = 0;
1b4bb16b 722 struct object_entry **write_order;
81a216a5 723
bcd7954e 724 if (progress > pack_to_stdout)
dc6a0757 725 progress_state = start_progress("Writing objects", nr_result);
79814f42 726 written_list = xmalloc(nr_objects * sizeof(*written_list));
1b4bb16b 727 write_order = compute_write_order();
183bdb2c 728
ebe27b13 729 do {
aa7e44bf 730 unsigned char sha1[20];
7ba502c4 731 char *pack_tmp_name = NULL;
aa7e44bf 732
cdf9db3c 733 if (pack_to_stdout)
2a128d63 734 f = sha1fd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
735 else
736 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 737
c0ad4657
JH
738 offset = write_pack_header(f, nr_remaining);
739 if (!offset)
740 die_errno("unable to write pack header");
ebe27b13 741 nr_written = 0;
720c9f7b 742 for (; i < nr_objects; i++) {
1b4bb16b 743 struct object_entry *e = write_order[i];
cddec4f8 744 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
745 break;
746 display_progress(progress_state, written);
747 }
c553ca25 748
ebe27b13
DH
749 /*
750 * Did we write the wrong # entries in the header?
751 * If so, rewrite it like in fast-import
752 */
54352bb2
LT
753 if (pack_to_stdout) {
754 sha1close(f, sha1, CSUM_CLOSE);
755 } else if (nr_written == nr_remaining) {
756 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 757 } else {
ac0463ed 758 int fd = sha1close(f, sha1, 0);
abeb40e5 759 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 760 nr_written, sha1, offset);
7ba502c4 761 close(fd);
ebe27b13
DH
762 }
763
764 if (!pack_to_stdout) {
f746bae8 765 struct stat st;
3bb72562 766 char tmpname[PATH_MAX];
d01fb92f 767
f746bae8
NP
768 /*
769 * Packs are runtime accessed in their mtime
770 * order since newer packs are more likely to contain
771 * younger objects. So if we are creating multiple
772 * packs then we should modify the mtime of later ones
773 * to preserve this property.
774 */
0e990530 775 if (stat(pack_tmp_name, &st) < 0) {
f746bae8 776 warning("failed to stat %s: %s",
0e990530 777 pack_tmp_name, strerror(errno));
f746bae8
NP
778 } else if (!last_mtime) {
779 last_mtime = st.st_mtime;
780 } else {
781 struct utimbuf utb;
782 utb.actime = st.st_atime;
783 utb.modtime = --last_mtime;
0e990530 784 if (utime(pack_tmp_name, &utb) < 0)
f746bae8
NP
785 warning("failed utime() on %s: %s",
786 tmpname, strerror(errno));
787 }
788
0e990530
JH
789 /* Enough space for "-<sha-1>.pack"? */
790 if (sizeof(tmpname) <= strlen(base_name) + 50)
791 die("pack base name '%s' too long", base_name);
792 snprintf(tmpname, sizeof(tmpname), "%s-", base_name);
793 finish_tmp_packfile(tmpname, pack_tmp_name,
794 written_list, nr_written,
795 &pack_idx_opts, sha1);
7ba502c4 796 free(pack_tmp_name);
aa7e44bf 797 puts(sha1_to_hex(sha1));
ebe27b13
DH
798 }
799
800 /* mark written objects as written to previous pack */
801 for (j = 0; j < nr_written; j++) {
79814f42 802 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
803 }
804 nr_remaining -= nr_written;
720c9f7b 805 } while (nr_remaining && i < nr_objects);
ebe27b13
DH
806
807 free(written_list);
1b4bb16b 808 free(write_order);
4d4fcc54 809 stop_progress(&progress_state);
67c08ce1 810 if (written != nr_result)
6e1c2344
RJ
811 die("wrote %"PRIu32" objects while expecting %"PRIu32,
812 written, nr_result);
c323ac7d
LT
813}
814
7a979d99
JH
815static int locate_object_entry_hash(const unsigned char *sha1)
816{
817 int i;
818 unsigned int ui;
819 memcpy(&ui, sha1, sizeof(unsigned int));
820 i = ui % object_ix_hashsz;
821 while (0 < object_ix[i]) {
aa7e44bf 822 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
823 return i;
824 if (++i == object_ix_hashsz)
825 i = 0;
826 }
827 return -1 - i;
828}
829
830static struct object_entry *locate_object_entry(const unsigned char *sha1)
831{
832 int i;
833
834 if (!object_ix_hashsz)
835 return NULL;
836
837 i = locate_object_entry_hash(sha1);
838 if (0 <= i)
839 return &objects[object_ix[i]-1];
840 return NULL;
841}
842
843static void rehash_objects(void)
c323ac7d 844{
7cadf491 845 uint32_t i;
7a979d99
JH
846 struct object_entry *oe;
847
848 object_ix_hashsz = nr_objects * 3;
849 if (object_ix_hashsz < 1024)
850 object_ix_hashsz = 1024;
851 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 852 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 853 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 854 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
855 if (0 <= ix)
856 continue;
857 ix = -1 - ix;
858 object_ix[ix] = i + 1;
859 }
860}
861
ce0bd642 862static unsigned name_hash(const char *name)
1d6b38cc 863{
48fb7deb 864 unsigned c, hash = 0;
ce0bd642 865
bc32fed5
JH
866 if (!name)
867 return 0;
868
eeef7135 869 /*
ce0bd642
LT
870 * This effectively just creates a sortable number from the
871 * last sixteen non-whitespace characters. Last characters
872 * count "most", so things that end in ".c" sort together.
eeef7135 873 */
ce0bd642
LT
874 while ((c = *name++) != 0) {
875 if (isspace(c))
876 continue;
877 hash = (hash >> 2) + (c << 24);
878 }
1d6b38cc
JH
879 return hash;
880}
881
a74db82e
JH
882static void setup_delta_attr_check(struct git_attr_check *check)
883{
884 static struct git_attr *attr_delta;
885
886 if (!attr_delta)
7fb0eaa2 887 attr_delta = git_attr("delta");
a74db82e
JH
888
889 check[0].attr = attr_delta;
890}
891
892static int no_try_delta(const char *path)
893{
894 struct git_attr_check check[1];
895
896 setup_delta_attr_check(check);
d932f4eb 897 if (git_check_attr(path, ARRAY_SIZE(check), check))
a74db82e
JH
898 return 0;
899 if (ATTR_FALSE(check->value))
900 return 1;
901 return 0;
902}
903
5c49c116 904static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 905 const char *name, int exclude)
c323ac7d 906{
c323ac7d 907 struct object_entry *entry;
29b734e4 908 struct packed_git *p, *found_pack = NULL;
6777a59f 909 off_t found_offset = 0;
29b734e4 910 int ix;
bc32fed5 911 unsigned hash = name_hash(name);
29b734e4
NP
912
913 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
914 if (ix >= 0) {
915 if (exclude) {
916 entry = objects + object_ix[ix] - 1;
81a216a5
NP
917 if (!entry->preferred_base)
918 nr_result--;
29b734e4
NP
919 entry->preferred_base = 1;
920 }
921 return 0;
922 }
c323ac7d 923
daae0625
BC
924 if (!exclude && local && has_loose_object_nonlocal(sha1))
925 return 0;
926
5c49c116
NP
927 for (p = packed_git; p; p = p->next) {
928 off_t offset = find_pack_entry_one(sha1, p);
929 if (offset) {
930 if (!found_pack) {
4c080182 931 if (!is_pack_valid(p)) {
58a6a9cc 932 warning("packfile %s cannot be accessed", p->pack_name);
4c080182
JK
933 continue;
934 }
5c49c116
NP
935 found_offset = offset;
936 found_pack = p;
64560374 937 }
5c49c116
NP
938 if (exclude)
939 break;
940 if (incremental)
941 return 0;
942 if (local && !p->pack_local)
943 return 0;
e96fb9b8
BC
944 if (ignore_packed_keep && p->pack_local && p->pack_keep)
945 return 0;
64560374
LT
946 }
947 }
eb019375 948
29b734e4
NP
949 if (nr_objects >= nr_alloc) {
950 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 951 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 952 }
29b734e4
NP
953
954 entry = objects + nr_objects++;
c323ac7d 955 memset(entry, 0, sizeof(*entry));
aa7e44bf 956 hashcpy(entry->idx.sha1, sha1);
27225f2e 957 entry->hash = hash;
5c49c116
NP
958 if (type)
959 entry->type = type;
29b734e4
NP
960 if (exclude)
961 entry->preferred_base = 1;
81a216a5
NP
962 else
963 nr_result++;
29b734e4
NP
964 if (found_pack) {
965 entry->in_pack = found_pack;
966 entry->in_pack_offset = found_offset;
967 }
7a979d99
JH
968
969 if (object_ix_hashsz * 3 <= nr_objects * 4)
970 rehash_objects();
29b734e4
NP
971 else
972 object_ix[-1 - ix] = nr_objects;
7a979d99 973
4d4fcc54 974 display_progress(progress_state, nr_objects);
29b734e4 975
a74db82e
JH
976 if (name && no_try_delta(name))
977 entry->no_try_delta = 1;
978
29b734e4 979 return 1;
c323ac7d
LT
980}
981
5379a5c5
JH
982struct pbase_tree_cache {
983 unsigned char sha1[20];
984 int ref;
985 int temporary;
986 void *tree_data;
987 unsigned long tree_size;
988};
989
990static struct pbase_tree_cache *(pbase_tree_cache[256]);
991static int pbase_tree_cache_ix(const unsigned char *sha1)
992{
993 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
994}
995static int pbase_tree_cache_ix_incr(int ix)
996{
997 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
998}
999
1000static struct pbase_tree {
1001 struct pbase_tree *next;
1002 /* This is a phony "cache" entry; we are not
1003 * going to evict it nor find it through _get()
1004 * mechanism -- this is for the toplevel node that
1005 * would almost always change with any commit.
1006 */
1007 struct pbase_tree_cache pcache;
1008} *pbase_tree;
1009
1010static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1011{
1012 struct pbase_tree_cache *ent, *nent;
1013 void *data;
1014 unsigned long size;
21666f1a 1015 enum object_type type;
5379a5c5
JH
1016 int neigh;
1017 int my_ix = pbase_tree_cache_ix(sha1);
1018 int available_ix = -1;
1019
1020 /* pbase-tree-cache acts as a limited hashtable.
1021 * your object will be found at your index or within a few
1022 * slots after that slot if it is cached.
1023 */
1024 for (neigh = 0; neigh < 8; neigh++) {
1025 ent = pbase_tree_cache[my_ix];
a89fccd2 1026 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
1027 ent->ref++;
1028 return ent;
1029 }
1030 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1031 ((0 <= available_ix) &&
1032 (!ent && pbase_tree_cache[available_ix])))
1033 available_ix = my_ix;
1034 if (!ent)
1035 break;
1036 my_ix = pbase_tree_cache_ix_incr(my_ix);
1037 }
1038
1039 /* Did not find one. Either we got a bogus request or
1040 * we need to read and perhaps cache.
1041 */
21666f1a 1042 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
1043 if (!data)
1044 return NULL;
21666f1a 1045 if (type != OBJ_TREE) {
5379a5c5
JH
1046 free(data);
1047 return NULL;
1048 }
1049
1050 /* We need to either cache or return a throwaway copy */
1051
1052 if (available_ix < 0)
1053 ent = NULL;
1054 else {
1055 ent = pbase_tree_cache[available_ix];
1056 my_ix = available_ix;
1057 }
1058
1059 if (!ent) {
1060 nent = xmalloc(sizeof(*nent));
1061 nent->temporary = (available_ix < 0);
1062 }
1063 else {
1064 /* evict and reuse */
1065 free(ent->tree_data);
1066 nent = ent;
1067 }
e702496e 1068 hashcpy(nent->sha1, sha1);
5379a5c5
JH
1069 nent->tree_data = data;
1070 nent->tree_size = size;
1071 nent->ref = 1;
1072 if (!nent->temporary)
1073 pbase_tree_cache[my_ix] = nent;
1074 return nent;
1075}
1076
1077static void pbase_tree_put(struct pbase_tree_cache *cache)
1078{
1079 if (!cache->temporary) {
1080 cache->ref--;
1081 return;
1082 }
1083 free(cache->tree_data);
1084 free(cache);
1085}
1086
1087static int name_cmp_len(const char *name)
1088{
1089 int i;
1090 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1091 ;
1092 return i;
1093}
1094
1095static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1096 const char *name,
ce0bd642
LT
1097 int cmplen,
1098 const char *fullname)
3f9ac8d2 1099{
4c068a98 1100 struct name_entry entry;
8a5a8d6c 1101 int cmp;
4c068a98
LT
1102
1103 while (tree_entry(tree,&entry)) {
1211be6b
LT
1104 if (S_ISGITLINK(entry.mode))
1105 continue;
0de16337 1106 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1107 memcmp(name, entry.path, cmplen);
1108 if (cmp > 0)
7a979d99 1109 continue;
8a5a8d6c
NP
1110 if (cmp < 0)
1111 return;
5379a5c5 1112 if (name[cmplen] != '/') {
5c49c116 1113 add_object_entry(entry.sha1,
4d1012c3 1114 object_type(entry.mode),
bc32fed5 1115 fullname, 1);
5379a5c5
JH
1116 return;
1117 }
8a5a8d6c 1118 if (S_ISDIR(entry.mode)) {
7a979d99 1119 struct tree_desc sub;
5379a5c5
JH
1120 struct pbase_tree_cache *tree;
1121 const char *down = name+cmplen+1;
1122 int downlen = name_cmp_len(down);
1123
4c068a98 1124 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
1125 if (!tree)
1126 return;
6fda5e51 1127 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1128
ce0bd642 1129 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1130 pbase_tree_put(tree);
1131 }
1132 }
1133}
1d6b38cc 1134
5379a5c5
JH
1135static unsigned *done_pbase_paths;
1136static int done_pbase_paths_num;
1137static int done_pbase_paths_alloc;
1138static int done_pbase_path_pos(unsigned hash)
1139{
1140 int lo = 0;
1141 int hi = done_pbase_paths_num;
1142 while (lo < hi) {
1143 int mi = (hi + lo) / 2;
1144 if (done_pbase_paths[mi] == hash)
1145 return mi;
1146 if (done_pbase_paths[mi] < hash)
1147 hi = mi;
1148 else
1149 lo = mi + 1;
1150 }
1151 return -lo-1;
1152}
1153
1154static int check_pbase_path(unsigned hash)
1155{
1156 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1157 if (0 <= pos)
1158 return 1;
1159 pos = -pos - 1;
1160 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1161 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1162 done_pbase_paths = xrealloc(done_pbase_paths,
1163 done_pbase_paths_alloc *
1164 sizeof(unsigned));
1165 }
1166 done_pbase_paths_num++;
1167 if (pos < done_pbase_paths_num)
1168 memmove(done_pbase_paths + pos + 1,
1169 done_pbase_paths + pos,
1170 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1171 done_pbase_paths[pos] = hash;
1172 return 0;
1173}
1174
bc32fed5 1175static void add_preferred_base_object(const char *name)
5379a5c5
JH
1176{
1177 struct pbase_tree *it;
8a5a8d6c 1178 int cmplen;
bc32fed5 1179 unsigned hash = name_hash(name);
5379a5c5 1180
8a5a8d6c 1181 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1182 return;
1183
8a5a8d6c 1184 cmplen = name_cmp_len(name);
5379a5c5
JH
1185 for (it = pbase_tree; it; it = it->next) {
1186 if (cmplen == 0) {
bc32fed5 1187 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
1188 }
1189 else {
1190 struct tree_desc tree;
6fda5e51 1191 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1192 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1193 }
3f9ac8d2 1194 }
3f9ac8d2
JH
1195}
1196
7a979d99 1197static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 1198{
5379a5c5
JH
1199 struct pbase_tree *it;
1200 void *data;
1201 unsigned long size;
1202 unsigned char tree_sha1[20];
1d6b38cc 1203
8d1d8f83
JH
1204 if (window <= num_preferred_base++)
1205 return;
1206
5379a5c5
JH
1207 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1208 if (!data)
7a979d99 1209 return;
5379a5c5
JH
1210
1211 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1212 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1213 free(data);
1214 return;
1215 }
1216 }
1217
1218 it = xcalloc(1, sizeof(*it));
1219 it->next = pbase_tree;
1220 pbase_tree = it;
1221
e702496e 1222 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1223 it->pcache.tree_data = data;
1224 it->pcache.tree_size = size;
3f9ac8d2
JH
1225}
1226
0ef95f72
NP
1227static void cleanup_preferred_base(void)
1228{
1229 struct pbase_tree *it;
1230 unsigned i;
1231
1232 it = pbase_tree;
1233 pbase_tree = NULL;
1234 while (it) {
1235 struct pbase_tree *this = it;
1236 it = this->next;
1237 free(this->pcache.tree_data);
1238 free(this);
1239 }
1240
1241 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1242 if (!pbase_tree_cache[i])
1243 continue;
1244 free(pbase_tree_cache[i]->tree_data);
1245 free(pbase_tree_cache[i]);
1246 pbase_tree_cache[i] = NULL;
1247 }
1248
1249 free(done_pbase_paths);
1250 done_pbase_paths = NULL;
1251 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1252}
1253
c323ac7d
LT
1254static void check_object(struct object_entry *entry)
1255{
5c49c116 1256 if (entry->in_pack) {
780e6e73 1257 struct packed_git *p = entry->in_pack;
03e79c88 1258 struct pack_window *w_curs = NULL;
5c49c116
NP
1259 const unsigned char *base_ref = NULL;
1260 struct object_entry *base_entry;
1261 unsigned long used, used_0;
ef49a7a0 1262 unsigned long avail;
5c49c116
NP
1263 off_t ofs;
1264 unsigned char *buf, c;
780e6e73 1265
6777a59f 1266 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1267
5c49c116 1268 /*
fa736f72
NP
1269 * We want in_pack_type even if we do not reuse delta
1270 * since non-delta representations could still be reused.
ab7cd7bb 1271 */
09ded04b 1272 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1273 &entry->in_pack_type,
1274 &entry->size);
03d66015
NP
1275 if (used == 0)
1276 goto give_up;
ab7cd7bb 1277
5c49c116
NP
1278 /*
1279 * Determine if this is a delta and if so whether we can
1280 * reuse it or not. Otherwise let's find out as cheaply as
1281 * possible what the actual type and size for this object is.
3f9ac8d2 1282 */
5c49c116
NP
1283 switch (entry->in_pack_type) {
1284 default:
1285 /* Not a delta hence we've already got all we need. */
1286 entry->type = entry->in_pack_type;
1287 entry->in_pack_header_size = used;
03d66015
NP
1288 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1289 goto give_up;
5c49c116
NP
1290 unuse_pack(&w_curs);
1291 return;
1292 case OBJ_REF_DELTA:
a7de7130 1293 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1294 base_ref = use_pack(p, &w_curs,
1295 entry->in_pack_offset + used, NULL);
1296 entry->in_pack_header_size = used + 20;
1297 break;
1298 case OBJ_OFS_DELTA:
1299 buf = use_pack(p, &w_curs,
1300 entry->in_pack_offset + used, NULL);
1301 used_0 = 0;
1302 c = buf[used_0++];
1303 ofs = c & 127;
1304 while (c & 128) {
1305 ofs += 1;
03d66015
NP
1306 if (!ofs || MSB(ofs, 7)) {
1307 error("delta base offset overflow in pack for %s",
1308 sha1_to_hex(entry->idx.sha1));
1309 goto give_up;
1310 }
5c49c116
NP
1311 c = buf[used_0++];
1312 ofs = (ofs << 7) + (c & 127);
780e6e73 1313 }
5c49c116 1314 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1315 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1316 error("delta base offset out of bound for %s",
1317 sha1_to_hex(entry->idx.sha1));
1318 goto give_up;
1319 }
a7de7130 1320 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1321 struct revindex_entry *revidx;
1322 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1323 if (!revidx)
1324 goto give_up;
3449f8c4
NP
1325 base_ref = nth_packed_object_sha1(p, revidx->nr);
1326 }
5c49c116
NP
1327 entry->in_pack_header_size = used + used_0;
1328 break;
780e6e73 1329 }
780e6e73 1330
5c49c116
NP
1331 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1332 /*
1333 * If base_ref was set above that means we wish to
1334 * reuse delta data, and we even found that base
1335 * in the list of objects we want to pack. Goodie!
1336 *
1337 * Depth value does not matter - find_deltas() will
1338 * never consider reused delta as the base object to
1339 * deltify other objects against, in order to avoid
1340 * circular deltas.
3f9ac8d2 1341 */
780e6e73 1342 entry->type = entry->in_pack_type;
5c49c116 1343 entry->delta = base_entry;
64bd76b1 1344 entry->delta_size = entry->size;
15b4d577
JH
1345 entry->delta_sibling = base_entry->delta_child;
1346 base_entry->delta_child = entry;
5c49c116
NP
1347 unuse_pack(&w_curs);
1348 return;
1349 }
ab7cd7bb 1350
5c49c116
NP
1351 if (entry->type) {
1352 /*
1353 * This must be a delta and we already know what the
1354 * final object type is. Let's extract the actual
1355 * object size from the delta header.
1356 */
1357 entry->size = get_size_from_delta(p, &w_curs,
1358 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1359 if (entry->size == 0)
1360 goto give_up;
5c49c116 1361 unuse_pack(&w_curs);
3f9ac8d2
JH
1362 return;
1363 }
5c49c116
NP
1364
1365 /*
1366 * No choice but to fall back to the recursive delta walk
1367 * with sha1_object_info() to find about the object type
1368 * at this point...
1369 */
03d66015 1370 give_up:
5c49c116 1371 unuse_pack(&w_curs);
36e4d74a 1372 }
3f9ac8d2 1373
aa7e44bf 1374 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1375 /*
1376 * The error condition is checked in prepare_pack(). This is
1377 * to permit a missing preferred base object to be ignored
1378 * as a preferred base. Doing so can result in a larger
1379 * pack file, but the transfer will still take place.
1380 */
3f9ac8d2
JH
1381}
1382
5c49c116
NP
1383static int pack_offset_sort(const void *_a, const void *_b)
1384{
1385 const struct object_entry *a = *(struct object_entry **)_a;
1386 const struct object_entry *b = *(struct object_entry **)_b;
1387
1388 /* avoid filesystem trashing with loose objects */
1389 if (!a->in_pack && !b->in_pack)
aa7e44bf 1390 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1391
1392 if (a->in_pack < b->in_pack)
1393 return -1;
1394 if (a->in_pack > b->in_pack)
1395 return 1;
1396 return a->in_pack_offset < b->in_pack_offset ? -1 :
1397 (a->in_pack_offset > b->in_pack_offset);
1398}
1399
c323ac7d
LT
1400static void get_object_details(void)
1401{
7cadf491 1402 uint32_t i;
5c49c116
NP
1403 struct object_entry **sorted_by_offset;
1404
1405 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1406 for (i = 0; i < nr_objects; i++)
1407 sorted_by_offset[i] = objects + i;
1408 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1409
15366280
JH
1410 for (i = 0; i < nr_objects; i++) {
1411 struct object_entry *entry = sorted_by_offset[i];
1412 check_object(entry);
754980d0 1413 if (big_file_threshold < entry->size)
15366280
JH
1414 entry->no_try_delta = 1;
1415 }
3449f8c4 1416
5c49c116 1417 free(sorted_by_offset);
c323ac7d
LT
1418}
1419
b904166c
NP
1420/*
1421 * We search for deltas in a list sorted by type, by filename hash, and then
1422 * by size, so that we see progressively smaller and smaller files.
1423 * That's because we prefer deltas to be from the bigger file
1424 * to the smaller -- deletes are potentially cheaper, but perhaps
1425 * more importantly, the bigger file is likely the more recent
1426 * one. The deepest deltas are therefore the oldest objects which are
1427 * less susceptible to be accessed often.
1428 */
9668cf59 1429static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1430{
9668cf59
NP
1431 const struct object_entry *a = *(struct object_entry **)_a;
1432 const struct object_entry *b = *(struct object_entry **)_b;
1433
c323ac7d 1434 if (a->type > b->type)
27225f2e 1435 return -1;
b904166c 1436 if (a->type < b->type)
27225f2e 1437 return 1;
b904166c 1438 if (a->hash > b->hash)
7a979d99 1439 return -1;
b904166c 1440 if (a->hash < b->hash)
7a979d99 1441 return 1;
b904166c 1442 if (a->preferred_base > b->preferred_base)
c323ac7d 1443 return -1;
b904166c
NP
1444 if (a->preferred_base < b->preferred_base)
1445 return 1;
c323ac7d 1446 if (a->size > b->size)
b904166c
NP
1447 return -1;
1448 if (a->size < b->size)
c323ac7d 1449 return 1;
b904166c 1450 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1451}
1452
1453struct unpacked {
1454 struct object_entry *entry;
1455 void *data;
f6c7081a 1456 struct delta_index *index;
5a235b5e 1457 unsigned depth;
c323ac7d
LT
1458};
1459
d250626c
NP
1460static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1461 unsigned long delta_size)
074b2eea
MK
1462{
1463 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1464 return 0;
1465
e3dfddb3
MK
1466 if (delta_size < cache_max_small_delta_size)
1467 return 1;
1468
074b2eea
MK
1469 /* cache delta, if objects are large enough compared to delta size */
1470 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1471 return 1;
1472
1473 return 0;
1474}
1475
7eb151d6 1476#ifndef NO_PTHREADS
8ecce684 1477
44626dc7 1478static pthread_mutex_t read_mutex;
8ecce684
NP
1479#define read_lock() pthread_mutex_lock(&read_mutex)
1480#define read_unlock() pthread_mutex_unlock(&read_mutex)
1481
44626dc7 1482static pthread_mutex_t cache_mutex;
3c701839
NP
1483#define cache_lock() pthread_mutex_lock(&cache_mutex)
1484#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1485
44626dc7 1486static pthread_mutex_t progress_mutex;
8ecce684
NP
1487#define progress_lock() pthread_mutex_lock(&progress_mutex)
1488#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1489
1490#else
1491
e1ef8673
JH
1492#define read_lock() (void)0
1493#define read_unlock() (void)0
1494#define cache_lock() (void)0
1495#define cache_unlock() (void)0
1496#define progress_lock() (void)0
1497#define progress_unlock() (void)0
8ecce684
NP
1498
1499#endif
1500
f6c7081a 1501static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1502 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1503{
f6c7081a
NP
1504 struct object_entry *trg_entry = trg->entry;
1505 struct object_entry *src_entry = src->entry;
560b25a8 1506 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1507 unsigned ref_depth;
21666f1a 1508 enum object_type type;
c323ac7d
LT
1509 void *delta_buf;
1510
1511 /* Don't bother doing diffs between different types */
f6c7081a 1512 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1513 return -1;
1514
51d1e83f 1515 /*
15f07e06
JK
1516 * We do not bother to try a delta that we discarded on an
1517 * earlier try, but only when reusing delta data. Note that
1518 * src_entry that is marked as the preferred_base should always
1519 * be considered, as even if we produce a suboptimal delta against
1520 * it, we will still save the transfer cost, as we already know
1521 * the other side has it and we won't send src_entry at all.
51d1e83f 1522 */
a7de7130 1523 if (reuse_delta && trg_entry->in_pack &&
e9195b58 1524 trg_entry->in_pack == src_entry->in_pack &&
15f07e06 1525 !src_entry->preferred_base &&
e9195b58
JH
1526 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1527 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1528 return 0;
1529
898b14ce 1530 /* Let's not bust the allowed depth. */
5a235b5e 1531 if (src->depth >= max_depth)
d116a45a 1532 return 0;
c323ac7d 1533
c3b06a69 1534 /* Now some size filtering heuristics. */
560b25a8 1535 trg_size = trg_entry->size;
c83f032e
NP
1536 if (!trg_entry->delta) {
1537 max_size = trg_size/2 - 20;
1538 ref_depth = 1;
1539 } else {
1540 max_size = trg_entry->delta_size;
5a235b5e 1541 ref_depth = trg->depth;
c83f032e 1542 }
720fe22d 1543 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1544 (max_depth - ref_depth + 1);
c3b06a69
NP
1545 if (max_size == 0)
1546 return 0;
f6c7081a 1547 src_size = src_entry->size;
560b25a8 1548 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1549 if (sizediff >= max_size)
f527cb8c 1550 return 0;
a1dab41a
BD
1551 if (trg_size < src_size / 32)
1552 return 0;
f6c7081a 1553
560b25a8
NP
1554 /* Load data if not already done */
1555 if (!trg->data) {
8ecce684 1556 read_lock();
aa7e44bf 1557 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1558 read_unlock();
2e3404c3
JH
1559 if (!trg->data)
1560 die("object %s cannot be read",
1561 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1562 if (sz != trg_size)
1563 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1564 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1565 *mem_usage += sz;
560b25a8
NP
1566 }
1567 if (!src->data) {
8ecce684 1568 read_lock();
aa7e44bf 1569 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1570 read_unlock();
71064a95
NP
1571 if (!src->data) {
1572 if (src_entry->preferred_base) {
1573 static int warned = 0;
1574 if (!warned++)
1575 warning("object %s cannot be read",
1576 sha1_to_hex(src_entry->idx.sha1));
1577 /*
1578 * Those objects are not included in the
1579 * resulting pack. Be resilient and ignore
1580 * them if they can't be read, in case the
1581 * pack could be created nevertheless.
1582 */
1583 return 0;
1584 }
2e3404c3
JH
1585 die("object %s cannot be read",
1586 sha1_to_hex(src_entry->idx.sha1));
71064a95 1587 }
560b25a8
NP
1588 if (sz != src_size)
1589 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1590 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1591 *mem_usage += sz;
560b25a8
NP
1592 }
1593 if (!src->index) {
1594 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1595 if (!src->index) {
1596 static int warned = 0;
1597 if (!warned++)
1598 warning("suboptimal pack - out of memory");
1599 return 0;
1600 }
ef0316fc 1601 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1602 }
1603
1604 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1605 if (!delta_buf)
75c42d8c 1606 return 0;
f6c7081a 1607
9e2d57a0 1608 if (trg_entry->delta) {
848d732c
BD
1609 /* Prefer only shallower same-sized deltas. */
1610 if (delta_size == trg_entry->delta_size &&
5a235b5e 1611 src->depth + 1 >= trg->depth) {
848d732c
BD
1612 free(delta_buf);
1613 return 0;
1614 }
074b2eea 1615 }
9e2d57a0 1616
3c701839
NP
1617 /*
1618 * Handle memory allocation outside of the cache
1619 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1620 * away when NO_PTHREADS is defined.
3c701839 1621 */
8e0f7003 1622 free(trg_entry->delta_data);
3c701839 1623 cache_lock();
9e2d57a0
NP
1624 if (trg_entry->delta_data) {
1625 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1626 trg_entry->delta_data = NULL;
1627 }
d250626c 1628 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1629 delta_cache_size += delta_size;
3c701839
NP
1630 cache_unlock();
1631 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1632 } else {
1633 cache_unlock();
074b2eea 1634 free(delta_buf);
3c701839
NP
1635 }
1636
b7a28f78
NP
1637 trg_entry->delta = src_entry;
1638 trg_entry->delta_size = delta_size;
1639 trg->depth = src->depth + 1;
1640
f6c7081a 1641 return 1;
c323ac7d
LT
1642}
1643
898b14ce 1644static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1645{
898b14ce
NP
1646 struct object_entry *child = me->delta_child;
1647 unsigned int m = n;
1648 while (child) {
1649 unsigned int c = check_delta_limit(child, n + 1);
1650 if (m < c)
1651 m = c;
1652 child = child->delta_sibling;
1653 }
1654 return m;
b2504a0d
NP
1655}
1656
75ad235c 1657static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1658{
ef0316fc 1659 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1660 free_delta_index(n->index);
1661 n->index = NULL;
1662 if (n->data) {
ef0316fc 1663 freed_mem += n->entry->size;
a97773ce
BD
1664 free(n->data);
1665 n->data = NULL;
a97773ce
BD
1666 }
1667 n->entry = NULL;
7d7baa5e 1668 n->depth = 0;
ef0316fc 1669 return freed_mem;
a97773ce
BD
1670}
1671
384b32c0 1672static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1673 int window, int depth, unsigned *processed)
c323ac7d 1674{
384b32c0 1675 uint32_t i, idx = 0, count = 0;
7cadf491 1676 struct unpacked *array;
ef0316fc 1677 unsigned long mem_usage = 0;
c323ac7d 1678
19d4b416 1679 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1680
384b32c0 1681 for (;;) {
421b488a 1682 struct object_entry *entry;
c323ac7d 1683 struct unpacked *n = array + idx;
ef0316fc 1684 int j, max_depth, best_base = -1;
c323ac7d 1685
384b32c0
NP
1686 progress_lock();
1687 if (!*list_size) {
1688 progress_unlock();
1689 break;
1690 }
421b488a 1691 entry = *list++;
384b32c0
NP
1692 (*list_size)--;
1693 if (!entry->preferred_base) {
1694 (*processed)++;
1695 display_progress(progress_state, *processed);
1696 }
1697 progress_unlock();
1698
ef0316fc 1699 mem_usage -= free_unpacked(n);
c323ac7d 1700 n->entry = entry;
ab7cd7bb 1701
a97773ce 1702 while (window_memory_limit &&
ef0316fc 1703 mem_usage > window_memory_limit &&
a97773ce
BD
1704 count > 1) {
1705 uint32_t tail = (idx + window - count) % window;
75ad235c 1706 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1707 count--;
1708 }
1709
75d39853
NP
1710 /* We do not compute delta to *create* objects we are not
1711 * going to pack.
1712 */
1713 if (entry->preferred_base)
1714 goto next;
1715
898b14ce
NP
1716 /*
1717 * If the current object is at pack edge, take the depth the
1718 * objects that depend on the current object into account
1719 * otherwise they would become too deep.
1720 */
1721 max_depth = depth;
1722 if (entry->delta_child) {
1723 max_depth -= check_delta_limit(entry, 0);
1724 if (max_depth <= 0)
1725 goto next;
1726 }
1727
78817c15
LT
1728 j = window;
1729 while (--j > 0) {
77639870 1730 int ret;
7cadf491 1731 uint32_t other_idx = idx + j;
c323ac7d 1732 struct unpacked *m;
78817c15
LT
1733 if (other_idx >= window)
1734 other_idx -= window;
c323ac7d
LT
1735 m = array + other_idx;
1736 if (!m->entry)
1737 break;
ef0316fc 1738 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1739 if (ret < 0)
c323ac7d 1740 break;
77639870
JH
1741 else if (ret > 0)
1742 best_base = other_idx;
c323ac7d 1743 }
898b14ce 1744
ed4a9031
NP
1745 /*
1746 * If we decided to cache the delta data, then it is best
1747 * to compress it right away. First because we have to do
1748 * it anyway, and doing it here while we're threaded will
1749 * save a lot of time in the non threaded write phase,
1750 * as well as allow for caching more deltas within
1751 * the same cache size limit.
1752 * ...
1753 * But only if not writing to stdout, since in that case
1754 * the network is most likely throttling writes anyway,
1755 * and therefore it is best to go to the write phase ASAP
1756 * instead, as we can afford spending more time compressing
1757 * between writes at that moment.
1758 */
1759 if (entry->delta_data && !pack_to_stdout) {
1760 entry->z_delta_size = do_compress(&entry->delta_data,
1761 entry->delta_size);
1762 cache_lock();
1763 delta_cache_size -= entry->delta_size;
1764 delta_cache_size += entry->z_delta_size;
1765 cache_unlock();
1766 }
1767
70ca1a3f
JH
1768 /* if we made n a delta, and if n is already at max
1769 * depth, leaving it in the window is pointless. we
1770 * should evict it first.
70ca1a3f 1771 */
70baf5d4 1772 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1773 continue;
ff45715c 1774
77639870
JH
1775 /*
1776 * Move the best delta base up in the window, after the
1777 * currently deltified object, to keep it longer. It will
1778 * be the first base object to be attempted next.
1779 */
1780 if (entry->delta) {
1781 struct unpacked swap = array[best_base];
1782 int dist = (window + idx - best_base) % window;
1783 int dst = best_base;
1784 while (dist--) {
1785 int src = (dst + 1) % window;
1786 array[dst] = array[src];
1787 dst = src;
1788 }
1789 array[dst] = swap;
1790 }
1791
898b14ce 1792 next:
521a4f4c 1793 idx++;
a97773ce
BD
1794 if (count + 1 < window)
1795 count++;
521a4f4c
LT
1796 if (idx >= window)
1797 idx = 0;
384b32c0 1798 }
adee7bdf 1799
f6c7081a 1800 for (i = 0; i < window; ++i) {
ff45715c 1801 free_delta_index(array[i].index);
adee7bdf 1802 free(array[i].data);
f6c7081a 1803 }
adee7bdf 1804 free(array);
c323ac7d
LT
1805}
1806
7eb151d6 1807#ifndef NO_PTHREADS
8ecce684 1808
a9a74636
NP
1809static void try_to_free_from_threads(size_t size)
1810{
1811 read_lock();
1812 release_pack_memory(size, -1);
1813 read_unlock();
1814}
1815
bc9b2175 1816static try_to_free_t old_try_to_free_routine;
851c34b0 1817
50f22ada
JS
1818/*
1819 * The main thread waits on the condition that (at least) one of the workers
1820 * has stopped working (which is indicated in the .working member of
1821 * struct thread_params).
1822 * When a work thread has completed its work, it sets .working to 0 and
1823 * signals the main thread and waits on the condition that .data_ready
1824 * becomes 1.
1825 */
1826
8ecce684
NP
1827struct thread_params {
1828 pthread_t thread;
1829 struct object_entry **list;
1830 unsigned list_size;
384b32c0 1831 unsigned remaining;
8ecce684
NP
1832 int window;
1833 int depth;
50f22ada
JS
1834 int working;
1835 int data_ready;
1836 pthread_mutex_t mutex;
1837 pthread_cond_t cond;
8ecce684
NP
1838 unsigned *processed;
1839};
1840
44626dc7
AH
1841static pthread_cond_t progress_cond;
1842
1843/*
1844 * Mutex and conditional variable can't be statically-initialized on Windows.
1845 */
1846static void init_threaded_search(void)
1847{
93749194 1848 init_recursive_mutex(&read_mutex);
44626dc7
AH
1849 pthread_mutex_init(&cache_mutex, NULL);
1850 pthread_mutex_init(&progress_mutex, NULL);
1851 pthread_cond_init(&progress_cond, NULL);
851c34b0 1852 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
1853}
1854
1855static void cleanup_threaded_search(void)
1856{
851c34b0 1857 set_try_to_free_routine(old_try_to_free_routine);
44626dc7
AH
1858 pthread_cond_destroy(&progress_cond);
1859 pthread_mutex_destroy(&read_mutex);
1860 pthread_mutex_destroy(&cache_mutex);
1861 pthread_mutex_destroy(&progress_mutex);
1862}
c2a33679 1863
8ecce684
NP
1864static void *threaded_find_deltas(void *arg)
1865{
c2a33679
NP
1866 struct thread_params *me = arg;
1867
50f22ada 1868 while (me->remaining) {
384b32c0 1869 find_deltas(me->list, &me->remaining,
c2a33679 1870 me->window, me->depth, me->processed);
50f22ada
JS
1871
1872 progress_lock();
1873 me->working = 0;
1874 pthread_cond_signal(&progress_cond);
1875 progress_unlock();
1876
1877 /*
1878 * We must not set ->data_ready before we wait on the
1879 * condition because the main thread may have set it to 1
1880 * before we get here. In order to be sure that new
1881 * work is available if we see 1 in ->data_ready, it
1882 * was initialized to 0 before this thread was spawned
1883 * and we reset it to 0 right away.
1884 */
1885 pthread_mutex_lock(&me->mutex);
1886 while (!me->data_ready)
1887 pthread_cond_wait(&me->cond, &me->mutex);
1888 me->data_ready = 0;
1889 pthread_mutex_unlock(&me->mutex);
c2a33679 1890 }
50f22ada
JS
1891 /* leave ->working 1 so that this doesn't get more work assigned */
1892 return NULL;
8ecce684
NP
1893}
1894
8ecce684
NP
1895static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1896 int window, int depth, unsigned *processed)
1897{
dcda3614 1898 struct thread_params *p;
384b32c0 1899 int i, ret, active_threads = 0;
c2a33679 1900
44626dc7
AH
1901 init_threaded_search();
1902
ef055571
NP
1903 if (!delta_search_threads) /* --threads=0 means autodetect */
1904 delta_search_threads = online_cpus();
367f4a43 1905 if (delta_search_threads <= 1) {
384b32c0 1906 find_deltas(list, &list_size, window, depth, processed);
44626dc7 1907 cleanup_threaded_search();
367f4a43
NP
1908 return;
1909 }
43cc2b42 1910 if (progress > pack_to_stdout)
b6c29915 1911 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1912 delta_search_threads);
dcda3614 1913 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 1914
50f22ada 1915 /* Partition the work amongst work threads. */
367f4a43 1916 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1917 unsigned sub_size = list_size / (delta_search_threads - i);
1918
bf874896
NP
1919 /* don't use too small segments or no deltas will be found */
1920 if (sub_size < 2*window && i+1 < delta_search_threads)
1921 sub_size = 0;
1922
8ecce684
NP
1923 p[i].window = window;
1924 p[i].depth = depth;
1925 p[i].processed = processed;
50f22ada
JS
1926 p[i].working = 1;
1927 p[i].data_ready = 0;
c2a33679 1928
59921b4b 1929 /* try to split chunks on "path" boundaries */
6fc74703
NP
1930 while (sub_size && sub_size < list_size &&
1931 list[sub_size]->hash &&
384b32c0
NP
1932 list[sub_size]->hash == list[sub_size-1]->hash)
1933 sub_size++;
1934
50f22ada
JS
1935 p[i].list = list;
1936 p[i].list_size = sub_size;
1937 p[i].remaining = sub_size;
59921b4b 1938
384b32c0
NP
1939 list += sub_size;
1940 list_size -= sub_size;
1941 }
1942
50f22ada
JS
1943 /* Start work threads. */
1944 for (i = 0; i < delta_search_threads; i++) {
1945 if (!p[i].list_size)
1946 continue;
68e6a4f8
JS
1947 pthread_mutex_init(&p[i].mutex, NULL);
1948 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
1949 ret = pthread_create(&p[i].thread, NULL,
1950 threaded_find_deltas, &p[i]);
1951 if (ret)
1952 die("unable to create thread: %s", strerror(ret));
1953 active_threads++;
1954 }
1955
384b32c0
NP
1956 /*
1957 * Now let's wait for work completion. Each time a thread is done
1958 * with its work, we steal half of the remaining work from the
1959 * thread with the largest number of unprocessed objects and give
1960 * it to that newly idle thread. This ensure good load balancing
1961 * until the remaining object list segments are simply too short
1962 * to be worth splitting anymore.
1963 */
50f22ada
JS
1964 while (active_threads) {
1965 struct thread_params *target = NULL;
384b32c0
NP
1966 struct thread_params *victim = NULL;
1967 unsigned sub_size = 0;
384b32c0
NP
1968
1969 progress_lock();
50f22ada
JS
1970 for (;;) {
1971 for (i = 0; !target && i < delta_search_threads; i++)
1972 if (!p[i].working)
1973 target = &p[i];
1974 if (target)
1975 break;
1976 pthread_cond_wait(&progress_cond, &progress_mutex);
1977 }
1978
384b32c0
NP
1979 for (i = 0; i < delta_search_threads; i++)
1980 if (p[i].remaining > 2*window &&
1981 (!victim || victim->remaining < p[i].remaining))
1982 victim = &p[i];
1983 if (victim) {
1984 sub_size = victim->remaining / 2;
1985 list = victim->list + victim->list_size - sub_size;
1986 while (sub_size && list[0]->hash &&
1987 list[0]->hash == list[-1]->hash) {
1988 list++;
1989 sub_size--;
1990 }
eb9688ff
NP
1991 if (!sub_size) {
1992 /*
1993 * It is possible for some "paths" to have
1994 * so many objects that no hash boundary
1995 * might be found. Let's just steal the
1996 * exact half in that case.
1997 */
1998 sub_size = victim->remaining / 2;
1999 list -= sub_size;
2000 }
384b32c0
NP
2001 target->list = list;
2002 victim->list_size -= sub_size;
2003 victim->remaining -= sub_size;
2004 }
384b32c0
NP
2005 target->list_size = sub_size;
2006 target->remaining = sub_size;
50f22ada
JS
2007 target->working = 1;
2008 progress_unlock();
2009
2010 pthread_mutex_lock(&target->mutex);
2011 target->data_ready = 1;
2012 pthread_cond_signal(&target->cond);
2013 pthread_mutex_unlock(&target->mutex);
c2a33679 2014
384b32c0 2015 if (!sub_size) {
b81d9af7 2016 pthread_join(target->thread, NULL);
50f22ada
JS
2017 pthread_cond_destroy(&target->cond);
2018 pthread_mutex_destroy(&target->mutex);
384b32c0 2019 active_threads--;
c2a33679 2020 }
50f22ada 2021 }
44626dc7 2022 cleanup_threaded_search();
dcda3614 2023 free(p);
8ecce684
NP
2024}
2025
2026#else
384b32c0 2027#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
2028#endif
2029
f0a24aa5
SP
2030static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
2031{
2032 unsigned char peeled[20];
2033
2034 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
2035 !peel_ref(path, peeled) && /* peelable? */
2036 !is_null_sha1(peeled) && /* annotated tag? */
2037 locate_object_entry(peeled)) /* object packed? */
2038 add_object_entry(sha1, OBJ_TAG, NULL, 0);
2039 return 0;
2040}
2041
f3123c4a
JH
2042static void prepare_pack(int window, int depth)
2043{
9668cf59 2044 struct object_entry **delta_list;
6e1c2344
RJ
2045 uint32_t i, nr_deltas;
2046 unsigned n;
9668cf59 2047
3f9ac8d2 2048 get_object_details();
9668cf59 2049
0e8189e2
NP
2050 /*
2051 * If we're locally repacking then we need to be doubly careful
2052 * from now on in order to make sure no stealth corruption gets
2053 * propagated to the new pack. Clients receiving streamed packs
2054 * should validate everything they get anyway so no need to incur
2055 * the additional cost here in that case.
2056 */
2057 if (!pack_to_stdout)
2058 do_check_packed_object_crc = 1;
2059
75d39853 2060 if (!nr_objects || !window || !depth)
9668cf59
NP
2061 return;
2062
2063 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
2064 nr_deltas = n = 0;
2065
2066 for (i = 0; i < nr_objects; i++) {
2067 struct object_entry *entry = objects + i;
2068
2069 if (entry->delta)
2070 /* This happens if we decided to reuse existing
a7de7130 2071 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2072 */
2073 continue;
2074
2075 if (entry->size < 50)
2076 continue;
2077
2078 if (entry->no_try_delta)
2079 continue;
2080
6d6f9cdd 2081 if (!entry->preferred_base) {
75d39853 2082 nr_deltas++;
6d6f9cdd
SP
2083 if (entry->type < 0)
2084 die("unable to get type of object %s",
2085 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
2086 } else {
2087 if (entry->type < 0) {
2088 /*
2089 * This object is not found, but we
2090 * don't have to include it anyway.
2091 */
2092 continue;
2093 }
6d6f9cdd 2094 }
75d39853
NP
2095
2096 delta_list[n++] = entry;
2097 }
2098
2f8b8947 2099 if (nr_deltas && n > 1) {
e334977d
NP
2100 unsigned nr_done = 0;
2101 if (progress)
dc6a0757
NP
2102 progress_state = start_progress("Compressing objects",
2103 nr_deltas);
75d39853 2104 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 2105 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2106 stop_progress(&progress_state);
e334977d
NP
2107 if (nr_done != nr_deltas)
2108 die("inconsistency with delta count");
75d39853 2109 }
9668cf59 2110 free(delta_list);
f3123c4a
JH
2111}
2112
ef90d6d4 2113static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2114{
eeefa7c9 2115 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2116 window = git_config_int(k, v);
2117 return 0;
2118 }
a97773ce
BD
2119 if (!strcmp(k, "pack.windowmemory")) {
2120 window_memory_limit = git_config_ulong(k, v);
2121 return 0;
2122 }
2123 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2124 depth = git_config_int(k, v);
2125 return 0;
2126 }
960ccca6
DH
2127 if (!strcmp(k, "pack.compression")) {
2128 int level = git_config_int(k, v);
2129 if (level == -1)
2130 level = Z_DEFAULT_COMPRESSION;
2131 else if (level < 0 || level > Z_BEST_COMPRESSION)
2132 die("bad pack compression level %d", level);
2133 pack_compression_level = level;
2134 pack_compression_seen = 1;
2135 return 0;
2136 }
074b2eea
MK
2137 if (!strcmp(k, "pack.deltacachesize")) {
2138 max_delta_cache_size = git_config_int(k, v);
2139 return 0;
2140 }
e3dfddb3
MK
2141 if (!strcmp(k, "pack.deltacachelimit")) {
2142 cache_max_small_delta_size = git_config_int(k, v);
2143 return 0;
2144 }
693b86ff
NP
2145 if (!strcmp(k, "pack.threads")) {
2146 delta_search_threads = git_config_int(k, v);
833e3df1 2147 if (delta_search_threads < 0)
693b86ff
NP
2148 die("invalid number of threads specified (%d)",
2149 delta_search_threads);
7eb151d6 2150#ifdef NO_PTHREADS
833e3df1 2151 if (delta_search_threads != 1)
693b86ff
NP
2152 warning("no threads support, ignoring %s", k);
2153#endif
2154 return 0;
2155 }
4d00bda2 2156 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2157 pack_idx_opts.version = git_config_int(k, v);
2158 if (pack_idx_opts.version > 2)
6e1c2344 2159 die("bad pack.indexversion=%"PRIu32,
ebcfb379 2160 pack_idx_opts.version);
4d00bda2
NP
2161 return 0;
2162 }
ef90d6d4 2163 return git_default_config(k, v, cb);
4812a93a
JK
2164}
2165
b5d97e6b 2166static void read_object_list_from_stdin(void)
c323ac7d 2167{
b5d97e6b
JH
2168 char line[40 + 1 + PATH_MAX + 2];
2169 unsigned char sha1[20];
b2504a0d 2170
da93d12b 2171 for (;;) {
da93d12b
LT
2172 if (!fgets(line, sizeof(line), stdin)) {
2173 if (feof(stdin))
2174 break;
2175 if (!ferror(stdin))
2176 die("fgets returned NULL, not EOF, not error!");
687dd75c 2177 if (errno != EINTR)
d824cbba 2178 die_errno("fgets");
687dd75c
JH
2179 clearerr(stdin);
2180 continue;
da93d12b 2181 }
7a979d99
JH
2182 if (line[0] == '-') {
2183 if (get_sha1_hex(line+1, sha1))
2184 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 2185 line);
8d1d8f83 2186 add_preferred_base(sha1);
7a979d99 2187 continue;
21fcd1bd 2188 }
c323ac7d 2189 if (get_sha1_hex(line, sha1))
ef07618f 2190 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 2191
bc32fed5
JH
2192 add_preferred_base_object(line+41);
2193 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 2194 }
b5d97e6b
JH
2195}
2196
08cdfb13
JH
2197#define OBJECT_ADDED (1u<<20)
2198
11c211fa 2199static void show_commit(struct commit *commit, void *data)
b5d97e6b 2200{
bc32fed5 2201 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 2202 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
2203}
2204
49473672
JH
2205static void show_object(struct object *obj,
2206 const struct name_path *path, const char *last,
2207 void *data)
b5d97e6b 2208{
cf2ab916
LT
2209 char *name = path_name(path, last);
2210
8d2dfc49
LT
2211 add_preferred_base_object(name);
2212 add_object_entry(obj->sha1, obj->type, name, 0);
2213 obj->flags |= OBJECT_ADDED;
2214
2215 /*
2216 * We will have generated the hash from the name,
2217 * but not saved a pointer to it - we can free it
2218 */
2219 free((char *)name);
b5d97e6b
JH
2220}
2221
8d1d8f83
JH
2222static void show_edge(struct commit *commit)
2223{
2224 add_preferred_base(commit->object.sha1);
2225}
2226
08cdfb13
JH
2227struct in_pack_object {
2228 off_t offset;
2229 struct object *object;
2230};
2231
2232struct in_pack {
2233 int alloc;
2234 int nr;
2235 struct in_pack_object *array;
2236};
2237
2238static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2239{
2240 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
2241 in_pack->array[in_pack->nr].object = object;
2242 in_pack->nr++;
2243}
2244
2245/*
2246 * Compare the objects in the offset order, in order to emulate the
f18d244a 2247 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2248 */
2249static int ofscmp(const void *a_, const void *b_)
2250{
2251 struct in_pack_object *a = (struct in_pack_object *)a_;
2252 struct in_pack_object *b = (struct in_pack_object *)b_;
2253
2254 if (a->offset < b->offset)
2255 return -1;
2256 else if (a->offset > b->offset)
2257 return 1;
2258 else
2259 return hashcmp(a->object->sha1, b->object->sha1);
2260}
2261
2262static void add_objects_in_unpacked_packs(struct rev_info *revs)
2263{
2264 struct packed_git *p;
2265 struct in_pack in_pack;
2266 uint32_t i;
2267
2268 memset(&in_pack, 0, sizeof(in_pack));
2269
2270 for (p = packed_git; p; p = p->next) {
2271 const unsigned char *sha1;
2272 struct object *o;
2273
79bc4c71 2274 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2275 continue;
2276 if (open_pack_index(p))
2277 die("cannot open pack index");
2278
2279 ALLOC_GROW(in_pack.array,
2280 in_pack.nr + p->num_objects,
2281 in_pack.alloc);
2282
2283 for (i = 0; i < p->num_objects; i++) {
2284 sha1 = nth_packed_object_sha1(p, i);
2285 o = lookup_unknown_object(sha1);
2286 if (!(o->flags & OBJECT_ADDED))
2287 mark_in_pack_object(o, p, &in_pack);
2288 o->flags |= OBJECT_ADDED;
2289 }
2290 }
2291
2292 if (in_pack.nr) {
2293 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2294 ofscmp);
2295 for (i = 0; i < in_pack.nr; i++) {
2296 struct object *o = in_pack.array[i].object;
2297 add_object_entry(o->sha1, o->type, "", 0);
2298 }
2299 }
2300 free(in_pack.array);
2301}
2302
094085e3
BC
2303static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2304{
2305 static struct packed_git *last_found = (void *)1;
2306 struct packed_git *p;
2307
2308 p = (last_found != (void *)1) ? last_found : packed_git;
2309
2310 while (p) {
2311 if ((!p->pack_local || p->pack_keep) &&
2312 find_pack_entry_one(sha1, p)) {
2313 last_found = p;
2314 return 1;
2315 }
2316 if (p == last_found)
2317 p = packed_git;
2318 else
2319 p = p->next;
2320 if (p == last_found)
2321 p = p->next;
2322 }
2323 return 0;
2324}
2325
ca11b212
NP
2326static void loosen_unused_packed_objects(struct rev_info *revs)
2327{
2328 struct packed_git *p;
2329 uint32_t i;
2330 const unsigned char *sha1;
2331
2332 for (p = packed_git; p; p = p->next) {
79bc4c71 2333 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2334 continue;
2335
7e52f566
JK
2336 if (unpack_unreachable_expiration &&
2337 p->mtime < unpack_unreachable_expiration)
2338 continue;
2339
ca11b212
NP
2340 if (open_pack_index(p))
2341 die("cannot open pack index");
2342
2343 for (i = 0; i < p->num_objects; i++) {
2344 sha1 = nth_packed_object_sha1(p, i);
094085e3
BC
2345 if (!locate_object_entry(sha1) &&
2346 !has_sha1_pack_kept_or_nonlocal(sha1))
ca11b212
NP
2347 if (force_object_loose(sha1, p->mtime))
2348 die("unable to force loose object");
2349 }
2350 }
2351}
2352
8d1d8f83 2353static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2354{
2355 struct rev_info revs;
2356 char line[1000];
b5d97e6b
JH
2357 int flags = 0;
2358
b5d97e6b
JH
2359 init_revisions(&revs, NULL);
2360 save_commit_buffer = 0;
b5d97e6b
JH
2361 setup_revisions(ac, av, &revs, NULL);
2362
b5d97e6b
JH
2363 while (fgets(line, sizeof(line), stdin) != NULL) {
2364 int len = strlen(line);
872c930d 2365 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2366 line[--len] = 0;
2367 if (!len)
2368 break;
2369 if (*line == '-') {
2370 if (!strcmp(line, "--not")) {
2371 flags ^= UNINTERESTING;
2372 continue;
2373 }
2374 die("not a rev '%s'", line);
2375 }
8e676e8b 2376 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
b5d97e6b
JH
2377 die("bad revision '%s'", line);
2378 }
2379
3d51e1b5
MK
2380 if (prepare_revision_walk(&revs))
2381 die("revision walk setup failed");
8d1d8f83 2382 mark_edges_uninteresting(revs.commits, &revs, show_edge);
11c211fa 2383 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13
JH
2384
2385 if (keep_unreachable)
2386 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2387 if (unpack_unreachable)
2388 loosen_unused_packed_objects(&revs);
b5d97e6b
JH
2389}
2390
99fb6e04
NTND
2391static int option_parse_index_version(const struct option *opt,
2392 const char *arg, int unset)
2393{
2394 char *c;
2395 const char *val = arg;
2396 pack_idx_opts.version = strtoul(val, &c, 10);
2397 if (pack_idx_opts.version > 2)
2398 die(_("unsupported index version %s"), val);
2399 if (*c == ',' && c[1])
2400 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2401 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2402 die(_("bad index version '%s'"), val);
2403 return 0;
2404}
2405
7e52f566
JK
2406static int option_parse_unpack_unreachable(const struct option *opt,
2407 const char *arg, int unset)
2408{
2409 if (unset) {
2410 unpack_unreachable = 0;
2411 unpack_unreachable_expiration = 0;
2412 }
2413 else {
2414 unpack_unreachable = 1;
2415 if (arg)
2416 unpack_unreachable_expiration = approxidate(arg);
2417 }
2418 return 0;
2419}
2420
99fb6e04
NTND
2421static int option_parse_ulong(const struct option *opt,
2422 const char *arg, int unset)
2423{
2424 if (unset)
2425 die(_("option %s does not accept negative form"),
2426 opt->long_name);
2427
2428 if (!git_parse_ulong(arg, opt->value))
2429 die(_("unable to parse value '%s' for option %s"),
2430 arg, opt->long_name);
2431 return 0;
2432}
2433
2434#define OPT_ULONG(s, l, v, h) \
2435 { OPTION_CALLBACK, (s), (l), (v), "n", (h), \
2436 PARSE_OPT_NONEG, option_parse_ulong }
2437
b5d97e6b
JH
2438int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2439{
b5d97e6b 2440 int use_internal_rev_list = 0;
8d1d8f83 2441 int thin = 0;
4f366275 2442 int all_progress_implied = 0;
99fb6e04
NTND
2443 const char *rp_av[6];
2444 int rp_ac = 0;
2445 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2446 struct option pack_objects_options[] = {
2447 OPT_SET_INT('q', "quiet", &progress,
2448 "do not show progress meter", 0),
2449 OPT_SET_INT(0, "progress", &progress,
2450 "show progress meter", 1),
2451 OPT_SET_INT(0, "all-progress", &progress,
2452 "show progress meter during object writing phase", 2),
2453 OPT_BOOL(0, "all-progress-implied",
2454 &all_progress_implied,
2455 "similar to --all-progress when progress meter is shown"),
2456 { OPTION_CALLBACK, 0, "index-version", NULL, "version[,offset]",
2457 "write the pack index file in the specified idx format version",
2458 0, option_parse_index_version },
2459 OPT_ULONG(0, "max-pack-size", &pack_size_limit,
2460 "maximum size of each output pack file"),
2461 OPT_BOOL(0, "local", &local,
2462 "ignore borrowed objects from alternate object store"),
2463 OPT_BOOL(0, "incremental", &incremental,
2464 "ignore packed objects"),
2465 OPT_INTEGER(0, "window", &window,
2466 "limit pack window by objects"),
2467 OPT_ULONG(0, "window-memory", &window_memory_limit,
2468 "limit pack window by memory in addition to object limit"),
2469 OPT_INTEGER(0, "depth", &depth,
2470 "maximum length of delta chain allowed in the resulting pack"),
2471 OPT_BOOL(0, "reuse-delta", &reuse_delta,
2472 "reuse existing deltas"),
2473 OPT_BOOL(0, "reuse-object", &reuse_object,
2474 "reuse existing objects"),
2475 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2476 "use OFS_DELTA objects"),
2477 OPT_INTEGER(0, "threads", &delta_search_threads,
2478 "use threads when searching for best delta matches"),
2479 OPT_BOOL(0, "non-empty", &non_empty,
2480 "do not create an empty pack output"),
2481 OPT_BOOL(0, "revs", &use_internal_rev_list,
2482 "read revision arguments from standard input"),
2483 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2484 "limit the objects to those that are not yet packed",
2485 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2486 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2487 "include objects reachable from any reference",
2488 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2489 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2490 "include objects referred by reflog entries",
2491 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2492 OPT_BOOL(0, "stdout", &pack_to_stdout,
2493 "output pack to stdout"),
2494 OPT_BOOL(0, "include-tag", &include_tag,
2495 "include tag objects that refer to objects to be packed"),
2496 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2497 "keep unreachable objects"),
7e52f566
JK
2498 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, "time",
2499 "unpack unreachable objects newer than <time>",
2500 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
99fb6e04
NTND
2501 OPT_BOOL(0, "thin", &thin,
2502 "create thin packs"),
2503 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2504 "ignore packs that have companion .keep file"),
2505 OPT_INTEGER(0, "compression", &pack_compression_level,
2506 "pack compression level"),
2507 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2508 "do not hide commits by grafts", 0),
2509 OPT_END(),
2510 };
8d1d8f83 2511
dae556bd
CC
2512 read_replace_refs = 0;
2513
ebcfb379 2514 reset_pack_idx_option(&pack_idx_opts);
ef90d6d4 2515 git_config(git_pack_config, NULL);
960ccca6
DH
2516 if (!pack_compression_seen && core_compression_seen)
2517 pack_compression_level = core_compression_level;
b5d97e6b
JH
2518
2519 progress = isatty(2);
99fb6e04
NTND
2520 argc = parse_options(argc, argv, prefix, pack_objects_options,
2521 pack_usage, 0);
b5d97e6b 2522
99fb6e04
NTND
2523 if (argc) {
2524 base_name = argv[0];
2525 argc--;
b5d97e6b 2526 }
99fb6e04
NTND
2527 if (pack_to_stdout != !base_name || argc)
2528 usage_with_options(pack_usage, pack_objects_options);
b5d97e6b 2529
99fb6e04
NTND
2530 rp_av[rp_ac++] = "pack-objects";
2531 if (thin) {
2532 use_internal_rev_list = 1;
2533 rp_av[rp_ac++] = "--objects-edge";
2534 } else
2535 rp_av[rp_ac++] = "--objects";
b5d97e6b 2536
99fb6e04
NTND
2537 if (rev_list_all) {
2538 use_internal_rev_list = 1;
2539 rp_av[rp_ac++] = "--all";
2540 }
2541 if (rev_list_reflog) {
2542 use_internal_rev_list = 1;
2543 rp_av[rp_ac++] = "--reflog";
2544 }
2545 if (rev_list_unpacked) {
2546 use_internal_rev_list = 1;
2547 rp_av[rp_ac++] = "--unpacked";
2548 }
b5d97e6b 2549
99fb6e04
NTND
2550 if (!reuse_object)
2551 reuse_delta = 0;
2552 if (pack_compression_level == -1)
2553 pack_compression_level = Z_DEFAULT_COMPRESSION;
2554 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2555 die("bad pack compression level %d", pack_compression_level);
2556#ifdef NO_PTHREADS
2557 if (delta_search_threads != 1)
2b34e486 2558 warning("no threads support, ignoring --threads");
99fb6e04 2559#endif
2b84b5a8
JS
2560 if (!pack_to_stdout && !pack_size_limit)
2561 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
2562 if (pack_to_stdout && pack_size_limit)
2563 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
2564 if (pack_size_limit && pack_size_limit < 1024*1024) {
2565 warning("minimum pack size limit is 1 MiB");
2566 pack_size_limit = 1024*1024;
2567 }
01c12a23 2568
8d1d8f83
JH
2569 if (!pack_to_stdout && thin)
2570 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2571
ca11b212
NP
2572 if (keep_unreachable && unpack_unreachable)
2573 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2574
4f366275
NP
2575 if (progress && all_progress_implied)
2576 progress = 2;
2577
b5d97e6b
JH
2578 prepare_packed_git();
2579
13aaf148 2580 if (progress)
dc6a0757 2581 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2582 if (!use_internal_rev_list)
2583 read_object_list_from_stdin();
8d1d8f83
JH
2584 else {
2585 rp_av[rp_ac] = NULL;
2586 get_object_list(rp_ac, rp_av);
2587 }
0ef95f72 2588 cleanup_preferred_base();
f0a24aa5
SP
2589 if (include_tag && nr_result)
2590 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2591 stop_progress(&progress_state);
96a02f8f 2592
f0b0af1b 2593 if (non_empty && !nr_result)
1c4a2912 2594 return 0;
f7ae6a93
NP
2595 if (nr_result)
2596 prepare_pack(window, depth);
d01fb92f 2597 write_pack_file();
ab7cd7bb 2598 if (progress)
6e1c2344
RJ
2599 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2600 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2601 written, written_delta, reused, reused_delta);
c323ac7d
LT
2602 return 0;
2603}