rev-list: support termination at promisor objects
[git/git.git] / builtin / pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d 2#include "cache.h"
b2141fc1 3#include "config.h"
a74db82e 4#include "attr.h"
c323ac7d 5#include "object.h"
8e440259
PE
6#include "blob.h"
7#include "commit.h"
8#include "tag.h"
9#include "tree.h"
c323ac7d 10#include "delta.h"
a733cb60 11#include "pack.h"
3449f8c4 12#include "pack-revindex.h"
c38138cd 13#include "csum-file.h"
1b0c7174 14#include "tree-walk.h"
b5d97e6b
JH
15#include "diff.h"
16#include "revision.h"
17#include "list-objects.h"
9535ce73
JH
18#include "list-objects-filter.h"
19#include "list-objects-filter-options.h"
2834bc27 20#include "pack-objects.h"
96a02f8f 21#include "progress.h"
f0a24aa5 22#include "refs.h"
cf2ba13a 23#include "streaming.h"
93749194 24#include "thread-utils.h"
6b8fda2d 25#include "pack-bitmap.h"
abcb8655
JK
26#include "reachable.h"
27#include "sha1-array.h"
edfbb2aa 28#include "argv-array.h"
c9af708b 29#include "mru.h"
0317f455 30#include "packfile.h"
8ecce684 31
99fb6e04 32static const char *pack_usage[] = {
b8c1d275
AH
33 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
34 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
99fb6e04
NTND
35 NULL
36};
c323ac7d 37
3f9ac8d2 38/*
2834bc27
VM
39 * Objects we are going to pack are collected in the `to_pack` structure.
40 * It contains an array (dynamically expanded) of the object data, and a map
41 * that can resolve SHA1s to their position in the array.
3f9ac8d2 42 */
2834bc27
VM
43static struct packing_data to_pack;
44
79814f42 45static struct pack_idx_entry **written_list;
2834bc27 46static uint32_t nr_result, nr_written;
3f9ac8d2 47
96f1e58f 48static int non_empty;
a7de7130 49static int reuse_delta = 1, reuse_object = 1;
e5e9714a 50static int keep_unreachable, unpack_unreachable, include_tag;
dddbad72 51static timestamp_t unpack_unreachable_expiration;
e26a8c47 52static int pack_loose_unreachable;
96f1e58f 53static int local;
56dfeb62 54static int have_non_local_packs;
96f1e58f 55static int incremental;
e96fb9b8 56static int ignore_packed_keep;
be6b1914 57static int allow_ofs_delta;
ebcfb379 58static struct pack_idx_option pack_idx_opts;
d01fb92f 59static const char *base_name;
024701f1 60static int progress = 1;
4812a93a 61static int window = 10;
568508e7 62static unsigned long pack_size_limit;
618e613a 63static int depth = 50;
43cc2b42 64static int delta_search_threads;
df6d6101 65static int pack_to_stdout;
8d1d8f83 66static int num_preferred_base;
dc6a0757 67static struct progress *progress_state;
c323ac7d 68
6b8fda2d
VM
69static struct packed_git *reuse_packfile;
70static uint32_t reuse_packfile_objects;
71static off_t reuse_packfile_offset;
72
645c432d
KS
73static int use_bitmap_index_default = 1;
74static int use_bitmap_index = -1;
7cc8f971 75static int write_bitmap_index;
ae4f07fb 76static uint16_t write_bitmap_options;
6b8fda2d 77
074b2eea 78static unsigned long delta_cache_size = 0;
5749b0b2 79static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
e3dfddb3 80static unsigned long cache_max_small_delta_size = 1000;
074b2eea 81
a97773ce
BD
82static unsigned long window_memory_limit = 0;
83
9535ce73
JH
84static struct list_objects_filter_options filter_options;
85
86enum missing_action {
87 MA_ERROR = 0, /* fail if any missing objects are encountered */
88 MA_ALLOW_ANY, /* silently allow ALL missing objects */
89};
90static enum missing_action arg_missing_action;
91static show_object_fn fn_show_object;
92
3f9ac8d2
JH
93/*
94 * stats
95 */
7cadf491
SP
96static uint32_t written, written_delta;
97static uint32_t reused, reused_delta;
3f9ac8d2 98
7cc8f971
VM
99/*
100 * Indexed commits
101 */
102static struct commit **indexed_commits;
103static unsigned int indexed_commits_nr;
104static unsigned int indexed_commits_alloc;
105
106static void index_commit_for_bitmap(struct commit *commit)
107{
108 if (indexed_commits_nr >= indexed_commits_alloc) {
109 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
2756ca43 110 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
7cc8f971
VM
111 }
112
113 indexed_commits[indexed_commits_nr++] = commit;
114}
780e6e73 115
3613f9b4 116static void *get_delta(struct object_entry *entry)
c323ac7d 117{
3613f9b4
NP
118 unsigned long size, base_size, delta_size;
119 void *buf, *base_buf, *delta_buf;
21666f1a 120 enum object_type type;
c323ac7d 121
e6a492b7 122 buf = read_sha1_file(entry->idx.oid.hash, &type, &size);
3613f9b4 123 if (!buf)
e6a492b7 124 die("unable to read %s", oid_to_hex(&entry->idx.oid));
125 base_buf = read_sha1_file(entry->delta->idx.oid.hash, &type,
126 &base_size);
3613f9b4 127 if (!base_buf)
e6a492b7 128 die("unable to read %s",
129 oid_to_hex(&entry->delta->idx.oid));
3613f9b4 130 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 131 buf, size, &delta_size, 0);
3613f9b4 132 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 133 die("delta size changed");
3613f9b4
NP
134 free(buf);
135 free(base_buf);
c323ac7d
LT
136 return delta_buf;
137}
138
30ebb40a
NP
139static unsigned long do_compress(void **pptr, unsigned long size)
140{
ef49a7a0 141 git_zstream stream;
30ebb40a
NP
142 void *in, *out;
143 unsigned long maxsize;
144
55bb5c91 145 git_deflate_init(&stream, pack_compression_level);
225a6f10 146 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
147
148 in = *pptr;
149 out = xmalloc(maxsize);
150 *pptr = out;
151
152 stream.next_in = in;
153 stream.avail_in = size;
154 stream.next_out = out;
155 stream.avail_out = maxsize;
55bb5c91 156 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 157 ; /* nothing */
55bb5c91 158 git_deflate_end(&stream);
30ebb40a
NP
159
160 free(in);
161 return stream.total_out;
162}
163
cf2ba13a
NTND
164static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
165 const unsigned char *sha1)
166{
167 git_zstream stream;
168 unsigned char ibuf[1024 * 16];
169 unsigned char obuf[1024 * 16];
170 unsigned long olen = 0;
171
cf2ba13a
NTND
172 git_deflate_init(&stream, pack_compression_level);
173
174 for (;;) {
175 ssize_t readlen;
176 int zret = Z_OK;
177 readlen = read_istream(st, ibuf, sizeof(ibuf));
178 if (readlen == -1)
179 die(_("unable to read %s"), sha1_to_hex(sha1));
180
181 stream.next_in = ibuf;
182 stream.avail_in = readlen;
183 while ((stream.avail_in || readlen == 0) &&
184 (zret == Z_OK || zret == Z_BUF_ERROR)) {
185 stream.next_out = obuf;
186 stream.avail_out = sizeof(obuf);
187 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
188 sha1write(f, obuf, stream.next_out - obuf);
189 olen += stream.next_out - obuf;
190 }
191 if (stream.avail_in)
192 die(_("deflate error (%d)"), zret);
193 if (readlen == 0) {
194 if (zret != Z_STREAM_END)
195 die(_("deflate error (%d)"), zret);
196 break;
197 }
198 }
199 git_deflate_end(&stream);
200 return olen;
201}
202
780e6e73
NP
203/*
204 * we are going to reuse the existing object data as is. make
205 * sure it is not corrupt.
206 */
079afb18
SP
207static int check_pack_inflate(struct packed_git *p,
208 struct pack_window **w_curs,
6777a59f
SP
209 off_t offset,
210 off_t len,
079afb18
SP
211 unsigned long expect)
212{
ef49a7a0 213 git_zstream stream;
079afb18
SP
214 unsigned char fakebuf[4096], *in;
215 int st;
216
217 memset(&stream, 0, sizeof(stream));
39c68542 218 git_inflate_init(&stream);
079afb18
SP
219 do {
220 in = use_pack(p, w_curs, offset, &stream.avail_in);
221 stream.next_in = in;
222 stream.next_out = fakebuf;
223 stream.avail_out = sizeof(fakebuf);
39c68542 224 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
225 offset += stream.next_in - in;
226 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 227 git_inflate_end(&stream);
079afb18
SP
228 return (st == Z_STREAM_END &&
229 stream.total_out == expect &&
230 stream.total_in == len) ? 0 : -1;
231}
232
233static void copy_pack_data(struct sha1file *f,
234 struct packed_git *p,
235 struct pack_window **w_curs,
6777a59f
SP
236 off_t offset,
237 off_t len)
079afb18
SP
238{
239 unsigned char *in;
ef49a7a0 240 unsigned long avail;
079afb18
SP
241
242 while (len) {
243 in = use_pack(p, w_curs, offset, &avail);
244 if (avail > len)
ef49a7a0 245 avail = (unsigned long)len;
079afb18
SP
246 sha1write(f, in, avail);
247 offset += avail;
248 len -= avail;
249 }
250}
251
1b4bb16b 252/* Return 0 if we will bust the pack-size limit */
c9018b03
NTND
253static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
254 unsigned long limit, int usable_delta)
c323ac7d 255{
c9018b03 256 unsigned long size, datalen;
2c5e2865
JK
257 unsigned char header[MAX_PACK_OBJECT_HEADER],
258 dheader[MAX_PACK_OBJECT_HEADER];
6777a59f 259 unsigned hdrlen;
2c5ef824 260 enum object_type type;
c9018b03 261 void *buf;
cf2ba13a 262 struct git_istream *st = NULL;
c9018b03
NTND
263
264 if (!usable_delta) {
cf2ba13a
NTND
265 if (entry->type == OBJ_BLOB &&
266 entry->size > big_file_threshold &&
e6a492b7 267 (st = open_istream(entry->idx.oid.hash, &type, &size, NULL)) != NULL)
cf2ba13a
NTND
268 buf = NULL;
269 else {
e6a492b7 270 buf = read_sha1_file(entry->idx.oid.hash, &type,
271 &size);
cf2ba13a 272 if (!buf)
e6a492b7 273 die(_("unable to read %s"),
274 oid_to_hex(&entry->idx.oid));
cf2ba13a 275 }
c9018b03
NTND
276 /*
277 * make sure no cached delta data remains from a
278 * previous attempt before a pack split occurred.
279 */
6a83d902 280 FREE_AND_NULL(entry->delta_data);
c9018b03
NTND
281 entry->z_delta_size = 0;
282 } else if (entry->delta_data) {
283 size = entry->delta_size;
284 buf = entry->delta_data;
285 entry->delta_data = NULL;
286 type = (allow_ofs_delta && entry->delta->idx.offset) ?
287 OBJ_OFS_DELTA : OBJ_REF_DELTA;
288 } else {
289 buf = get_delta(entry);
290 size = entry->delta_size;
291 type = (allow_ofs_delta && entry->delta->idx.offset) ?
292 OBJ_OFS_DELTA : OBJ_REF_DELTA;
293 }
294
cf2ba13a
NTND
295 if (st) /* large blob case, just assume we don't compress well */
296 datalen = size;
297 else if (entry->z_delta_size)
c9018b03
NTND
298 datalen = entry->z_delta_size;
299 else
300 datalen = do_compress(&buf, size);
301
302 /*
303 * The object header is a byte of 'type' followed by zero or
304 * more bytes of length.
305 */
7202a6fa
JK
306 hdrlen = encode_in_pack_object_header(header, sizeof(header),
307 type, size);
c9018b03
NTND
308
309 if (type == OBJ_OFS_DELTA) {
310 /*
311 * Deltas with relative base contain an additional
312 * encoding of the relative offset for the delta
313 * base from this object's position in the pack.
314 */
315 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
316 unsigned pos = sizeof(dheader) - 1;
317 dheader[pos] = ofs & 127;
318 while (ofs >>= 7)
319 dheader[--pos] = 128 | (--ofs & 127);
320 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
cf2ba13a
NTND
321 if (st)
322 close_istream(st);
c9018b03
NTND
323 free(buf);
324 return 0;
325 }
326 sha1write(f, header, hdrlen);
327 sha1write(f, dheader + pos, sizeof(dheader) - pos);
328 hdrlen += sizeof(dheader) - pos;
329 } else if (type == OBJ_REF_DELTA) {
330 /*
331 * Deltas with a base reference contain
332 * an additional 20 bytes for the base sha1.
333 */
334 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
cf2ba13a
NTND
335 if (st)
336 close_istream(st);
c9018b03
NTND
337 free(buf);
338 return 0;
339 }
340 sha1write(f, header, hdrlen);
e6a492b7 341 sha1write(f, entry->delta->idx.oid.hash, 20);
c9018b03
NTND
342 hdrlen += 20;
343 } else {
344 if (limit && hdrlen + datalen + 20 >= limit) {
cf2ba13a
NTND
345 if (st)
346 close_istream(st);
c9018b03
NTND
347 free(buf);
348 return 0;
349 }
350 sha1write(f, header, hdrlen);
351 }
cf2ba13a 352 if (st) {
e6a492b7 353 datalen = write_large_blob_data(st, f, entry->idx.oid.hash);
cf2ba13a
NTND
354 close_istream(st);
355 } else {
356 sha1write(f, buf, datalen);
357 free(buf);
358 }
c9018b03
NTND
359
360 return hdrlen + datalen;
361}
362
363/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
364static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
365 unsigned long limit, int usable_delta)
c9018b03
NTND
366{
367 struct packed_git *p = entry->in_pack;
368 struct pack_window *w_curs = NULL;
369 struct revindex_entry *revidx;
370 off_t offset;
371 enum object_type type = entry->type;
211c61c6 372 off_t datalen;
2c5e2865
JK
373 unsigned char header[MAX_PACK_OBJECT_HEADER],
374 dheader[MAX_PACK_OBJECT_HEADER];
c9018b03
NTND
375 unsigned hdrlen;
376
377 if (entry->delta)
378 type = (allow_ofs_delta && entry->delta->idx.offset) ?
379 OBJ_OFS_DELTA : OBJ_REF_DELTA;
7202a6fa
JK
380 hdrlen = encode_in_pack_object_header(header, sizeof(header),
381 type, entry->size);
c9018b03
NTND
382
383 offset = entry->in_pack_offset;
384 revidx = find_pack_revindex(p, offset);
385 datalen = revidx[1].offset - offset;
386 if (!pack_to_stdout && p->index_version > 1 &&
387 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
e6a492b7 388 error("bad packed object CRC for %s",
389 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
390 unuse_pack(&w_curs);
391 return write_no_reuse_object(f, entry, limit, usable_delta);
392 }
393
394 offset += entry->in_pack_header_size;
395 datalen -= entry->in_pack_header_size;
396
397 if (!pack_to_stdout && p->index_version == 1 &&
398 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
e6a492b7 399 error("corrupt packed object for %s",
400 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
401 unuse_pack(&w_curs);
402 return write_no_reuse_object(f, entry, limit, usable_delta);
403 }
404
405 if (type == OBJ_OFS_DELTA) {
406 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
407 unsigned pos = sizeof(dheader) - 1;
408 dheader[pos] = ofs & 127;
409 while (ofs >>= 7)
410 dheader[--pos] = 128 | (--ofs & 127);
411 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
412 unuse_pack(&w_curs);
413 return 0;
414 }
415 sha1write(f, header, hdrlen);
416 sha1write(f, dheader + pos, sizeof(dheader) - pos);
417 hdrlen += sizeof(dheader) - pos;
418 reused_delta++;
419 } else if (type == OBJ_REF_DELTA) {
420 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
421 unuse_pack(&w_curs);
422 return 0;
423 }
424 sha1write(f, header, hdrlen);
e6a492b7 425 sha1write(f, entry->delta->idx.oid.hash, 20);
c9018b03
NTND
426 hdrlen += 20;
427 reused_delta++;
428 } else {
429 if (limit && hdrlen + datalen + 20 >= limit) {
430 unuse_pack(&w_curs);
431 return 0;
432 }
433 sha1write(f, header, hdrlen);
434 }
435 copy_pack_data(f, p, &w_curs, offset, datalen);
436 unuse_pack(&w_curs);
437 reused++;
438 return hdrlen + datalen;
439}
440
441/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
442static off_t write_object(struct sha1file *f,
443 struct object_entry *entry,
444 off_t write_offset)
c9018b03 445{
af92a645
NTND
446 unsigned long limit;
447 off_t len;
2c5ef824 448 int usable_delta, to_reuse;
c323ac7d 449
78d1e84f
NP
450 if (!pack_to_stdout)
451 crc32_begin(f);
452
a2430dde 453 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
454 if (!pack_size_limit || !nr_written)
455 limit = 0;
456 else if (pack_size_limit <= write_offset)
457 /*
458 * the earlier object did not fit the limit; avoid
459 * mistaking this with unlimited (i.e. limit = 0).
460 */
461 limit = 1;
462 else
463 limit = pack_size_limit - write_offset;
2c5ef824
NP
464
465 if (!entry->delta)
466 usable_delta = 0; /* no delta */
467 else if (!pack_size_limit)
468 usable_delta = 1; /* unlimited packfile */
469 else if (entry->delta->idx.offset == (off_t)-1)
470 usable_delta = 0; /* base was written to another pack */
471 else if (entry->delta->idx.offset)
472 usable_delta = 1; /* base already exists in this pack */
473 else
474 usable_delta = 0; /* base could end up in another pack */
475
a7de7130 476 if (!reuse_object)
fa736f72
NP
477 to_reuse = 0; /* explicit */
478 else if (!entry->in_pack)
ab7cd7bb 479 to_reuse = 0; /* can't reuse what we don't have */
c9018b03 480 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
17b08f2c
DH
481 /* check_object() decided it for us ... */
482 to_reuse = usable_delta;
483 /* ... but pack split may override that */
c9018b03 484 else if (entry->type != entry->in_pack_type)
ab7cd7bb
JH
485 to_reuse = 0; /* pack has delta which is unusable */
486 else if (entry->delta)
487 to_reuse = 0; /* we want to pack afresh */
488 else
489 to_reuse = 1; /* we have it in-pack undeltified,
490 * and we do not need to deltify it.
491 */
492
c9018b03
NTND
493 if (!to_reuse)
494 len = write_no_reuse_object(f, entry, limit, usable_delta);
495 else
496 len = write_reuse_object(f, entry, limit, usable_delta);
497 if (!len)
498 return 0;
64bd76b1 499
17b08f2c 500 if (usable_delta)
ab7cd7bb 501 written_delta++;
3f9ac8d2 502 written++;
78d1e84f 503 if (!pack_to_stdout)
aa7e44bf 504 entry->idx.crc32 = crc32_end(f);
c9018b03 505 return len;
c323ac7d
LT
506}
507
f63c79db
JH
508enum write_one_status {
509 WRITE_ONE_SKIP = -1, /* already written */
510 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
511 WRITE_ONE_WRITTEN = 1, /* normal */
512 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
513};
514
515static enum write_one_status write_one(struct sha1file *f,
516 struct object_entry *e,
517 off_t *offset)
9d5ab962 518{
af92a645 519 off_t size;
f63c79db 520 int recursing;
d7dd0223 521
f63c79db
JH
522 /*
523 * we set offset to 1 (which is an impossible value) to mark
524 * the fact that this object is involved in "write its base
525 * first before writing a deltified object" recursion.
526 */
527 recursing = (e->idx.offset == 1);
528 if (recursing) {
529 warning("recursive delta detected for object %s",
e6a492b7 530 oid_to_hex(&e->idx.oid));
f63c79db
JH
531 return WRITE_ONE_RECURSIVE;
532 } else if (e->idx.offset || e->preferred_base) {
533 /* offset is non zero if object is written already. */
534 return WRITE_ONE_SKIP;
535 }
d7dd0223 536
720c9f7b 537 /* if we are deltified, write out base object first. */
f63c79db
JH
538 if (e->delta) {
539 e->idx.offset = 1; /* now recurse */
540 switch (write_one(f, e->delta, offset)) {
541 case WRITE_ONE_RECURSIVE:
542 /* we cannot depend on this one */
543 e->delta = NULL;
544 break;
545 default:
546 break;
547 case WRITE_ONE_BREAK:
548 e->idx.offset = recursing;
549 return WRITE_ONE_BREAK;
550 }
551 }
d7dd0223 552
6ed7f25e
NP
553 e->idx.offset = *offset;
554 size = write_object(f, e, *offset);
17b08f2c 555 if (!size) {
f63c79db
JH
556 e->idx.offset = recursing;
557 return WRITE_ONE_BREAK;
17b08f2c 558 }
79814f42 559 written_list[nr_written++] = &e->idx;
d7dd0223
NP
560
561 /* make sure off_t is sufficiently large not to wrap */
c03c8315 562 if (signed_add_overflows(*offset, size))
d7dd0223 563 die("pack too large for current definition of off_t");
6ed7f25e 564 *offset += size;
f63c79db 565 return WRITE_ONE_WRITTEN;
9d5ab962
JH
566}
567
d155254c 568static int mark_tagged(const char *path, const struct object_id *oid, int flag,
1b4bb16b
JH
569 void *cb_data)
570{
571 unsigned char peeled[20];
d155254c 572 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
1b4bb16b
JH
573
574 if (entry)
575 entry->tagged = 1;
576 if (!peel_ref(path, peeled)) {
2834bc27 577 entry = packlist_find(&to_pack, peeled, NULL);
1b4bb16b
JH
578 if (entry)
579 entry->tagged = 1;
580 }
581 return 0;
582}
583
be126818 584static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 585 unsigned int *endp,
1b4bb16b
JH
586 struct object_entry *e)
587{
588 if (e->filled)
589 return;
590 wo[(*endp)++] = e;
591 e->filled = 1;
592}
593
594static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 595 unsigned int *endp,
1b4bb16b
JH
596 struct object_entry *e)
597{
f380872f
DM
598 int add_to_order = 1;
599 while (e) {
600 if (add_to_order) {
601 struct object_entry *s;
602 /* add this node... */
603 add_to_write_order(wo, endp, e);
604 /* all its siblings... */
605 for (s = e->delta_sibling; s; s = s->delta_sibling) {
606 add_to_write_order(wo, endp, s);
607 }
608 }
609 /* drop down a level to add left subtree nodes if possible */
610 if (e->delta_child) {
611 add_to_order = 1;
612 e = e->delta_child;
613 } else {
614 add_to_order = 0;
615 /* our sibling might have some children, it is next */
616 if (e->delta_sibling) {
617 e = e->delta_sibling;
618 continue;
619 }
620 /* go back to our parent node */
621 e = e->delta;
622 while (e && !e->delta_sibling) {
623 /* we're on the right side of a subtree, keep
624 * going up until we can go right again */
625 e = e->delta;
626 }
627 if (!e) {
628 /* done- we hit our original root node */
629 return;
630 }
631 /* pass it off to sibling at this level */
632 e = e->delta_sibling;
633 }
634 };
1b4bb16b
JH
635}
636
637static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 638 unsigned int *endp,
1b4bb16b
JH
639 struct object_entry *e)
640{
641 struct object_entry *root;
642
643 for (root = e; root->delta; root = root->delta)
644 ; /* nothing */
1b4bb16b
JH
645 add_descendants_to_write_order(wo, endp, root);
646}
647
648static struct object_entry **compute_write_order(void)
649{
38d4debb 650 unsigned int i, wo_end, last_untagged;
1b4bb16b 651
b32fa95f 652 struct object_entry **wo;
2834bc27 653 struct object_entry *objects = to_pack.objects;
1b4bb16b 654
2834bc27 655 for (i = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
656 objects[i].tagged = 0;
657 objects[i].filled = 0;
658 objects[i].delta_child = NULL;
659 objects[i].delta_sibling = NULL;
660 }
661
662 /*
663 * Fully connect delta_child/delta_sibling network.
664 * Make sure delta_sibling is sorted in the original
665 * recency order.
666 */
2834bc27 667 for (i = to_pack.nr_objects; i > 0;) {
92bef1a1 668 struct object_entry *e = &objects[--i];
1b4bb16b
JH
669 if (!e->delta)
670 continue;
671 /* Mark me as the first child */
672 e->delta_sibling = e->delta->delta_child;
673 e->delta->delta_child = e;
674 }
675
676 /*
677 * Mark objects that are at the tip of tags.
678 */
d155254c 679 for_each_tag_ref(mark_tagged, NULL);
1b4bb16b
JH
680
681 /*
38d4debb 682 * Give the objects in the original recency order until
1b4bb16b
JH
683 * we see a tagged tip.
684 */
b32fa95f 685 ALLOC_ARRAY(wo, to_pack.nr_objects);
2834bc27 686 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
687 if (objects[i].tagged)
688 break;
689 add_to_write_order(wo, &wo_end, &objects[i]);
690 }
38d4debb 691 last_untagged = i;
1b4bb16b
JH
692
693 /*
694 * Then fill all the tagged tips.
695 */
2834bc27 696 for (; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
697 if (objects[i].tagged)
698 add_to_write_order(wo, &wo_end, &objects[i]);
699 }
700
701 /*
702 * And then all remaining commits and tags.
703 */
2834bc27 704 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
705 if (objects[i].type != OBJ_COMMIT &&
706 objects[i].type != OBJ_TAG)
707 continue;
708 add_to_write_order(wo, &wo_end, &objects[i]);
709 }
710
711 /*
712 * And then all the trees.
713 */
2834bc27 714 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
715 if (objects[i].type != OBJ_TREE)
716 continue;
717 add_to_write_order(wo, &wo_end, &objects[i]);
718 }
719
720 /*
721 * Finally all the rest in really tight order
722 */
2834bc27 723 for (i = last_untagged; i < to_pack.nr_objects; i++) {
38d4debb
DM
724 if (!objects[i].filled)
725 add_family_to_write_order(wo, &wo_end, &objects[i]);
726 }
727
2834bc27
VM
728 if (wo_end != to_pack.nr_objects)
729 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
1b4bb16b
JH
730
731 return wo;
732}
733
6b8fda2d
VM
734static off_t write_reused_pack(struct sha1file *f)
735{
736 unsigned char buffer[8192];
657673f1 737 off_t to_write, total;
6b8fda2d
VM
738 int fd;
739
740 if (!is_pack_valid(reuse_packfile))
741 die("packfile is invalid: %s", reuse_packfile->pack_name);
742
a5436b57 743 fd = git_open(reuse_packfile->pack_name);
6b8fda2d
VM
744 if (fd < 0)
745 die_errno("unable to open packfile for reuse: %s",
746 reuse_packfile->pack_name);
747
748 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
749 die_errno("unable to seek in reused packfile");
750
751 if (reuse_packfile_offset < 0)
752 reuse_packfile_offset = reuse_packfile->pack_size - 20;
753
657673f1 754 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
6b8fda2d
VM
755
756 while (to_write) {
757 int read_pack = xread(fd, buffer, sizeof(buffer));
758
759 if (read_pack <= 0)
760 die_errno("unable to read from reused packfile");
761
762 if (read_pack > to_write)
763 read_pack = to_write;
764
765 sha1write(f, buffer, read_pack);
766 to_write -= read_pack;
657673f1
JK
767
768 /*
769 * We don't know the actual number of objects written,
770 * only how many bytes written, how many bytes total, and
771 * how many objects total. So we can fake it by pretending all
772 * objects we are writing are the same size. This gives us a
773 * smooth progress meter, and at the end it matches the true
774 * answer.
775 */
776 written = reuse_packfile_objects *
777 (((double)(total - to_write)) / total);
778 display_progress(progress_state, written);
6b8fda2d
VM
779 }
780
781 close(fd);
657673f1
JK
782 written = reuse_packfile_objects;
783 display_progress(progress_state, written);
6b8fda2d
VM
784 return reuse_packfile_offset - sizeof(struct pack_header);
785}
786
9cea46cd
EW
787static const char no_split_warning[] = N_(
788"disabling bitmap writing, packs are split due to pack.packSizeLimit"
789);
790
d01fb92f 791static void write_pack_file(void)
c323ac7d 792{
ebe27b13 793 uint32_t i = 0, j;
d22b9290 794 struct sha1file *f;
6ed7f25e 795 off_t offset;
ebe27b13 796 uint32_t nr_remaining = nr_result;
f746bae8 797 time_t last_mtime = 0;
1b4bb16b 798 struct object_entry **write_order;
81a216a5 799
bcd7954e 800 if (progress > pack_to_stdout)
754dbc43 801 progress_state = start_progress(_("Writing objects"), nr_result);
b32fa95f 802 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1b4bb16b 803 write_order = compute_write_order();
183bdb2c 804
ebe27b13 805 do {
aa7e44bf 806 unsigned char sha1[20];
7ba502c4 807 char *pack_tmp_name = NULL;
aa7e44bf 808
cdf9db3c 809 if (pack_to_stdout)
2a128d63 810 f = sha1fd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
811 else
812 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 813
c0ad4657 814 offset = write_pack_header(f, nr_remaining);
6b8fda2d
VM
815
816 if (reuse_packfile) {
817 off_t packfile_size;
818 assert(pack_to_stdout);
819
820 packfile_size = write_reused_pack(f);
821 offset += packfile_size;
822 }
823
ebe27b13 824 nr_written = 0;
2834bc27 825 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 826 struct object_entry *e = write_order[i];
cddec4f8 827 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
828 break;
829 display_progress(progress_state, written);
830 }
c553ca25 831
ebe27b13
DH
832 /*
833 * Did we write the wrong # entries in the header?
834 * If so, rewrite it like in fast-import
835 */
54352bb2
LT
836 if (pack_to_stdout) {
837 sha1close(f, sha1, CSUM_CLOSE);
838 } else if (nr_written == nr_remaining) {
839 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 840 } else {
ac0463ed 841 int fd = sha1close(f, sha1, 0);
abeb40e5 842 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 843 nr_written, sha1, offset);
7ba502c4 844 close(fd);
9cea46cd
EW
845 if (write_bitmap_index) {
846 warning(_(no_split_warning));
847 write_bitmap_index = 0;
848 }
ebe27b13
DH
849 }
850
851 if (!pack_to_stdout) {
f746bae8 852 struct stat st;
58892711 853 struct strbuf tmpname = STRBUF_INIT;
d01fb92f 854
f746bae8
NP
855 /*
856 * Packs are runtime accessed in their mtime
857 * order since newer packs are more likely to contain
858 * younger objects. So if we are creating multiple
859 * packs then we should modify the mtime of later ones
860 * to preserve this property.
861 */
0e990530 862 if (stat(pack_tmp_name, &st) < 0) {
54d47394 863 warning_errno("failed to stat %s", pack_tmp_name);
f746bae8
NP
864 } else if (!last_mtime) {
865 last_mtime = st.st_mtime;
866 } else {
867 struct utimbuf utb;
868 utb.actime = st.st_atime;
869 utb.modtime = --last_mtime;
0e990530 870 if (utime(pack_tmp_name, &utb) < 0)
54d47394 871 warning_errno("failed utime() on %s", pack_tmp_name);
f746bae8
NP
872 }
873
58892711 874 strbuf_addf(&tmpname, "%s-", base_name);
7cc8f971
VM
875
876 if (write_bitmap_index) {
877 bitmap_writer_set_checksum(sha1);
878 bitmap_writer_build_type_index(written_list, nr_written);
879 }
880
58892711 881 finish_tmp_packfile(&tmpname, pack_tmp_name,
0e990530
JH
882 written_list, nr_written,
883 &pack_idx_opts, sha1);
7cc8f971
VM
884
885 if (write_bitmap_index) {
58892711 886 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
7cc8f971
VM
887
888 stop_progress(&progress_state);
889
890 bitmap_writer_show_progress(progress);
891 bitmap_writer_reuse_bitmaps(&to_pack);
892 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
893 bitmap_writer_build(&to_pack);
ae4f07fb 894 bitmap_writer_finish(written_list, nr_written,
58892711 895 tmpname.buf, write_bitmap_options);
7cc8f971
VM
896 write_bitmap_index = 0;
897 }
898
58892711 899 strbuf_release(&tmpname);
7ba502c4 900 free(pack_tmp_name);
aa7e44bf 901 puts(sha1_to_hex(sha1));
ebe27b13
DH
902 }
903
904 /* mark written objects as written to previous pack */
905 for (j = 0; j < nr_written; j++) {
79814f42 906 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
907 }
908 nr_remaining -= nr_written;
2834bc27 909 } while (nr_remaining && i < to_pack.nr_objects);
ebe27b13
DH
910
911 free(written_list);
1b4bb16b 912 free(write_order);
4d4fcc54 913 stop_progress(&progress_state);
67c08ce1 914 if (written != nr_result)
6e1c2344
RJ
915 die("wrote %"PRIu32" objects while expecting %"PRIu32,
916 written, nr_result);
c323ac7d
LT
917}
918
a74db82e
JH
919static int no_try_delta(const char *path)
920{
2aef63d3 921 static struct attr_check *check;
a74db82e 922
2aef63d3
JH
923 if (!check)
924 check = attr_check_initl("delta", NULL);
925 if (git_check_attr(path, check))
a74db82e 926 return 0;
2aef63d3 927 if (ATTR_FALSE(check->items[0].value))
a74db82e
JH
928 return 1;
929 return 0;
930}
931
ce2bc424
JK
932/*
933 * When adding an object, check whether we have already added it
934 * to our packing list. If so, we can skip. However, if we are
935 * being asked to excludei t, but the previous mention was to include
936 * it, make sure to adjust its flags and tweak our numbers accordingly.
937 *
938 * As an optimization, we pass out the index position where we would have
939 * found the item, since that saves us from having to look it up again a
940 * few lines later when we want to add the new entry.
941 */
942static int have_duplicate_entry(const unsigned char *sha1,
943 int exclude,
944 uint32_t *index_pos)
c323ac7d 945{
c323ac7d 946 struct object_entry *entry;
29b734e4 947
ce2bc424
JK
948 entry = packlist_find(&to_pack, sha1, index_pos);
949 if (!entry)
29b734e4 950 return 0;
ce2bc424
JK
951
952 if (exclude) {
953 if (!entry->preferred_base)
954 nr_result--;
955 entry->preferred_base = 1;
29b734e4 956 }
c323ac7d 957
ce2bc424
JK
958 return 1;
959}
960
702d1b95
KS
961static int want_found_object(int exclude, struct packed_git *p)
962{
963 if (exclude)
964 return 1;
965 if (incremental)
966 return 0;
967
968 /*
969 * When asked to do --local (do not include an object that appears in a
970 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
971 * an object that appears in a pack marked with .keep), finding a pack
972 * that matches the criteria is sufficient for us to decide to omit it.
973 * However, even if this pack does not satisfy the criteria, we need to
974 * make sure no copy of this object appears in _any_ pack that makes us
975 * to omit the object, so we need to check all the packs.
976 *
977 * We can however first check whether these options can possible matter;
978 * if they do not matter we know we want the object in generated pack.
979 * Otherwise, we signal "-1" at the end to tell the caller that we do
980 * not know either way, and it needs to check more packs.
981 */
982 if (!ignore_packed_keep &&
983 (!local || !have_non_local_packs))
984 return 1;
985
986 if (local && !p->pack_local)
987 return 0;
988 if (ignore_packed_keep && p->pack_local && p->pack_keep)
989 return 0;
990
991 /* we don't know yet; keep looking for more packs */
992 return -1;
993}
994
ce2bc424
JK
995/*
996 * Check whether we want the object in the pack (e.g., we do not want
997 * objects found in non-local stores if the "--local" option was used).
998 *
702d1b95
KS
999 * If the caller already knows an existing pack it wants to take the object
1000 * from, that is passed in *found_pack and *found_offset; otherwise this
1001 * function finds if there is any pack that has the object and returns the pack
1002 * and its offset in these variables.
ce2bc424
JK
1003 */
1004static int want_object_in_pack(const unsigned char *sha1,
1005 int exclude,
1006 struct packed_git **found_pack,
1007 off_t *found_offset)
1008{
c9af708b 1009 struct mru_entry *entry;
702d1b95 1010 int want;
ce2bc424 1011
daae0625
BC
1012 if (!exclude && local && has_loose_object_nonlocal(sha1))
1013 return 0;
1014
702d1b95
KS
1015 /*
1016 * If we already know the pack object lives in, start checks from that
1017 * pack - in the usual case when neither --local was given nor .keep files
1018 * are present we will determine the answer right now.
1019 */
1020 if (*found_pack) {
1021 want = want_found_object(exclude, *found_pack);
1022 if (want != -1)
1023 return want;
1024 }
ce2bc424 1025
607bd831 1026 for (entry = packed_git_mru.head; entry; entry = entry->next) {
c9af708b 1027 struct packed_git *p = entry->item;
702d1b95
KS
1028 off_t offset;
1029
1030 if (p == *found_pack)
1031 offset = *found_offset;
1032 else
1033 offset = find_pack_entry_one(sha1, p);
1034
5c49c116 1035 if (offset) {
ce2bc424 1036 if (!*found_pack) {
319b678a 1037 if (!is_pack_valid(p))
4c080182 1038 continue;
ce2bc424
JK
1039 *found_offset = offset;
1040 *found_pack = p;
64560374 1041 }
702d1b95 1042 want = want_found_object(exclude, p);
e6e24c94 1043 if (!exclude && want > 0)
607bd831 1044 mru_mark(&packed_git_mru, entry);
702d1b95
KS
1045 if (want != -1)
1046 return want;
64560374
LT
1047 }
1048 }
eb019375 1049
ce2bc424
JK
1050 return 1;
1051}
1052
1053static void create_object_entry(const unsigned char *sha1,
1054 enum object_type type,
1055 uint32_t hash,
1056 int exclude,
1057 int no_try_delta,
1058 uint32_t index_pos,
1059 struct packed_git *found_pack,
1060 off_t found_offset)
1061{
1062 struct object_entry *entry;
29b734e4 1063
2834bc27 1064 entry = packlist_alloc(&to_pack, sha1, index_pos);
27225f2e 1065 entry->hash = hash;
5c49c116
NP
1066 if (type)
1067 entry->type = type;
29b734e4
NP
1068 if (exclude)
1069 entry->preferred_base = 1;
81a216a5
NP
1070 else
1071 nr_result++;
29b734e4
NP
1072 if (found_pack) {
1073 entry->in_pack = found_pack;
1074 entry->in_pack_offset = found_offset;
1075 }
7a979d99 1076
ce2bc424
JK
1077 entry->no_try_delta = no_try_delta;
1078}
29b734e4 1079
373c67da
JK
1080static const char no_closure_warning[] = N_(
1081"disabling bitmap writing, as some objects are not being packed"
1082);
1083
ce2bc424
JK
1084static int add_object_entry(const unsigned char *sha1, enum object_type type,
1085 const char *name, int exclude)
1086{
702d1b95
KS
1087 struct packed_git *found_pack = NULL;
1088 off_t found_offset = 0;
ce2bc424 1089 uint32_t index_pos;
7a979d99 1090
ce2bc424
JK
1091 if (have_duplicate_entry(sha1, exclude, &index_pos))
1092 return 0;
a74db82e 1093
373c67da
JK
1094 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1095 /* The pack is missing an object, so it will not have closure */
1096 if (write_bitmap_index) {
1097 warning(_(no_closure_warning));
1098 write_bitmap_index = 0;
1099 }
ce2bc424 1100 return 0;
373c67da 1101 }
29b734e4 1102
ce2bc424
JK
1103 create_object_entry(sha1, type, pack_name_hash(name),
1104 exclude, name && no_try_delta(name),
1105 index_pos, found_pack, found_offset);
a74db82e 1106
78d2214e 1107 display_progress(progress_state, nr_result);
29b734e4 1108 return 1;
c323ac7d
LT
1109}
1110
6b8fda2d
VM
1111static int add_object_entry_from_bitmap(const unsigned char *sha1,
1112 enum object_type type,
1113 int flags, uint32_t name_hash,
1114 struct packed_git *pack, off_t offset)
1115{
1116 uint32_t index_pos;
1117
1118 if (have_duplicate_entry(sha1, 0, &index_pos))
1119 return 0;
1120
702d1b95
KS
1121 if (!want_object_in_pack(sha1, 0, &pack, &offset))
1122 return 0;
1123
6b8fda2d
VM
1124 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1125
78d2214e 1126 display_progress(progress_state, nr_result);
29b734e4 1127 return 1;
c323ac7d
LT
1128}
1129
5379a5c5
JH
1130struct pbase_tree_cache {
1131 unsigned char sha1[20];
1132 int ref;
1133 int temporary;
1134 void *tree_data;
1135 unsigned long tree_size;
1136};
1137
1138static struct pbase_tree_cache *(pbase_tree_cache[256]);
1139static int pbase_tree_cache_ix(const unsigned char *sha1)
1140{
1141 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1142}
1143static int pbase_tree_cache_ix_incr(int ix)
1144{
1145 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1146}
1147
1148static struct pbase_tree {
1149 struct pbase_tree *next;
1150 /* This is a phony "cache" entry; we are not
01689909 1151 * going to evict it or find it through _get()
5379a5c5
JH
1152 * mechanism -- this is for the toplevel node that
1153 * would almost always change with any commit.
1154 */
1155 struct pbase_tree_cache pcache;
1156} *pbase_tree;
1157
1158static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1159{
1160 struct pbase_tree_cache *ent, *nent;
1161 void *data;
1162 unsigned long size;
21666f1a 1163 enum object_type type;
5379a5c5
JH
1164 int neigh;
1165 int my_ix = pbase_tree_cache_ix(sha1);
1166 int available_ix = -1;
1167
1168 /* pbase-tree-cache acts as a limited hashtable.
1169 * your object will be found at your index or within a few
1170 * slots after that slot if it is cached.
1171 */
1172 for (neigh = 0; neigh < 8; neigh++) {
1173 ent = pbase_tree_cache[my_ix];
a89fccd2 1174 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
1175 ent->ref++;
1176 return ent;
1177 }
1178 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1179 ((0 <= available_ix) &&
1180 (!ent && pbase_tree_cache[available_ix])))
1181 available_ix = my_ix;
1182 if (!ent)
1183 break;
1184 my_ix = pbase_tree_cache_ix_incr(my_ix);
1185 }
1186
1187 /* Did not find one. Either we got a bogus request or
1188 * we need to read and perhaps cache.
1189 */
21666f1a 1190 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
1191 if (!data)
1192 return NULL;
21666f1a 1193 if (type != OBJ_TREE) {
5379a5c5
JH
1194 free(data);
1195 return NULL;
1196 }
1197
1198 /* We need to either cache or return a throwaway copy */
1199
1200 if (available_ix < 0)
1201 ent = NULL;
1202 else {
1203 ent = pbase_tree_cache[available_ix];
1204 my_ix = available_ix;
1205 }
1206
1207 if (!ent) {
1208 nent = xmalloc(sizeof(*nent));
1209 nent->temporary = (available_ix < 0);
1210 }
1211 else {
1212 /* evict and reuse */
1213 free(ent->tree_data);
1214 nent = ent;
1215 }
e702496e 1216 hashcpy(nent->sha1, sha1);
5379a5c5
JH
1217 nent->tree_data = data;
1218 nent->tree_size = size;
1219 nent->ref = 1;
1220 if (!nent->temporary)
1221 pbase_tree_cache[my_ix] = nent;
1222 return nent;
1223}
1224
1225static void pbase_tree_put(struct pbase_tree_cache *cache)
1226{
1227 if (!cache->temporary) {
1228 cache->ref--;
1229 return;
1230 }
1231 free(cache->tree_data);
1232 free(cache);
1233}
1234
1235static int name_cmp_len(const char *name)
1236{
1237 int i;
1238 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1239 ;
1240 return i;
1241}
1242
1243static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1244 const char *name,
ce0bd642
LT
1245 int cmplen,
1246 const char *fullname)
3f9ac8d2 1247{
4c068a98 1248 struct name_entry entry;
8a5a8d6c 1249 int cmp;
4c068a98
LT
1250
1251 while (tree_entry(tree,&entry)) {
1211be6b
LT
1252 if (S_ISGITLINK(entry.mode))
1253 continue;
0de16337 1254 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1255 memcmp(name, entry.path, cmplen);
1256 if (cmp > 0)
7a979d99 1257 continue;
8a5a8d6c
NP
1258 if (cmp < 0)
1259 return;
5379a5c5 1260 if (name[cmplen] != '/') {
7d924c91 1261 add_object_entry(entry.oid->hash,
4d1012c3 1262 object_type(entry.mode),
bc32fed5 1263 fullname, 1);
5379a5c5
JH
1264 return;
1265 }
8a5a8d6c 1266 if (S_ISDIR(entry.mode)) {
7a979d99 1267 struct tree_desc sub;
5379a5c5
JH
1268 struct pbase_tree_cache *tree;
1269 const char *down = name+cmplen+1;
1270 int downlen = name_cmp_len(down);
1271
7d924c91 1272 tree = pbase_tree_get(entry.oid->hash);
5379a5c5
JH
1273 if (!tree)
1274 return;
6fda5e51 1275 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1276
ce0bd642 1277 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1278 pbase_tree_put(tree);
1279 }
1280 }
1281}
1d6b38cc 1282
5379a5c5
JH
1283static unsigned *done_pbase_paths;
1284static int done_pbase_paths_num;
1285static int done_pbase_paths_alloc;
1286static int done_pbase_path_pos(unsigned hash)
1287{
1288 int lo = 0;
1289 int hi = done_pbase_paths_num;
1290 while (lo < hi) {
19716b21 1291 int mi = lo + (hi - lo) / 2;
5379a5c5
JH
1292 if (done_pbase_paths[mi] == hash)
1293 return mi;
1294 if (done_pbase_paths[mi] < hash)
1295 hi = mi;
1296 else
1297 lo = mi + 1;
1298 }
1299 return -lo-1;
1300}
1301
1302static int check_pbase_path(unsigned hash)
1303{
c7b07805 1304 int pos = done_pbase_path_pos(hash);
5379a5c5
JH
1305 if (0 <= pos)
1306 return 1;
1307 pos = -pos - 1;
25e19407
DD
1308 ALLOC_GROW(done_pbase_paths,
1309 done_pbase_paths_num + 1,
1310 done_pbase_paths_alloc);
5379a5c5
JH
1311 done_pbase_paths_num++;
1312 if (pos < done_pbase_paths_num)
f331ab9d
RS
1313 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1314 done_pbase_paths_num - pos - 1);
5379a5c5
JH
1315 done_pbase_paths[pos] = hash;
1316 return 0;
1317}
1318
bc32fed5 1319static void add_preferred_base_object(const char *name)
5379a5c5
JH
1320{
1321 struct pbase_tree *it;
8a5a8d6c 1322 int cmplen;
68fb36eb 1323 unsigned hash = pack_name_hash(name);
5379a5c5 1324
8a5a8d6c 1325 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1326 return;
1327
8a5a8d6c 1328 cmplen = name_cmp_len(name);
5379a5c5
JH
1329 for (it = pbase_tree; it; it = it->next) {
1330 if (cmplen == 0) {
bc32fed5 1331 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
1332 }
1333 else {
1334 struct tree_desc tree;
6fda5e51 1335 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1336 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1337 }
3f9ac8d2 1338 }
3f9ac8d2
JH
1339}
1340
7a979d99 1341static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 1342{
5379a5c5
JH
1343 struct pbase_tree *it;
1344 void *data;
1345 unsigned long size;
1346 unsigned char tree_sha1[20];
1d6b38cc 1347
8d1d8f83
JH
1348 if (window <= num_preferred_base++)
1349 return;
1350
5379a5c5
JH
1351 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1352 if (!data)
7a979d99 1353 return;
5379a5c5
JH
1354
1355 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1356 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1357 free(data);
1358 return;
1359 }
1360 }
1361
1362 it = xcalloc(1, sizeof(*it));
1363 it->next = pbase_tree;
1364 pbase_tree = it;
1365
e702496e 1366 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1367 it->pcache.tree_data = data;
1368 it->pcache.tree_size = size;
3f9ac8d2
JH
1369}
1370
0ef95f72
NP
1371static void cleanup_preferred_base(void)
1372{
1373 struct pbase_tree *it;
1374 unsigned i;
1375
1376 it = pbase_tree;
1377 pbase_tree = NULL;
1378 while (it) {
1379 struct pbase_tree *this = it;
1380 it = this->next;
1381 free(this->pcache.tree_data);
1382 free(this);
1383 }
1384
1385 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1386 if (!pbase_tree_cache[i])
1387 continue;
1388 free(pbase_tree_cache[i]->tree_data);
6a83d902 1389 FREE_AND_NULL(pbase_tree_cache[i]);
0ef95f72
NP
1390 }
1391
6a83d902 1392 FREE_AND_NULL(done_pbase_paths);
0ef95f72
NP
1393 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1394}
1395
c323ac7d
LT
1396static void check_object(struct object_entry *entry)
1397{
5c49c116 1398 if (entry->in_pack) {
780e6e73 1399 struct packed_git *p = entry->in_pack;
03e79c88 1400 struct pack_window *w_curs = NULL;
5c49c116
NP
1401 const unsigned char *base_ref = NULL;
1402 struct object_entry *base_entry;
1403 unsigned long used, used_0;
ef49a7a0 1404 unsigned long avail;
5c49c116
NP
1405 off_t ofs;
1406 unsigned char *buf, c;
780e6e73 1407
6777a59f 1408 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1409
5c49c116 1410 /*
fa736f72
NP
1411 * We want in_pack_type even if we do not reuse delta
1412 * since non-delta representations could still be reused.
ab7cd7bb 1413 */
09ded04b 1414 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1415 &entry->in_pack_type,
1416 &entry->size);
03d66015
NP
1417 if (used == 0)
1418 goto give_up;
ab7cd7bb 1419
5c49c116
NP
1420 /*
1421 * Determine if this is a delta and if so whether we can
1422 * reuse it or not. Otherwise let's find out as cheaply as
1423 * possible what the actual type and size for this object is.
3f9ac8d2 1424 */
5c49c116
NP
1425 switch (entry->in_pack_type) {
1426 default:
1427 /* Not a delta hence we've already got all we need. */
1428 entry->type = entry->in_pack_type;
1429 entry->in_pack_header_size = used;
03d66015
NP
1430 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1431 goto give_up;
5c49c116
NP
1432 unuse_pack(&w_curs);
1433 return;
1434 case OBJ_REF_DELTA:
a7de7130 1435 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1436 base_ref = use_pack(p, &w_curs,
1437 entry->in_pack_offset + used, NULL);
1438 entry->in_pack_header_size = used + 20;
1439 break;
1440 case OBJ_OFS_DELTA:
1441 buf = use_pack(p, &w_curs,
1442 entry->in_pack_offset + used, NULL);
1443 used_0 = 0;
1444 c = buf[used_0++];
1445 ofs = c & 127;
1446 while (c & 128) {
1447 ofs += 1;
03d66015
NP
1448 if (!ofs || MSB(ofs, 7)) {
1449 error("delta base offset overflow in pack for %s",
e6a492b7 1450 oid_to_hex(&entry->idx.oid));
03d66015
NP
1451 goto give_up;
1452 }
5c49c116
NP
1453 c = buf[used_0++];
1454 ofs = (ofs << 7) + (c & 127);
780e6e73 1455 }
5c49c116 1456 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1457 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1458 error("delta base offset out of bound for %s",
e6a492b7 1459 oid_to_hex(&entry->idx.oid));
03d66015
NP
1460 goto give_up;
1461 }
a7de7130 1462 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1463 struct revindex_entry *revidx;
1464 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1465 if (!revidx)
1466 goto give_up;
3449f8c4
NP
1467 base_ref = nth_packed_object_sha1(p, revidx->nr);
1468 }
5c49c116
NP
1469 entry->in_pack_header_size = used + used_0;
1470 break;
780e6e73 1471 }
780e6e73 1472
2834bc27 1473 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
5c49c116
NP
1474 /*
1475 * If base_ref was set above that means we wish to
1476 * reuse delta data, and we even found that base
1477 * in the list of objects we want to pack. Goodie!
1478 *
1479 * Depth value does not matter - find_deltas() will
1480 * never consider reused delta as the base object to
1481 * deltify other objects against, in order to avoid
1482 * circular deltas.
3f9ac8d2 1483 */
780e6e73 1484 entry->type = entry->in_pack_type;
5c49c116 1485 entry->delta = base_entry;
64bd76b1 1486 entry->delta_size = entry->size;
15b4d577
JH
1487 entry->delta_sibling = base_entry->delta_child;
1488 base_entry->delta_child = entry;
5c49c116
NP
1489 unuse_pack(&w_curs);
1490 return;
1491 }
ab7cd7bb 1492
5c49c116
NP
1493 if (entry->type) {
1494 /*
1495 * This must be a delta and we already know what the
1496 * final object type is. Let's extract the actual
1497 * object size from the delta header.
1498 */
1499 entry->size = get_size_from_delta(p, &w_curs,
1500 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1501 if (entry->size == 0)
1502 goto give_up;
5c49c116 1503 unuse_pack(&w_curs);
3f9ac8d2
JH
1504 return;
1505 }
5c49c116
NP
1506
1507 /*
1508 * No choice but to fall back to the recursive delta walk
1509 * with sha1_object_info() to find about the object type
1510 * at this point...
1511 */
03d66015 1512 give_up:
5c49c116 1513 unuse_pack(&w_curs);
36e4d74a 1514 }
3f9ac8d2 1515
e6a492b7 1516 entry->type = sha1_object_info(entry->idx.oid.hash, &entry->size);
6d6f9cdd
SP
1517 /*
1518 * The error condition is checked in prepare_pack(). This is
1519 * to permit a missing preferred base object to be ignored
1520 * as a preferred base. Doing so can result in a larger
1521 * pack file, but the transfer will still take place.
1522 */
3f9ac8d2
JH
1523}
1524
5c49c116
NP
1525static int pack_offset_sort(const void *_a, const void *_b)
1526{
1527 const struct object_entry *a = *(struct object_entry **)_a;
1528 const struct object_entry *b = *(struct object_entry **)_b;
1529
1530 /* avoid filesystem trashing with loose objects */
1531 if (!a->in_pack && !b->in_pack)
e6a492b7 1532 return oidcmp(&a->idx.oid, &b->idx.oid);
5c49c116
NP
1533
1534 if (a->in_pack < b->in_pack)
1535 return -1;
1536 if (a->in_pack > b->in_pack)
1537 return 1;
1538 return a->in_pack_offset < b->in_pack_offset ? -1 :
1539 (a->in_pack_offset > b->in_pack_offset);
1540}
1541
4cf2143e
JK
1542/*
1543 * Drop an on-disk delta we were planning to reuse. Naively, this would
1544 * just involve blanking out the "delta" field, but we have to deal
1545 * with some extra book-keeping:
1546 *
1547 * 1. Removing ourselves from the delta_sibling linked list.
1548 *
1549 * 2. Updating our size/type to the non-delta representation. These were
1550 * either not recorded initially (size) or overwritten with the delta type
1551 * (type) when check_object() decided to reuse the delta.
7dbabbbe
JK
1552 *
1553 * 3. Resetting our delta depth, as we are now a base object.
4cf2143e
JK
1554 */
1555static void drop_reused_delta(struct object_entry *entry)
1556{
1557 struct object_entry **p = &entry->delta->delta_child;
1558 struct object_info oi = OBJECT_INFO_INIT;
1559
1560 while (*p) {
1561 if (*p == entry)
1562 *p = (*p)->delta_sibling;
1563 else
1564 p = &(*p)->delta_sibling;
1565 }
1566 entry->delta = NULL;
7dbabbbe 1567 entry->depth = 0;
4cf2143e
JK
1568
1569 oi.sizep = &entry->size;
1570 oi.typep = &entry->type;
1571 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1572 /*
1573 * We failed to get the info from this pack for some reason;
1574 * fall back to sha1_object_info, which may find another copy.
1575 * And if that fails, the error will be recorded in entry->type
1576 * and dealt with in prepare_pack().
1577 */
e6a492b7 1578 entry->type = sha1_object_info(entry->idx.oid.hash,
1579 &entry->size);
4cf2143e
JK
1580 }
1581}
1582
1583/*
1584 * Follow the chain of deltas from this entry onward, throwing away any links
1585 * that cause us to hit a cycle (as determined by the DFS state flags in
1586 * the entries).
7dbabbbe
JK
1587 *
1588 * We also detect too-long reused chains that would violate our --depth
1589 * limit.
4cf2143e
JK
1590 */
1591static void break_delta_chains(struct object_entry *entry)
1592{
42b766d7
JK
1593 /*
1594 * The actual depth of each object we will write is stored as an int,
1595 * as it cannot exceed our int "depth" limit. But before we break
1596 * changes based no that limit, we may potentially go as deep as the
1597 * number of objects, which is elsewhere bounded to a uint32_t.
1598 */
1599 uint32_t total_depth;
1600 struct object_entry *cur, *next;
1601
1602 for (cur = entry, total_depth = 0;
1603 cur;
1604 cur = cur->delta, total_depth++) {
1605 if (cur->dfs_state == DFS_DONE) {
1606 /*
1607 * We've already seen this object and know it isn't
1608 * part of a cycle. We do need to append its depth
1609 * to our count.
1610 */
1611 total_depth += cur->depth;
1612 break;
1613 }
4cf2143e 1614
4cf2143e 1615 /*
42b766d7
JK
1616 * We break cycles before looping, so an ACTIVE state (or any
1617 * other cruft which made its way into the state variable)
1618 * is a bug.
4cf2143e 1619 */
42b766d7
JK
1620 if (cur->dfs_state != DFS_NONE)
1621 die("BUG: confusing delta dfs state in first pass: %d",
1622 cur->dfs_state);
4cf2143e 1623
4cf2143e 1624 /*
42b766d7
JK
1625 * Now we know this is the first time we've seen the object. If
1626 * it's not a delta, we're done traversing, but we'll mark it
1627 * done to save time on future traversals.
4cf2143e 1628 */
42b766d7
JK
1629 if (!cur->delta) {
1630 cur->dfs_state = DFS_DONE;
1631 break;
1632 }
4cf2143e 1633
4cf2143e 1634 /*
42b766d7
JK
1635 * Mark ourselves as active and see if the next step causes
1636 * us to cycle to another active object. It's important to do
1637 * this _before_ we loop, because it impacts where we make the
1638 * cut, and thus how our total_depth counter works.
1639 * E.g., We may see a partial loop like:
1640 *
1641 * A -> B -> C -> D -> B
1642 *
1643 * Cutting B->C breaks the cycle. But now the depth of A is
1644 * only 1, and our total_depth counter is at 3. The size of the
1645 * error is always one less than the size of the cycle we
1646 * broke. Commits C and D were "lost" from A's chain.
1647 *
1648 * If we instead cut D->B, then the depth of A is correct at 3.
1649 * We keep all commits in the chain that we examined.
4cf2143e 1650 */
42b766d7
JK
1651 cur->dfs_state = DFS_ACTIVE;
1652 if (cur->delta->dfs_state == DFS_ACTIVE) {
1653 drop_reused_delta(cur);
1654 cur->dfs_state = DFS_DONE;
1655 break;
7dbabbbe 1656 }
42b766d7 1657 }
7dbabbbe 1658
42b766d7
JK
1659 /*
1660 * And now that we've gone all the way to the bottom of the chain, we
1661 * need to clear the active flags and set the depth fields as
1662 * appropriate. Unlike the loop above, which can quit when it drops a
1663 * delta, we need to keep going to look for more depth cuts. So we need
1664 * an extra "next" pointer to keep going after we reset cur->delta.
1665 */
1666 for (cur = entry; cur; cur = next) {
1667 next = cur->delta;
4cf2143e 1668
42b766d7
JK
1669 /*
1670 * We should have a chain of zero or more ACTIVE states down to
1671 * a final DONE. We can quit after the DONE, because either it
1672 * has no bases, or we've already handled them in a previous
1673 * call.
1674 */
1675 if (cur->dfs_state == DFS_DONE)
1676 break;
1677 else if (cur->dfs_state != DFS_ACTIVE)
1678 die("BUG: confusing delta dfs state in second pass: %d",
1679 cur->dfs_state);
4cf2143e 1680
4cf2143e 1681 /*
42b766d7
JK
1682 * If the total_depth is more than depth, then we need to snip
1683 * the chain into two or more smaller chains that don't exceed
1684 * the maximum depth. Most of the resulting chains will contain
1685 * (depth + 1) entries (i.e., depth deltas plus one base), and
1686 * the last chain (i.e., the one containing entry) will contain
1687 * whatever entries are left over, namely
1688 * (total_depth % (depth + 1)) of them.
1689 *
1690 * Since we are iterating towards decreasing depth, we need to
1691 * decrement total_depth as we go, and we need to write to the
1692 * entry what its final depth will be after all of the
1693 * snipping. Since we're snipping into chains of length (depth
1694 * + 1) entries, the final depth of an entry will be its
1695 * original depth modulo (depth + 1). Any time we encounter an
1696 * entry whose final depth is supposed to be zero, we snip it
1697 * from its delta base, thereby making it so.
4cf2143e 1698 */
42b766d7
JK
1699 cur->depth = (total_depth--) % (depth + 1);
1700 if (!cur->depth)
1701 drop_reused_delta(cur);
1702
1703 cur->dfs_state = DFS_DONE;
4cf2143e
JK
1704 }
1705}
1706
c323ac7d
LT
1707static void get_object_details(void)
1708{
7cadf491 1709 uint32_t i;
5c49c116
NP
1710 struct object_entry **sorted_by_offset;
1711
2834bc27
VM
1712 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1713 for (i = 0; i < to_pack.nr_objects; i++)
1714 sorted_by_offset[i] = to_pack.objects + i;
9ed0d8d6 1715 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
c323ac7d 1716
2834bc27 1717 for (i = 0; i < to_pack.nr_objects; i++) {
15366280
JH
1718 struct object_entry *entry = sorted_by_offset[i];
1719 check_object(entry);
754980d0 1720 if (big_file_threshold < entry->size)
15366280
JH
1721 entry->no_try_delta = 1;
1722 }
3449f8c4 1723
4cf2143e
JK
1724 /*
1725 * This must happen in a second pass, since we rely on the delta
1726 * information for the whole list being completed.
1727 */
1728 for (i = 0; i < to_pack.nr_objects; i++)
1729 break_delta_chains(&to_pack.objects[i]);
1730
5c49c116 1731 free(sorted_by_offset);
c323ac7d
LT
1732}
1733
b904166c
NP
1734/*
1735 * We search for deltas in a list sorted by type, by filename hash, and then
1736 * by size, so that we see progressively smaller and smaller files.
1737 * That's because we prefer deltas to be from the bigger file
1738 * to the smaller -- deletes are potentially cheaper, but perhaps
1739 * more importantly, the bigger file is likely the more recent
1740 * one. The deepest deltas are therefore the oldest objects which are
1741 * less susceptible to be accessed often.
1742 */
9668cf59 1743static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1744{
9668cf59
NP
1745 const struct object_entry *a = *(struct object_entry **)_a;
1746 const struct object_entry *b = *(struct object_entry **)_b;
1747
c323ac7d 1748 if (a->type > b->type)
27225f2e 1749 return -1;
b904166c 1750 if (a->type < b->type)
27225f2e 1751 return 1;
b904166c 1752 if (a->hash > b->hash)
7a979d99 1753 return -1;
b904166c 1754 if (a->hash < b->hash)
7a979d99 1755 return 1;
b904166c 1756 if (a->preferred_base > b->preferred_base)
c323ac7d 1757 return -1;
b904166c
NP
1758 if (a->preferred_base < b->preferred_base)
1759 return 1;
c323ac7d 1760 if (a->size > b->size)
b904166c
NP
1761 return -1;
1762 if (a->size < b->size)
c323ac7d 1763 return 1;
b904166c 1764 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1765}
1766
1767struct unpacked {
1768 struct object_entry *entry;
1769 void *data;
f6c7081a 1770 struct delta_index *index;
5a235b5e 1771 unsigned depth;
c323ac7d
LT
1772};
1773
d250626c
NP
1774static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1775 unsigned long delta_size)
074b2eea
MK
1776{
1777 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1778 return 0;
1779
e3dfddb3
MK
1780 if (delta_size < cache_max_small_delta_size)
1781 return 1;
1782
074b2eea
MK
1783 /* cache delta, if objects are large enough compared to delta size */
1784 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1785 return 1;
1786
1787 return 0;
1788}
1789
7eb151d6 1790#ifndef NO_PTHREADS
8ecce684 1791
44626dc7 1792static pthread_mutex_t read_mutex;
8ecce684
NP
1793#define read_lock() pthread_mutex_lock(&read_mutex)
1794#define read_unlock() pthread_mutex_unlock(&read_mutex)
1795
44626dc7 1796static pthread_mutex_t cache_mutex;
3c701839
NP
1797#define cache_lock() pthread_mutex_lock(&cache_mutex)
1798#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1799
44626dc7 1800static pthread_mutex_t progress_mutex;
8ecce684
NP
1801#define progress_lock() pthread_mutex_lock(&progress_mutex)
1802#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1803
1804#else
1805
e1ef8673
JH
1806#define read_lock() (void)0
1807#define read_unlock() (void)0
1808#define cache_lock() (void)0
1809#define cache_unlock() (void)0
1810#define progress_lock() (void)0
1811#define progress_unlock() (void)0
8ecce684
NP
1812
1813#endif
1814
f6c7081a 1815static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1816 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1817{
f6c7081a
NP
1818 struct object_entry *trg_entry = trg->entry;
1819 struct object_entry *src_entry = src->entry;
560b25a8 1820 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1821 unsigned ref_depth;
21666f1a 1822 enum object_type type;
c323ac7d
LT
1823 void *delta_buf;
1824
1825 /* Don't bother doing diffs between different types */
f6c7081a 1826 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1827 return -1;
1828
51d1e83f 1829 /*
15f07e06
JK
1830 * We do not bother to try a delta that we discarded on an
1831 * earlier try, but only when reusing delta data. Note that
1832 * src_entry that is marked as the preferred_base should always
1833 * be considered, as even if we produce a suboptimal delta against
1834 * it, we will still save the transfer cost, as we already know
1835 * the other side has it and we won't send src_entry at all.
51d1e83f 1836 */
a7de7130 1837 if (reuse_delta && trg_entry->in_pack &&
e9195b58 1838 trg_entry->in_pack == src_entry->in_pack &&
15f07e06 1839 !src_entry->preferred_base &&
e9195b58
JH
1840 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1841 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1842 return 0;
1843
898b14ce 1844 /* Let's not bust the allowed depth. */
5a235b5e 1845 if (src->depth >= max_depth)
d116a45a 1846 return 0;
c323ac7d 1847
c3b06a69 1848 /* Now some size filtering heuristics. */
560b25a8 1849 trg_size = trg_entry->size;
c83f032e
NP
1850 if (!trg_entry->delta) {
1851 max_size = trg_size/2 - 20;
1852 ref_depth = 1;
1853 } else {
1854 max_size = trg_entry->delta_size;
5a235b5e 1855 ref_depth = trg->depth;
c83f032e 1856 }
720fe22d 1857 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1858 (max_depth - ref_depth + 1);
c3b06a69
NP
1859 if (max_size == 0)
1860 return 0;
f6c7081a 1861 src_size = src_entry->size;
560b25a8 1862 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1863 if (sizediff >= max_size)
f527cb8c 1864 return 0;
a1dab41a
BD
1865 if (trg_size < src_size / 32)
1866 return 0;
f6c7081a 1867
560b25a8
NP
1868 /* Load data if not already done */
1869 if (!trg->data) {
8ecce684 1870 read_lock();
e6a492b7 1871 trg->data = read_sha1_file(trg_entry->idx.oid.hash, &type,
1872 &sz);
8ecce684 1873 read_unlock();
2e3404c3
JH
1874 if (!trg->data)
1875 die("object %s cannot be read",
e6a492b7 1876 oid_to_hex(&trg_entry->idx.oid));
560b25a8
NP
1877 if (sz != trg_size)
1878 die("object %s inconsistent object length (%lu vs %lu)",
e6a492b7 1879 oid_to_hex(&trg_entry->idx.oid), sz,
1880 trg_size);
ef0316fc 1881 *mem_usage += sz;
560b25a8
NP
1882 }
1883 if (!src->data) {
8ecce684 1884 read_lock();
e6a492b7 1885 src->data = read_sha1_file(src_entry->idx.oid.hash, &type,
1886 &sz);
8ecce684 1887 read_unlock();
71064a95
NP
1888 if (!src->data) {
1889 if (src_entry->preferred_base) {
1890 static int warned = 0;
1891 if (!warned++)
1892 warning("object %s cannot be read",
e6a492b7 1893 oid_to_hex(&src_entry->idx.oid));
71064a95
NP
1894 /*
1895 * Those objects are not included in the
1896 * resulting pack. Be resilient and ignore
1897 * them if they can't be read, in case the
1898 * pack could be created nevertheless.
1899 */
1900 return 0;
1901 }
2e3404c3 1902 die("object %s cannot be read",
e6a492b7 1903 oid_to_hex(&src_entry->idx.oid));
71064a95 1904 }
560b25a8
NP
1905 if (sz != src_size)
1906 die("object %s inconsistent object length (%lu vs %lu)",
e6a492b7 1907 oid_to_hex(&src_entry->idx.oid), sz,
1908 src_size);
ef0316fc 1909 *mem_usage += sz;
560b25a8
NP
1910 }
1911 if (!src->index) {
1912 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1913 if (!src->index) {
1914 static int warned = 0;
1915 if (!warned++)
1916 warning("suboptimal pack - out of memory");
1917 return 0;
1918 }
ef0316fc 1919 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1920 }
1921
1922 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1923 if (!delta_buf)
75c42d8c 1924 return 0;
f6c7081a 1925
9e2d57a0 1926 if (trg_entry->delta) {
848d732c
BD
1927 /* Prefer only shallower same-sized deltas. */
1928 if (delta_size == trg_entry->delta_size &&
5a235b5e 1929 src->depth + 1 >= trg->depth) {
848d732c
BD
1930 free(delta_buf);
1931 return 0;
1932 }
074b2eea 1933 }
9e2d57a0 1934
3c701839
NP
1935 /*
1936 * Handle memory allocation outside of the cache
1937 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1938 * away when NO_PTHREADS is defined.
3c701839 1939 */
8e0f7003 1940 free(trg_entry->delta_data);
3c701839 1941 cache_lock();
9e2d57a0
NP
1942 if (trg_entry->delta_data) {
1943 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1944 trg_entry->delta_data = NULL;
1945 }
d250626c 1946 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1947 delta_cache_size += delta_size;
3c701839
NP
1948 cache_unlock();
1949 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1950 } else {
1951 cache_unlock();
074b2eea 1952 free(delta_buf);
3c701839
NP
1953 }
1954
b7a28f78
NP
1955 trg_entry->delta = src_entry;
1956 trg_entry->delta_size = delta_size;
1957 trg->depth = src->depth + 1;
1958
f6c7081a 1959 return 1;
c323ac7d
LT
1960}
1961
898b14ce 1962static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1963{
898b14ce
NP
1964 struct object_entry *child = me->delta_child;
1965 unsigned int m = n;
1966 while (child) {
1967 unsigned int c = check_delta_limit(child, n + 1);
1968 if (m < c)
1969 m = c;
1970 child = child->delta_sibling;
1971 }
1972 return m;
b2504a0d
NP
1973}
1974
75ad235c 1975static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1976{
ef0316fc 1977 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1978 free_delta_index(n->index);
1979 n->index = NULL;
1980 if (n->data) {
ef0316fc 1981 freed_mem += n->entry->size;
6a83d902 1982 FREE_AND_NULL(n->data);
a97773ce
BD
1983 }
1984 n->entry = NULL;
7d7baa5e 1985 n->depth = 0;
ef0316fc 1986 return freed_mem;
a97773ce
BD
1987}
1988
384b32c0 1989static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1990 int window, int depth, unsigned *processed)
c323ac7d 1991{
384b32c0 1992 uint32_t i, idx = 0, count = 0;
7cadf491 1993 struct unpacked *array;
ef0316fc 1994 unsigned long mem_usage = 0;
c323ac7d 1995
19d4b416 1996 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1997
384b32c0 1998 for (;;) {
421b488a 1999 struct object_entry *entry;
c323ac7d 2000 struct unpacked *n = array + idx;
ef0316fc 2001 int j, max_depth, best_base = -1;
c323ac7d 2002
384b32c0
NP
2003 progress_lock();
2004 if (!*list_size) {
2005 progress_unlock();
2006 break;
2007 }
421b488a 2008 entry = *list++;
384b32c0
NP
2009 (*list_size)--;
2010 if (!entry->preferred_base) {
2011 (*processed)++;
2012 display_progress(progress_state, *processed);
2013 }
2014 progress_unlock();
2015
ef0316fc 2016 mem_usage -= free_unpacked(n);
c323ac7d 2017 n->entry = entry;
ab7cd7bb 2018
a97773ce 2019 while (window_memory_limit &&
ef0316fc 2020 mem_usage > window_memory_limit &&
a97773ce
BD
2021 count > 1) {
2022 uint32_t tail = (idx + window - count) % window;
75ad235c 2023 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
2024 count--;
2025 }
2026
75d39853
NP
2027 /* We do not compute delta to *create* objects we are not
2028 * going to pack.
2029 */
2030 if (entry->preferred_base)
2031 goto next;
2032
898b14ce
NP
2033 /*
2034 * If the current object is at pack edge, take the depth the
2035 * objects that depend on the current object into account
2036 * otherwise they would become too deep.
2037 */
2038 max_depth = depth;
2039 if (entry->delta_child) {
2040 max_depth -= check_delta_limit(entry, 0);
2041 if (max_depth <= 0)
2042 goto next;
2043 }
2044
78817c15
LT
2045 j = window;
2046 while (--j > 0) {
77639870 2047 int ret;
7cadf491 2048 uint32_t other_idx = idx + j;
c323ac7d 2049 struct unpacked *m;
78817c15
LT
2050 if (other_idx >= window)
2051 other_idx -= window;
c323ac7d
LT
2052 m = array + other_idx;
2053 if (!m->entry)
2054 break;
ef0316fc 2055 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 2056 if (ret < 0)
c323ac7d 2057 break;
77639870
JH
2058 else if (ret > 0)
2059 best_base = other_idx;
c323ac7d 2060 }
898b14ce 2061
ed4a9031
NP
2062 /*
2063 * If we decided to cache the delta data, then it is best
2064 * to compress it right away. First because we have to do
2065 * it anyway, and doing it here while we're threaded will
2066 * save a lot of time in the non threaded write phase,
2067 * as well as allow for caching more deltas within
2068 * the same cache size limit.
2069 * ...
2070 * But only if not writing to stdout, since in that case
2071 * the network is most likely throttling writes anyway,
2072 * and therefore it is best to go to the write phase ASAP
2073 * instead, as we can afford spending more time compressing
2074 * between writes at that moment.
2075 */
2076 if (entry->delta_data && !pack_to_stdout) {
2077 entry->z_delta_size = do_compress(&entry->delta_data,
2078 entry->delta_size);
2079 cache_lock();
2080 delta_cache_size -= entry->delta_size;
2081 delta_cache_size += entry->z_delta_size;
2082 cache_unlock();
2083 }
2084
70ca1a3f
JH
2085 /* if we made n a delta, and if n is already at max
2086 * depth, leaving it in the window is pointless. we
2087 * should evict it first.
70ca1a3f 2088 */
70baf5d4 2089 if (entry->delta && max_depth <= n->depth)
70ca1a3f 2090 continue;
ff45715c 2091
77639870
JH
2092 /*
2093 * Move the best delta base up in the window, after the
2094 * currently deltified object, to keep it longer. It will
2095 * be the first base object to be attempted next.
2096 */
2097 if (entry->delta) {
2098 struct unpacked swap = array[best_base];
2099 int dist = (window + idx - best_base) % window;
2100 int dst = best_base;
2101 while (dist--) {
2102 int src = (dst + 1) % window;
2103 array[dst] = array[src];
2104 dst = src;
2105 }
2106 array[dst] = swap;
2107 }
2108
898b14ce 2109 next:
521a4f4c 2110 idx++;
a97773ce
BD
2111 if (count + 1 < window)
2112 count++;
521a4f4c
LT
2113 if (idx >= window)
2114 idx = 0;
384b32c0 2115 }
adee7bdf 2116
f6c7081a 2117 for (i = 0; i < window; ++i) {
ff45715c 2118 free_delta_index(array[i].index);
adee7bdf 2119 free(array[i].data);
f6c7081a 2120 }
adee7bdf 2121 free(array);
c323ac7d
LT
2122}
2123
7eb151d6 2124#ifndef NO_PTHREADS
8ecce684 2125
a9a74636
NP
2126static void try_to_free_from_threads(size_t size)
2127{
2128 read_lock();
7c3ecb32 2129 release_pack_memory(size);
a9a74636
NP
2130 read_unlock();
2131}
2132
bc9b2175 2133static try_to_free_t old_try_to_free_routine;
851c34b0 2134
50f22ada
JS
2135/*
2136 * The main thread waits on the condition that (at least) one of the workers
2137 * has stopped working (which is indicated in the .working member of
2138 * struct thread_params).
2139 * When a work thread has completed its work, it sets .working to 0 and
2140 * signals the main thread and waits on the condition that .data_ready
2141 * becomes 1.
2142 */
2143
8ecce684
NP
2144struct thread_params {
2145 pthread_t thread;
2146 struct object_entry **list;
2147 unsigned list_size;
384b32c0 2148 unsigned remaining;
8ecce684
NP
2149 int window;
2150 int depth;
50f22ada
JS
2151 int working;
2152 int data_ready;
2153 pthread_mutex_t mutex;
2154 pthread_cond_t cond;
8ecce684
NP
2155 unsigned *processed;
2156};
2157
44626dc7
AH
2158static pthread_cond_t progress_cond;
2159
2160/*
2161 * Mutex and conditional variable can't be statically-initialized on Windows.
2162 */
2163static void init_threaded_search(void)
2164{
93749194 2165 init_recursive_mutex(&read_mutex);
44626dc7
AH
2166 pthread_mutex_init(&cache_mutex, NULL);
2167 pthread_mutex_init(&progress_mutex, NULL);
2168 pthread_cond_init(&progress_cond, NULL);
851c34b0 2169 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
2170}
2171
2172static void cleanup_threaded_search(void)
2173{
851c34b0 2174 set_try_to_free_routine(old_try_to_free_routine);
44626dc7
AH
2175 pthread_cond_destroy(&progress_cond);
2176 pthread_mutex_destroy(&read_mutex);
2177 pthread_mutex_destroy(&cache_mutex);
2178 pthread_mutex_destroy(&progress_mutex);
2179}
c2a33679 2180
8ecce684
NP
2181static void *threaded_find_deltas(void *arg)
2182{
c2a33679
NP
2183 struct thread_params *me = arg;
2184
0c2ad00b 2185 progress_lock();
50f22ada 2186 while (me->remaining) {
0c2ad00b
2187 progress_unlock();
2188
384b32c0 2189 find_deltas(me->list, &me->remaining,
c2a33679 2190 me->window, me->depth, me->processed);
50f22ada
JS
2191
2192 progress_lock();
2193 me->working = 0;
2194 pthread_cond_signal(&progress_cond);
2195 progress_unlock();
2196
2197 /*
2198 * We must not set ->data_ready before we wait on the
2199 * condition because the main thread may have set it to 1
2200 * before we get here. In order to be sure that new
2201 * work is available if we see 1 in ->data_ready, it
2202 * was initialized to 0 before this thread was spawned
2203 * and we reset it to 0 right away.
2204 */
2205 pthread_mutex_lock(&me->mutex);
2206 while (!me->data_ready)
2207 pthread_cond_wait(&me->cond, &me->mutex);
2208 me->data_ready = 0;
2209 pthread_mutex_unlock(&me->mutex);
0c2ad00b
2210
2211 progress_lock();
c2a33679 2212 }
0c2ad00b 2213 progress_unlock();
50f22ada
JS
2214 /* leave ->working 1 so that this doesn't get more work assigned */
2215 return NULL;
8ecce684
NP
2216}
2217
8ecce684
NP
2218static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2219 int window, int depth, unsigned *processed)
2220{
dcda3614 2221 struct thread_params *p;
384b32c0 2222 int i, ret, active_threads = 0;
c2a33679 2223
44626dc7
AH
2224 init_threaded_search();
2225
367f4a43 2226 if (delta_search_threads <= 1) {
384b32c0 2227 find_deltas(list, &list_size, window, depth, processed);
44626dc7 2228 cleanup_threaded_search();
367f4a43
NP
2229 return;
2230 }
43cc2b42 2231 if (progress > pack_to_stdout)
b6c29915 2232 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 2233 delta_search_threads);
dcda3614 2234 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 2235
50f22ada 2236 /* Partition the work amongst work threads. */
367f4a43 2237 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
2238 unsigned sub_size = list_size / (delta_search_threads - i);
2239
bf874896
NP
2240 /* don't use too small segments or no deltas will be found */
2241 if (sub_size < 2*window && i+1 < delta_search_threads)
2242 sub_size = 0;
2243
8ecce684
NP
2244 p[i].window = window;
2245 p[i].depth = depth;
2246 p[i].processed = processed;
50f22ada
JS
2247 p[i].working = 1;
2248 p[i].data_ready = 0;
c2a33679 2249
59921b4b 2250 /* try to split chunks on "path" boundaries */
6fc74703
NP
2251 while (sub_size && sub_size < list_size &&
2252 list[sub_size]->hash &&
384b32c0
NP
2253 list[sub_size]->hash == list[sub_size-1]->hash)
2254 sub_size++;
2255
50f22ada
JS
2256 p[i].list = list;
2257 p[i].list_size = sub_size;
2258 p[i].remaining = sub_size;
59921b4b 2259
384b32c0
NP
2260 list += sub_size;
2261 list_size -= sub_size;
2262 }
2263
50f22ada
JS
2264 /* Start work threads. */
2265 for (i = 0; i < delta_search_threads; i++) {
2266 if (!p[i].list_size)
2267 continue;
68e6a4f8
JS
2268 pthread_mutex_init(&p[i].mutex, NULL);
2269 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
2270 ret = pthread_create(&p[i].thread, NULL,
2271 threaded_find_deltas, &p[i]);
2272 if (ret)
2273 die("unable to create thread: %s", strerror(ret));
2274 active_threads++;
2275 }
2276
384b32c0
NP
2277 /*
2278 * Now let's wait for work completion. Each time a thread is done
2279 * with its work, we steal half of the remaining work from the
2280 * thread with the largest number of unprocessed objects and give
2281 * it to that newly idle thread. This ensure good load balancing
2282 * until the remaining object list segments are simply too short
2283 * to be worth splitting anymore.
2284 */
50f22ada
JS
2285 while (active_threads) {
2286 struct thread_params *target = NULL;
384b32c0
NP
2287 struct thread_params *victim = NULL;
2288 unsigned sub_size = 0;
384b32c0
NP
2289
2290 progress_lock();
50f22ada
JS
2291 for (;;) {
2292 for (i = 0; !target && i < delta_search_threads; i++)
2293 if (!p[i].working)
2294 target = &p[i];
2295 if (target)
2296 break;
2297 pthread_cond_wait(&progress_cond, &progress_mutex);
2298 }
2299
384b32c0
NP
2300 for (i = 0; i < delta_search_threads; i++)
2301 if (p[i].remaining > 2*window &&
2302 (!victim || victim->remaining < p[i].remaining))
2303 victim = &p[i];
2304 if (victim) {
2305 sub_size = victim->remaining / 2;
2306 list = victim->list + victim->list_size - sub_size;
2307 while (sub_size && list[0]->hash &&
2308 list[0]->hash == list[-1]->hash) {
2309 list++;
2310 sub_size--;
2311 }
eb9688ff
NP
2312 if (!sub_size) {
2313 /*
2314 * It is possible for some "paths" to have
2315 * so many objects that no hash boundary
2316 * might be found. Let's just steal the
2317 * exact half in that case.
2318 */
2319 sub_size = victim->remaining / 2;
2320 list -= sub_size;
2321 }
384b32c0
NP
2322 target->list = list;
2323 victim->list_size -= sub_size;
2324 victim->remaining -= sub_size;
2325 }
384b32c0
NP
2326 target->list_size = sub_size;
2327 target->remaining = sub_size;
50f22ada
JS
2328 target->working = 1;
2329 progress_unlock();
2330
2331 pthread_mutex_lock(&target->mutex);
2332 target->data_ready = 1;
2333 pthread_cond_signal(&target->cond);
2334 pthread_mutex_unlock(&target->mutex);
c2a33679 2335
384b32c0 2336 if (!sub_size) {
b81d9af7 2337 pthread_join(target->thread, NULL);
50f22ada
JS
2338 pthread_cond_destroy(&target->cond);
2339 pthread_mutex_destroy(&target->mutex);
384b32c0 2340 active_threads--;
c2a33679 2341 }
50f22ada 2342 }
44626dc7 2343 cleanup_threaded_search();
dcda3614 2344 free(p);
8ecce684
NP
2345}
2346
2347#else
384b32c0 2348#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
2349#endif
2350
b773ddea
JK
2351static void add_tag_chain(const struct object_id *oid)
2352{
2353 struct tag *tag;
2354
2355 /*
2356 * We catch duplicates already in add_object_entry(), but we'd
2357 * prefer to do this extra check to avoid having to parse the
2358 * tag at all if we already know that it's being packed (e.g., if
2359 * it was included via bitmaps, we would not have parsed it
2360 * previously).
2361 */
2362 if (packlist_find(&to_pack, oid->hash, NULL))
2363 return;
2364
d3101b53 2365 tag = lookup_tag(oid);
b773ddea
JK
2366 while (1) {
2367 if (!tag || parse_tag(tag) || !tag->tagged)
2368 die("unable to pack objects reachable from tag %s",
2369 oid_to_hex(oid));
2370
2371 add_object_entry(tag->object.oid.hash, OBJ_TAG, NULL, 0);
2372
2373 if (tag->tagged->type != OBJ_TAG)
2374 return;
2375
2376 tag = (struct tag *)tag->tagged;
2377 }
2378}
2379
d155254c 2380static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
f0a24aa5 2381{
d155254c 2382 struct object_id peeled;
f0a24aa5 2383
59556548 2384 if (starts_with(path, "refs/tags/") && /* is a tag? */
d155254c
MH
2385 !peel_ref(path, peeled.hash) && /* peelable? */
2386 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
b773ddea 2387 add_tag_chain(oid);
f0a24aa5
SP
2388 return 0;
2389}
2390
f3123c4a
JH
2391static void prepare_pack(int window, int depth)
2392{
9668cf59 2393 struct object_entry **delta_list;
6e1c2344
RJ
2394 uint32_t i, nr_deltas;
2395 unsigned n;
9668cf59 2396
3f9ac8d2 2397 get_object_details();
9668cf59 2398
0e8189e2
NP
2399 /*
2400 * If we're locally repacking then we need to be doubly careful
2401 * from now on in order to make sure no stealth corruption gets
2402 * propagated to the new pack. Clients receiving streamed packs
2403 * should validate everything they get anyway so no need to incur
2404 * the additional cost here in that case.
2405 */
2406 if (!pack_to_stdout)
2407 do_check_packed_object_crc = 1;
2408
2834bc27 2409 if (!to_pack.nr_objects || !window || !depth)
9668cf59
NP
2410 return;
2411
b32fa95f 2412 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
75d39853
NP
2413 nr_deltas = n = 0;
2414
2834bc27
VM
2415 for (i = 0; i < to_pack.nr_objects; i++) {
2416 struct object_entry *entry = to_pack.objects + i;
75d39853
NP
2417
2418 if (entry->delta)
2419 /* This happens if we decided to reuse existing
a7de7130 2420 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2421 */
2422 continue;
2423
2424 if (entry->size < 50)
2425 continue;
2426
2427 if (entry->no_try_delta)
2428 continue;
2429
6d6f9cdd 2430 if (!entry->preferred_base) {
75d39853 2431 nr_deltas++;
6d6f9cdd
SP
2432 if (entry->type < 0)
2433 die("unable to get type of object %s",
e6a492b7 2434 oid_to_hex(&entry->idx.oid));
eede9f42
NP
2435 } else {
2436 if (entry->type < 0) {
2437 /*
2438 * This object is not found, but we
2439 * don't have to include it anyway.
2440 */
2441 continue;
2442 }
6d6f9cdd 2443 }
75d39853
NP
2444
2445 delta_list[n++] = entry;
2446 }
2447
2f8b8947 2448 if (nr_deltas && n > 1) {
e334977d
NP
2449 unsigned nr_done = 0;
2450 if (progress)
754dbc43 2451 progress_state = start_progress(_("Compressing objects"),
dc6a0757 2452 nr_deltas);
9ed0d8d6 2453 QSORT(delta_list, n, type_size_sort);
8ecce684 2454 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2455 stop_progress(&progress_state);
e334977d
NP
2456 if (nr_done != nr_deltas)
2457 die("inconsistency with delta count");
75d39853 2458 }
9668cf59 2459 free(delta_list);
f3123c4a
JH
2460}
2461
ef90d6d4 2462static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2463{
eeefa7c9 2464 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2465 window = git_config_int(k, v);
2466 return 0;
2467 }
a97773ce
BD
2468 if (!strcmp(k, "pack.windowmemory")) {
2469 window_memory_limit = git_config_ulong(k, v);
2470 return 0;
2471 }
2472 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2473 depth = git_config_int(k, v);
2474 return 0;
2475 }
074b2eea
MK
2476 if (!strcmp(k, "pack.deltacachesize")) {
2477 max_delta_cache_size = git_config_int(k, v);
2478 return 0;
2479 }
e3dfddb3
MK
2480 if (!strcmp(k, "pack.deltacachelimit")) {
2481 cache_max_small_delta_size = git_config_int(k, v);
2482 return 0;
2483 }
ae4f07fb
VM
2484 if (!strcmp(k, "pack.writebitmaphashcache")) {
2485 if (git_config_bool(k, v))
2486 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2487 else
2488 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2489 }
6b8fda2d 2490 if (!strcmp(k, "pack.usebitmaps")) {
645c432d 2491 use_bitmap_index_default = git_config_bool(k, v);
6b8fda2d
VM
2492 return 0;
2493 }
693b86ff
NP
2494 if (!strcmp(k, "pack.threads")) {
2495 delta_search_threads = git_config_int(k, v);
833e3df1 2496 if (delta_search_threads < 0)
693b86ff
NP
2497 die("invalid number of threads specified (%d)",
2498 delta_search_threads);
7eb151d6 2499#ifdef NO_PTHREADS
2e96d815 2500 if (delta_search_threads != 1) {
693b86ff 2501 warning("no threads support, ignoring %s", k);
2e96d815
ÆAB
2502 delta_search_threads = 0;
2503 }
693b86ff
NP
2504#endif
2505 return 0;
2506 }
4d00bda2 2507 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2508 pack_idx_opts.version = git_config_int(k, v);
2509 if (pack_idx_opts.version > 2)
6e1c2344 2510 die("bad pack.indexversion=%"PRIu32,
ebcfb379 2511 pack_idx_opts.version);
4d00bda2
NP
2512 return 0;
2513 }
ef90d6d4 2514 return git_default_config(k, v, cb);
4812a93a
JK
2515}
2516
b5d97e6b 2517static void read_object_list_from_stdin(void)
c323ac7d 2518{
b5d97e6b
JH
2519 char line[40 + 1 + PATH_MAX + 2];
2520 unsigned char sha1[20];
b2504a0d 2521
da93d12b 2522 for (;;) {
da93d12b
LT
2523 if (!fgets(line, sizeof(line), stdin)) {
2524 if (feof(stdin))
2525 break;
2526 if (!ferror(stdin))
2527 die("fgets returned NULL, not EOF, not error!");
687dd75c 2528 if (errno != EINTR)
d824cbba 2529 die_errno("fgets");
687dd75c
JH
2530 clearerr(stdin);
2531 continue;
da93d12b 2532 }
7a979d99
JH
2533 if (line[0] == '-') {
2534 if (get_sha1_hex(line+1, sha1))
2535 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 2536 line);
8d1d8f83 2537 add_preferred_base(sha1);
7a979d99 2538 continue;
21fcd1bd 2539 }
c323ac7d 2540 if (get_sha1_hex(line, sha1))
ef07618f 2541 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 2542
bc32fed5
JH
2543 add_preferred_base_object(line+41);
2544 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 2545 }
b5d97e6b
JH
2546}
2547
08cdfb13
JH
2548#define OBJECT_ADDED (1u<<20)
2549
11c211fa 2550static void show_commit(struct commit *commit, void *data)
b5d97e6b 2551{
ed1c9977 2552 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
08cdfb13 2553 commit->object.flags |= OBJECT_ADDED;
7cc8f971
VM
2554
2555 if (write_bitmap_index)
2556 index_commit_for_bitmap(commit);
b5d97e6b
JH
2557}
2558
de1e67d0 2559static void show_object(struct object *obj, const char *name, void *data)
b5d97e6b 2560{
8d2dfc49 2561 add_preferred_base_object(name);
ed1c9977 2562 add_object_entry(obj->oid.hash, obj->type, name, 0);
8d2dfc49 2563 obj->flags |= OBJECT_ADDED;
b5d97e6b
JH
2564}
2565
9535ce73
JH
2566static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2567{
2568 assert(arg_missing_action == MA_ALLOW_ANY);
2569
2570 /*
2571 * Quietly ignore ALL missing objects. This avoids problems with
2572 * staging them now and getting an odd error later.
2573 */
2574 if (!has_object_file(&obj->oid))
2575 return;
2576
2577 show_object(obj, name, data);
2578}
2579
2580static int option_parse_missing_action(const struct option *opt,
2581 const char *arg, int unset)
2582{
2583 assert(arg);
2584 assert(!unset);
2585
2586 if (!strcmp(arg, "error")) {
2587 arg_missing_action = MA_ERROR;
2588 fn_show_object = show_object;
2589 return 0;
2590 }
2591
2592 if (!strcmp(arg, "allow-any")) {
2593 arg_missing_action = MA_ALLOW_ANY;
2594 fn_show_object = show_object__ma_allow_any;
2595 return 0;
2596 }
2597
2598 die(_("invalid value for --missing"));
2599 return 0;
2600}
2601
8d1d8f83
JH
2602static void show_edge(struct commit *commit)
2603{
ed1c9977 2604 add_preferred_base(commit->object.oid.hash);
8d1d8f83
JH
2605}
2606
08cdfb13
JH
2607struct in_pack_object {
2608 off_t offset;
2609 struct object *object;
2610};
2611
2612struct in_pack {
071bcaab
RJ
2613 unsigned int alloc;
2614 unsigned int nr;
08cdfb13
JH
2615 struct in_pack_object *array;
2616};
2617
2618static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2619{
ed1c9977 2620 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
08cdfb13
JH
2621 in_pack->array[in_pack->nr].object = object;
2622 in_pack->nr++;
2623}
2624
2625/*
2626 * Compare the objects in the offset order, in order to emulate the
f18d244a 2627 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2628 */
2629static int ofscmp(const void *a_, const void *b_)
2630{
2631 struct in_pack_object *a = (struct in_pack_object *)a_;
2632 struct in_pack_object *b = (struct in_pack_object *)b_;
2633
2634 if (a->offset < b->offset)
2635 return -1;
2636 else if (a->offset > b->offset)
2637 return 1;
2638 else
f2fd0760 2639 return oidcmp(&a->object->oid, &b->object->oid);
08cdfb13
JH
2640}
2641
2642static void add_objects_in_unpacked_packs(struct rev_info *revs)
2643{
2644 struct packed_git *p;
2645 struct in_pack in_pack;
2646 uint32_t i;
2647
2648 memset(&in_pack, 0, sizeof(in_pack));
2649
2650 for (p = packed_git; p; p = p->next) {
2651 const unsigned char *sha1;
2652 struct object *o;
2653
79bc4c71 2654 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2655 continue;
2656 if (open_pack_index(p))
2657 die("cannot open pack index");
2658
2659 ALLOC_GROW(in_pack.array,
2660 in_pack.nr + p->num_objects,
2661 in_pack.alloc);
2662
2663 for (i = 0; i < p->num_objects; i++) {
2664 sha1 = nth_packed_object_sha1(p, i);
2665 o = lookup_unknown_object(sha1);
2666 if (!(o->flags & OBJECT_ADDED))
2667 mark_in_pack_object(o, p, &in_pack);
2668 o->flags |= OBJECT_ADDED;
2669 }
2670 }
2671
2672 if (in_pack.nr) {
9ed0d8d6 2673 QSORT(in_pack.array, in_pack.nr, ofscmp);
08cdfb13
JH
2674 for (i = 0; i < in_pack.nr; i++) {
2675 struct object *o = in_pack.array[i].object;
ed1c9977 2676 add_object_entry(o->oid.hash, o->type, "", 0);
08cdfb13
JH
2677 }
2678 }
2679 free(in_pack.array);
2680}
2681
76c1d9a0 2682static int add_loose_object(const struct object_id *oid, const char *path,
e26a8c47
JK
2683 void *data)
2684{
76c1d9a0 2685 enum object_type type = sha1_object_info(oid->hash, NULL);
e26a8c47
JK
2686
2687 if (type < 0) {
2688 warning("loose object at %s could not be examined", path);
2689 return 0;
2690 }
2691
76c1d9a0 2692 add_object_entry(oid->hash, type, "", 0);
e26a8c47
JK
2693 return 0;
2694}
2695
2696/*
2697 * We actually don't even have to worry about reachability here.
2698 * add_object_entry will weed out duplicates, so we just add every
2699 * loose object we find.
2700 */
2701static void add_unreachable_loose_objects(void)
2702{
2703 for_each_loose_file_in_objdir(get_object_directory(),
2704 add_loose_object,
2705 NULL, NULL, NULL);
2706}
2707
094085e3
BC
2708static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2709{
2710 static struct packed_git *last_found = (void *)1;
2711 struct packed_git *p;
2712
2713 p = (last_found != (void *)1) ? last_found : packed_git;
2714
2715 while (p) {
2716 if ((!p->pack_local || p->pack_keep) &&
2717 find_pack_entry_one(sha1, p)) {
2718 last_found = p;
2719 return 1;
2720 }
2721 if (p == last_found)
2722 p = packed_git;
2723 else
2724 p = p->next;
2725 if (p == last_found)
2726 p = p->next;
2727 }
2728 return 0;
2729}
2730
abcb8655
JK
2731/*
2732 * Store a list of sha1s that are should not be discarded
2733 * because they are either written too recently, or are
2734 * reachable from another object that was.
2735 *
2736 * This is filled by get_object_list.
2737 */
910650d2 2738static struct oid_array recent_objects;
abcb8655 2739
4ce3621a 2740static int loosened_object_can_be_discarded(const struct object_id *oid,
dddbad72 2741 timestamp_t mtime)
d0d46abc
JK
2742{
2743 if (!unpack_unreachable_expiration)
2744 return 0;
2745 if (mtime > unpack_unreachable_expiration)
2746 return 0;
910650d2 2747 if (oid_array_lookup(&recent_objects, oid) >= 0)
abcb8655 2748 return 0;
d0d46abc
JK
2749 return 1;
2750}
2751
ca11b212
NP
2752static void loosen_unused_packed_objects(struct rev_info *revs)
2753{
2754 struct packed_git *p;
2755 uint32_t i;
4ce3621a 2756 struct object_id oid;
ca11b212
NP
2757
2758 for (p = packed_git; p; p = p->next) {
79bc4c71 2759 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2760 continue;
2761
2762 if (open_pack_index(p))
2763 die("cannot open pack index");
2764
2765 for (i = 0; i < p->num_objects; i++) {
4ce3621a 2766 nth_packed_object_oid(&oid, p, i);
2767 if (!packlist_find(&to_pack, oid.hash, NULL) &&
2768 !has_sha1_pack_kept_or_nonlocal(oid.hash) &&
2769 !loosened_object_can_be_discarded(&oid, p->mtime))
2770 if (force_object_loose(oid.hash, p->mtime))
ca11b212
NP
2771 die("unable to force loose object");
2772 }
2773 }
2774}
2775
69e4b342 2776/*
645c432d
KS
2777 * This tracks any options which pack-reuse code expects to be on, or which a
2778 * reader of the pack might not understand, and which would therefore prevent
2779 * blind reuse of what we have on disk.
69e4b342
JK
2780 */
2781static int pack_options_allow_reuse(void)
2782{
9df4a607
JK
2783 return pack_to_stdout &&
2784 allow_ofs_delta &&
2785 !ignore_packed_keep &&
2786 (!local || !have_non_local_packs) &&
2787 !incremental;
69e4b342
JK
2788}
2789
6b8fda2d
VM
2790static int get_object_list_from_bitmap(struct rev_info *revs)
2791{
2792 if (prepare_bitmap_walk(revs) < 0)
2793 return -1;
2794
69e4b342
JK
2795 if (pack_options_allow_reuse() &&
2796 !reuse_partial_packfile_from_bitmap(
6b8fda2d
VM
2797 &reuse_packfile,
2798 &reuse_packfile_objects,
2799 &reuse_packfile_offset)) {
2800 assert(reuse_packfile_objects);
2801 nr_result += reuse_packfile_objects;
78d2214e 2802 display_progress(progress_state, nr_result);
6b8fda2d
VM
2803 }
2804
2805 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2806 return 0;
2807}
2808
abcb8655 2809static void record_recent_object(struct object *obj,
de1e67d0 2810 const char *name,
abcb8655
JK
2811 void *data)
2812{
910650d2 2813 oid_array_append(&recent_objects, &obj->oid);
abcb8655
JK
2814}
2815
2816static void record_recent_commit(struct commit *commit, void *data)
2817{
910650d2 2818 oid_array_append(&recent_objects, &commit->object.oid);
abcb8655
JK
2819}
2820
8d1d8f83 2821static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2822{
2823 struct rev_info revs;
2824 char line[1000];
b5d97e6b
JH
2825 int flags = 0;
2826
b5d97e6b
JH
2827 init_revisions(&revs, NULL);
2828 save_commit_buffer = 0;
b5d97e6b
JH
2829 setup_revisions(ac, av, &revs, NULL);
2830
b790e0f6
NTND
2831 /* make sure shallows are read */
2832 is_repository_shallow();
2833
b5d97e6b
JH
2834 while (fgets(line, sizeof(line), stdin) != NULL) {
2835 int len = strlen(line);
872c930d 2836 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2837 line[--len] = 0;
2838 if (!len)
2839 break;
2840 if (*line == '-') {
2841 if (!strcmp(line, "--not")) {
2842 flags ^= UNINTERESTING;
7cc8f971 2843 write_bitmap_index = 0;
b5d97e6b
JH
2844 continue;
2845 }
b790e0f6 2846 if (starts_with(line, "--shallow ")) {
e92b848c 2847 struct object_id oid;
2848 if (get_oid_hex(line + 10, &oid))
b790e0f6 2849 die("not an SHA-1 '%s'", line + 10);
e92b848c 2850 register_shallow(&oid);
f7f91086 2851 use_bitmap_index = 0;
b790e0f6
NTND
2852 continue;
2853 }
b5d97e6b
JH
2854 die("not a rev '%s'", line);
2855 }
8e676e8b 2856 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
b5d97e6b
JH
2857 die("bad revision '%s'", line);
2858 }
2859
6b8fda2d
VM
2860 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2861 return;
2862
3d51e1b5
MK
2863 if (prepare_revision_walk(&revs))
2864 die("revision walk setup failed");
e76a5fb4 2865 mark_edges_uninteresting(&revs, show_edge);
9535ce73
JH
2866
2867 if (!fn_show_object)
2868 fn_show_object = show_object;
2869 traverse_commit_list_filtered(&filter_options, &revs,
2870 show_commit, fn_show_object, NULL,
2871 NULL);
08cdfb13 2872
abcb8655
JK
2873 if (unpack_unreachable_expiration) {
2874 revs.ignore_missing_links = 1;
2875 if (add_unseen_recent_objects_to_traversal(&revs,
2876 unpack_unreachable_expiration))
2877 die("unable to add recent objects");
2878 if (prepare_revision_walk(&revs))
2879 die("revision walk setup failed");
2880 traverse_commit_list(&revs, record_recent_commit,
2881 record_recent_object, NULL);
2882 }
2883
08cdfb13
JH
2884 if (keep_unreachable)
2885 add_objects_in_unpacked_packs(&revs);
e26a8c47
JK
2886 if (pack_loose_unreachable)
2887 add_unreachable_loose_objects();
ca11b212
NP
2888 if (unpack_unreachable)
2889 loosen_unused_packed_objects(&revs);
abcb8655 2890
910650d2 2891 oid_array_clear(&recent_objects);
b5d97e6b
JH
2892}
2893
99fb6e04
NTND
2894static int option_parse_index_version(const struct option *opt,
2895 const char *arg, int unset)
2896{
2897 char *c;
2898 const char *val = arg;
2899 pack_idx_opts.version = strtoul(val, &c, 10);
2900 if (pack_idx_opts.version > 2)
2901 die(_("unsupported index version %s"), val);
2902 if (*c == ',' && c[1])
2903 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2904 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2905 die(_("bad index version '%s'"), val);
2906 return 0;
2907}
2908
7e52f566
JK
2909static int option_parse_unpack_unreachable(const struct option *opt,
2910 const char *arg, int unset)
2911{
2912 if (unset) {
2913 unpack_unreachable = 0;
2914 unpack_unreachable_expiration = 0;
2915 }
2916 else {
2917 unpack_unreachable = 1;
2918 if (arg)
2919 unpack_unreachable_expiration = approxidate(arg);
2920 }
2921 return 0;
2922}
2923
b5d97e6b
JH
2924int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2925{
b5d97e6b 2926 int use_internal_rev_list = 0;
8d1d8f83 2927 int thin = 0;
2dacf26d 2928 int shallow = 0;
4f366275 2929 int all_progress_implied = 0;
edfbb2aa 2930 struct argv_array rp = ARGV_ARRAY_INIT;
99fb6e04 2931 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
c90f9e13 2932 int rev_list_index = 0;
99fb6e04
NTND
2933 struct option pack_objects_options[] = {
2934 OPT_SET_INT('q', "quiet", &progress,
4c688120 2935 N_("do not show progress meter"), 0),
99fb6e04 2936 OPT_SET_INT(0, "progress", &progress,
4c688120 2937 N_("show progress meter"), 1),
99fb6e04 2938 OPT_SET_INT(0, "all-progress", &progress,
4c688120 2939 N_("show progress meter during object writing phase"), 2),
99fb6e04
NTND
2940 OPT_BOOL(0, "all-progress-implied",
2941 &all_progress_implied,
4c688120
NTND
2942 N_("similar to --all-progress when progress meter is shown")),
2943 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2944 N_("write the pack index file in the specified idx format version"),
99fb6e04 2945 0, option_parse_index_version },
2a514ed8
CB
2946 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2947 N_("maximum size of each output pack file")),
99fb6e04 2948 OPT_BOOL(0, "local", &local,
4c688120 2949 N_("ignore borrowed objects from alternate object store")),
99fb6e04 2950 OPT_BOOL(0, "incremental", &incremental,
4c688120 2951 N_("ignore packed objects")),
99fb6e04 2952 OPT_INTEGER(0, "window", &window,
4c688120 2953 N_("limit pack window by objects")),
2a514ed8
CB
2954 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2955 N_("limit pack window by memory in addition to object limit")),
99fb6e04 2956 OPT_INTEGER(0, "depth", &depth,
4c688120 2957 N_("maximum length of delta chain allowed in the resulting pack")),
99fb6e04 2958 OPT_BOOL(0, "reuse-delta", &reuse_delta,
4c688120 2959 N_("reuse existing deltas")),
99fb6e04 2960 OPT_BOOL(0, "reuse-object", &reuse_object,
4c688120 2961 N_("reuse existing objects")),
99fb6e04 2962 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
4c688120 2963 N_("use OFS_DELTA objects")),
99fb6e04 2964 OPT_INTEGER(0, "threads", &delta_search_threads,
4c688120 2965 N_("use threads when searching for best delta matches")),
99fb6e04 2966 OPT_BOOL(0, "non-empty", &non_empty,
4c688120 2967 N_("do not create an empty pack output")),
99fb6e04 2968 OPT_BOOL(0, "revs", &use_internal_rev_list,
4c688120 2969 N_("read revision arguments from standard input")),
99fb6e04 2970 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
4c688120 2971 N_("limit the objects to those that are not yet packed"),
99fb6e04
NTND
2972 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2973 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
4c688120 2974 N_("include objects reachable from any reference"),
99fb6e04
NTND
2975 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2976 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
4c688120 2977 N_("include objects referred by reflog entries"),
99fb6e04 2978 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
c90f9e13
JK
2979 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2980 N_("include objects referred to by the index"),
2981 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
99fb6e04 2982 OPT_BOOL(0, "stdout", &pack_to_stdout,
4c688120 2983 N_("output pack to stdout")),
99fb6e04 2984 OPT_BOOL(0, "include-tag", &include_tag,
4c688120 2985 N_("include tag objects that refer to objects to be packed")),
99fb6e04 2986 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
4c688120 2987 N_("keep unreachable objects")),
e26a8c47
JK
2988 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2989 N_("pack loose unreachable objects")),
4c688120
NTND
2990 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2991 N_("unpack unreachable objects newer than <time>"),
7e52f566 2992 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
99fb6e04 2993 OPT_BOOL(0, "thin", &thin,
4c688120 2994 N_("create thin packs")),
2dacf26d 2995 OPT_BOOL(0, "shallow", &shallow,
2996 N_("create packs suitable for shallow fetches")),
99fb6e04 2997 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
4c688120 2998 N_("ignore packs that have companion .keep file")),
99fb6e04 2999 OPT_INTEGER(0, "compression", &pack_compression_level,
4c688120 3000 N_("pack compression level")),
99fb6e04 3001 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
4c688120 3002 N_("do not hide commits by grafts"), 0),
6b8fda2d
VM
3003 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3004 N_("use a bitmap index if available to speed up counting objects")),
7cc8f971
VM
3005 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3006 N_("write a bitmap index together with the pack index")),
9535ce73
JH
3007 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3008 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3009 N_("handling for missing objects"), PARSE_OPT_NONEG,
3010 option_parse_missing_action },
99fb6e04
NTND
3011 OPT_END(),
3012 };
8d1d8f83 3013
afc711b8 3014 check_replace_refs = 0;
dae556bd 3015
ebcfb379 3016 reset_pack_idx_option(&pack_idx_opts);
ef90d6d4 3017 git_config(git_pack_config, NULL);
b5d97e6b
JH
3018
3019 progress = isatty(2);
99fb6e04
NTND
3020 argc = parse_options(argc, argv, prefix, pack_objects_options,
3021 pack_usage, 0);
b5d97e6b 3022
99fb6e04
NTND
3023 if (argc) {
3024 base_name = argv[0];
3025 argc--;
b5d97e6b 3026 }
99fb6e04
NTND
3027 if (pack_to_stdout != !base_name || argc)
3028 usage_with_options(pack_usage, pack_objects_options);
b5d97e6b 3029
edfbb2aa 3030 argv_array_push(&rp, "pack-objects");
99fb6e04
NTND
3031 if (thin) {
3032 use_internal_rev_list = 1;
2dacf26d 3033 argv_array_push(&rp, shallow
3034 ? "--objects-edge-aggressive"
3035 : "--objects-edge");
99fb6e04 3036 } else
edfbb2aa 3037 argv_array_push(&rp, "--objects");
b5d97e6b 3038
99fb6e04
NTND
3039 if (rev_list_all) {
3040 use_internal_rev_list = 1;
edfbb2aa 3041 argv_array_push(&rp, "--all");
99fb6e04
NTND
3042 }
3043 if (rev_list_reflog) {
3044 use_internal_rev_list = 1;
edfbb2aa 3045 argv_array_push(&rp, "--reflog");
99fb6e04 3046 }
c90f9e13
JK
3047 if (rev_list_index) {
3048 use_internal_rev_list = 1;
3049 argv_array_push(&rp, "--indexed-objects");
99fb6e04
NTND
3050 }
3051 if (rev_list_unpacked) {
3052 use_internal_rev_list = 1;
edfbb2aa 3053 argv_array_push(&rp, "--unpacked");
99fb6e04 3054 }
b5d97e6b 3055
99fb6e04
NTND
3056 if (!reuse_object)
3057 reuse_delta = 0;
3058 if (pack_compression_level == -1)
3059 pack_compression_level = Z_DEFAULT_COMPRESSION;
3060 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3061 die("bad pack compression level %d", pack_compression_level);
0c45d258
JH
3062
3063 if (!delta_search_threads) /* --threads=0 means autodetect */
3064 delta_search_threads = online_cpus();
3065
99fb6e04
NTND
3066#ifdef NO_PTHREADS
3067 if (delta_search_threads != 1)
2b34e486 3068 warning("no threads support, ignoring --threads");
99fb6e04 3069#endif
2b84b5a8
JS
3070 if (!pack_to_stdout && !pack_size_limit)
3071 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
3072 if (pack_to_stdout && pack_size_limit)
3073 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
3074 if (pack_size_limit && pack_size_limit < 1024*1024) {
3075 warning("minimum pack size limit is 1 MiB");
3076 pack_size_limit = 1024*1024;
3077 }
01c12a23 3078
8d1d8f83
JH
3079 if (!pack_to_stdout && thin)
3080 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 3081
ca11b212
NP
3082 if (keep_unreachable && unpack_unreachable)
3083 die("--keep-unreachable and --unpack-unreachable are incompatible.");
b1e757f3
JK
3084 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3085 unpack_unreachable_expiration = 0;
ca11b212 3086
9535ce73
JH
3087 if (filter_options.choice) {
3088 if (!pack_to_stdout)
3089 die("cannot use --filter without --stdout.");
3090 use_bitmap_index = 0;
3091 }
3092
645c432d
KS
3093 /*
3094 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3095 *
3096 * - to produce good pack (with bitmap index not-yet-packed objects are
3097 * packed in suboptimal order).
3098 *
3099 * - to use more robust pack-generation codepath (avoiding possible
3100 * bugs in bitmap code and possible bitmap index corruption).
3101 */
3102 if (!pack_to_stdout)
3103 use_bitmap_index_default = 0;
3104
3105 if (use_bitmap_index < 0)
3106 use_bitmap_index = use_bitmap_index_default;
3107
3108 /* "hard" reasons not to use bitmaps; these just won't work at all */
3109 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
6b8fda2d
VM
3110 use_bitmap_index = 0;
3111
7cc8f971
VM
3112 if (pack_to_stdout || !rev_list_all)
3113 write_bitmap_index = 0;
3114
4f366275
NP
3115 if (progress && all_progress_implied)
3116 progress = 2;
3117
b5d97e6b 3118 prepare_packed_git();
56dfeb62
JK
3119 if (ignore_packed_keep) {
3120 struct packed_git *p;
3121 for (p = packed_git; p; p = p->next)
3122 if (p->pack_local && p->pack_keep)
3123 break;
3124 if (!p) /* no keep-able packs found */
3125 ignore_packed_keep = 0;
3126 }
3127 if (local) {
3128 /*
3129 * unlike ignore_packed_keep above, we do not want to
3130 * unset "local" based on looking at packs, as it
3131 * also covers non-local objects
3132 */
3133 struct packed_git *p;
3134 for (p = packed_git; p; p = p->next) {
3135 if (!p->pack_local) {
3136 have_non_local_packs = 1;
3137 break;
3138 }
3139 }
3140 }
b5d97e6b 3141
13aaf148 3142 if (progress)
754dbc43 3143 progress_state = start_progress(_("Counting objects"), 0);
b5d97e6b
JH
3144 if (!use_internal_rev_list)
3145 read_object_list_from_stdin();
8d1d8f83 3146 else {
edfbb2aa
JK
3147 get_object_list(rp.argc, rp.argv);
3148 argv_array_clear(&rp);
8d1d8f83 3149 }
0ef95f72 3150 cleanup_preferred_base();
d155254c
MH
3151 if (include_tag && nr_result)
3152 for_each_ref(add_ref_tag, NULL);
4d4fcc54 3153 stop_progress(&progress_state);
96a02f8f 3154
f0b0af1b 3155 if (non_empty && !nr_result)
1c4a2912 3156 return 0;
f7ae6a93
NP
3157 if (nr_result)
3158 prepare_pack(window, depth);
d01fb92f 3159 write_pack_file();
ab7cd7bb 3160 if (progress)
6e1c2344
RJ
3161 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3162 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 3163 written, written_delta, reused, reused_delta);
c323ac7d
LT
3164 return 0;
3165}