repo-settings: consolidate some config settings
[git/git.git] / builtin / pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d 2#include "cache.h"
a80d72db 3#include "repository.h"
b2141fc1 4#include "config.h"
a74db82e 5#include "attr.h"
c323ac7d 6#include "object.h"
8e440259
PE
7#include "blob.h"
8#include "commit.h"
9#include "tag.h"
10#include "tree.h"
c323ac7d 11#include "delta.h"
a733cb60 12#include "pack.h"
3449f8c4 13#include "pack-revindex.h"
c38138cd 14#include "csum-file.h"
1b0c7174 15#include "tree-walk.h"
b5d97e6b
JH
16#include "diff.h"
17#include "revision.h"
18#include "list-objects.h"
9535ce73
JH
19#include "list-objects-filter.h"
20#include "list-objects-filter-options.h"
2834bc27 21#include "pack-objects.h"
96a02f8f 22#include "progress.h"
f0a24aa5 23#include "refs.h"
cf2ba13a 24#include "streaming.h"
93749194 25#include "thread-utils.h"
6b8fda2d 26#include "pack-bitmap.h"
28b8a730 27#include "delta-islands.h"
abcb8655
JK
28#include "reachable.h"
29#include "sha1-array.h"
edfbb2aa 30#include "argv-array.h"
ec2dd32c 31#include "list.h"
0317f455 32#include "packfile.h"
a80d72db 33#include "object-store.h"
ed7e5fc3 34#include "dir.h"
6a22d521 35#include "midx.h"
ae417807 36#include "trace2.h"
8ecce684 37
43fa44fa 38#define IN_PACK(obj) oe_in_pack(&to_pack, obj)
ac77d0c3
NTND
39#define SIZE(obj) oe_size(&to_pack, obj)
40#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
0aca34e8 41#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
898eba5e
NTND
42#define DELTA(obj) oe_delta(&to_pack, obj)
43#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
44#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
45#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
6a1e32d5 46#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
0aca34e8 47#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
898eba5e
NTND
48#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
49#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
43fa44fa 50
99fb6e04 51static const char *pack_usage[] = {
b8c1d275
AH
52 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
53 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
99fb6e04
NTND
54 NULL
55};
c323ac7d 56
3f9ac8d2 57/*
2834bc27
VM
58 * Objects we are going to pack are collected in the `to_pack` structure.
59 * It contains an array (dynamically expanded) of the object data, and a map
60 * that can resolve SHA1s to their position in the array.
3f9ac8d2 61 */
2834bc27
VM
62static struct packing_data to_pack;
63
79814f42 64static struct pack_idx_entry **written_list;
5af05043 65static uint32_t nr_result, nr_written, nr_seen;
6a1e32d5 66static struct bitmap_index *bitmap_git;
28b8a730 67static uint32_t write_layer;
3f9ac8d2 68
96f1e58f 69static int non_empty;
a7de7130 70static int reuse_delta = 1, reuse_object = 1;
e5e9714a 71static int keep_unreachable, unpack_unreachable, include_tag;
dddbad72 72static timestamp_t unpack_unreachable_expiration;
e26a8c47 73static int pack_loose_unreachable;
96f1e58f 74static int local;
56dfeb62 75static int have_non_local_packs;
96f1e58f 76static int incremental;
ed7e5fc3
NTND
77static int ignore_packed_keep_on_disk;
78static int ignore_packed_keep_in_core;
be6b1914 79static int allow_ofs_delta;
ebcfb379 80static struct pack_idx_option pack_idx_opts;
d01fb92f 81static const char *base_name;
024701f1 82static int progress = 1;
4812a93a 83static int window = 10;
568508e7 84static unsigned long pack_size_limit;
618e613a 85static int depth = 50;
43cc2b42 86static int delta_search_threads;
df6d6101 87static int pack_to_stdout;
4f6d26b1 88static int sparse;
6a1e32d5 89static int thin;
8d1d8f83 90static int num_preferred_base;
dc6a0757 91static struct progress *progress_state;
c323ac7d 92
6b8fda2d
VM
93static struct packed_git *reuse_packfile;
94static uint32_t reuse_packfile_objects;
95static off_t reuse_packfile_offset;
96
645c432d
KS
97static int use_bitmap_index_default = 1;
98static int use_bitmap_index = -1;
7cc8f971 99static int write_bitmap_index;
d4316604 100static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
6b8fda2d 101
0c16cd49
JT
102static int exclude_promisor_objects;
103
28b8a730
JK
104static int use_delta_islands;
105
074b2eea 106static unsigned long delta_cache_size = 0;
9806f5a7 107static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
e3dfddb3 108static unsigned long cache_max_small_delta_size = 1000;
074b2eea 109
a97773ce
BD
110static unsigned long window_memory_limit = 0;
111
9535ce73
JH
112static struct list_objects_filter_options filter_options;
113
114enum missing_action {
0c16cd49
JT
115 MA_ERROR = 0, /* fail if any missing objects are encountered */
116 MA_ALLOW_ANY, /* silently allow ALL missing objects */
117 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
9535ce73
JH
118};
119static enum missing_action arg_missing_action;
120static show_object_fn fn_show_object;
121
3f9ac8d2
JH
122/*
123 * stats
124 */
7cadf491
SP
125static uint32_t written, written_delta;
126static uint32_t reused, reused_delta;
3f9ac8d2 127
7cc8f971
VM
128/*
129 * Indexed commits
130 */
131static struct commit **indexed_commits;
132static unsigned int indexed_commits_nr;
133static unsigned int indexed_commits_alloc;
134
135static void index_commit_for_bitmap(struct commit *commit)
136{
137 if (indexed_commits_nr >= indexed_commits_alloc) {
138 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
2756ca43 139 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
7cc8f971
VM
140 }
141
142 indexed_commits[indexed_commits_nr++] = commit;
143}
780e6e73 144
3613f9b4 145static void *get_delta(struct object_entry *entry)
c323ac7d 146{
3613f9b4
NP
147 unsigned long size, base_size, delta_size;
148 void *buf, *base_buf, *delta_buf;
21666f1a 149 enum object_type type;
c323ac7d 150
b4f5aca4 151 buf = read_object_file(&entry->idx.oid, &type, &size);
3613f9b4 152 if (!buf)
f616db6a 153 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
898eba5e
NTND
154 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
155 &base_size);
3613f9b4 156 if (!base_buf)
e6a492b7 157 die("unable to read %s",
898eba5e 158 oid_to_hex(&DELTA(entry)->idx.oid));
3613f9b4 159 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 160 buf, size, &delta_size, 0);
1a07e59c
NTND
161 /*
162 * We succesfully computed this delta once but dropped it for
163 * memory reasons. Something is very wrong if this time we
164 * recompute and create a different delta.
165 */
0aca34e8 166 if (!delta_buf || delta_size != DELTA_SIZE(entry))
1a07e59c 167 BUG("delta size changed");
3613f9b4
NP
168 free(buf);
169 free(base_buf);
c323ac7d
LT
170 return delta_buf;
171}
172
30ebb40a
NP
173static unsigned long do_compress(void **pptr, unsigned long size)
174{
ef49a7a0 175 git_zstream stream;
30ebb40a
NP
176 void *in, *out;
177 unsigned long maxsize;
178
55bb5c91 179 git_deflate_init(&stream, pack_compression_level);
225a6f10 180 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
181
182 in = *pptr;
183 out = xmalloc(maxsize);
184 *pptr = out;
185
186 stream.next_in = in;
187 stream.avail_in = size;
188 stream.next_out = out;
189 stream.avail_out = maxsize;
55bb5c91 190 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 191 ; /* nothing */
55bb5c91 192 git_deflate_end(&stream);
30ebb40a
NP
193
194 free(in);
195 return stream.total_out;
196}
197
98a3beab 198static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
188960b4 199 const struct object_id *oid)
cf2ba13a
NTND
200{
201 git_zstream stream;
202 unsigned char ibuf[1024 * 16];
203 unsigned char obuf[1024 * 16];
204 unsigned long olen = 0;
205
cf2ba13a
NTND
206 git_deflate_init(&stream, pack_compression_level);
207
208 for (;;) {
209 ssize_t readlen;
210 int zret = Z_OK;
211 readlen = read_istream(st, ibuf, sizeof(ibuf));
212 if (readlen == -1)
188960b4 213 die(_("unable to read %s"), oid_to_hex(oid));
cf2ba13a
NTND
214
215 stream.next_in = ibuf;
216 stream.avail_in = readlen;
217 while ((stream.avail_in || readlen == 0) &&
218 (zret == Z_OK || zret == Z_BUF_ERROR)) {
219 stream.next_out = obuf;
220 stream.avail_out = sizeof(obuf);
221 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
98a3beab 222 hashwrite(f, obuf, stream.next_out - obuf);
cf2ba13a
NTND
223 olen += stream.next_out - obuf;
224 }
225 if (stream.avail_in)
226 die(_("deflate error (%d)"), zret);
227 if (readlen == 0) {
228 if (zret != Z_STREAM_END)
229 die(_("deflate error (%d)"), zret);
230 break;
231 }
232 }
233 git_deflate_end(&stream);
234 return olen;
235}
236
780e6e73
NP
237/*
238 * we are going to reuse the existing object data as is. make
239 * sure it is not corrupt.
240 */
079afb18
SP
241static int check_pack_inflate(struct packed_git *p,
242 struct pack_window **w_curs,
6777a59f
SP
243 off_t offset,
244 off_t len,
079afb18
SP
245 unsigned long expect)
246{
ef49a7a0 247 git_zstream stream;
079afb18
SP
248 unsigned char fakebuf[4096], *in;
249 int st;
250
251 memset(&stream, 0, sizeof(stream));
39c68542 252 git_inflate_init(&stream);
079afb18
SP
253 do {
254 in = use_pack(p, w_curs, offset, &stream.avail_in);
255 stream.next_in = in;
256 stream.next_out = fakebuf;
257 stream.avail_out = sizeof(fakebuf);
39c68542 258 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
259 offset += stream.next_in - in;
260 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 261 git_inflate_end(&stream);
079afb18
SP
262 return (st == Z_STREAM_END &&
263 stream.total_out == expect &&
264 stream.total_in == len) ? 0 : -1;
265}
266
98a3beab 267static void copy_pack_data(struct hashfile *f,
079afb18
SP
268 struct packed_git *p,
269 struct pack_window **w_curs,
6777a59f
SP
270 off_t offset,
271 off_t len)
079afb18
SP
272{
273 unsigned char *in;
ef49a7a0 274 unsigned long avail;
079afb18
SP
275
276 while (len) {
277 in = use_pack(p, w_curs, offset, &avail);
278 if (avail > len)
ef49a7a0 279 avail = (unsigned long)len;
98a3beab 280 hashwrite(f, in, avail);
079afb18
SP
281 offset += avail;
282 len -= avail;
283 }
284}
285
1b4bb16b 286/* Return 0 if we will bust the pack-size limit */
98a3beab 287static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
c9018b03 288 unsigned long limit, int usable_delta)
c323ac7d 289{
c9018b03 290 unsigned long size, datalen;
2c5e2865
JK
291 unsigned char header[MAX_PACK_OBJECT_HEADER],
292 dheader[MAX_PACK_OBJECT_HEADER];
6777a59f 293 unsigned hdrlen;
2c5ef824 294 enum object_type type;
c9018b03 295 void *buf;
cf2ba13a 296 struct git_istream *st = NULL;
41179100 297 const unsigned hashsz = the_hash_algo->rawsz;
c9018b03
NTND
298
299 if (!usable_delta) {
fd9b1bae 300 if (oe_type(entry) == OBJ_BLOB &&
ac77d0c3 301 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
ef7b5195 302 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
cf2ba13a
NTND
303 buf = NULL;
304 else {
b4f5aca4 305 buf = read_object_file(&entry->idx.oid, &type, &size);
cf2ba13a 306 if (!buf)
e6a492b7 307 die(_("unable to read %s"),
308 oid_to_hex(&entry->idx.oid));
cf2ba13a 309 }
c9018b03
NTND
310 /*
311 * make sure no cached delta data remains from a
312 * previous attempt before a pack split occurred.
313 */
6a83d902 314 FREE_AND_NULL(entry->delta_data);
c9018b03
NTND
315 entry->z_delta_size = 0;
316 } else if (entry->delta_data) {
0aca34e8 317 size = DELTA_SIZE(entry);
c9018b03
NTND
318 buf = entry->delta_data;
319 entry->delta_data = NULL;
898eba5e 320 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03
NTND
321 OBJ_OFS_DELTA : OBJ_REF_DELTA;
322 } else {
323 buf = get_delta(entry);
0aca34e8 324 size = DELTA_SIZE(entry);
898eba5e 325 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03
NTND
326 OBJ_OFS_DELTA : OBJ_REF_DELTA;
327 }
328
cf2ba13a
NTND
329 if (st) /* large blob case, just assume we don't compress well */
330 datalen = size;
331 else if (entry->z_delta_size)
c9018b03
NTND
332 datalen = entry->z_delta_size;
333 else
334 datalen = do_compress(&buf, size);
335
336 /*
337 * The object header is a byte of 'type' followed by zero or
338 * more bytes of length.
339 */
7202a6fa
JK
340 hdrlen = encode_in_pack_object_header(header, sizeof(header),
341 type, size);
c9018b03
NTND
342
343 if (type == OBJ_OFS_DELTA) {
344 /*
345 * Deltas with relative base contain an additional
346 * encoding of the relative offset for the delta
347 * base from this object's position in the pack.
348 */
898eba5e 349 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
c9018b03
NTND
350 unsigned pos = sizeof(dheader) - 1;
351 dheader[pos] = ofs & 127;
352 while (ofs >>= 7)
353 dheader[--pos] = 128 | (--ofs & 127);
41179100 354 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
cf2ba13a
NTND
355 if (st)
356 close_istream(st);
c9018b03
NTND
357 free(buf);
358 return 0;
359 }
98a3beab 360 hashwrite(f, header, hdrlen);
361 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
c9018b03
NTND
362 hdrlen += sizeof(dheader) - pos;
363 } else if (type == OBJ_REF_DELTA) {
364 /*
365 * Deltas with a base reference contain
41179100 366 * additional bytes for the base object ID.
c9018b03 367 */
41179100 368 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
cf2ba13a
NTND
369 if (st)
370 close_istream(st);
c9018b03
NTND
371 free(buf);
372 return 0;
373 }
98a3beab 374 hashwrite(f, header, hdrlen);
42c8ce1c 375 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
41179100 376 hdrlen += hashsz;
c9018b03 377 } else {
41179100 378 if (limit && hdrlen + datalen + hashsz >= limit) {
cf2ba13a
NTND
379 if (st)
380 close_istream(st);
c9018b03
NTND
381 free(buf);
382 return 0;
383 }
98a3beab 384 hashwrite(f, header, hdrlen);
c9018b03 385 }
cf2ba13a 386 if (st) {
188960b4 387 datalen = write_large_blob_data(st, f, &entry->idx.oid);
cf2ba13a
NTND
388 close_istream(st);
389 } else {
98a3beab 390 hashwrite(f, buf, datalen);
cf2ba13a
NTND
391 free(buf);
392 }
c9018b03
NTND
393
394 return hdrlen + datalen;
395}
396
397/* Return 0 if we will bust the pack-size limit */
98a3beab 398static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
af92a645 399 unsigned long limit, int usable_delta)
c9018b03 400{
43fa44fa 401 struct packed_git *p = IN_PACK(entry);
c9018b03
NTND
402 struct pack_window *w_curs = NULL;
403 struct revindex_entry *revidx;
404 off_t offset;
fd9b1bae 405 enum object_type type = oe_type(entry);
211c61c6 406 off_t datalen;
2c5e2865
JK
407 unsigned char header[MAX_PACK_OBJECT_HEADER],
408 dheader[MAX_PACK_OBJECT_HEADER];
c9018b03 409 unsigned hdrlen;
41179100 410 const unsigned hashsz = the_hash_algo->rawsz;
ac77d0c3 411 unsigned long entry_size = SIZE(entry);
c9018b03 412
898eba5e
NTND
413 if (DELTA(entry))
414 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03 415 OBJ_OFS_DELTA : OBJ_REF_DELTA;
7202a6fa 416 hdrlen = encode_in_pack_object_header(header, sizeof(header),
ac77d0c3 417 type, entry_size);
c9018b03
NTND
418
419 offset = entry->in_pack_offset;
420 revidx = find_pack_revindex(p, offset);
421 datalen = revidx[1].offset - offset;
422 if (!pack_to_stdout && p->index_version > 1 &&
423 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
f616db6a 424 error(_("bad packed object CRC for %s"),
e6a492b7 425 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
426 unuse_pack(&w_curs);
427 return write_no_reuse_object(f, entry, limit, usable_delta);
428 }
429
430 offset += entry->in_pack_header_size;
431 datalen -= entry->in_pack_header_size;
432
433 if (!pack_to_stdout && p->index_version == 1 &&
ac77d0c3 434 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
f616db6a 435 error(_("corrupt packed object for %s"),
e6a492b7 436 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
437 unuse_pack(&w_curs);
438 return write_no_reuse_object(f, entry, limit, usable_delta);
439 }
440
441 if (type == OBJ_OFS_DELTA) {
898eba5e 442 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
c9018b03
NTND
443 unsigned pos = sizeof(dheader) - 1;
444 dheader[pos] = ofs & 127;
445 while (ofs >>= 7)
446 dheader[--pos] = 128 | (--ofs & 127);
41179100 447 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
c9018b03
NTND
448 unuse_pack(&w_curs);
449 return 0;
450 }
98a3beab 451 hashwrite(f, header, hdrlen);
452 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
c9018b03
NTND
453 hdrlen += sizeof(dheader) - pos;
454 reused_delta++;
455 } else if (type == OBJ_REF_DELTA) {
41179100 456 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
c9018b03
NTND
457 unuse_pack(&w_curs);
458 return 0;
459 }
98a3beab 460 hashwrite(f, header, hdrlen);
42c8ce1c 461 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
41179100 462 hdrlen += hashsz;
c9018b03
NTND
463 reused_delta++;
464 } else {
41179100 465 if (limit && hdrlen + datalen + hashsz >= limit) {
c9018b03
NTND
466 unuse_pack(&w_curs);
467 return 0;
468 }
98a3beab 469 hashwrite(f, header, hdrlen);
c9018b03
NTND
470 }
471 copy_pack_data(f, p, &w_curs, offset, datalen);
472 unuse_pack(&w_curs);
473 reused++;
474 return hdrlen + datalen;
475}
476
477/* Return 0 if we will bust the pack-size limit */
98a3beab 478static off_t write_object(struct hashfile *f,
af92a645
NTND
479 struct object_entry *entry,
480 off_t write_offset)
c9018b03 481{
af92a645
NTND
482 unsigned long limit;
483 off_t len;
2c5ef824 484 int usable_delta, to_reuse;
c323ac7d 485
78d1e84f
NP
486 if (!pack_to_stdout)
487 crc32_begin(f);
488
a2430dde 489 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
490 if (!pack_size_limit || !nr_written)
491 limit = 0;
492 else if (pack_size_limit <= write_offset)
493 /*
494 * the earlier object did not fit the limit; avoid
495 * mistaking this with unlimited (i.e. limit = 0).
496 */
497 limit = 1;
498 else
499 limit = pack_size_limit - write_offset;
2c5ef824 500
898eba5e 501 if (!DELTA(entry))
2c5ef824
NP
502 usable_delta = 0; /* no delta */
503 else if (!pack_size_limit)
504 usable_delta = 1; /* unlimited packfile */
898eba5e 505 else if (DELTA(entry)->idx.offset == (off_t)-1)
2c5ef824 506 usable_delta = 0; /* base was written to another pack */
898eba5e 507 else if (DELTA(entry)->idx.offset)
2c5ef824
NP
508 usable_delta = 1; /* base already exists in this pack */
509 else
510 usable_delta = 0; /* base could end up in another pack */
511
a7de7130 512 if (!reuse_object)
fa736f72 513 to_reuse = 0; /* explicit */
43fa44fa 514 else if (!IN_PACK(entry))
ab7cd7bb 515 to_reuse = 0; /* can't reuse what we don't have */
fd9b1bae
NTND
516 else if (oe_type(entry) == OBJ_REF_DELTA ||
517 oe_type(entry) == OBJ_OFS_DELTA)
17b08f2c
DH
518 /* check_object() decided it for us ... */
519 to_reuse = usable_delta;
520 /* ... but pack split may override that */
fd9b1bae 521 else if (oe_type(entry) != entry->in_pack_type)
ab7cd7bb 522 to_reuse = 0; /* pack has delta which is unusable */
898eba5e 523 else if (DELTA(entry))
ab7cd7bb
JH
524 to_reuse = 0; /* we want to pack afresh */
525 else
526 to_reuse = 1; /* we have it in-pack undeltified,
527 * and we do not need to deltify it.
528 */
529
c9018b03
NTND
530 if (!to_reuse)
531 len = write_no_reuse_object(f, entry, limit, usable_delta);
532 else
533 len = write_reuse_object(f, entry, limit, usable_delta);
534 if (!len)
535 return 0;
64bd76b1 536
17b08f2c 537 if (usable_delta)
ab7cd7bb 538 written_delta++;
3f9ac8d2 539 written++;
78d1e84f 540 if (!pack_to_stdout)
aa7e44bf 541 entry->idx.crc32 = crc32_end(f);
c9018b03 542 return len;
c323ac7d
LT
543}
544
f63c79db
JH
545enum write_one_status {
546 WRITE_ONE_SKIP = -1, /* already written */
547 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
548 WRITE_ONE_WRITTEN = 1, /* normal */
549 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
550};
551
98a3beab 552static enum write_one_status write_one(struct hashfile *f,
f63c79db
JH
553 struct object_entry *e,
554 off_t *offset)
9d5ab962 555{
af92a645 556 off_t size;
f63c79db 557 int recursing;
d7dd0223 558
f63c79db
JH
559 /*
560 * we set offset to 1 (which is an impossible value) to mark
561 * the fact that this object is involved in "write its base
562 * first before writing a deltified object" recursion.
563 */
564 recursing = (e->idx.offset == 1);
565 if (recursing) {
f616db6a 566 warning(_("recursive delta detected for object %s"),
e6a492b7 567 oid_to_hex(&e->idx.oid));
f63c79db
JH
568 return WRITE_ONE_RECURSIVE;
569 } else if (e->idx.offset || e->preferred_base) {
570 /* offset is non zero if object is written already. */
571 return WRITE_ONE_SKIP;
572 }
d7dd0223 573
720c9f7b 574 /* if we are deltified, write out base object first. */
898eba5e 575 if (DELTA(e)) {
f63c79db 576 e->idx.offset = 1; /* now recurse */
898eba5e 577 switch (write_one(f, DELTA(e), offset)) {
f63c79db
JH
578 case WRITE_ONE_RECURSIVE:
579 /* we cannot depend on this one */
898eba5e 580 SET_DELTA(e, NULL);
f63c79db
JH
581 break;
582 default:
583 break;
584 case WRITE_ONE_BREAK:
585 e->idx.offset = recursing;
586 return WRITE_ONE_BREAK;
587 }
588 }
d7dd0223 589
6ed7f25e
NP
590 e->idx.offset = *offset;
591 size = write_object(f, e, *offset);
17b08f2c 592 if (!size) {
f63c79db
JH
593 e->idx.offset = recursing;
594 return WRITE_ONE_BREAK;
17b08f2c 595 }
79814f42 596 written_list[nr_written++] = &e->idx;
d7dd0223
NP
597
598 /* make sure off_t is sufficiently large not to wrap */
c03c8315 599 if (signed_add_overflows(*offset, size))
f616db6a 600 die(_("pack too large for current definition of off_t"));
6ed7f25e 601 *offset += size;
f63c79db 602 return WRITE_ONE_WRITTEN;
9d5ab962
JH
603}
604
d155254c 605static int mark_tagged(const char *path, const struct object_id *oid, int flag,
1b4bb16b
JH
606 void *cb_data)
607{
188960b4 608 struct object_id peeled;
3df28cae 609 struct object_entry *entry = packlist_find(&to_pack, oid, NULL);
1b4bb16b
JH
610
611 if (entry)
612 entry->tagged = 1;
b420d909 613 if (!peel_ref(path, &peeled)) {
3df28cae 614 entry = packlist_find(&to_pack, &peeled, NULL);
1b4bb16b
JH
615 if (entry)
616 entry->tagged = 1;
617 }
618 return 0;
619}
620
be126818 621static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 622 unsigned int *endp,
1b4bb16b
JH
623 struct object_entry *e)
624{
fe0ac2fb 625 if (e->filled || oe_layer(&to_pack, e) != write_layer)
1b4bb16b
JH
626 return;
627 wo[(*endp)++] = e;
628 e->filled = 1;
629}
630
631static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 632 unsigned int *endp,
1b4bb16b
JH
633 struct object_entry *e)
634{
f380872f
DM
635 int add_to_order = 1;
636 while (e) {
637 if (add_to_order) {
638 struct object_entry *s;
639 /* add this node... */
640 add_to_write_order(wo, endp, e);
641 /* all its siblings... */
898eba5e 642 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
f380872f
DM
643 add_to_write_order(wo, endp, s);
644 }
645 }
646 /* drop down a level to add left subtree nodes if possible */
898eba5e 647 if (DELTA_CHILD(e)) {
f380872f 648 add_to_order = 1;
898eba5e 649 e = DELTA_CHILD(e);
f380872f
DM
650 } else {
651 add_to_order = 0;
652 /* our sibling might have some children, it is next */
898eba5e
NTND
653 if (DELTA_SIBLING(e)) {
654 e = DELTA_SIBLING(e);
f380872f
DM
655 continue;
656 }
657 /* go back to our parent node */
898eba5e
NTND
658 e = DELTA(e);
659 while (e && !DELTA_SIBLING(e)) {
f380872f
DM
660 /* we're on the right side of a subtree, keep
661 * going up until we can go right again */
898eba5e 662 e = DELTA(e);
f380872f
DM
663 }
664 if (!e) {
665 /* done- we hit our original root node */
666 return;
667 }
668 /* pass it off to sibling at this level */
898eba5e 669 e = DELTA_SIBLING(e);
f380872f
DM
670 }
671 };
1b4bb16b
JH
672}
673
674static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 675 unsigned int *endp,
1b4bb16b
JH
676 struct object_entry *e)
677{
678 struct object_entry *root;
679
898eba5e 680 for (root = e; DELTA(root); root = DELTA(root))
1b4bb16b 681 ; /* nothing */
1b4bb16b
JH
682 add_descendants_to_write_order(wo, endp, root);
683}
684
f64ba53e 685static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
1b4bb16b 686{
f64ba53e 687 unsigned int i, last_untagged;
2834bc27 688 struct object_entry *objects = to_pack.objects;
1b4bb16b 689
2834bc27 690 for (i = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
691 if (objects[i].tagged)
692 break;
f64ba53e 693 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b 694 }
38d4debb 695 last_untagged = i;
1b4bb16b
JH
696
697 /*
698 * Then fill all the tagged tips.
699 */
2834bc27 700 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 701 if (objects[i].tagged)
f64ba53e 702 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
703 }
704
705 /*
706 * And then all remaining commits and tags.
707 */
2834bc27 708 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fd9b1bae
NTND
709 if (oe_type(&objects[i]) != OBJ_COMMIT &&
710 oe_type(&objects[i]) != OBJ_TAG)
1b4bb16b 711 continue;
f64ba53e 712 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
713 }
714
715 /*
716 * And then all the trees.
717 */
2834bc27 718 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fd9b1bae 719 if (oe_type(&objects[i]) != OBJ_TREE)
1b4bb16b 720 continue;
f64ba53e 721 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
722 }
723
724 /*
725 * Finally all the rest in really tight order
726 */
2834bc27 727 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fe0ac2fb 728 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
f64ba53e 729 add_family_to_write_order(wo, wo_end, &objects[i]);
38d4debb 730 }
f64ba53e
CC
731}
732
733static struct object_entry **compute_write_order(void)
734{
28b8a730 735 uint32_t max_layers = 1;
f64ba53e
CC
736 unsigned int i, wo_end;
737
738 struct object_entry **wo;
739 struct object_entry *objects = to_pack.objects;
740
741 for (i = 0; i < to_pack.nr_objects; i++) {
742 objects[i].tagged = 0;
743 objects[i].filled = 0;
744 SET_DELTA_CHILD(&objects[i], NULL);
745 SET_DELTA_SIBLING(&objects[i], NULL);
746 }
747
748 /*
749 * Fully connect delta_child/delta_sibling network.
750 * Make sure delta_sibling is sorted in the original
751 * recency order.
752 */
753 for (i = to_pack.nr_objects; i > 0;) {
754 struct object_entry *e = &objects[--i];
755 if (!DELTA(e))
756 continue;
757 /* Mark me as the first child */
758 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
759 SET_DELTA_CHILD(DELTA(e), e);
38d4debb
DM
760 }
761
f64ba53e
CC
762 /*
763 * Mark objects that are at the tip of tags.
764 */
765 for_each_tag_ref(mark_tagged, NULL);
766
28b8a730
JK
767 if (use_delta_islands)
768 max_layers = compute_pack_layers(&to_pack);
769
f64ba53e
CC
770 ALLOC_ARRAY(wo, to_pack.nr_objects);
771 wo_end = 0;
772
28b8a730
JK
773 for (; write_layer < max_layers; ++write_layer)
774 compute_layer_order(wo, &wo_end);
38d4debb 775
2834bc27 776 if (wo_end != to_pack.nr_objects)
f616db6a
NTND
777 die(_("ordered %u objects, expected %"PRIu32),
778 wo_end, to_pack.nr_objects);
1b4bb16b
JH
779
780 return wo;
781}
782
98a3beab 783static off_t write_reused_pack(struct hashfile *f)
6b8fda2d
VM
784{
785 unsigned char buffer[8192];
657673f1 786 off_t to_write, total;
6b8fda2d
VM
787 int fd;
788
789 if (!is_pack_valid(reuse_packfile))
f616db6a 790 die(_("packfile is invalid: %s"), reuse_packfile->pack_name);
6b8fda2d 791
a5436b57 792 fd = git_open(reuse_packfile->pack_name);
6b8fda2d 793 if (fd < 0)
f616db6a 794 die_errno(_("unable to open packfile for reuse: %s"),
6b8fda2d
VM
795 reuse_packfile->pack_name);
796
797 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
f616db6a 798 die_errno(_("unable to seek in reused packfile"));
6b8fda2d
VM
799
800 if (reuse_packfile_offset < 0)
41179100 801 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
6b8fda2d 802
657673f1 803 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
6b8fda2d
VM
804
805 while (to_write) {
806 int read_pack = xread(fd, buffer, sizeof(buffer));
807
808 if (read_pack <= 0)
f616db6a 809 die_errno(_("unable to read from reused packfile"));
6b8fda2d
VM
810
811 if (read_pack > to_write)
812 read_pack = to_write;
813
98a3beab 814 hashwrite(f, buffer, read_pack);
6b8fda2d 815 to_write -= read_pack;
657673f1
JK
816
817 /*
818 * We don't know the actual number of objects written,
819 * only how many bytes written, how many bytes total, and
820 * how many objects total. So we can fake it by pretending all
821 * objects we are writing are the same size. This gives us a
822 * smooth progress meter, and at the end it matches the true
823 * answer.
824 */
825 written = reuse_packfile_objects *
826 (((double)(total - to_write)) / total);
827 display_progress(progress_state, written);
6b8fda2d
VM
828 }
829
830 close(fd);
657673f1
JK
831 written = reuse_packfile_objects;
832 display_progress(progress_state, written);
6b8fda2d
VM
833 return reuse_packfile_offset - sizeof(struct pack_header);
834}
835
9cea46cd
EW
836static const char no_split_warning[] = N_(
837"disabling bitmap writing, packs are split due to pack.packSizeLimit"
838);
839
d01fb92f 840static void write_pack_file(void)
c323ac7d 841{
ebe27b13 842 uint32_t i = 0, j;
98a3beab 843 struct hashfile *f;
6ed7f25e 844 off_t offset;
ebe27b13 845 uint32_t nr_remaining = nr_result;
f746bae8 846 time_t last_mtime = 0;
1b4bb16b 847 struct object_entry **write_order;
81a216a5 848
bcd7954e 849 if (progress > pack_to_stdout)
754dbc43 850 progress_state = start_progress(_("Writing objects"), nr_result);
b32fa95f 851 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1b4bb16b 852 write_order = compute_write_order();
183bdb2c 853
ebe27b13 854 do {
188960b4 855 struct object_id oid;
7ba502c4 856 char *pack_tmp_name = NULL;
aa7e44bf 857
cdf9db3c 858 if (pack_to_stdout)
98a3beab 859 f = hashfd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
860 else
861 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 862
c0ad4657 863 offset = write_pack_header(f, nr_remaining);
6b8fda2d
VM
864
865 if (reuse_packfile) {
866 off_t packfile_size;
867 assert(pack_to_stdout);
868
869 packfile_size = write_reused_pack(f);
870 offset += packfile_size;
871 }
872
ebe27b13 873 nr_written = 0;
2834bc27 874 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 875 struct object_entry *e = write_order[i];
cddec4f8 876 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
877 break;
878 display_progress(progress_state, written);
879 }
c553ca25 880
ebe27b13
DH
881 /*
882 * Did we write the wrong # entries in the header?
883 * If so, rewrite it like in fast-import
884 */
54352bb2 885 if (pack_to_stdout) {
cfe83216 886 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
54352bb2 887 } else if (nr_written == nr_remaining) {
cfe83216 888 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
ebe27b13 889 } else {
f2af9f5e 890 int fd = finalize_hashfile(f, oid.hash, 0);
188960b4 891 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
892 nr_written, oid.hash, offset);
7ba502c4 893 close(fd);
9cea46cd
EW
894 if (write_bitmap_index) {
895 warning(_(no_split_warning));
896 write_bitmap_index = 0;
897 }
ebe27b13
DH
898 }
899
900 if (!pack_to_stdout) {
f746bae8 901 struct stat st;
58892711 902 struct strbuf tmpname = STRBUF_INIT;
d01fb92f 903
f746bae8
NP
904 /*
905 * Packs are runtime accessed in their mtime
906 * order since newer packs are more likely to contain
907 * younger objects. So if we are creating multiple
908 * packs then we should modify the mtime of later ones
909 * to preserve this property.
910 */
0e990530 911 if (stat(pack_tmp_name, &st) < 0) {
f616db6a 912 warning_errno(_("failed to stat %s"), pack_tmp_name);
f746bae8
NP
913 } else if (!last_mtime) {
914 last_mtime = st.st_mtime;
915 } else {
916 struct utimbuf utb;
917 utb.actime = st.st_atime;
918 utb.modtime = --last_mtime;
0e990530 919 if (utime(pack_tmp_name, &utb) < 0)
f616db6a 920 warning_errno(_("failed utime() on %s"), pack_tmp_name);
f746bae8
NP
921 }
922
58892711 923 strbuf_addf(&tmpname, "%s-", base_name);
7cc8f971
VM
924
925 if (write_bitmap_index) {
188960b4 926 bitmap_writer_set_checksum(oid.hash);
06af3bba
NTND
927 bitmap_writer_build_type_index(
928 &to_pack, written_list, nr_written);
7cc8f971
VM
929 }
930
58892711 931 finish_tmp_packfile(&tmpname, pack_tmp_name,
0e990530 932 written_list, nr_written,
188960b4 933 &pack_idx_opts, oid.hash);
7cc8f971
VM
934
935 if (write_bitmap_index) {
188960b4 936 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
7cc8f971
VM
937
938 stop_progress(&progress_state);
939
940 bitmap_writer_show_progress(progress);
941 bitmap_writer_reuse_bitmaps(&to_pack);
942 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
943 bitmap_writer_build(&to_pack);
ae4f07fb 944 bitmap_writer_finish(written_list, nr_written,
58892711 945 tmpname.buf, write_bitmap_options);
7cc8f971
VM
946 write_bitmap_index = 0;
947 }
948
58892711 949 strbuf_release(&tmpname);
7ba502c4 950 free(pack_tmp_name);
188960b4 951 puts(oid_to_hex(&oid));
ebe27b13
DH
952 }
953
954 /* mark written objects as written to previous pack */
955 for (j = 0; j < nr_written; j++) {
79814f42 956 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
957 }
958 nr_remaining -= nr_written;
2834bc27 959 } while (nr_remaining && i < to_pack.nr_objects);
ebe27b13
DH
960
961 free(written_list);
1b4bb16b 962 free(write_order);
4d4fcc54 963 stop_progress(&progress_state);
67c08ce1 964 if (written != nr_result)
f616db6a
NTND
965 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
966 written, nr_result);
9ed87902
JT
967 trace2_data_intmax("pack-objects", the_repository,
968 "write_pack_file/wrote", nr_result);
c323ac7d
LT
969}
970
a74db82e
JH
971static int no_try_delta(const char *path)
972{
2aef63d3 973 static struct attr_check *check;
a74db82e 974
2aef63d3
JH
975 if (!check)
976 check = attr_check_initl("delta", NULL);
f8adbec9 977 git_check_attr(the_repository->index, path, check);
2aef63d3 978 if (ATTR_FALSE(check->items[0].value))
a74db82e
JH
979 return 1;
980 return 0;
981}
982
ce2bc424
JK
983/*
984 * When adding an object, check whether we have already added it
985 * to our packing list. If so, we can skip. However, if we are
986 * being asked to excludei t, but the previous mention was to include
987 * it, make sure to adjust its flags and tweak our numbers accordingly.
988 *
989 * As an optimization, we pass out the index position where we would have
990 * found the item, since that saves us from having to look it up again a
991 * few lines later when we want to add the new entry.
992 */
188960b4 993static int have_duplicate_entry(const struct object_id *oid,
ce2bc424
JK
994 int exclude,
995 uint32_t *index_pos)
c323ac7d 996{
c323ac7d 997 struct object_entry *entry;
29b734e4 998
3df28cae 999 entry = packlist_find(&to_pack, oid, index_pos);
ce2bc424 1000 if (!entry)
29b734e4 1001 return 0;
ce2bc424
JK
1002
1003 if (exclude) {
1004 if (!entry->preferred_base)
1005 nr_result--;
1006 entry->preferred_base = 1;
29b734e4 1007 }
c323ac7d 1008
ce2bc424
JK
1009 return 1;
1010}
1011
702d1b95
KS
1012static int want_found_object(int exclude, struct packed_git *p)
1013{
1014 if (exclude)
1015 return 1;
1016 if (incremental)
1017 return 0;
1018
1019 /*
1020 * When asked to do --local (do not include an object that appears in a
1021 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1022 * an object that appears in a pack marked with .keep), finding a pack
1023 * that matches the criteria is sufficient for us to decide to omit it.
1024 * However, even if this pack does not satisfy the criteria, we need to
1025 * make sure no copy of this object appears in _any_ pack that makes us
1026 * to omit the object, so we need to check all the packs.
1027 *
1028 * We can however first check whether these options can possible matter;
1029 * if they do not matter we know we want the object in generated pack.
1030 * Otherwise, we signal "-1" at the end to tell the caller that we do
1031 * not know either way, and it needs to check more packs.
1032 */
ed7e5fc3
NTND
1033 if (!ignore_packed_keep_on_disk &&
1034 !ignore_packed_keep_in_core &&
702d1b95
KS
1035 (!local || !have_non_local_packs))
1036 return 1;
1037
1038 if (local && !p->pack_local)
1039 return 0;
ed7e5fc3
NTND
1040 if (p->pack_local &&
1041 ((ignore_packed_keep_on_disk && p->pack_keep) ||
1042 (ignore_packed_keep_in_core && p->pack_keep_in_core)))
702d1b95
KS
1043 return 0;
1044
1045 /* we don't know yet; keep looking for more packs */
1046 return -1;
1047}
1048
ce2bc424
JK
1049/*
1050 * Check whether we want the object in the pack (e.g., we do not want
1051 * objects found in non-local stores if the "--local" option was used).
1052 *
702d1b95
KS
1053 * If the caller already knows an existing pack it wants to take the object
1054 * from, that is passed in *found_pack and *found_offset; otherwise this
1055 * function finds if there is any pack that has the object and returns the pack
1056 * and its offset in these variables.
ce2bc424 1057 */
188960b4 1058static int want_object_in_pack(const struct object_id *oid,
ce2bc424
JK
1059 int exclude,
1060 struct packed_git **found_pack,
1061 off_t *found_offset)
1062{
702d1b95 1063 int want;
8865859d 1064 struct list_head *pos;
6a22d521 1065 struct multi_pack_index *m;
ce2bc424 1066
6862ebbf 1067 if (!exclude && local && has_loose_object_nonlocal(oid))
daae0625
BC
1068 return 0;
1069
702d1b95
KS
1070 /*
1071 * If we already know the pack object lives in, start checks from that
1072 * pack - in the usual case when neither --local was given nor .keep files
1073 * are present we will determine the answer right now.
1074 */
1075 if (*found_pack) {
1076 want = want_found_object(exclude, *found_pack);
1077 if (want != -1)
1078 return want;
1079 }
6a22d521
DS
1080
1081 for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1082 struct pack_entry e;
64404a24 1083 if (fill_midx_entry(the_repository, oid, &e, m)) {
6a22d521
DS
1084 struct packed_git *p = e.p;
1085 off_t offset;
1086
1087 if (p == *found_pack)
1088 offset = *found_offset;
1089 else
1090 offset = find_pack_entry_one(oid->hash, p);
1091
1092 if (offset) {
1093 if (!*found_pack) {
1094 if (!is_pack_valid(p))
1095 continue;
1096 *found_offset = offset;
1097 *found_pack = p;
1098 }
1099 want = want_found_object(exclude, p);
1100 if (want != -1)
1101 return want;
1102 }
1103 }
1104 }
1105
a80d72db 1106 list_for_each(pos, get_packed_git_mru(the_repository)) {
ec2dd32c 1107 struct packed_git *p = list_entry(pos, struct packed_git, mru);
702d1b95
KS
1108 off_t offset;
1109
1110 if (p == *found_pack)
1111 offset = *found_offset;
1112 else
188960b4 1113 offset = find_pack_entry_one(oid->hash, p);
702d1b95 1114
5c49c116 1115 if (offset) {
ce2bc424 1116 if (!*found_pack) {
319b678a 1117 if (!is_pack_valid(p))
4c080182 1118 continue;
ce2bc424
JK
1119 *found_offset = offset;
1120 *found_pack = p;
64560374 1121 }
702d1b95 1122 want = want_found_object(exclude, p);
e6e24c94 1123 if (!exclude && want > 0)
a80d72db
SB
1124 list_move(&p->mru,
1125 get_packed_git_mru(the_repository));
702d1b95
KS
1126 if (want != -1)
1127 return want;
64560374
LT
1128 }
1129 }
eb019375 1130
ce2bc424
JK
1131 return 1;
1132}
1133
188960b4 1134static void create_object_entry(const struct object_id *oid,
ce2bc424
JK
1135 enum object_type type,
1136 uint32_t hash,
1137 int exclude,
1138 int no_try_delta,
1139 uint32_t index_pos,
1140 struct packed_git *found_pack,
1141 off_t found_offset)
1142{
1143 struct object_entry *entry;
29b734e4 1144
188960b4 1145 entry = packlist_alloc(&to_pack, oid->hash, index_pos);
27225f2e 1146 entry->hash = hash;
fd9b1bae 1147 oe_set_type(entry, type);
29b734e4
NP
1148 if (exclude)
1149 entry->preferred_base = 1;
81a216a5
NP
1150 else
1151 nr_result++;
29b734e4 1152 if (found_pack) {
43fa44fa 1153 oe_set_in_pack(&to_pack, entry, found_pack);
29b734e4
NP
1154 entry->in_pack_offset = found_offset;
1155 }
7a979d99 1156
ce2bc424
JK
1157 entry->no_try_delta = no_try_delta;
1158}
29b734e4 1159
373c67da
JK
1160static const char no_closure_warning[] = N_(
1161"disabling bitmap writing, as some objects are not being packed"
1162);
1163
188960b4 1164static int add_object_entry(const struct object_id *oid, enum object_type type,
ce2bc424
JK
1165 const char *name, int exclude)
1166{
702d1b95
KS
1167 struct packed_git *found_pack = NULL;
1168 off_t found_offset = 0;
ce2bc424 1169 uint32_t index_pos;
7a979d99 1170
5af05043
NTND
1171 display_progress(progress_state, ++nr_seen);
1172
188960b4 1173 if (have_duplicate_entry(oid, exclude, &index_pos))
ce2bc424 1174 return 0;
a74db82e 1175
188960b4 1176 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
373c67da
JK
1177 /* The pack is missing an object, so it will not have closure */
1178 if (write_bitmap_index) {
1179 warning(_(no_closure_warning));
1180 write_bitmap_index = 0;
1181 }
ce2bc424 1182 return 0;
373c67da 1183 }
29b734e4 1184
188960b4 1185 create_object_entry(oid, type, pack_name_hash(name),
ce2bc424
JK
1186 exclude, name && no_try_delta(name),
1187 index_pos, found_pack, found_offset);
29b734e4 1188 return 1;
c323ac7d
LT
1189}
1190
20664967 1191static int add_object_entry_from_bitmap(const struct object_id *oid,
6b8fda2d
VM
1192 enum object_type type,
1193 int flags, uint32_t name_hash,
1194 struct packed_git *pack, off_t offset)
1195{
1196 uint32_t index_pos;
1197
5af05043
NTND
1198 display_progress(progress_state, ++nr_seen);
1199
188960b4 1200 if (have_duplicate_entry(oid, 0, &index_pos))
6b8fda2d
VM
1201 return 0;
1202
188960b4 1203 if (!want_object_in_pack(oid, 0, &pack, &offset))
702d1b95
KS
1204 return 0;
1205
188960b4 1206 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
29b734e4 1207 return 1;
c323ac7d
LT
1208}
1209
5379a5c5 1210struct pbase_tree_cache {
188960b4 1211 struct object_id oid;
5379a5c5
JH
1212 int ref;
1213 int temporary;
1214 void *tree_data;
1215 unsigned long tree_size;
1216};
1217
1218static struct pbase_tree_cache *(pbase_tree_cache[256]);
188960b4 1219static int pbase_tree_cache_ix(const struct object_id *oid)
5379a5c5 1220{
188960b4 1221 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
5379a5c5
JH
1222}
1223static int pbase_tree_cache_ix_incr(int ix)
1224{
1225 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1226}
1227
1228static struct pbase_tree {
1229 struct pbase_tree *next;
1230 /* This is a phony "cache" entry; we are not
01689909 1231 * going to evict it or find it through _get()
5379a5c5
JH
1232 * mechanism -- this is for the toplevel node that
1233 * would almost always change with any commit.
1234 */
1235 struct pbase_tree_cache pcache;
1236} *pbase_tree;
1237
188960b4 1238static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
5379a5c5
JH
1239{
1240 struct pbase_tree_cache *ent, *nent;
1241 void *data;
1242 unsigned long size;
21666f1a 1243 enum object_type type;
5379a5c5 1244 int neigh;
188960b4 1245 int my_ix = pbase_tree_cache_ix(oid);
5379a5c5
JH
1246 int available_ix = -1;
1247
1248 /* pbase-tree-cache acts as a limited hashtable.
1249 * your object will be found at your index or within a few
1250 * slots after that slot if it is cached.
1251 */
1252 for (neigh = 0; neigh < 8; neigh++) {
1253 ent = pbase_tree_cache[my_ix];
4a7e27e9 1254 if (ent && oideq(&ent->oid, oid)) {
5379a5c5
JH
1255 ent->ref++;
1256 return ent;
1257 }
1258 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1259 ((0 <= available_ix) &&
1260 (!ent && pbase_tree_cache[available_ix])))
1261 available_ix = my_ix;
1262 if (!ent)
1263 break;
1264 my_ix = pbase_tree_cache_ix_incr(my_ix);
1265 }
1266
1267 /* Did not find one. Either we got a bogus request or
1268 * we need to read and perhaps cache.
1269 */
b4f5aca4 1270 data = read_object_file(oid, &type, &size);
5379a5c5
JH
1271 if (!data)
1272 return NULL;
21666f1a 1273 if (type != OBJ_TREE) {
5379a5c5
JH
1274 free(data);
1275 return NULL;
1276 }
1277
1278 /* We need to either cache or return a throwaway copy */
1279
1280 if (available_ix < 0)
1281 ent = NULL;
1282 else {
1283 ent = pbase_tree_cache[available_ix];
1284 my_ix = available_ix;
1285 }
1286
1287 if (!ent) {
1288 nent = xmalloc(sizeof(*nent));
1289 nent->temporary = (available_ix < 0);
1290 }
1291 else {
1292 /* evict and reuse */
1293 free(ent->tree_data);
1294 nent = ent;
1295 }
188960b4 1296 oidcpy(&nent->oid, oid);
5379a5c5
JH
1297 nent->tree_data = data;
1298 nent->tree_size = size;
1299 nent->ref = 1;
1300 if (!nent->temporary)
1301 pbase_tree_cache[my_ix] = nent;
1302 return nent;
1303}
1304
1305static void pbase_tree_put(struct pbase_tree_cache *cache)
1306{
1307 if (!cache->temporary) {
1308 cache->ref--;
1309 return;
1310 }
1311 free(cache->tree_data);
1312 free(cache);
1313}
1314
1315static int name_cmp_len(const char *name)
1316{
1317 int i;
1318 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1319 ;
1320 return i;
1321}
1322
1323static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1324 const char *name,
ce0bd642
LT
1325 int cmplen,
1326 const char *fullname)
3f9ac8d2 1327{
4c068a98 1328 struct name_entry entry;
8a5a8d6c 1329 int cmp;
4c068a98
LT
1330
1331 while (tree_entry(tree,&entry)) {
1211be6b
LT
1332 if (S_ISGITLINK(entry.mode))
1333 continue;
0de16337 1334 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1335 memcmp(name, entry.path, cmplen);
1336 if (cmp > 0)
7a979d99 1337 continue;
8a5a8d6c
NP
1338 if (cmp < 0)
1339 return;
5379a5c5 1340 if (name[cmplen] != '/') {
ea82b2a0 1341 add_object_entry(&entry.oid,
4d1012c3 1342 object_type(entry.mode),
bc32fed5 1343 fullname, 1);
5379a5c5
JH
1344 return;
1345 }
8a5a8d6c 1346 if (S_ISDIR(entry.mode)) {
7a979d99 1347 struct tree_desc sub;
5379a5c5
JH
1348 struct pbase_tree_cache *tree;
1349 const char *down = name+cmplen+1;
1350 int downlen = name_cmp_len(down);
1351
ea82b2a0 1352 tree = pbase_tree_get(&entry.oid);
5379a5c5
JH
1353 if (!tree)
1354 return;
6fda5e51 1355 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1356
ce0bd642 1357 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1358 pbase_tree_put(tree);
1359 }
1360 }
1361}
1d6b38cc 1362
5379a5c5
JH
1363static unsigned *done_pbase_paths;
1364static int done_pbase_paths_num;
1365static int done_pbase_paths_alloc;
1366static int done_pbase_path_pos(unsigned hash)
1367{
1368 int lo = 0;
1369 int hi = done_pbase_paths_num;
1370 while (lo < hi) {
19716b21 1371 int mi = lo + (hi - lo) / 2;
5379a5c5
JH
1372 if (done_pbase_paths[mi] == hash)
1373 return mi;
1374 if (done_pbase_paths[mi] < hash)
1375 hi = mi;
1376 else
1377 lo = mi + 1;
1378 }
1379 return -lo-1;
1380}
1381
1382static int check_pbase_path(unsigned hash)
1383{
c7b07805 1384 int pos = done_pbase_path_pos(hash);
5379a5c5
JH
1385 if (0 <= pos)
1386 return 1;
1387 pos = -pos - 1;
25e19407
DD
1388 ALLOC_GROW(done_pbase_paths,
1389 done_pbase_paths_num + 1,
1390 done_pbase_paths_alloc);
5379a5c5
JH
1391 done_pbase_paths_num++;
1392 if (pos < done_pbase_paths_num)
f331ab9d
RS
1393 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1394 done_pbase_paths_num - pos - 1);
5379a5c5
JH
1395 done_pbase_paths[pos] = hash;
1396 return 0;
1397}
1398
bc32fed5 1399static void add_preferred_base_object(const char *name)
5379a5c5
JH
1400{
1401 struct pbase_tree *it;
8a5a8d6c 1402 int cmplen;
68fb36eb 1403 unsigned hash = pack_name_hash(name);
5379a5c5 1404
8a5a8d6c 1405 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1406 return;
1407
8a5a8d6c 1408 cmplen = name_cmp_len(name);
5379a5c5
JH
1409 for (it = pbase_tree; it; it = it->next) {
1410 if (cmplen == 0) {
188960b4 1411 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
5379a5c5
JH
1412 }
1413 else {
1414 struct tree_desc tree;
6fda5e51 1415 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1416 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1417 }
3f9ac8d2 1418 }
3f9ac8d2
JH
1419}
1420
188960b4 1421static void add_preferred_base(struct object_id *oid)
3f9ac8d2 1422{
5379a5c5
JH
1423 struct pbase_tree *it;
1424 void *data;
1425 unsigned long size;
188960b4 1426 struct object_id tree_oid;
1d6b38cc 1427
8d1d8f83
JH
1428 if (window <= num_preferred_base++)
1429 return;
1430
d3b4705a
NTND
1431 data = read_object_with_reference(the_repository, oid,
1432 tree_type, &size, &tree_oid);
5379a5c5 1433 if (!data)
7a979d99 1434 return;
5379a5c5
JH
1435
1436 for (it = pbase_tree; it; it = it->next) {
4a7e27e9 1437 if (oideq(&it->pcache.oid, &tree_oid)) {
5379a5c5
JH
1438 free(data);
1439 return;
1440 }
1441 }
1442
1443 it = xcalloc(1, sizeof(*it));
1444 it->next = pbase_tree;
1445 pbase_tree = it;
1446
188960b4 1447 oidcpy(&it->pcache.oid, &tree_oid);
5379a5c5
JH
1448 it->pcache.tree_data = data;
1449 it->pcache.tree_size = size;
3f9ac8d2
JH
1450}
1451
0ef95f72
NP
1452static void cleanup_preferred_base(void)
1453{
1454 struct pbase_tree *it;
1455 unsigned i;
1456
1457 it = pbase_tree;
1458 pbase_tree = NULL;
1459 while (it) {
095b3b2c
BW
1460 struct pbase_tree *tmp = it;
1461 it = tmp->next;
1462 free(tmp->pcache.tree_data);
1463 free(tmp);
0ef95f72
NP
1464 }
1465
1466 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1467 if (!pbase_tree_cache[i])
1468 continue;
1469 free(pbase_tree_cache[i]->tree_data);
6a83d902 1470 FREE_AND_NULL(pbase_tree_cache[i]);
0ef95f72
NP
1471 }
1472
6a83d902 1473 FREE_AND_NULL(done_pbase_paths);
0ef95f72
NP
1474 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1475}
1476
2fa233a5
JK
1477/*
1478 * Return 1 iff the object specified by "delta" can be sent
1479 * literally as a delta against the base in "base_sha1". If
1480 * so, then *base_out will point to the entry in our packing
1481 * list, or NULL if we must use the external-base list.
1482 *
1483 * Depth value does not matter - find_deltas() will
1484 * never consider reused delta as the base object to
1485 * deltify other objects against, in order to avoid
1486 * circular deltas.
1487 */
1488static int can_reuse_delta(const unsigned char *base_sha1,
1489 struct object_entry *delta,
1490 struct object_entry **base_out)
1491{
1492 struct object_entry *base;
3c771448 1493 struct object_id base_oid;
2fa233a5
JK
1494
1495 if (!base_sha1)
1496 return 0;
1497
3df28cae
JK
1498 oidread(&base_oid, base_sha1);
1499
2fa233a5
JK
1500 /*
1501 * First see if we're already sending the base (or it's explicitly in
1502 * our "excluded" list).
1503 */
3df28cae 1504 base = packlist_find(&to_pack, &base_oid, NULL);
2fa233a5
JK
1505 if (base) {
1506 if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1507 return 0;
1508 *base_out = base;
1509 return 1;
1510 }
1511
1512 /*
1513 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1514 * even if it was buried too deep in history to make it into the
1515 * packing list.
1516 */
3c771448 1517 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {
2fa233a5 1518 if (use_delta_islands) {
2fa233a5
JK
1519 if (!in_same_island(&delta->idx.oid, &base_oid))
1520 return 0;
1521 }
1522 *base_out = NULL;
1523 return 1;
1524 }
1525
1526 return 0;
1527}
1528
c323ac7d
LT
1529static void check_object(struct object_entry *entry)
1530{
ac77d0c3
NTND
1531 unsigned long canonical_size;
1532
43fa44fa
NTND
1533 if (IN_PACK(entry)) {
1534 struct packed_git *p = IN_PACK(entry);
03e79c88 1535 struct pack_window *w_curs = NULL;
5c49c116
NP
1536 const unsigned char *base_ref = NULL;
1537 struct object_entry *base_entry;
1538 unsigned long used, used_0;
ef49a7a0 1539 unsigned long avail;
5c49c116
NP
1540 off_t ofs;
1541 unsigned char *buf, c;
fd9b1bae 1542 enum object_type type;
27a7d067 1543 unsigned long in_pack_size;
780e6e73 1544
6777a59f 1545 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1546
5c49c116 1547 /*
fa736f72
NP
1548 * We want in_pack_type even if we do not reuse delta
1549 * since non-delta representations could still be reused.
ab7cd7bb 1550 */
09ded04b 1551 used = unpack_object_header_buffer(buf, avail,
fd9b1bae 1552 &type,
27a7d067 1553 &in_pack_size);
03d66015
NP
1554 if (used == 0)
1555 goto give_up;
ab7cd7bb 1556
fd9b1bae
NTND
1557 if (type < 0)
1558 BUG("invalid type %d", type);
1559 entry->in_pack_type = type;
1560
5c49c116
NP
1561 /*
1562 * Determine if this is a delta and if so whether we can
1563 * reuse it or not. Otherwise let's find out as cheaply as
1564 * possible what the actual type and size for this object is.
3f9ac8d2 1565 */
5c49c116
NP
1566 switch (entry->in_pack_type) {
1567 default:
1568 /* Not a delta hence we've already got all we need. */
fd9b1bae 1569 oe_set_type(entry, entry->in_pack_type);
ac77d0c3 1570 SET_SIZE(entry, in_pack_size);
5c49c116 1571 entry->in_pack_header_size = used;
fd9b1bae 1572 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
03d66015 1573 goto give_up;
5c49c116
NP
1574 unuse_pack(&w_curs);
1575 return;
1576 case OBJ_REF_DELTA:
a7de7130 1577 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1578 base_ref = use_pack(p, &w_curs,
1579 entry->in_pack_offset + used, NULL);
41179100 1580 entry->in_pack_header_size = used + the_hash_algo->rawsz;
5c49c116
NP
1581 break;
1582 case OBJ_OFS_DELTA:
1583 buf = use_pack(p, &w_curs,
1584 entry->in_pack_offset + used, NULL);
1585 used_0 = 0;
1586 c = buf[used_0++];
1587 ofs = c & 127;
1588 while (c & 128) {
1589 ofs += 1;
03d66015 1590 if (!ofs || MSB(ofs, 7)) {
f616db6a 1591 error(_("delta base offset overflow in pack for %s"),
e6a492b7 1592 oid_to_hex(&entry->idx.oid));
03d66015
NP
1593 goto give_up;
1594 }
5c49c116
NP
1595 c = buf[used_0++];
1596 ofs = (ofs << 7) + (c & 127);
780e6e73 1597 }
5c49c116 1598 ofs = entry->in_pack_offset - ofs;
03d66015 1599 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
f616db6a 1600 error(_("delta base offset out of bound for %s"),
e6a492b7 1601 oid_to_hex(&entry->idx.oid));
03d66015
NP
1602 goto give_up;
1603 }
a7de7130 1604 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1605 struct revindex_entry *revidx;
1606 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1607 if (!revidx)
1608 goto give_up;
3449f8c4
NP
1609 base_ref = nth_packed_object_sha1(p, revidx->nr);
1610 }
5c49c116
NP
1611 entry->in_pack_header_size = used + used_0;
1612 break;
780e6e73 1613 }
780e6e73 1614
2fa233a5 1615 if (can_reuse_delta(base_ref, entry, &base_entry)) {
fd9b1bae 1616 oe_set_type(entry, entry->in_pack_type);
ac77d0c3 1617 SET_SIZE(entry, in_pack_size); /* delta size */
0aca34e8 1618 SET_DELTA_SIZE(entry, in_pack_size);
6a1e32d5
JK
1619
1620 if (base_entry) {
1621 SET_DELTA(entry, base_entry);
1622 entry->delta_sibling_idx = base_entry->delta_child_idx;
1623 SET_DELTA_CHILD(base_entry, entry);
1624 } else {
1625 SET_DELTA_EXT(entry, base_ref);
1626 }
1627
5c49c116
NP
1628 unuse_pack(&w_curs);
1629 return;
1630 }
ab7cd7bb 1631
fd9b1bae 1632 if (oe_type(entry)) {
27a7d067
NTND
1633 off_t delta_pos;
1634
5c49c116
NP
1635 /*
1636 * This must be a delta and we already know what the
1637 * final object type is. Let's extract the actual
1638 * object size from the delta header.
1639 */
27a7d067 1640 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
ac77d0c3
NTND
1641 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1642 if (canonical_size == 0)
03d66015 1643 goto give_up;
ac77d0c3 1644 SET_SIZE(entry, canonical_size);
5c49c116 1645 unuse_pack(&w_curs);
3f9ac8d2
JH
1646 return;
1647 }
5c49c116
NP
1648
1649 /*
1650 * No choice but to fall back to the recursive delta walk
c93206b4 1651 * with oid_object_info() to find about the object type
5c49c116
NP
1652 * at this point...
1653 */
03d66015 1654 give_up:
5c49c116 1655 unuse_pack(&w_curs);
36e4d74a 1656 }
3f9ac8d2 1657
ad635e82
JH
1658 oe_set_type(entry,
1659 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
ac77d0c3
NTND
1660 if (entry->type_valid) {
1661 SET_SIZE(entry, canonical_size);
1662 } else {
1663 /*
1664 * Bad object type is checked in prepare_pack(). This is
1665 * to permit a missing preferred base object to be ignored
1666 * as a preferred base. Doing so can result in a larger
1667 * pack file, but the transfer will still take place.
1668 */
1669 }
3f9ac8d2
JH
1670}
1671
5c49c116
NP
1672static int pack_offset_sort(const void *_a, const void *_b)
1673{
1674 const struct object_entry *a = *(struct object_entry **)_a;
1675 const struct object_entry *b = *(struct object_entry **)_b;
43fa44fa
NTND
1676 const struct packed_git *a_in_pack = IN_PACK(a);
1677 const struct packed_git *b_in_pack = IN_PACK(b);
5c49c116
NP
1678
1679 /* avoid filesystem trashing with loose objects */
43fa44fa 1680 if (!a_in_pack && !b_in_pack)
e6a492b7 1681 return oidcmp(&a->idx.oid, &b->idx.oid);
5c49c116 1682
43fa44fa 1683 if (a_in_pack < b_in_pack)
5c49c116 1684 return -1;
43fa44fa 1685 if (a_in_pack > b_in_pack)
5c49c116
NP
1686 return 1;
1687 return a->in_pack_offset < b->in_pack_offset ? -1 :
1688 (a->in_pack_offset > b->in_pack_offset);
1689}
1690
4cf2143e
JK
1691/*
1692 * Drop an on-disk delta we were planning to reuse. Naively, this would
1693 * just involve blanking out the "delta" field, but we have to deal
1694 * with some extra book-keeping:
1695 *
1696 * 1. Removing ourselves from the delta_sibling linked list.
1697 *
1698 * 2. Updating our size/type to the non-delta representation. These were
1699 * either not recorded initially (size) or overwritten with the delta type
1700 * (type) when check_object() decided to reuse the delta.
7dbabbbe
JK
1701 *
1702 * 3. Resetting our delta depth, as we are now a base object.
4cf2143e
JK
1703 */
1704static void drop_reused_delta(struct object_entry *entry)
1705{
898eba5e 1706 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
4cf2143e 1707 struct object_info oi = OBJECT_INFO_INIT;
fd9b1bae 1708 enum object_type type;
ac77d0c3 1709 unsigned long size;
4cf2143e 1710
898eba5e
NTND
1711 while (*idx) {
1712 struct object_entry *oe = &to_pack.objects[*idx - 1];
4cf2143e 1713
898eba5e
NTND
1714 if (oe == entry)
1715 *idx = oe->delta_sibling_idx;
4cf2143e 1716 else
898eba5e 1717 idx = &oe->delta_sibling_idx;
4cf2143e 1718 }
898eba5e 1719 SET_DELTA(entry, NULL);
7dbabbbe 1720 entry->depth = 0;
4cf2143e 1721
ac77d0c3 1722 oi.sizep = &size;
fd9b1bae 1723 oi.typep = &type;
ad635e82 1724 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
4cf2143e
JK
1725 /*
1726 * We failed to get the info from this pack for some reason;
c93206b4 1727 * fall back to oid_object_info, which may find another copy.
fd9b1bae 1728 * And if that fails, the error will be recorded in oe_type(entry)
4cf2143e
JK
1729 * and dealt with in prepare_pack().
1730 */
ad635e82
JH
1731 oe_set_type(entry,
1732 oid_object_info(the_repository, &entry->idx.oid, &size));
fd9b1bae
NTND
1733 } else {
1734 oe_set_type(entry, type);
4cf2143e 1735 }
ac77d0c3 1736 SET_SIZE(entry, size);
4cf2143e
JK
1737}
1738
1739/*
1740 * Follow the chain of deltas from this entry onward, throwing away any links
1741 * that cause us to hit a cycle (as determined by the DFS state flags in
1742 * the entries).
7dbabbbe
JK
1743 *
1744 * We also detect too-long reused chains that would violate our --depth
1745 * limit.
4cf2143e
JK
1746 */
1747static void break_delta_chains(struct object_entry *entry)
1748{
42b766d7
JK
1749 /*
1750 * The actual depth of each object we will write is stored as an int,
1751 * as it cannot exceed our int "depth" limit. But before we break
1752 * changes based no that limit, we may potentially go as deep as the
1753 * number of objects, which is elsewhere bounded to a uint32_t.
1754 */
1755 uint32_t total_depth;
1756 struct object_entry *cur, *next;
1757
1758 for (cur = entry, total_depth = 0;
1759 cur;
898eba5e 1760 cur = DELTA(cur), total_depth++) {
42b766d7
JK
1761 if (cur->dfs_state == DFS_DONE) {
1762 /*
1763 * We've already seen this object and know it isn't
1764 * part of a cycle. We do need to append its depth
1765 * to our count.
1766 */
1767 total_depth += cur->depth;
1768 break;
1769 }
4cf2143e 1770
4cf2143e 1771 /*
42b766d7
JK
1772 * We break cycles before looping, so an ACTIVE state (or any
1773 * other cruft which made its way into the state variable)
1774 * is a bug.
4cf2143e 1775 */
42b766d7 1776 if (cur->dfs_state != DFS_NONE)
033abf97 1777 BUG("confusing delta dfs state in first pass: %d",
42b766d7 1778 cur->dfs_state);
4cf2143e 1779
4cf2143e 1780 /*
42b766d7
JK
1781 * Now we know this is the first time we've seen the object. If
1782 * it's not a delta, we're done traversing, but we'll mark it
1783 * done to save time on future traversals.
4cf2143e 1784 */
898eba5e 1785 if (!DELTA(cur)) {
42b766d7
JK
1786 cur->dfs_state = DFS_DONE;
1787 break;
1788 }
4cf2143e 1789
4cf2143e 1790 /*
42b766d7
JK
1791 * Mark ourselves as active and see if the next step causes
1792 * us to cycle to another active object. It's important to do
1793 * this _before_ we loop, because it impacts where we make the
1794 * cut, and thus how our total_depth counter works.
1795 * E.g., We may see a partial loop like:
1796 *
1797 * A -> B -> C -> D -> B
1798 *
1799 * Cutting B->C breaks the cycle. But now the depth of A is
1800 * only 1, and our total_depth counter is at 3. The size of the
1801 * error is always one less than the size of the cycle we
1802 * broke. Commits C and D were "lost" from A's chain.
1803 *
1804 * If we instead cut D->B, then the depth of A is correct at 3.
1805 * We keep all commits in the chain that we examined.
4cf2143e 1806 */
42b766d7 1807 cur->dfs_state = DFS_ACTIVE;
898eba5e 1808 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
42b766d7
JK
1809 drop_reused_delta(cur);
1810 cur->dfs_state = DFS_DONE;
1811 break;
7dbabbbe 1812 }
42b766d7 1813 }
7dbabbbe 1814
42b766d7
JK
1815 /*
1816 * And now that we've gone all the way to the bottom of the chain, we
1817 * need to clear the active flags and set the depth fields as
1818 * appropriate. Unlike the loop above, which can quit when it drops a
1819 * delta, we need to keep going to look for more depth cuts. So we need
1820 * an extra "next" pointer to keep going after we reset cur->delta.
1821 */
1822 for (cur = entry; cur; cur = next) {
898eba5e 1823 next = DELTA(cur);
4cf2143e 1824
42b766d7
JK
1825 /*
1826 * We should have a chain of zero or more ACTIVE states down to
1827 * a final DONE. We can quit after the DONE, because either it
1828 * has no bases, or we've already handled them in a previous
1829 * call.
1830 */
1831 if (cur->dfs_state == DFS_DONE)
1832 break;
1833 else if (cur->dfs_state != DFS_ACTIVE)
033abf97 1834 BUG("confusing delta dfs state in second pass: %d",
42b766d7 1835 cur->dfs_state);
4cf2143e 1836
4cf2143e 1837 /*
42b766d7
JK
1838 * If the total_depth is more than depth, then we need to snip
1839 * the chain into two or more smaller chains that don't exceed
1840 * the maximum depth. Most of the resulting chains will contain
1841 * (depth + 1) entries (i.e., depth deltas plus one base), and
1842 * the last chain (i.e., the one containing entry) will contain
1843 * whatever entries are left over, namely
1844 * (total_depth % (depth + 1)) of them.
1845 *
1846 * Since we are iterating towards decreasing depth, we need to
1847 * decrement total_depth as we go, and we need to write to the
1848 * entry what its final depth will be after all of the
1849 * snipping. Since we're snipping into chains of length (depth
1850 * + 1) entries, the final depth of an entry will be its
1851 * original depth modulo (depth + 1). Any time we encounter an
1852 * entry whose final depth is supposed to be zero, we snip it
1853 * from its delta base, thereby making it so.
4cf2143e 1854 */
42b766d7
JK
1855 cur->depth = (total_depth--) % (depth + 1);
1856 if (!cur->depth)
1857 drop_reused_delta(cur);
1858
1859 cur->dfs_state = DFS_DONE;
4cf2143e
JK
1860 }
1861}
1862
c323ac7d
LT
1863static void get_object_details(void)
1864{
7cadf491 1865 uint32_t i;
5c49c116
NP
1866 struct object_entry **sorted_by_offset;
1867
5af05043
NTND
1868 if (progress)
1869 progress_state = start_progress(_("Counting objects"),
1870 to_pack.nr_objects);
1871
2834bc27
VM
1872 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1873 for (i = 0; i < to_pack.nr_objects; i++)
1874 sorted_by_offset[i] = to_pack.objects + i;
9ed0d8d6 1875 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
c323ac7d 1876
2834bc27 1877 for (i = 0; i < to_pack.nr_objects; i++) {
15366280
JH
1878 struct object_entry *entry = sorted_by_offset[i];
1879 check_object(entry);
ac77d0c3
NTND
1880 if (entry->type_valid &&
1881 oe_size_greater_than(&to_pack, entry, big_file_threshold))
15366280 1882 entry->no_try_delta = 1;
5af05043 1883 display_progress(progress_state, i + 1);
15366280 1884 }
5af05043 1885 stop_progress(&progress_state);
3449f8c4 1886
4cf2143e
JK
1887 /*
1888 * This must happen in a second pass, since we rely on the delta
1889 * information for the whole list being completed.
1890 */
1891 for (i = 0; i < to_pack.nr_objects; i++)
1892 break_delta_chains(&to_pack.objects[i]);
1893
5c49c116 1894 free(sorted_by_offset);
c323ac7d
LT
1895}
1896
b904166c
NP
1897/*
1898 * We search for deltas in a list sorted by type, by filename hash, and then
1899 * by size, so that we see progressively smaller and smaller files.
1900 * That's because we prefer deltas to be from the bigger file
1901 * to the smaller -- deletes are potentially cheaper, but perhaps
1902 * more importantly, the bigger file is likely the more recent
1903 * one. The deepest deltas are therefore the oldest objects which are
1904 * less susceptible to be accessed often.
1905 */
9668cf59 1906static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1907{
9668cf59
NP
1908 const struct object_entry *a = *(struct object_entry **)_a;
1909 const struct object_entry *b = *(struct object_entry **)_b;
33de80b1
SL
1910 const enum object_type a_type = oe_type(a);
1911 const enum object_type b_type = oe_type(b);
1912 const unsigned long a_size = SIZE(a);
1913 const unsigned long b_size = SIZE(b);
9668cf59 1914
fd9b1bae 1915 if (a_type > b_type)
27225f2e 1916 return -1;
fd9b1bae 1917 if (a_type < b_type)
27225f2e 1918 return 1;
b904166c 1919 if (a->hash > b->hash)
7a979d99 1920 return -1;
b904166c 1921 if (a->hash < b->hash)
7a979d99 1922 return 1;
b904166c 1923 if (a->preferred_base > b->preferred_base)
c323ac7d 1924 return -1;
b904166c
NP
1925 if (a->preferred_base < b->preferred_base)
1926 return 1;
28b8a730 1927 if (use_delta_islands) {
33de80b1 1928 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
28b8a730
JK
1929 if (island_cmp)
1930 return island_cmp;
1931 }
ac77d0c3 1932 if (a_size > b_size)
b904166c 1933 return -1;
ac77d0c3 1934 if (a_size < b_size)
c323ac7d 1935 return 1;
b904166c 1936 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1937}
1938
1939struct unpacked {
1940 struct object_entry *entry;
1941 void *data;
f6c7081a 1942 struct delta_index *index;
5a235b5e 1943 unsigned depth;
c323ac7d
LT
1944};
1945
d250626c
NP
1946static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1947 unsigned long delta_size)
074b2eea
MK
1948{
1949 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1950 return 0;
1951
e3dfddb3
MK
1952 if (delta_size < cache_max_small_delta_size)
1953 return 1;
1954
074b2eea
MK
1955 /* cache delta, if objects are large enough compared to delta size */
1956 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1957 return 1;
1958
1959 return 0;
1960}
1961
ffbd51cc 1962/* Protect delta_cache_size */
44626dc7 1963static pthread_mutex_t cache_mutex;
3c701839
NP
1964#define cache_lock() pthread_mutex_lock(&cache_mutex)
1965#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1966
ffbd51cc
NTND
1967/*
1968 * Protect object list partitioning (e.g. struct thread_param) and
1969 * progress_state
1970 */
44626dc7 1971static pthread_mutex_t progress_mutex;
8ecce684
NP
1972#define progress_lock() pthread_mutex_lock(&progress_mutex)
1973#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1974
ffbd51cc
NTND
1975/*
1976 * Access to struct object_entry is unprotected since each thread owns
1977 * a portion of the main object list. Just don't access object entries
1978 * ahead in the list because they can be stolen and would need
1979 * progress_mutex for protection.
1980 */
8ecce684 1981
ac77d0c3
NTND
1982/*
1983 * Return the size of the object without doing any delta
1984 * reconstruction (so non-deltas are true object sizes, but deltas
1985 * return the size of the delta data).
1986 */
1987unsigned long oe_get_size_slow(struct packing_data *pack,
1988 const struct object_entry *e)
1989{
1990 struct packed_git *p;
1991 struct pack_window *w_curs;
1992 unsigned char *buf;
1993 enum object_type type;
1994 unsigned long used, avail, size;
1995
1996 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
edb673cf 1997 packing_data_lock(&to_pack);
ad635e82 1998 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
ac77d0c3
NTND
1999 die(_("unable to get size of %s"),
2000 oid_to_hex(&e->idx.oid));
edb673cf 2001 packing_data_unlock(&to_pack);
ac77d0c3
NTND
2002 return size;
2003 }
2004
2005 p = oe_in_pack(pack, e);
2006 if (!p)
2007 BUG("when e->type is a delta, it must belong to a pack");
2008
edb673cf 2009 packing_data_lock(&to_pack);
ac77d0c3
NTND
2010 w_curs = NULL;
2011 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2012 used = unpack_object_header_buffer(buf, avail, &type, &size);
2013 if (used == 0)
2014 die(_("unable to parse object header of %s"),
2015 oid_to_hex(&e->idx.oid));
2016
2017 unuse_pack(&w_curs);
edb673cf 2018 packing_data_unlock(&to_pack);
ac77d0c3
NTND
2019 return size;
2020}
2021
f6c7081a 2022static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 2023 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 2024{
f6c7081a
NP
2025 struct object_entry *trg_entry = trg->entry;
2026 struct object_entry *src_entry = src->entry;
560b25a8 2027 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 2028 unsigned ref_depth;
21666f1a 2029 enum object_type type;
c323ac7d
LT
2030 void *delta_buf;
2031
2032 /* Don't bother doing diffs between different types */
fd9b1bae 2033 if (oe_type(trg_entry) != oe_type(src_entry))
c323ac7d
LT
2034 return -1;
2035
51d1e83f 2036 /*
15f07e06
JK
2037 * We do not bother to try a delta that we discarded on an
2038 * earlier try, but only when reusing delta data. Note that
2039 * src_entry that is marked as the preferred_base should always
2040 * be considered, as even if we produce a suboptimal delta against
2041 * it, we will still save the transfer cost, as we already know
2042 * the other side has it and we won't send src_entry at all.
51d1e83f 2043 */
43fa44fa
NTND
2044 if (reuse_delta && IN_PACK(trg_entry) &&
2045 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
15f07e06 2046 !src_entry->preferred_base &&
e9195b58
JH
2047 trg_entry->in_pack_type != OBJ_REF_DELTA &&
2048 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
2049 return 0;
2050
898b14ce 2051 /* Let's not bust the allowed depth. */
5a235b5e 2052 if (src->depth >= max_depth)
d116a45a 2053 return 0;
c323ac7d 2054
c3b06a69 2055 /* Now some size filtering heuristics. */
ac77d0c3 2056 trg_size = SIZE(trg_entry);
898eba5e 2057 if (!DELTA(trg_entry)) {
41179100 2058 max_size = trg_size/2 - the_hash_algo->rawsz;
c83f032e
NP
2059 ref_depth = 1;
2060 } else {
0aca34e8 2061 max_size = DELTA_SIZE(trg_entry);
5a235b5e 2062 ref_depth = trg->depth;
c83f032e 2063 }
720fe22d 2064 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 2065 (max_depth - ref_depth + 1);
c3b06a69
NP
2066 if (max_size == 0)
2067 return 0;
ac77d0c3 2068 src_size = SIZE(src_entry);
560b25a8 2069 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 2070 if (sizediff >= max_size)
f527cb8c 2071 return 0;
a1dab41a
BD
2072 if (trg_size < src_size / 32)
2073 return 0;
f6c7081a 2074
28b8a730
JK
2075 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2076 return 0;
2077
560b25a8
NP
2078 /* Load data if not already done */
2079 if (!trg->data) {
edb673cf 2080 packing_data_lock(&to_pack);
b4f5aca4 2081 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
edb673cf 2082 packing_data_unlock(&to_pack);
2e3404c3 2083 if (!trg->data)
f616db6a 2084 die(_("object %s cannot be read"),
e6a492b7 2085 oid_to_hex(&trg_entry->idx.oid));
560b25a8 2086 if (sz != trg_size)
ca473cef
TB
2087 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2088 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2089 (uintmax_t)trg_size);
ef0316fc 2090 *mem_usage += sz;
560b25a8
NP
2091 }
2092 if (!src->data) {
edb673cf 2093 packing_data_lock(&to_pack);
b4f5aca4 2094 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
edb673cf 2095 packing_data_unlock(&to_pack);
71064a95
NP
2096 if (!src->data) {
2097 if (src_entry->preferred_base) {
2098 static int warned = 0;
2099 if (!warned++)
f616db6a 2100 warning(_("object %s cannot be read"),
e6a492b7 2101 oid_to_hex(&src_entry->idx.oid));
71064a95
NP
2102 /*
2103 * Those objects are not included in the
2104 * resulting pack. Be resilient and ignore
2105 * them if they can't be read, in case the
2106 * pack could be created nevertheless.
2107 */
2108 return 0;
2109 }
f616db6a 2110 die(_("object %s cannot be read"),
e6a492b7 2111 oid_to_hex(&src_entry->idx.oid));
71064a95 2112 }
560b25a8 2113 if (sz != src_size)
ca473cef
TB
2114 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2115 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2116 (uintmax_t)src_size);
ef0316fc 2117 *mem_usage += sz;
560b25a8
NP
2118 }
2119 if (!src->index) {
2120 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
2121 if (!src->index) {
2122 static int warned = 0;
2123 if (!warned++)
f616db6a 2124 warning(_("suboptimal pack - out of memory"));
a588d88a
MK
2125 return 0;
2126 }
ef0316fc 2127 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
2128 }
2129
2130 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 2131 if (!delta_buf)
75c42d8c 2132 return 0;
f6c7081a 2133
898eba5e 2134 if (DELTA(trg_entry)) {
848d732c 2135 /* Prefer only shallower same-sized deltas. */
0aca34e8 2136 if (delta_size == DELTA_SIZE(trg_entry) &&
5a235b5e 2137 src->depth + 1 >= trg->depth) {
848d732c
BD
2138 free(delta_buf);
2139 return 0;
2140 }
074b2eea 2141 }
9e2d57a0 2142
3c701839
NP
2143 /*
2144 * Handle memory allocation outside of the cache
2145 * accounting lock. Compiler will optimize the strangeness
7eb151d6 2146 * away when NO_PTHREADS is defined.
3c701839 2147 */
8e0f7003 2148 free(trg_entry->delta_data);
3c701839 2149 cache_lock();
9e2d57a0 2150 if (trg_entry->delta_data) {
0aca34e8 2151 delta_cache_size -= DELTA_SIZE(trg_entry);
9e2d57a0
NP
2152 trg_entry->delta_data = NULL;
2153 }
d250626c 2154 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 2155 delta_cache_size += delta_size;
3c701839
NP
2156 cache_unlock();
2157 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2158 } else {
2159 cache_unlock();
074b2eea 2160 free(delta_buf);
3c701839
NP
2161 }
2162
898eba5e 2163 SET_DELTA(trg_entry, src_entry);
0aca34e8 2164 SET_DELTA_SIZE(trg_entry, delta_size);
b7a28f78
NP
2165 trg->depth = src->depth + 1;
2166
f6c7081a 2167 return 1;
c323ac7d
LT
2168}
2169
898b14ce 2170static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 2171{
898eba5e 2172 struct object_entry *child = DELTA_CHILD(me);
898b14ce
NP
2173 unsigned int m = n;
2174 while (child) {
33de80b1 2175 const unsigned int c = check_delta_limit(child, n + 1);
898b14ce
NP
2176 if (m < c)
2177 m = c;
898eba5e 2178 child = DELTA_SIBLING(child);
898b14ce
NP
2179 }
2180 return m;
b2504a0d
NP
2181}
2182
75ad235c 2183static unsigned long free_unpacked(struct unpacked *n)
a97773ce 2184{
ef0316fc 2185 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
2186 free_delta_index(n->index);
2187 n->index = NULL;
2188 if (n->data) {
ac77d0c3 2189 freed_mem += SIZE(n->entry);
6a83d902 2190 FREE_AND_NULL(n->data);
a97773ce
BD
2191 }
2192 n->entry = NULL;
7d7baa5e 2193 n->depth = 0;
ef0316fc 2194 return freed_mem;
a97773ce
BD
2195}
2196
384b32c0 2197static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 2198 int window, int depth, unsigned *processed)
c323ac7d 2199{
384b32c0 2200 uint32_t i, idx = 0, count = 0;
7cadf491 2201 struct unpacked *array;
ef0316fc 2202 unsigned long mem_usage = 0;
c323ac7d 2203
19d4b416 2204 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 2205
384b32c0 2206 for (;;) {
421b488a 2207 struct object_entry *entry;
c323ac7d 2208 struct unpacked *n = array + idx;
ef0316fc 2209 int j, max_depth, best_base = -1;
c323ac7d 2210
384b32c0
NP
2211 progress_lock();
2212 if (!*list_size) {
2213 progress_unlock();
2214 break;
2215 }
421b488a 2216 entry = *list++;
384b32c0
NP
2217 (*list_size)--;
2218 if (!entry->preferred_base) {
2219 (*processed)++;
2220 display_progress(progress_state, *processed);
2221 }
2222 progress_unlock();
2223
ef0316fc 2224 mem_usage -= free_unpacked(n);
c323ac7d 2225 n->entry = entry;
ab7cd7bb 2226
a97773ce 2227 while (window_memory_limit &&
ef0316fc 2228 mem_usage > window_memory_limit &&
a97773ce 2229 count > 1) {
33de80b1 2230 const uint32_t tail = (idx + window - count) % window;
75ad235c 2231 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
2232 count--;
2233 }
2234
75d39853
NP
2235 /* We do not compute delta to *create* objects we are not
2236 * going to pack.
2237 */
2238 if (entry->preferred_base)
2239 goto next;
2240
898b14ce
NP
2241 /*
2242 * If the current object is at pack edge, take the depth the
2243 * objects that depend on the current object into account
2244 * otherwise they would become too deep.
2245 */
2246 max_depth = depth;
898eba5e 2247 if (DELTA_CHILD(entry)) {
898b14ce
NP
2248 max_depth -= check_delta_limit(entry, 0);
2249 if (max_depth <= 0)
2250 goto next;
2251 }
2252
78817c15
LT
2253 j = window;
2254 while (--j > 0) {
77639870 2255 int ret;
7cadf491 2256 uint32_t other_idx = idx + j;
c323ac7d 2257 struct unpacked *m;
78817c15
LT
2258 if (other_idx >= window)
2259 other_idx -= window;
c323ac7d
LT
2260 m = array + other_idx;
2261 if (!m->entry)
2262 break;
ef0316fc 2263 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 2264 if (ret < 0)
c323ac7d 2265 break;
77639870
JH
2266 else if (ret > 0)
2267 best_base = other_idx;
c323ac7d 2268 }
898b14ce 2269
ed4a9031
NP
2270 /*
2271 * If we decided to cache the delta data, then it is best
2272 * to compress it right away. First because we have to do
2273 * it anyway, and doing it here while we're threaded will
2274 * save a lot of time in the non threaded write phase,
2275 * as well as allow for caching more deltas within
2276 * the same cache size limit.
2277 * ...
2278 * But only if not writing to stdout, since in that case
2279 * the network is most likely throttling writes anyway,
2280 * and therefore it is best to go to the write phase ASAP
2281 * instead, as we can afford spending more time compressing
2282 * between writes at that moment.
2283 */
2284 if (entry->delta_data && !pack_to_stdout) {
0cb3c142
NTND
2285 unsigned long size;
2286
0aca34e8 2287 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
0cb3c142
NTND
2288 if (size < (1U << OE_Z_DELTA_BITS)) {
2289 entry->z_delta_size = size;
2290 cache_lock();
0aca34e8 2291 delta_cache_size -= DELTA_SIZE(entry);
0cb3c142
NTND
2292 delta_cache_size += entry->z_delta_size;
2293 cache_unlock();
2294 } else {
2295 FREE_AND_NULL(entry->delta_data);
2296 entry->z_delta_size = 0;
2297 }
ed4a9031
NP
2298 }
2299
70ca1a3f
JH
2300 /* if we made n a delta, and if n is already at max
2301 * depth, leaving it in the window is pointless. we
2302 * should evict it first.
70ca1a3f 2303 */
898eba5e 2304 if (DELTA(entry) && max_depth <= n->depth)
70ca1a3f 2305 continue;
ff45715c 2306
77639870
JH
2307 /*
2308 * Move the best delta base up in the window, after the
2309 * currently deltified object, to keep it longer. It will
2310 * be the first base object to be attempted next.
2311 */
898eba5e 2312 if (DELTA(entry)) {
77639870
JH
2313 struct unpacked swap = array[best_base];
2314 int dist = (window + idx - best_base) % window;
2315 int dst = best_base;
2316 while (dist--) {
2317 int src = (dst + 1) % window;
2318 array[dst] = array[src];
2319 dst = src;
2320 }
2321 array[dst] = swap;
2322 }
2323
898b14ce 2324 next:
521a4f4c 2325 idx++;
a97773ce
BD
2326 if (count + 1 < window)
2327 count++;
521a4f4c
LT
2328 if (idx >= window)
2329 idx = 0;
384b32c0 2330 }
adee7bdf 2331
f6c7081a 2332 for (i = 0; i < window; ++i) {
ff45715c 2333 free_delta_index(array[i].index);
adee7bdf 2334 free(array[i].data);
f6c7081a 2335 }
adee7bdf 2336 free(array);
c323ac7d
LT
2337}
2338
a9a74636
NP
2339static void try_to_free_from_threads(size_t size)
2340{
edb673cf 2341 packing_data_lock(&to_pack);
7c3ecb32 2342 release_pack_memory(size);
edb673cf 2343 packing_data_unlock(&to_pack);
a9a74636
NP
2344}
2345
bc9b2175 2346static try_to_free_t old_try_to_free_routine;
851c34b0 2347
50f22ada 2348/*
ffbd51cc
NTND
2349 * The main object list is split into smaller lists, each is handed to
2350 * one worker.
2351 *
50f22ada
JS
2352 * The main thread waits on the condition that (at least) one of the workers
2353 * has stopped working (which is indicated in the .working member of
2354 * struct thread_params).
ffbd51cc 2355 *
50f22ada
JS
2356 * When a work thread has completed its work, it sets .working to 0 and
2357 * signals the main thread and waits on the condition that .data_ready
2358 * becomes 1.
ffbd51cc
NTND
2359 *
2360 * The main thread steals half of the work from the worker that has
2361 * most work left to hand it to the idle worker.
50f22ada
JS
2362 */
2363
8ecce684
NP
2364struct thread_params {
2365 pthread_t thread;
2366 struct object_entry **list;
2367 unsigned list_size;
384b32c0 2368 unsigned remaining;
8ecce684
NP
2369 int window;
2370 int depth;
50f22ada
JS
2371 int working;
2372 int data_ready;
2373 pthread_mutex_t mutex;
2374 pthread_cond_t cond;
8ecce684
NP
2375 unsigned *processed;
2376};
2377
44626dc7
AH
2378static pthread_cond_t progress_cond;
2379
2380/*
2381 * Mutex and conditional variable can't be statically-initialized on Windows.
2382 */
2383static void init_threaded_search(void)
2384{
44626dc7
AH
2385 pthread_mutex_init(&cache_mutex, NULL);
2386 pthread_mutex_init(&progress_mutex, NULL);
2387 pthread_cond_init(&progress_cond, NULL);
851c34b0 2388 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
2389}
2390
2391static void cleanup_threaded_search(void)
2392{
851c34b0 2393 set_try_to_free_routine(old_try_to_free_routine);
44626dc7 2394 pthread_cond_destroy(&progress_cond);
44626dc7
AH
2395 pthread_mutex_destroy(&cache_mutex);
2396 pthread_mutex_destroy(&progress_mutex);
2397}
c2a33679 2398
8ecce684
NP
2399static void *threaded_find_deltas(void *arg)
2400{
c2a33679
NP
2401 struct thread_params *me = arg;
2402
0c2ad00b 2403 progress_lock();
50f22ada 2404 while (me->remaining) {
0c2ad00b
2405 progress_unlock();
2406
384b32c0 2407 find_deltas(me->list, &me->remaining,
c2a33679 2408 me->window, me->depth, me->processed);
50f22ada
JS
2409
2410 progress_lock();
2411 me->working = 0;
2412 pthread_cond_signal(&progress_cond);
2413 progress_unlock();
2414
2415 /*
2416 * We must not set ->data_ready before we wait on the
2417 * condition because the main thread may have set it to 1
2418 * before we get here. In order to be sure that new
2419 * work is available if we see 1 in ->data_ready, it
2420 * was initialized to 0 before this thread was spawned
2421 * and we reset it to 0 right away.
2422 */
2423 pthread_mutex_lock(&me->mutex);
2424 while (!me->data_ready)
2425 pthread_cond_wait(&me->cond, &me->mutex);
2426 me->data_ready = 0;
2427 pthread_mutex_unlock(&me->mutex);
0c2ad00b
2428
2429 progress_lock();
c2a33679 2430 }
0c2ad00b 2431 progress_unlock();
50f22ada
JS
2432 /* leave ->working 1 so that this doesn't get more work assigned */
2433 return NULL;
8ecce684
NP
2434}
2435
8ecce684
NP
2436static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2437 int window, int depth, unsigned *processed)
2438{
dcda3614 2439 struct thread_params *p;
384b32c0 2440 int i, ret, active_threads = 0;
c2a33679 2441
44626dc7
AH
2442 init_threaded_search();
2443
367f4a43 2444 if (delta_search_threads <= 1) {
384b32c0 2445 find_deltas(list, &list_size, window, depth, processed);
44626dc7 2446 cleanup_threaded_search();
367f4a43
NP
2447 return;
2448 }
43cc2b42 2449 if (progress > pack_to_stdout)
f616db6a 2450 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
1a07e59c 2451 delta_search_threads);
dcda3614 2452 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 2453
50f22ada 2454 /* Partition the work amongst work threads. */
367f4a43 2455 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
2456 unsigned sub_size = list_size / (delta_search_threads - i);
2457
bf874896
NP
2458 /* don't use too small segments or no deltas will be found */
2459 if (sub_size < 2*window && i+1 < delta_search_threads)
2460 sub_size = 0;
2461
8ecce684
NP
2462 p[i].window = window;
2463 p[i].depth = depth;
2464 p[i].processed = processed;
50f22ada
JS
2465 p[i].working = 1;
2466 p[i].data_ready = 0;
c2a33679 2467
59921b4b 2468 /* try to split chunks on "path" boundaries */
6fc74703
NP
2469 while (sub_size && sub_size < list_size &&
2470 list[sub_size]->hash &&
384b32c0
NP
2471 list[sub_size]->hash == list[sub_size-1]->hash)
2472 sub_size++;
2473
50f22ada
JS
2474 p[i].list = list;
2475 p[i].list_size = sub_size;
2476 p[i].remaining = sub_size;
59921b4b 2477
384b32c0
NP
2478 list += sub_size;
2479 list_size -= sub_size;
2480 }
2481
50f22ada
JS
2482 /* Start work threads. */
2483 for (i = 0; i < delta_search_threads; i++) {
2484 if (!p[i].list_size)
2485 continue;
68e6a4f8
JS
2486 pthread_mutex_init(&p[i].mutex, NULL);
2487 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
2488 ret = pthread_create(&p[i].thread, NULL,
2489 threaded_find_deltas, &p[i]);
2490 if (ret)
f616db6a 2491 die(_("unable to create thread: %s"), strerror(ret));
50f22ada
JS
2492 active_threads++;
2493 }
2494
384b32c0
NP
2495 /*
2496 * Now let's wait for work completion. Each time a thread is done
2497 * with its work, we steal half of the remaining work from the
2498 * thread with the largest number of unprocessed objects and give
2499 * it to that newly idle thread. This ensure good load balancing
2500 * until the remaining object list segments are simply too short
2501 * to be worth splitting anymore.
2502 */
50f22ada
JS
2503 while (active_threads) {
2504 struct thread_params *target = NULL;
384b32c0
NP
2505 struct thread_params *victim = NULL;
2506 unsigned sub_size = 0;
384b32c0
NP
2507
2508 progress_lock();
50f22ada
JS
2509 for (;;) {
2510 for (i = 0; !target && i < delta_search_threads; i++)
2511 if (!p[i].working)
2512 target = &p[i];
2513 if (target)
2514 break;
2515 pthread_cond_wait(&progress_cond, &progress_mutex);
2516 }
2517
384b32c0
NP
2518 for (i = 0; i < delta_search_threads; i++)
2519 if (p[i].remaining > 2*window &&
2520 (!victim || victim->remaining < p[i].remaining))
2521 victim = &p[i];
2522 if (victim) {
2523 sub_size = victim->remaining / 2;
2524 list = victim->list + victim->list_size - sub_size;
2525 while (sub_size && list[0]->hash &&
2526 list[0]->hash == list[-1]->hash) {
2527 list++;
2528 sub_size--;
2529 }
eb9688ff
NP
2530 if (!sub_size) {
2531 /*
2532 * It is possible for some "paths" to have
2533 * so many objects that no hash boundary
2534 * might be found. Let's just steal the
2535 * exact half in that case.
2536 */
2537 sub_size = victim->remaining / 2;
2538 list -= sub_size;
2539 }
384b32c0
NP
2540 target->list = list;
2541 victim->list_size -= sub_size;
2542 victim->remaining -= sub_size;
2543 }
384b32c0
NP
2544 target->list_size = sub_size;
2545 target->remaining = sub_size;
50f22ada
JS
2546 target->working = 1;
2547 progress_unlock();
2548
2549 pthread_mutex_lock(&target->mutex);
2550 target->data_ready = 1;
2551 pthread_cond_signal(&target->cond);
2552 pthread_mutex_unlock(&target->mutex);
c2a33679 2553
384b32c0 2554 if (!sub_size) {
b81d9af7 2555 pthread_join(target->thread, NULL);
50f22ada
JS
2556 pthread_cond_destroy(&target->cond);
2557 pthread_mutex_destroy(&target->mutex);
384b32c0 2558 active_threads--;
c2a33679 2559 }
50f22ada 2560 }
44626dc7 2561 cleanup_threaded_search();
dcda3614 2562 free(p);
8ecce684
NP
2563}
2564
b773ddea
JK
2565static void add_tag_chain(const struct object_id *oid)
2566{
2567 struct tag *tag;
2568
2569 /*
2570 * We catch duplicates already in add_object_entry(), but we'd
2571 * prefer to do this extra check to avoid having to parse the
2572 * tag at all if we already know that it's being packed (e.g., if
2573 * it was included via bitmaps, we would not have parsed it
2574 * previously).
2575 */
3df28cae 2576 if (packlist_find(&to_pack, oid, NULL))
b773ddea
JK
2577 return;
2578
ce71efb7 2579 tag = lookup_tag(the_repository, oid);
b773ddea
JK
2580 while (1) {
2581 if (!tag || parse_tag(tag) || !tag->tagged)
f616db6a 2582 die(_("unable to pack objects reachable from tag %s"),
b773ddea
JK
2583 oid_to_hex(oid));
2584
188960b4 2585 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
b773ddea
JK
2586
2587 if (tag->tagged->type != OBJ_TAG)
2588 return;
2589
2590 tag = (struct tag *)tag->tagged;
2591 }
2592}
2593
d155254c 2594static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
f0a24aa5 2595{
d155254c 2596 struct object_id peeled;
f0a24aa5 2597
59556548 2598 if (starts_with(path, "refs/tags/") && /* is a tag? */
b420d909 2599 !peel_ref(path, &peeled) && /* peelable? */
3df28cae 2600 packlist_find(&to_pack, &peeled, NULL)) /* object packed? */
b773ddea 2601 add_tag_chain(oid);
f0a24aa5
SP
2602 return 0;
2603}
2604
f3123c4a
JH
2605static void prepare_pack(int window, int depth)
2606{
9668cf59 2607 struct object_entry **delta_list;
6e1c2344
RJ
2608 uint32_t i, nr_deltas;
2609 unsigned n;
9668cf59 2610
28b8a730 2611 if (use_delta_islands)
385cb64f 2612 resolve_tree_islands(the_repository, progress, &to_pack);
28b8a730 2613
3f9ac8d2 2614 get_object_details();
9668cf59 2615
0e8189e2
NP
2616 /*
2617 * If we're locally repacking then we need to be doubly careful
2618 * from now on in order to make sure no stealth corruption gets
2619 * propagated to the new pack. Clients receiving streamed packs
2620 * should validate everything they get anyway so no need to incur
2621 * the additional cost here in that case.
2622 */
2623 if (!pack_to_stdout)
2624 do_check_packed_object_crc = 1;
2625
2834bc27 2626 if (!to_pack.nr_objects || !window || !depth)
9668cf59
NP
2627 return;
2628
b32fa95f 2629 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
75d39853
NP
2630 nr_deltas = n = 0;
2631
2834bc27
VM
2632 for (i = 0; i < to_pack.nr_objects; i++) {
2633 struct object_entry *entry = to_pack.objects + i;
75d39853 2634
898eba5e 2635 if (DELTA(entry))
75d39853 2636 /* This happens if we decided to reuse existing
a7de7130 2637 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2638 */
2639 continue;
2640
ac77d0c3
NTND
2641 if (!entry->type_valid ||
2642 oe_size_less_than(&to_pack, entry, 50))
75d39853
NP
2643 continue;
2644
2645 if (entry->no_try_delta)
2646 continue;
2647
6d6f9cdd 2648 if (!entry->preferred_base) {
75d39853 2649 nr_deltas++;
fd9b1bae 2650 if (oe_type(entry) < 0)
f616db6a 2651 die(_("unable to get type of object %s"),
e6a492b7 2652 oid_to_hex(&entry->idx.oid));
eede9f42 2653 } else {
fd9b1bae 2654 if (oe_type(entry) < 0) {
eede9f42
NP
2655 /*
2656 * This object is not found, but we
2657 * don't have to include it anyway.
2658 */
2659 continue;
2660 }
6d6f9cdd 2661 }
75d39853
NP
2662
2663 delta_list[n++] = entry;
2664 }
2665
2f8b8947 2666 if (nr_deltas && n > 1) {
e334977d
NP
2667 unsigned nr_done = 0;
2668 if (progress)
754dbc43 2669 progress_state = start_progress(_("Compressing objects"),
dc6a0757 2670 nr_deltas);
9ed0d8d6 2671 QSORT(delta_list, n, type_size_sort);
8ecce684 2672 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2673 stop_progress(&progress_state);
e334977d 2674 if (nr_done != nr_deltas)
f616db6a 2675 die(_("inconsistency with delta count"));
75d39853 2676 }
9668cf59 2677 free(delta_list);
f3123c4a
JH
2678}
2679
ef90d6d4 2680static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2681{
eeefa7c9 2682 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2683 window = git_config_int(k, v);
2684 return 0;
2685 }
a97773ce
BD
2686 if (!strcmp(k, "pack.windowmemory")) {
2687 window_memory_limit = git_config_ulong(k, v);
2688 return 0;
2689 }
2690 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2691 depth = git_config_int(k, v);
2692 return 0;
2693 }
074b2eea
MK
2694 if (!strcmp(k, "pack.deltacachesize")) {
2695 max_delta_cache_size = git_config_int(k, v);
2696 return 0;
2697 }
e3dfddb3
MK
2698 if (!strcmp(k, "pack.deltacachelimit")) {
2699 cache_max_small_delta_size = git_config_int(k, v);
2700 return 0;
2701 }
ae4f07fb
VM
2702 if (!strcmp(k, "pack.writebitmaphashcache")) {
2703 if (git_config_bool(k, v))
2704 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2705 else
2706 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2707 }
6b8fda2d 2708 if (!strcmp(k, "pack.usebitmaps")) {
645c432d 2709 use_bitmap_index_default = git_config_bool(k, v);
6b8fda2d
VM
2710 return 0;
2711 }
693b86ff
NP
2712 if (!strcmp(k, "pack.threads")) {
2713 delta_search_threads = git_config_int(k, v);
833e3df1 2714 if (delta_search_threads < 0)
f616db6a 2715 die(_("invalid number of threads specified (%d)"),
693b86ff 2716 delta_search_threads);
9c897c5c 2717 if (!HAVE_THREADS && delta_search_threads != 1) {
f616db6a 2718 warning(_("no threads support, ignoring %s"), k);
2e96d815
ÆAB
2719 delta_search_threads = 0;
2720 }
693b86ff
NP
2721 return 0;
2722 }
4d00bda2 2723 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2724 pack_idx_opts.version = git_config_int(k, v);
2725 if (pack_idx_opts.version > 2)
f616db6a 2726 die(_("bad pack.indexversion=%"PRIu32),
ebcfb379 2727 pack_idx_opts.version);
4d00bda2
NP
2728 return 0;
2729 }
ef90d6d4 2730 return git_default_config(k, v, cb);
4812a93a
JK
2731}
2732
b5d97e6b 2733static void read_object_list_from_stdin(void)
c323ac7d 2734{
188960b4 2735 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2736 struct object_id oid;
2737 const char *p;
b2504a0d 2738
da93d12b 2739 for (;;) {
da93d12b
LT
2740 if (!fgets(line, sizeof(line), stdin)) {
2741 if (feof(stdin))
2742 break;
2743 if (!ferror(stdin))
1a07e59c 2744 die("BUG: fgets returned NULL, not EOF, not error!");
687dd75c 2745 if (errno != EINTR)
d824cbba 2746 die_errno("fgets");
687dd75c
JH
2747 clearerr(stdin);
2748 continue;
da93d12b 2749 }
7a979d99 2750 if (line[0] == '-') {
188960b4 2751 if (get_oid_hex(line+1, &oid))
f616db6a 2752 die(_("expected edge object ID, got garbage:\n %s"),
b5d97e6b 2753 line);
188960b4 2754 add_preferred_base(&oid);
7a979d99 2755 continue;
21fcd1bd 2756 }
188960b4 2757 if (parse_oid_hex(line, &oid, &p))
f616db6a 2758 die(_("expected object ID, got garbage:\n %s"), line);
b5d97e6b 2759
188960b4 2760 add_preferred_base_object(p + 1);
fd9b1bae 2761 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
c323ac7d 2762 }
b5d97e6b
JH
2763}
2764
95308d64 2765/* Remember to update object flag allocation in object.h */
08cdfb13
JH
2766#define OBJECT_ADDED (1u<<20)
2767
11c211fa 2768static void show_commit(struct commit *commit, void *data)
b5d97e6b 2769{
188960b4 2770 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
08cdfb13 2771 commit->object.flags |= OBJECT_ADDED;
7cc8f971
VM
2772
2773 if (write_bitmap_index)
2774 index_commit_for_bitmap(commit);
28b8a730
JK
2775
2776 if (use_delta_islands)
2777 propagate_island_marks(commit);
b5d97e6b
JH
2778}
2779
de1e67d0 2780static void show_object(struct object *obj, const char *name, void *data)
b5d97e6b 2781{
8d2dfc49 2782 add_preferred_base_object(name);
188960b4 2783 add_object_entry(&obj->oid, obj->type, name, 0);
8d2dfc49 2784 obj->flags |= OBJECT_ADDED;
28b8a730
JK
2785
2786 if (use_delta_islands) {
2787 const char *p;
39490536 2788 unsigned depth;
28b8a730
JK
2789 struct object_entry *ent;
2790
39490536
JK
2791 /* the empty string is a root tree, which is depth 0 */
2792 depth = *name ? 1 : 0;
28b8a730
JK
2793 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2794 depth++;
2795
3df28cae 2796 ent = packlist_find(&to_pack, &obj->oid, NULL);
108f5303
CC
2797 if (ent && depth > oe_tree_depth(&to_pack, ent))
2798 oe_set_tree_depth(&to_pack, ent, depth);
28b8a730 2799 }
b5d97e6b
JH
2800}
2801
9535ce73
JH
2802static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2803{
2804 assert(arg_missing_action == MA_ALLOW_ANY);
2805
2806 /*
2807 * Quietly ignore ALL missing objects. This avoids problems with
2808 * staging them now and getting an odd error later.
2809 */
2810 if (!has_object_file(&obj->oid))
2811 return;
2812
2813 show_object(obj, name, data);
2814}
2815
0c16cd49
JT
2816static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2817{
2818 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2819
2820 /*
2821 * Quietly ignore EXPECTED missing objects. This avoids problems with
2822 * staging them now and getting an odd error later.
2823 */
2824 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2825 return;
2826
2827 show_object(obj, name, data);
2828}
2829
9535ce73
JH
2830static int option_parse_missing_action(const struct option *opt,
2831 const char *arg, int unset)
2832{
2833 assert(arg);
2834 assert(!unset);
2835
2836 if (!strcmp(arg, "error")) {
2837 arg_missing_action = MA_ERROR;
2838 fn_show_object = show_object;
2839 return 0;
2840 }
2841
2842 if (!strcmp(arg, "allow-any")) {
2843 arg_missing_action = MA_ALLOW_ANY;
0c16cd49 2844 fetch_if_missing = 0;
9535ce73
JH
2845 fn_show_object = show_object__ma_allow_any;
2846 return 0;
2847 }
2848
0c16cd49
JT
2849 if (!strcmp(arg, "allow-promisor")) {
2850 arg_missing_action = MA_ALLOW_PROMISOR;
2851 fetch_if_missing = 0;
2852 fn_show_object = show_object__ma_allow_promisor;
2853 return 0;
2854 }
2855
9535ce73
JH
2856 die(_("invalid value for --missing"));
2857 return 0;
2858}
2859
8d1d8f83
JH
2860static void show_edge(struct commit *commit)
2861{
188960b4 2862 add_preferred_base(&commit->object.oid);
8d1d8f83
JH
2863}
2864
08cdfb13
JH
2865struct in_pack_object {
2866 off_t offset;
2867 struct object *object;
2868};
2869
2870struct in_pack {
071bcaab
RJ
2871 unsigned int alloc;
2872 unsigned int nr;
08cdfb13
JH
2873 struct in_pack_object *array;
2874};
2875
2876static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2877{
ed1c9977 2878 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
08cdfb13
JH
2879 in_pack->array[in_pack->nr].object = object;
2880 in_pack->nr++;
2881}
2882
2883/*
2884 * Compare the objects in the offset order, in order to emulate the
f18d244a 2885 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2886 */
2887static int ofscmp(const void *a_, const void *b_)
2888{
2889 struct in_pack_object *a = (struct in_pack_object *)a_;
2890 struct in_pack_object *b = (struct in_pack_object *)b_;
2891
2892 if (a->offset < b->offset)
2893 return -1;
2894 else if (a->offset > b->offset)
2895 return 1;
2896 else
f2fd0760 2897 return oidcmp(&a->object->oid, &b->object->oid);
08cdfb13
JH
2898}
2899
7a2a7216 2900static void add_objects_in_unpacked_packs(void)
08cdfb13
JH
2901{
2902 struct packed_git *p;
2903 struct in_pack in_pack;
2904 uint32_t i;
2905
2906 memset(&in_pack, 0, sizeof(in_pack));
2907
454ea2e4 2908 for (p = get_all_packs(the_repository); p; p = p->next) {
188960b4 2909 struct object_id oid;
08cdfb13
JH
2910 struct object *o;
2911
ed7e5fc3 2912 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
08cdfb13
JH
2913 continue;
2914 if (open_pack_index(p))
f616db6a 2915 die(_("cannot open pack index"));
08cdfb13
JH
2916
2917 ALLOC_GROW(in_pack.array,
2918 in_pack.nr + p->num_objects,
2919 in_pack.alloc);
2920
2921 for (i = 0; i < p->num_objects; i++) {
188960b4 2922 nth_packed_object_oid(&oid, p, i);
0ebbcf70 2923 o = lookup_unknown_object(&oid);
08cdfb13
JH
2924 if (!(o->flags & OBJECT_ADDED))
2925 mark_in_pack_object(o, p, &in_pack);
2926 o->flags |= OBJECT_ADDED;
2927 }
2928 }
2929
2930 if (in_pack.nr) {
9ed0d8d6 2931 QSORT(in_pack.array, in_pack.nr, ofscmp);
08cdfb13
JH
2932 for (i = 0; i < in_pack.nr; i++) {
2933 struct object *o = in_pack.array[i].object;
188960b4 2934 add_object_entry(&o->oid, o->type, "", 0);
08cdfb13
JH
2935 }
2936 }
2937 free(in_pack.array);
2938}
2939
76c1d9a0 2940static int add_loose_object(const struct object_id *oid, const char *path,
e26a8c47
JK
2941 void *data)
2942{
0df8e965 2943 enum object_type type = oid_object_info(the_repository, oid, NULL);
e26a8c47
JK
2944
2945 if (type < 0) {
f616db6a 2946 warning(_("loose object at %s could not be examined"), path);
e26a8c47
JK
2947 return 0;
2948 }
2949
188960b4 2950 add_object_entry(oid, type, "", 0);
e26a8c47
JK
2951 return 0;
2952}
2953
2954/*
2955 * We actually don't even have to worry about reachability here.
2956 * add_object_entry will weed out duplicates, so we just add every
2957 * loose object we find.
2958 */
2959static void add_unreachable_loose_objects(void)
2960{
2961 for_each_loose_file_in_objdir(get_object_directory(),
2962 add_loose_object,
2963 NULL, NULL, NULL);
2964}
2965
188960b4 2966static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
094085e3
BC
2967{
2968 static struct packed_git *last_found = (void *)1;
2969 struct packed_git *p;
2970
a80d72db 2971 p = (last_found != (void *)1) ? last_found :
454ea2e4 2972 get_all_packs(the_repository);
094085e3
BC
2973
2974 while (p) {
ed7e5fc3
NTND
2975 if ((!p->pack_local || p->pack_keep ||
2976 p->pack_keep_in_core) &&
188960b4 2977 find_pack_entry_one(oid->hash, p)) {
094085e3
BC
2978 last_found = p;
2979 return 1;
2980 }
2981 if (p == last_found)
454ea2e4 2982 p = get_all_packs(the_repository);
094085e3
BC
2983 else
2984 p = p->next;
2985 if (p == last_found)
2986 p = p->next;
2987 }
2988 return 0;
2989}
2990
abcb8655
JK
2991/*
2992 * Store a list of sha1s that are should not be discarded
2993 * because they are either written too recently, or are
2994 * reachable from another object that was.
2995 *
2996 * This is filled by get_object_list.
2997 */
910650d2 2998static struct oid_array recent_objects;
abcb8655 2999
4ce3621a 3000static int loosened_object_can_be_discarded(const struct object_id *oid,
dddbad72 3001 timestamp_t mtime)
d0d46abc
JK
3002{
3003 if (!unpack_unreachable_expiration)
3004 return 0;
3005 if (mtime > unpack_unreachable_expiration)
3006 return 0;
910650d2 3007 if (oid_array_lookup(&recent_objects, oid) >= 0)
abcb8655 3008 return 0;
d0d46abc
JK
3009 return 1;
3010}
3011
7a2a7216 3012static void loosen_unused_packed_objects(void)
ca11b212
NP
3013{
3014 struct packed_git *p;
3015 uint32_t i;
4ce3621a 3016 struct object_id oid;
ca11b212 3017
454ea2e4 3018 for (p = get_all_packs(the_repository); p; p = p->next) {
ed7e5fc3 3019 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
ca11b212
NP
3020 continue;
3021
3022 if (open_pack_index(p))
f616db6a 3023 die(_("cannot open pack index"));
ca11b212
NP
3024
3025 for (i = 0; i < p->num_objects; i++) {
4ce3621a 3026 nth_packed_object_oid(&oid, p, i);
3df28cae 3027 if (!packlist_find(&to_pack, &oid, NULL) &&
188960b4 3028 !has_sha1_pack_kept_or_nonlocal(&oid) &&
4ce3621a 3029 !loosened_object_can_be_discarded(&oid, p->mtime))
4bdb70a4 3030 if (force_object_loose(&oid, p->mtime))
f616db6a 3031 die(_("unable to force loose object"));
ca11b212
NP
3032 }
3033 }
3034}
3035
69e4b342 3036/*
645c432d
KS
3037 * This tracks any options which pack-reuse code expects to be on, or which a
3038 * reader of the pack might not understand, and which would therefore prevent
3039 * blind reuse of what we have on disk.
69e4b342
JK
3040 */
3041static int pack_options_allow_reuse(void)
3042{
9df4a607
JK
3043 return pack_to_stdout &&
3044 allow_ofs_delta &&
ed7e5fc3
NTND
3045 !ignore_packed_keep_on_disk &&
3046 !ignore_packed_keep_in_core &&
9df4a607
JK
3047 (!local || !have_non_local_packs) &&
3048 !incremental;
69e4b342
JK
3049}
3050
6b8fda2d
VM
3051static int get_object_list_from_bitmap(struct rev_info *revs)
3052{
3ae5fa07 3053 if (!(bitmap_git = prepare_bitmap_walk(revs)))
6b8fda2d
VM
3054 return -1;
3055
69e4b342
JK
3056 if (pack_options_allow_reuse() &&
3057 !reuse_partial_packfile_from_bitmap(
3ae5fa07 3058 bitmap_git,
6b8fda2d
VM
3059 &reuse_packfile,
3060 &reuse_packfile_objects,
3061 &reuse_packfile_offset)) {
3062 assert(reuse_packfile_objects);
3063 nr_result += reuse_packfile_objects;
78d2214e 3064 display_progress(progress_state, nr_result);
6b8fda2d
VM
3065 }
3066
3ae5fa07 3067 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
6b8fda2d
VM
3068 return 0;
3069}
3070
abcb8655 3071static void record_recent_object(struct object *obj,
de1e67d0 3072 const char *name,
abcb8655
JK
3073 void *data)
3074{
910650d2 3075 oid_array_append(&recent_objects, &obj->oid);
abcb8655
JK
3076}
3077
3078static void record_recent_commit(struct commit *commit, void *data)
3079{
910650d2 3080 oid_array_append(&recent_objects, &commit->object.oid);
abcb8655
JK
3081}
3082
8d1d8f83 3083static void get_object_list(int ac, const char **av)
b5d97e6b
JH
3084{
3085 struct rev_info revs;
bbcde41a
MD
3086 struct setup_revision_opt s_r_opt = {
3087 .allow_exclude_promisor_objects = 1,
3088 };
b5d97e6b 3089 char line[1000];
b5d97e6b 3090 int flags = 0;
a4544b31 3091 int save_warning;
b5d97e6b 3092
2abf3503 3093 repo_init_revisions(the_repository, &revs, NULL);
b5d97e6b 3094 save_commit_buffer = 0;
bbcde41a 3095 setup_revisions(ac, av, &revs, &s_r_opt);
b5d97e6b 3096
b790e0f6 3097 /* make sure shallows are read */
c8813487 3098 is_repository_shallow(the_repository);
b790e0f6 3099
a4544b31
DS
3100 save_warning = warn_on_object_refname_ambiguity;
3101 warn_on_object_refname_ambiguity = 0;
3102
b5d97e6b
JH
3103 while (fgets(line, sizeof(line), stdin) != NULL) {
3104 int len = strlen(line);
872c930d 3105 if (len && line[len - 1] == '\n')
b5d97e6b
JH
3106 line[--len] = 0;
3107 if (!len)
3108 break;
3109 if (*line == '-') {
3110 if (!strcmp(line, "--not")) {
3111 flags ^= UNINTERESTING;
7cc8f971 3112 write_bitmap_index = 0;
b5d97e6b
JH
3113 continue;
3114 }
b790e0f6 3115 if (starts_with(line, "--shallow ")) {
e92b848c 3116 struct object_id oid;
3117 if (get_oid_hex(line + 10, &oid))
b790e0f6 3118 die("not an SHA-1 '%s'", line + 10);
19143f13 3119 register_shallow(the_repository, &oid);
f7f91086 3120 use_bitmap_index = 0;
b790e0f6
NTND
3121 continue;
3122 }
f616db6a 3123 die(_("not a rev '%s'"), line);
b5d97e6b 3124 }
8e676e8b 3125 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
f616db6a 3126 die(_("bad revision '%s'"), line);
b5d97e6b
JH
3127 }
3128
a4544b31
DS
3129 warn_on_object_refname_ambiguity = save_warning;
3130
6b8fda2d
VM
3131 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3132 return;
3133
28b8a730 3134 if (use_delta_islands)
bdbdf42f 3135 load_delta_islands(the_repository, progress);
28b8a730 3136
3d51e1b5 3137 if (prepare_revision_walk(&revs))
f616db6a 3138 die(_("revision walk setup failed"));
4f6d26b1 3139 mark_edges_uninteresting(&revs, show_edge, sparse);
9535ce73
JH
3140
3141 if (!fn_show_object)
3142 fn_show_object = show_object;
3143 traverse_commit_list_filtered(&filter_options, &revs,
3144 show_commit, fn_show_object, NULL,
3145 NULL);
08cdfb13 3146
abcb8655
JK
3147 if (unpack_unreachable_expiration) {
3148 revs.ignore_missing_links = 1;
3149 if (add_unseen_recent_objects_to_traversal(&revs,
3150 unpack_unreachable_expiration))
f616db6a 3151 die(_("unable to add recent objects"));
abcb8655 3152 if (prepare_revision_walk(&revs))
f616db6a 3153 die(_("revision walk setup failed"));
abcb8655
JK
3154 traverse_commit_list(&revs, record_recent_commit,
3155 record_recent_object, NULL);
3156 }
3157
08cdfb13 3158 if (keep_unreachable)
7a2a7216 3159 add_objects_in_unpacked_packs();
e26a8c47
JK
3160 if (pack_loose_unreachable)
3161 add_unreachable_loose_objects();
ca11b212 3162 if (unpack_unreachable)
7a2a7216 3163 loosen_unused_packed_objects();
abcb8655 3164
910650d2 3165 oid_array_clear(&recent_objects);
b5d97e6b
JH
3166}
3167
ed7e5fc3
NTND
3168static void add_extra_kept_packs(const struct string_list *names)
3169{
3170 struct packed_git *p;
3171
3172 if (!names->nr)
3173 return;
3174
454ea2e4