pack-objects: drop packlist index_pos optimization
[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;
25575015
JK
99static enum {
100 WRITE_BITMAP_FALSE = 0,
101 WRITE_BITMAP_QUIET,
102 WRITE_BITMAP_TRUE,
103} write_bitmap_index;
d4316604 104static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE;
6b8fda2d 105
0c16cd49
JT
106static int exclude_promisor_objects;
107
28b8a730
JK
108static int use_delta_islands;
109
074b2eea 110static unsigned long delta_cache_size = 0;
9806f5a7 111static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
e3dfddb3 112static unsigned long cache_max_small_delta_size = 1000;
074b2eea 113
a97773ce
BD
114static unsigned long window_memory_limit = 0;
115
9535ce73
JH
116static struct list_objects_filter_options filter_options;
117
118enum missing_action {
0c16cd49
JT
119 MA_ERROR = 0, /* fail if any missing objects are encountered */
120 MA_ALLOW_ANY, /* silently allow ALL missing objects */
121 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
9535ce73
JH
122};
123static enum missing_action arg_missing_action;
124static show_object_fn fn_show_object;
125
3f9ac8d2
JH
126/*
127 * stats
128 */
7cadf491
SP
129static uint32_t written, written_delta;
130static uint32_t reused, reused_delta;
3f9ac8d2 131
7cc8f971
VM
132/*
133 * Indexed commits
134 */
135static struct commit **indexed_commits;
136static unsigned int indexed_commits_nr;
137static unsigned int indexed_commits_alloc;
138
139static void index_commit_for_bitmap(struct commit *commit)
140{
141 if (indexed_commits_nr >= indexed_commits_alloc) {
142 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
2756ca43 143 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
7cc8f971
VM
144 }
145
146 indexed_commits[indexed_commits_nr++] = commit;
147}
780e6e73 148
3613f9b4 149static void *get_delta(struct object_entry *entry)
c323ac7d 150{
3613f9b4
NP
151 unsigned long size, base_size, delta_size;
152 void *buf, *base_buf, *delta_buf;
21666f1a 153 enum object_type type;
c323ac7d 154
b4f5aca4 155 buf = read_object_file(&entry->idx.oid, &type, &size);
3613f9b4 156 if (!buf)
f616db6a 157 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
898eba5e
NTND
158 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
159 &base_size);
3613f9b4 160 if (!base_buf)
e6a492b7 161 die("unable to read %s",
898eba5e 162 oid_to_hex(&DELTA(entry)->idx.oid));
3613f9b4 163 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 164 buf, size, &delta_size, 0);
1a07e59c
NTND
165 /*
166 * We succesfully computed this delta once but dropped it for
167 * memory reasons. Something is very wrong if this time we
168 * recompute and create a different delta.
169 */
0aca34e8 170 if (!delta_buf || delta_size != DELTA_SIZE(entry))
1a07e59c 171 BUG("delta size changed");
3613f9b4
NP
172 free(buf);
173 free(base_buf);
c323ac7d
LT
174 return delta_buf;
175}
176
30ebb40a
NP
177static unsigned long do_compress(void **pptr, unsigned long size)
178{
ef49a7a0 179 git_zstream stream;
30ebb40a
NP
180 void *in, *out;
181 unsigned long maxsize;
182
55bb5c91 183 git_deflate_init(&stream, pack_compression_level);
225a6f10 184 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
185
186 in = *pptr;
187 out = xmalloc(maxsize);
188 *pptr = out;
189
190 stream.next_in = in;
191 stream.avail_in = size;
192 stream.next_out = out;
193 stream.avail_out = maxsize;
55bb5c91 194 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 195 ; /* nothing */
55bb5c91 196 git_deflate_end(&stream);
30ebb40a
NP
197
198 free(in);
199 return stream.total_out;
200}
201
98a3beab 202static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
188960b4 203 const struct object_id *oid)
cf2ba13a
NTND
204{
205 git_zstream stream;
206 unsigned char ibuf[1024 * 16];
207 unsigned char obuf[1024 * 16];
208 unsigned long olen = 0;
209
cf2ba13a
NTND
210 git_deflate_init(&stream, pack_compression_level);
211
212 for (;;) {
213 ssize_t readlen;
214 int zret = Z_OK;
215 readlen = read_istream(st, ibuf, sizeof(ibuf));
216 if (readlen == -1)
188960b4 217 die(_("unable to read %s"), oid_to_hex(oid));
cf2ba13a
NTND
218
219 stream.next_in = ibuf;
220 stream.avail_in = readlen;
221 while ((stream.avail_in || readlen == 0) &&
222 (zret == Z_OK || zret == Z_BUF_ERROR)) {
223 stream.next_out = obuf;
224 stream.avail_out = sizeof(obuf);
225 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
98a3beab 226 hashwrite(f, obuf, stream.next_out - obuf);
cf2ba13a
NTND
227 olen += stream.next_out - obuf;
228 }
229 if (stream.avail_in)
230 die(_("deflate error (%d)"), zret);
231 if (readlen == 0) {
232 if (zret != Z_STREAM_END)
233 die(_("deflate error (%d)"), zret);
234 break;
235 }
236 }
237 git_deflate_end(&stream);
238 return olen;
239}
240
780e6e73
NP
241/*
242 * we are going to reuse the existing object data as is. make
243 * sure it is not corrupt.
244 */
079afb18
SP
245static int check_pack_inflate(struct packed_git *p,
246 struct pack_window **w_curs,
6777a59f
SP
247 off_t offset,
248 off_t len,
079afb18
SP
249 unsigned long expect)
250{
ef49a7a0 251 git_zstream stream;
079afb18
SP
252 unsigned char fakebuf[4096], *in;
253 int st;
254
255 memset(&stream, 0, sizeof(stream));
39c68542 256 git_inflate_init(&stream);
079afb18
SP
257 do {
258 in = use_pack(p, w_curs, offset, &stream.avail_in);
259 stream.next_in = in;
260 stream.next_out = fakebuf;
261 stream.avail_out = sizeof(fakebuf);
39c68542 262 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
263 offset += stream.next_in - in;
264 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 265 git_inflate_end(&stream);
079afb18
SP
266 return (st == Z_STREAM_END &&
267 stream.total_out == expect &&
268 stream.total_in == len) ? 0 : -1;
269}
270
98a3beab 271static void copy_pack_data(struct hashfile *f,
079afb18
SP
272 struct packed_git *p,
273 struct pack_window **w_curs,
6777a59f
SP
274 off_t offset,
275 off_t len)
079afb18
SP
276{
277 unsigned char *in;
ef49a7a0 278 unsigned long avail;
079afb18
SP
279
280 while (len) {
281 in = use_pack(p, w_curs, offset, &avail);
282 if (avail > len)
ef49a7a0 283 avail = (unsigned long)len;
98a3beab 284 hashwrite(f, in, avail);
079afb18
SP
285 offset += avail;
286 len -= avail;
287 }
288}
289
1b4bb16b 290/* Return 0 if we will bust the pack-size limit */
98a3beab 291static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
c9018b03 292 unsigned long limit, int usable_delta)
c323ac7d 293{
c9018b03 294 unsigned long size, datalen;
2c5e2865
JK
295 unsigned char header[MAX_PACK_OBJECT_HEADER],
296 dheader[MAX_PACK_OBJECT_HEADER];
6777a59f 297 unsigned hdrlen;
2c5ef824 298 enum object_type type;
c9018b03 299 void *buf;
cf2ba13a 300 struct git_istream *st = NULL;
41179100 301 const unsigned hashsz = the_hash_algo->rawsz;
c9018b03
NTND
302
303 if (!usable_delta) {
fd9b1bae 304 if (oe_type(entry) == OBJ_BLOB &&
ac77d0c3 305 oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
ef7b5195 306 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
cf2ba13a
NTND
307 buf = NULL;
308 else {
b4f5aca4 309 buf = read_object_file(&entry->idx.oid, &type, &size);
cf2ba13a 310 if (!buf)
e6a492b7 311 die(_("unable to read %s"),
312 oid_to_hex(&entry->idx.oid));
cf2ba13a 313 }
c9018b03
NTND
314 /*
315 * make sure no cached delta data remains from a
316 * previous attempt before a pack split occurred.
317 */
6a83d902 318 FREE_AND_NULL(entry->delta_data);
c9018b03
NTND
319 entry->z_delta_size = 0;
320 } else if (entry->delta_data) {
0aca34e8 321 size = DELTA_SIZE(entry);
c9018b03
NTND
322 buf = entry->delta_data;
323 entry->delta_data = NULL;
898eba5e 324 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03
NTND
325 OBJ_OFS_DELTA : OBJ_REF_DELTA;
326 } else {
327 buf = get_delta(entry);
0aca34e8 328 size = DELTA_SIZE(entry);
898eba5e 329 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03
NTND
330 OBJ_OFS_DELTA : OBJ_REF_DELTA;
331 }
332
cf2ba13a
NTND
333 if (st) /* large blob case, just assume we don't compress well */
334 datalen = size;
335 else if (entry->z_delta_size)
c9018b03
NTND
336 datalen = entry->z_delta_size;
337 else
338 datalen = do_compress(&buf, size);
339
340 /*
341 * The object header is a byte of 'type' followed by zero or
342 * more bytes of length.
343 */
7202a6fa
JK
344 hdrlen = encode_in_pack_object_header(header, sizeof(header),
345 type, size);
c9018b03
NTND
346
347 if (type == OBJ_OFS_DELTA) {
348 /*
349 * Deltas with relative base contain an additional
350 * encoding of the relative offset for the delta
351 * base from this object's position in the pack.
352 */
898eba5e 353 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
c9018b03
NTND
354 unsigned pos = sizeof(dheader) - 1;
355 dheader[pos] = ofs & 127;
356 while (ofs >>= 7)
357 dheader[--pos] = 128 | (--ofs & 127);
41179100 358 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
cf2ba13a
NTND
359 if (st)
360 close_istream(st);
c9018b03
NTND
361 free(buf);
362 return 0;
363 }
98a3beab 364 hashwrite(f, header, hdrlen);
365 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
c9018b03
NTND
366 hdrlen += sizeof(dheader) - pos;
367 } else if (type == OBJ_REF_DELTA) {
368 /*
369 * Deltas with a base reference contain
41179100 370 * additional bytes for the base object ID.
c9018b03 371 */
41179100 372 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
cf2ba13a
NTND
373 if (st)
374 close_istream(st);
c9018b03
NTND
375 free(buf);
376 return 0;
377 }
98a3beab 378 hashwrite(f, header, hdrlen);
42c8ce1c 379 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
41179100 380 hdrlen += hashsz;
c9018b03 381 } else {
41179100 382 if (limit && hdrlen + datalen + hashsz >= limit) {
cf2ba13a
NTND
383 if (st)
384 close_istream(st);
c9018b03
NTND
385 free(buf);
386 return 0;
387 }
98a3beab 388 hashwrite(f, header, hdrlen);
c9018b03 389 }
cf2ba13a 390 if (st) {
188960b4 391 datalen = write_large_blob_data(st, f, &entry->idx.oid);
cf2ba13a
NTND
392 close_istream(st);
393 } else {
98a3beab 394 hashwrite(f, buf, datalen);
cf2ba13a
NTND
395 free(buf);
396 }
c9018b03
NTND
397
398 return hdrlen + datalen;
399}
400
401/* Return 0 if we will bust the pack-size limit */
98a3beab 402static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
af92a645 403 unsigned long limit, int usable_delta)
c9018b03 404{
43fa44fa 405 struct packed_git *p = IN_PACK(entry);
c9018b03
NTND
406 struct pack_window *w_curs = NULL;
407 struct revindex_entry *revidx;
408 off_t offset;
fd9b1bae 409 enum object_type type = oe_type(entry);
211c61c6 410 off_t datalen;
2c5e2865
JK
411 unsigned char header[MAX_PACK_OBJECT_HEADER],
412 dheader[MAX_PACK_OBJECT_HEADER];
c9018b03 413 unsigned hdrlen;
41179100 414 const unsigned hashsz = the_hash_algo->rawsz;
ac77d0c3 415 unsigned long entry_size = SIZE(entry);
c9018b03 416
898eba5e
NTND
417 if (DELTA(entry))
418 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
c9018b03 419 OBJ_OFS_DELTA : OBJ_REF_DELTA;
7202a6fa 420 hdrlen = encode_in_pack_object_header(header, sizeof(header),
ac77d0c3 421 type, entry_size);
c9018b03
NTND
422
423 offset = entry->in_pack_offset;
424 revidx = find_pack_revindex(p, offset);
425 datalen = revidx[1].offset - offset;
426 if (!pack_to_stdout && p->index_version > 1 &&
427 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
f616db6a 428 error(_("bad packed object CRC for %s"),
e6a492b7 429 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
430 unuse_pack(&w_curs);
431 return write_no_reuse_object(f, entry, limit, usable_delta);
432 }
433
434 offset += entry->in_pack_header_size;
435 datalen -= entry->in_pack_header_size;
436
437 if (!pack_to_stdout && p->index_version == 1 &&
ac77d0c3 438 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
f616db6a 439 error(_("corrupt packed object for %s"),
e6a492b7 440 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
441 unuse_pack(&w_curs);
442 return write_no_reuse_object(f, entry, limit, usable_delta);
443 }
444
445 if (type == OBJ_OFS_DELTA) {
898eba5e 446 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
c9018b03
NTND
447 unsigned pos = sizeof(dheader) - 1;
448 dheader[pos] = ofs & 127;
449 while (ofs >>= 7)
450 dheader[--pos] = 128 | (--ofs & 127);
41179100 451 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
c9018b03
NTND
452 unuse_pack(&w_curs);
453 return 0;
454 }
98a3beab 455 hashwrite(f, header, hdrlen);
456 hashwrite(f, dheader + pos, sizeof(dheader) - pos);
c9018b03
NTND
457 hdrlen += sizeof(dheader) - pos;
458 reused_delta++;
459 } else if (type == OBJ_REF_DELTA) {
41179100 460 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
c9018b03
NTND
461 unuse_pack(&w_curs);
462 return 0;
463 }
98a3beab 464 hashwrite(f, header, hdrlen);
42c8ce1c 465 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
41179100 466 hdrlen += hashsz;
c9018b03
NTND
467 reused_delta++;
468 } else {
41179100 469 if (limit && hdrlen + datalen + hashsz >= limit) {
c9018b03
NTND
470 unuse_pack(&w_curs);
471 return 0;
472 }
98a3beab 473 hashwrite(f, header, hdrlen);
c9018b03
NTND
474 }
475 copy_pack_data(f, p, &w_curs, offset, datalen);
476 unuse_pack(&w_curs);
477 reused++;
478 return hdrlen + datalen;
479}
480
481/* Return 0 if we will bust the pack-size limit */
98a3beab 482static off_t write_object(struct hashfile *f,
af92a645
NTND
483 struct object_entry *entry,
484 off_t write_offset)
c9018b03 485{
af92a645
NTND
486 unsigned long limit;
487 off_t len;
2c5ef824 488 int usable_delta, to_reuse;
c323ac7d 489
78d1e84f
NP
490 if (!pack_to_stdout)
491 crc32_begin(f);
492
a2430dde 493 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
494 if (!pack_size_limit || !nr_written)
495 limit = 0;
496 else if (pack_size_limit <= write_offset)
497 /*
498 * the earlier object did not fit the limit; avoid
499 * mistaking this with unlimited (i.e. limit = 0).
500 */
501 limit = 1;
502 else
503 limit = pack_size_limit - write_offset;
2c5ef824 504
898eba5e 505 if (!DELTA(entry))
2c5ef824
NP
506 usable_delta = 0; /* no delta */
507 else if (!pack_size_limit)
508 usable_delta = 1; /* unlimited packfile */
898eba5e 509 else if (DELTA(entry)->idx.offset == (off_t)-1)
2c5ef824 510 usable_delta = 0; /* base was written to another pack */
898eba5e 511 else if (DELTA(entry)->idx.offset)
2c5ef824
NP
512 usable_delta = 1; /* base already exists in this pack */
513 else
514 usable_delta = 0; /* base could end up in another pack */
515
a7de7130 516 if (!reuse_object)
fa736f72 517 to_reuse = 0; /* explicit */
43fa44fa 518 else if (!IN_PACK(entry))
ab7cd7bb 519 to_reuse = 0; /* can't reuse what we don't have */
fd9b1bae
NTND
520 else if (oe_type(entry) == OBJ_REF_DELTA ||
521 oe_type(entry) == OBJ_OFS_DELTA)
17b08f2c
DH
522 /* check_object() decided it for us ... */
523 to_reuse = usable_delta;
524 /* ... but pack split may override that */
fd9b1bae 525 else if (oe_type(entry) != entry->in_pack_type)
ab7cd7bb 526 to_reuse = 0; /* pack has delta which is unusable */
898eba5e 527 else if (DELTA(entry))
ab7cd7bb
JH
528 to_reuse = 0; /* we want to pack afresh */
529 else
530 to_reuse = 1; /* we have it in-pack undeltified,
531 * and we do not need to deltify it.
532 */
533
c9018b03
NTND
534 if (!to_reuse)
535 len = write_no_reuse_object(f, entry, limit, usable_delta);
536 else
537 len = write_reuse_object(f, entry, limit, usable_delta);
538 if (!len)
539 return 0;
64bd76b1 540
17b08f2c 541 if (usable_delta)
ab7cd7bb 542 written_delta++;
3f9ac8d2 543 written++;
78d1e84f 544 if (!pack_to_stdout)
aa7e44bf 545 entry->idx.crc32 = crc32_end(f);
c9018b03 546 return len;
c323ac7d
LT
547}
548
f63c79db
JH
549enum write_one_status {
550 WRITE_ONE_SKIP = -1, /* already written */
551 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
552 WRITE_ONE_WRITTEN = 1, /* normal */
553 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
554};
555
98a3beab 556static enum write_one_status write_one(struct hashfile *f,
f63c79db
JH
557 struct object_entry *e,
558 off_t *offset)
9d5ab962 559{
af92a645 560 off_t size;
f63c79db 561 int recursing;
d7dd0223 562
f63c79db
JH
563 /*
564 * we set offset to 1 (which is an impossible value) to mark
565 * the fact that this object is involved in "write its base
566 * first before writing a deltified object" recursion.
567 */
568 recursing = (e->idx.offset == 1);
569 if (recursing) {
f616db6a 570 warning(_("recursive delta detected for object %s"),
e6a492b7 571 oid_to_hex(&e->idx.oid));
f63c79db
JH
572 return WRITE_ONE_RECURSIVE;
573 } else if (e->idx.offset || e->preferred_base) {
574 /* offset is non zero if object is written already. */
575 return WRITE_ONE_SKIP;
576 }
d7dd0223 577
720c9f7b 578 /* if we are deltified, write out base object first. */
898eba5e 579 if (DELTA(e)) {
f63c79db 580 e->idx.offset = 1; /* now recurse */
898eba5e 581 switch (write_one(f, DELTA(e), offset)) {
f63c79db
JH
582 case WRITE_ONE_RECURSIVE:
583 /* we cannot depend on this one */
898eba5e 584 SET_DELTA(e, NULL);
f63c79db
JH
585 break;
586 default:
587 break;
588 case WRITE_ONE_BREAK:
589 e->idx.offset = recursing;
590 return WRITE_ONE_BREAK;
591 }
592 }
d7dd0223 593
6ed7f25e
NP
594 e->idx.offset = *offset;
595 size = write_object(f, e, *offset);
17b08f2c 596 if (!size) {
f63c79db
JH
597 e->idx.offset = recursing;
598 return WRITE_ONE_BREAK;
17b08f2c 599 }
79814f42 600 written_list[nr_written++] = &e->idx;
d7dd0223
NP
601
602 /* make sure off_t is sufficiently large not to wrap */
c03c8315 603 if (signed_add_overflows(*offset, size))
f616db6a 604 die(_("pack too large for current definition of off_t"));
6ed7f25e 605 *offset += size;
f63c79db 606 return WRITE_ONE_WRITTEN;
9d5ab962
JH
607}
608
d155254c 609static int mark_tagged(const char *path, const struct object_id *oid, int flag,
1b4bb16b
JH
610 void *cb_data)
611{
188960b4 612 struct object_id peeled;
3a37876b 613 struct object_entry *entry = packlist_find(&to_pack, oid);
1b4bb16b
JH
614
615 if (entry)
616 entry->tagged = 1;
b420d909 617 if (!peel_ref(path, &peeled)) {
3a37876b 618 entry = packlist_find(&to_pack, &peeled);
1b4bb16b
JH
619 if (entry)
620 entry->tagged = 1;
621 }
622 return 0;
623}
624
be126818 625static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 626 unsigned int *endp,
1b4bb16b
JH
627 struct object_entry *e)
628{
fe0ac2fb 629 if (e->filled || oe_layer(&to_pack, e) != write_layer)
1b4bb16b
JH
630 return;
631 wo[(*endp)++] = e;
632 e->filled = 1;
633}
634
635static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 636 unsigned int *endp,
1b4bb16b
JH
637 struct object_entry *e)
638{
f380872f
DM
639 int add_to_order = 1;
640 while (e) {
641 if (add_to_order) {
642 struct object_entry *s;
643 /* add this node... */
644 add_to_write_order(wo, endp, e);
645 /* all its siblings... */
898eba5e 646 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
f380872f
DM
647 add_to_write_order(wo, endp, s);
648 }
649 }
650 /* drop down a level to add left subtree nodes if possible */
898eba5e 651 if (DELTA_CHILD(e)) {
f380872f 652 add_to_order = 1;
898eba5e 653 e = DELTA_CHILD(e);
f380872f
DM
654 } else {
655 add_to_order = 0;
656 /* our sibling might have some children, it is next */
898eba5e
NTND
657 if (DELTA_SIBLING(e)) {
658 e = DELTA_SIBLING(e);
f380872f
DM
659 continue;
660 }
661 /* go back to our parent node */
898eba5e
NTND
662 e = DELTA(e);
663 while (e && !DELTA_SIBLING(e)) {
f380872f
DM
664 /* we're on the right side of a subtree, keep
665 * going up until we can go right again */
898eba5e 666 e = DELTA(e);
f380872f
DM
667 }
668 if (!e) {
669 /* done- we hit our original root node */
670 return;
671 }
672 /* pass it off to sibling at this level */
898eba5e 673 e = DELTA_SIBLING(e);
f380872f
DM
674 }
675 };
1b4bb16b
JH
676}
677
678static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 679 unsigned int *endp,
1b4bb16b
JH
680 struct object_entry *e)
681{
682 struct object_entry *root;
683
898eba5e 684 for (root = e; DELTA(root); root = DELTA(root))
1b4bb16b 685 ; /* nothing */
1b4bb16b
JH
686 add_descendants_to_write_order(wo, endp, root);
687}
688
f64ba53e 689static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
1b4bb16b 690{
f64ba53e 691 unsigned int i, last_untagged;
2834bc27 692 struct object_entry *objects = to_pack.objects;
1b4bb16b 693
2834bc27 694 for (i = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
695 if (objects[i].tagged)
696 break;
f64ba53e 697 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b 698 }
38d4debb 699 last_untagged = i;
1b4bb16b
JH
700
701 /*
702 * Then fill all the tagged tips.
703 */
2834bc27 704 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 705 if (objects[i].tagged)
f64ba53e 706 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
707 }
708
709 /*
710 * And then all remaining commits and tags.
711 */
2834bc27 712 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fd9b1bae
NTND
713 if (oe_type(&objects[i]) != OBJ_COMMIT &&
714 oe_type(&objects[i]) != OBJ_TAG)
1b4bb16b 715 continue;
f64ba53e 716 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
717 }
718
719 /*
720 * And then all the trees.
721 */
2834bc27 722 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fd9b1bae 723 if (oe_type(&objects[i]) != OBJ_TREE)
1b4bb16b 724 continue;
f64ba53e 725 add_to_write_order(wo, wo_end, &objects[i]);
1b4bb16b
JH
726 }
727
728 /*
729 * Finally all the rest in really tight order
730 */
2834bc27 731 for (i = last_untagged; i < to_pack.nr_objects; i++) {
fe0ac2fb 732 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
f64ba53e 733 add_family_to_write_order(wo, wo_end, &objects[i]);
38d4debb 734 }
f64ba53e
CC
735}
736
737static struct object_entry **compute_write_order(void)
738{
28b8a730 739 uint32_t max_layers = 1;
f64ba53e
CC
740 unsigned int i, wo_end;
741
742 struct object_entry **wo;
743 struct object_entry *objects = to_pack.objects;
744
745 for (i = 0; i < to_pack.nr_objects; i++) {
746 objects[i].tagged = 0;
747 objects[i].filled = 0;
748 SET_DELTA_CHILD(&objects[i], NULL);
749 SET_DELTA_SIBLING(&objects[i], NULL);
750 }
751
752 /*
753 * Fully connect delta_child/delta_sibling network.
754 * Make sure delta_sibling is sorted in the original
755 * recency order.
756 */
757 for (i = to_pack.nr_objects; i > 0;) {
758 struct object_entry *e = &objects[--i];
759 if (!DELTA(e))
760 continue;
761 /* Mark me as the first child */
762 e->delta_sibling_idx = DELTA(e)->delta_child_idx;
763 SET_DELTA_CHILD(DELTA(e), e);
38d4debb
DM
764 }
765
f64ba53e
CC
766 /*
767 * Mark objects that are at the tip of tags.
768 */
769 for_each_tag_ref(mark_tagged, NULL);
770
28b8a730
JK
771 if (use_delta_islands)
772 max_layers = compute_pack_layers(&to_pack);
773
f64ba53e
CC
774 ALLOC_ARRAY(wo, to_pack.nr_objects);
775 wo_end = 0;
776
28b8a730
JK
777 for (; write_layer < max_layers; ++write_layer)
778 compute_layer_order(wo, &wo_end);
38d4debb 779
2834bc27 780 if (wo_end != to_pack.nr_objects)
f616db6a
NTND
781 die(_("ordered %u objects, expected %"PRIu32),
782 wo_end, to_pack.nr_objects);
1b4bb16b
JH
783
784 return wo;
785}
786
98a3beab 787static off_t write_reused_pack(struct hashfile *f)
6b8fda2d
VM
788{
789 unsigned char buffer[8192];
657673f1 790 off_t to_write, total;
6b8fda2d
VM
791 int fd;
792
793 if (!is_pack_valid(reuse_packfile))
f616db6a 794 die(_("packfile is invalid: %s"), reuse_packfile->pack_name);
6b8fda2d 795
a5436b57 796 fd = git_open(reuse_packfile->pack_name);
6b8fda2d 797 if (fd < 0)
f616db6a 798 die_errno(_("unable to open packfile for reuse: %s"),
6b8fda2d
VM
799 reuse_packfile->pack_name);
800
801 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
f616db6a 802 die_errno(_("unable to seek in reused packfile"));
6b8fda2d
VM
803
804 if (reuse_packfile_offset < 0)
41179100 805 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
6b8fda2d 806
657673f1 807 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
6b8fda2d
VM
808
809 while (to_write) {
810 int read_pack = xread(fd, buffer, sizeof(buffer));
811
812 if (read_pack <= 0)
f616db6a 813 die_errno(_("unable to read from reused packfile"));
6b8fda2d
VM
814
815 if (read_pack > to_write)
816 read_pack = to_write;
817
98a3beab 818 hashwrite(f, buffer, read_pack);
6b8fda2d 819 to_write -= read_pack;
657673f1
JK
820
821 /*
822 * We don't know the actual number of objects written,
823 * only how many bytes written, how many bytes total, and
824 * how many objects total. So we can fake it by pretending all
825 * objects we are writing are the same size. This gives us a
826 * smooth progress meter, and at the end it matches the true
827 * answer.
828 */
829 written = reuse_packfile_objects *
830 (((double)(total - to_write)) / total);
831 display_progress(progress_state, written);
6b8fda2d
VM
832 }
833
834 close(fd);
657673f1
JK
835 written = reuse_packfile_objects;
836 display_progress(progress_state, written);
6b8fda2d
VM
837 return reuse_packfile_offset - sizeof(struct pack_header);
838}
839
9cea46cd
EW
840static const char no_split_warning[] = N_(
841"disabling bitmap writing, packs are split due to pack.packSizeLimit"
842);
843
d01fb92f 844static void write_pack_file(void)
c323ac7d 845{
ebe27b13 846 uint32_t i = 0, j;
98a3beab 847 struct hashfile *f;
6ed7f25e 848 off_t offset;
ebe27b13 849 uint32_t nr_remaining = nr_result;
f746bae8 850 time_t last_mtime = 0;
1b4bb16b 851 struct object_entry **write_order;
81a216a5 852
bcd7954e 853 if (progress > pack_to_stdout)
754dbc43 854 progress_state = start_progress(_("Writing objects"), nr_result);
b32fa95f 855 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1b4bb16b 856 write_order = compute_write_order();
183bdb2c 857
ebe27b13 858 do {
188960b4 859 struct object_id oid;
7ba502c4 860 char *pack_tmp_name = NULL;
aa7e44bf 861
cdf9db3c 862 if (pack_to_stdout)
98a3beab 863 f = hashfd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
864 else
865 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 866
c0ad4657 867 offset = write_pack_header(f, nr_remaining);
6b8fda2d
VM
868
869 if (reuse_packfile) {
870 off_t packfile_size;
871 assert(pack_to_stdout);
872
873 packfile_size = write_reused_pack(f);
874 offset += packfile_size;
875 }
876
ebe27b13 877 nr_written = 0;
2834bc27 878 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 879 struct object_entry *e = write_order[i];
cddec4f8 880 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
881 break;
882 display_progress(progress_state, written);
883 }
c553ca25 884
ebe27b13
DH
885 /*
886 * Did we write the wrong # entries in the header?
887 * If so, rewrite it like in fast-import
888 */
54352bb2 889 if (pack_to_stdout) {
cfe83216 890 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
54352bb2 891 } else if (nr_written == nr_remaining) {
cfe83216 892 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
ebe27b13 893 } else {
f2af9f5e 894 int fd = finalize_hashfile(f, oid.hash, 0);
188960b4 895 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
896 nr_written, oid.hash, offset);
7ba502c4 897 close(fd);
9cea46cd 898 if (write_bitmap_index) {
25575015
JK
899 if (write_bitmap_index != WRITE_BITMAP_QUIET)
900 warning(_(no_split_warning));
9cea46cd
EW
901 write_bitmap_index = 0;
902 }
ebe27b13
DH
903 }
904
905 if (!pack_to_stdout) {
f746bae8 906 struct stat st;
58892711 907 struct strbuf tmpname = STRBUF_INIT;
d01fb92f 908
f746bae8
NP
909 /*
910 * Packs are runtime accessed in their mtime
911 * order since newer packs are more likely to contain
912 * younger objects. So if we are creating multiple
913 * packs then we should modify the mtime of later ones
914 * to preserve this property.
915 */
0e990530 916 if (stat(pack_tmp_name, &st) < 0) {
f616db6a 917 warning_errno(_("failed to stat %s"), pack_tmp_name);
f746bae8
NP
918 } else if (!last_mtime) {
919 last_mtime = st.st_mtime;
920 } else {
921 struct utimbuf utb;
922 utb.actime = st.st_atime;
923 utb.modtime = --last_mtime;
0e990530 924 if (utime(pack_tmp_name, &utb) < 0)
f616db6a 925 warning_errno(_("failed utime() on %s"), pack_tmp_name);
f746bae8
NP
926 }
927
58892711 928 strbuf_addf(&tmpname, "%s-", base_name);
7cc8f971
VM
929
930 if (write_bitmap_index) {
188960b4 931 bitmap_writer_set_checksum(oid.hash);
06af3bba
NTND
932 bitmap_writer_build_type_index(
933 &to_pack, written_list, nr_written);
7cc8f971
VM
934 }
935
58892711 936 finish_tmp_packfile(&tmpname, pack_tmp_name,
0e990530 937 written_list, nr_written,
188960b4 938 &pack_idx_opts, oid.hash);
7cc8f971
VM
939
940 if (write_bitmap_index) {
188960b4 941 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
7cc8f971
VM
942
943 stop_progress(&progress_state);
944
945 bitmap_writer_show_progress(progress);
946 bitmap_writer_reuse_bitmaps(&to_pack);
947 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
948 bitmap_writer_build(&to_pack);
ae4f07fb 949 bitmap_writer_finish(written_list, nr_written,
58892711 950 tmpname.buf, write_bitmap_options);
7cc8f971
VM
951 write_bitmap_index = 0;
952 }
953
58892711 954 strbuf_release(&tmpname);
7ba502c4 955 free(pack_tmp_name);
188960b4 956 puts(oid_to_hex(&oid));
ebe27b13
DH
957 }
958
959 /* mark written objects as written to previous pack */
960 for (j = 0; j < nr_written; j++) {
79814f42 961 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
962 }
963 nr_remaining -= nr_written;
2834bc27 964 } while (nr_remaining && i < to_pack.nr_objects);
ebe27b13
DH
965
966 free(written_list);
1b4bb16b 967 free(write_order);
4d4fcc54 968 stop_progress(&progress_state);
67c08ce1 969 if (written != nr_result)
f616db6a
NTND
970 die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
971 written, nr_result);
9ed87902
JT
972 trace2_data_intmax("pack-objects", the_repository,
973 "write_pack_file/wrote", nr_result);
c323ac7d
LT
974}
975
a74db82e
JH
976static int no_try_delta(const char *path)
977{
2aef63d3 978 static struct attr_check *check;
a74db82e 979
2aef63d3
JH
980 if (!check)
981 check = attr_check_initl("delta", NULL);
f8adbec9 982 git_check_attr(the_repository->index, path, check);
2aef63d3 983 if (ATTR_FALSE(check->items[0].value))
a74db82e
JH
984 return 1;
985 return 0;
986}
987
ce2bc424
JK
988/*
989 * When adding an object, check whether we have already added it
990 * to our packing list. If so, we can skip. However, if we are
991 * being asked to excludei t, but the previous mention was to include
992 * it, make sure to adjust its flags and tweak our numbers accordingly.
993 *
994 * As an optimization, we pass out the index position where we would have
995 * found the item, since that saves us from having to look it up again a
996 * few lines later when we want to add the new entry.
997 */
188960b4 998static int have_duplicate_entry(const struct object_id *oid,
3a37876b 999 int exclude)
c323ac7d 1000{
c323ac7d 1001 struct object_entry *entry;
29b734e4 1002
3a37876b 1003 entry = packlist_find(&to_pack, oid);
ce2bc424 1004 if (!entry)
29b734e4 1005 return 0;
ce2bc424
JK
1006
1007 if (exclude) {
1008 if (!entry->preferred_base)
1009 nr_result--;
1010 entry->preferred_base = 1;
29b734e4 1011 }
c323ac7d 1012
ce2bc424
JK
1013 return 1;
1014}
1015
702d1b95
KS
1016static int want_found_object(int exclude, struct packed_git *p)
1017{
1018 if (exclude)
1019 return 1;
1020 if (incremental)
1021 return 0;
1022
1023 /*
1024 * When asked to do --local (do not include an object that appears in a
1025 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1026 * an object that appears in a pack marked with .keep), finding a pack
1027 * that matches the criteria is sufficient for us to decide to omit it.
1028 * However, even if this pack does not satisfy the criteria, we need to
1029 * make sure no copy of this object appears in _any_ pack that makes us
1030 * to omit the object, so we need to check all the packs.
1031 *
1032 * We can however first check whether these options can possible matter;
1033 * if they do not matter we know we want the object in generated pack.
1034 * Otherwise, we signal "-1" at the end to tell the caller that we do
1035 * not know either way, and it needs to check more packs.
1036 */
ed7e5fc3
NTND
1037 if (!ignore_packed_keep_on_disk &&
1038 !ignore_packed_keep_in_core &&
702d1b95
KS
1039 (!local || !have_non_local_packs))
1040 return 1;
1041
1042 if (local && !p->pack_local)
1043 return 0;
ed7e5fc3
NTND
1044 if (p->pack_local &&
1045 ((ignore_packed_keep_on_disk && p->pack_keep) ||
1046 (ignore_packed_keep_in_core && p->pack_keep_in_core)))
702d1b95
KS
1047 return 0;
1048
1049 /* we don't know yet; keep looking for more packs */
1050 return -1;
1051}
1052
ce2bc424
JK
1053/*
1054 * Check whether we want the object in the pack (e.g., we do not want
1055 * objects found in non-local stores if the "--local" option was used).
1056 *
702d1b95
KS
1057 * If the caller already knows an existing pack it wants to take the object
1058 * from, that is passed in *found_pack and *found_offset; otherwise this
1059 * function finds if there is any pack that has the object and returns the pack
1060 * and its offset in these variables.
ce2bc424 1061 */
188960b4 1062static int want_object_in_pack(const struct object_id *oid,
ce2bc424
JK
1063 int exclude,
1064 struct packed_git **found_pack,
1065 off_t *found_offset)
1066{
702d1b95 1067 int want;
8865859d 1068 struct list_head *pos;
6a22d521 1069 struct multi_pack_index *m;
ce2bc424 1070
6862ebbf 1071 if (!exclude && local && has_loose_object_nonlocal(oid))
daae0625
BC
1072 return 0;
1073
702d1b95
KS
1074 /*
1075 * If we already know the pack object lives in, start checks from that
1076 * pack - in the usual case when neither --local was given nor .keep files
1077 * are present we will determine the answer right now.
1078 */
1079 if (*found_pack) {
1080 want = want_found_object(exclude, *found_pack);
1081 if (want != -1)
1082 return want;
1083 }
6a22d521
DS
1084
1085 for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1086 struct pack_entry e;
64404a24 1087 if (fill_midx_entry(the_repository, oid, &e, m)) {
6a22d521
DS
1088 struct packed_git *p = e.p;
1089 off_t offset;
1090
1091 if (p == *found_pack)
1092 offset = *found_offset;
1093 else
1094 offset = find_pack_entry_one(oid->hash, p);
1095
1096 if (offset) {
1097 if (!*found_pack) {
1098 if (!is_pack_valid(p))
1099 continue;
1100 *found_offset = offset;
1101 *found_pack = p;
1102 }
1103 want = want_found_object(exclude, p);
1104 if (want != -1)
1105 return want;
1106 }
1107 }
1108 }
1109
a80d72db 1110 list_for_each(pos, get_packed_git_mru(the_repository)) {
ec2dd32c 1111 struct packed_git *p = list_entry(pos, struct packed_git, mru);
702d1b95
KS
1112 off_t offset;
1113
1114 if (p == *found_pack)
1115 offset = *found_offset;
1116 else
188960b4 1117 offset = find_pack_entry_one(oid->hash, p);
702d1b95 1118
5c49c116 1119 if (offset) {
ce2bc424 1120 if (!*found_pack) {
319b678a 1121 if (!is_pack_valid(p))
4c080182 1122 continue;
ce2bc424
JK
1123 *found_offset = offset;
1124 *found_pack = p;
64560374 1125 }
702d1b95 1126 want = want_found_object(exclude, p);
e6e24c94 1127 if (!exclude && want > 0)
a80d72db
SB
1128 list_move(&p->mru,
1129 get_packed_git_mru(the_repository));
702d1b95
KS
1130 if (want != -1)
1131 return want;
64560374
LT
1132 }
1133 }
eb019375 1134
ce2bc424
JK
1135 return 1;
1136}
1137
188960b4 1138static void create_object_entry(const struct object_id *oid,
ce2bc424
JK
1139 enum object_type type,
1140 uint32_t hash,
1141 int exclude,
1142 int no_try_delta,
ce2bc424
JK
1143 struct packed_git *found_pack,
1144 off_t found_offset)
1145{
1146 struct object_entry *entry;
29b734e4 1147
3a37876b 1148 entry = packlist_alloc(&to_pack, oid);
27225f2e 1149 entry->hash = hash;
fd9b1bae 1150 oe_set_type(entry, type);
29b734e4
NP
1151 if (exclude)
1152 entry->preferred_base = 1;
81a216a5
NP
1153 else
1154 nr_result++;
29b734e4 1155 if (found_pack) {
43fa44fa 1156 oe_set_in_pack(&to_pack, entry, found_pack);
29b734e4
NP
1157 entry->in_pack_offset = found_offset;
1158 }
7a979d99 1159
ce2bc424
JK
1160 entry->no_try_delta = no_try_delta;
1161}
29b734e4 1162
373c67da
JK
1163static const char no_closure_warning[] = N_(
1164"disabling bitmap writing, as some objects are not being packed"
1165);
1166
188960b4 1167static int add_object_entry(const struct object_id *oid, enum object_type type,
ce2bc424
JK
1168 const char *name, int exclude)
1169{
702d1b95
KS
1170 struct packed_git *found_pack = NULL;
1171 off_t found_offset = 0;
7a979d99 1172
5af05043
NTND
1173 display_progress(progress_state, ++nr_seen);
1174
3a37876b 1175 if (have_duplicate_entry(oid, exclude))
ce2bc424 1176 return 0;
a74db82e 1177
188960b4 1178 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
373c67da
JK
1179 /* The pack is missing an object, so it will not have closure */
1180 if (write_bitmap_index) {
25575015
JK
1181 if (write_bitmap_index != WRITE_BITMAP_QUIET)
1182 warning(_(no_closure_warning));
373c67da
JK
1183 write_bitmap_index = 0;
1184 }
ce2bc424 1185 return 0;
373c67da 1186 }
29b734e4 1187
188960b4 1188 create_object_entry(oid, type, pack_name_hash(name),
ce2bc424 1189 exclude, name && no_try_delta(name),
3a37876b 1190 found_pack, found_offset);
29b734e4 1191 return 1;
c323ac7d
LT
1192}
1193
20664967 1194static int add_object_entry_from_bitmap(const struct object_id *oid,
6b8fda2d
VM
1195 enum object_type type,
1196 int flags, uint32_t name_hash,
1197 struct packed_git *pack, off_t offset)
1198{
5af05043
NTND
1199 display_progress(progress_state, ++nr_seen);
1200
3a37876b 1201 if (have_duplicate_entry(oid, 0))
6b8fda2d
VM
1202 return 0;
1203
188960b4 1204 if (!want_object_in_pack(oid, 0, &pack, &offset))
702d1b95
KS
1205 return 0;
1206
3a37876b 1207 create_object_entry(oid, type, name_hash, 0, 0, pack, offset);
29b734e4 1208 return 1;
c323ac7d
LT
1209}
1210
5379a5c5 1211struct pbase_tree_cache {
188960b4 1212 struct object_id oid;
5379a5c5
JH
1213 int ref;
1214 int temporary;
1215 void *tree_data;
1216 unsigned long tree_size;
1217};
1218
1219static struct pbase_tree_cache *(pbase_tree_cache[256]);
188960b4 1220static int pbase_tree_cache_ix(const struct object_id *oid)
5379a5c5 1221{
188960b4 1222 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
5379a5c5
JH
1223}
1224static int pbase_tree_cache_ix_incr(int ix)
1225{
1226 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1227}
1228
1229static struct pbase_tree {
1230 struct pbase_tree *next;
1231 /* This is a phony "cache" entry; we are not
01689909 1232 * going to evict it or find it through _get()
5379a5c5
JH
1233 * mechanism -- this is for the toplevel node that
1234 * would almost always change with any commit.
1235 */
1236 struct pbase_tree_cache pcache;
1237} *pbase_tree;
1238
188960b4 1239static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
5379a5c5
JH
1240{
1241 struct pbase_tree_cache *ent, *nent;
1242 void *data;
1243 unsigned long size;
21666f1a 1244 enum object_type type;
5379a5c5 1245 int neigh;
188960b4 1246 int my_ix = pbase_tree_cache_ix(oid);
5379a5c5
JH
1247 int available_ix = -1;
1248
1249 /* pbase-tree-cache acts as a limited hashtable.
1250 * your object will be found at your index or within a few
1251 * slots after that slot if it is cached.
1252 */
1253 for (neigh = 0; neigh < 8; neigh++) {
1254 ent = pbase_tree_cache[my_ix];
4a7e27e9 1255 if (ent && oideq(&ent->oid, oid)) {
5379a5c5
JH
1256 ent->ref++;
1257 return ent;
1258 }
1259 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1260 ((0 <= available_ix) &&
1261 (!ent && pbase_tree_cache[available_ix])))
1262 available_ix = my_ix;
1263 if (!ent)
1264 break;
1265 my_ix = pbase_tree_cache_ix_incr(my_ix);
1266 }
1267
1268 /* Did not find one. Either we got a bogus request or
1269 * we need to read and perhaps cache.
1270 */
b4f5aca4 1271 data = read_object_file(oid, &type, &size);
5379a5c5
JH
1272 if (!data)
1273 return NULL;
21666f1a 1274 if (type != OBJ_TREE) {
5379a5c5
JH
1275 free(data);
1276 return NULL;
1277 }
1278
1279 /* We need to either cache or return a throwaway copy */
1280
1281 if (available_ix < 0)
1282 ent = NULL;
1283 else {
1284 ent = pbase_tree_cache[available_ix];
1285 my_ix = available_ix;
1286 }
1287
1288 if (!ent) {
1289 nent = xmalloc(sizeof(*nent));
1290 nent->temporary = (available_ix < 0);
1291 }
1292 else {
1293 /* evict and reuse */
1294 free(ent->tree_data);
1295 nent = ent;
1296 }
188960b4 1297 oidcpy(&nent->oid, oid);
5379a5c5
JH
1298 nent->tree_data = data;
1299 nent->tree_size = size;
1300 nent->ref = 1;
1301 if (!nent->temporary)
1302 pbase_tree_cache[my_ix] = nent;
1303 return nent;
1304}
1305
1306static void pbase_tree_put(struct pbase_tree_cache *cache)
1307{
1308 if (!cache->temporary) {
1309 cache->ref--;
1310 return;
1311 }
1312 free(cache->tree_data);
1313 free(cache);
1314}
1315
1316static int name_cmp_len(const char *name)
1317{
1318 int i;
1319 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1320 ;
1321 return i;
1322}
1323
1324static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1325 const char *name,
ce0bd642
LT
1326 int cmplen,
1327 const char *fullname)
3f9ac8d2 1328{
4c068a98 1329 struct name_entry entry;
8a5a8d6c 1330 int cmp;
4c068a98
LT
1331
1332 while (tree_entry(tree,&entry)) {
1211be6b
LT
1333 if (S_ISGITLINK(entry.mode))
1334 continue;
0de16337 1335 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1336 memcmp(name, entry.path, cmplen);
1337 if (cmp > 0)
7a979d99 1338 continue;
8a5a8d6c
NP
1339 if (cmp < 0)
1340 return;
5379a5c5 1341 if (name[cmplen] != '/') {
ea82b2a0 1342 add_object_entry(&entry.oid,
4d1012c3 1343 object_type(entry.mode),
bc32fed5 1344 fullname, 1);
5379a5c5
JH
1345 return;
1346 }
8a5a8d6c 1347 if (S_ISDIR(entry.mode)) {
7a979d99 1348 struct tree_desc sub;
5379a5c5
JH
1349 struct pbase_tree_cache *tree;
1350 const char *down = name+cmplen+1;
1351 int downlen = name_cmp_len(down);
1352
ea82b2a0 1353 tree = pbase_tree_get(&entry.oid);
5379a5c5
JH
1354 if (!tree)
1355 return;
6fda5e51 1356 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1357
ce0bd642 1358 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1359 pbase_tree_put(tree);
1360 }
1361 }
1362}
1d6b38cc 1363
5379a5c5
JH
1364static unsigned *done_pbase_paths;
1365static int done_pbase_paths_num;
1366static int done_pbase_paths_alloc;
1367static int done_pbase_path_pos(unsigned hash)
1368{
1369 int lo = 0;
1370 int hi = done_pbase_paths_num;
1371 while (lo < hi) {
19716b21 1372 int mi = lo + (hi - lo) / 2;
5379a5c5
JH
1373 if (done_pbase_paths[mi] == hash)
1374 return mi;
1375 if (done_pbase_paths[mi] < hash)
1376 hi = mi;
1377 else
1378 lo = mi + 1;
1379 }
1380 return -lo-1;
1381}
1382
1383static int check_pbase_path(unsigned hash)
1384{
c7b07805 1385 int pos = done_pbase_path_pos(hash);
5379a5c5
JH
1386 if (0 <= pos)
1387 return 1;
1388 pos = -pos - 1;
25e19407
DD
1389 ALLOC_GROW(done_pbase_paths,
1390 done_pbase_paths_num + 1,
1391 done_pbase_paths_alloc);
5379a5c5
JH
1392 done_pbase_paths_num++;
1393 if (pos < done_pbase_paths_num)
f331ab9d
RS
1394 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1395 done_pbase_paths_num - pos - 1);
5379a5c5
JH
1396 done_pbase_paths[pos] = hash;
1397 return 0;
1398}
1399
bc32fed5 1400static void add_preferred_base_object(const char *name)
5379a5c5
JH
1401{
1402 struct pbase_tree *it;
8a5a8d6c 1403 int cmplen;
68fb36eb 1404 unsigned hash = pack_name_hash(name);
5379a5c5 1405
8a5a8d6c 1406 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1407 return;
1408
8a5a8d6c 1409 cmplen = name_cmp_len(name);
5379a5c5
JH
1410 for (it = pbase_tree; it; it = it->next) {
1411 if (cmplen == 0) {
188960b4 1412 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
5379a5c5
JH
1413 }
1414 else {
1415 struct tree_desc tree;
6fda5e51 1416 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1417 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1418 }
3f9ac8d2 1419 }
3f9ac8d2
JH
1420}
1421
188960b4 1422static void add_preferred_base(struct object_id *oid)
3f9ac8d2 1423{
5379a5c5
JH
1424 struct pbase_tree *it;
1425 void *data;
1426 unsigned long size;
188960b4 1427 struct object_id tree_oid;
1d6b38cc 1428
8d1d8f83
JH
1429 if (window <= num_preferred_base++)
1430 return;
1431
d3b4705a
NTND
1432 data = read_object_with_reference(the_repository, oid,
1433 tree_type, &size, &tree_oid);
5379a5c5 1434 if (!data)
7a979d99 1435 return;
5379a5c5
JH
1436
1437 for (it = pbase_tree; it; it = it->next) {
4a7e27e9 1438 if (oideq(&it->pcache.oid, &tree_oid)) {
5379a5c5
JH
1439 free(data);
1440 return;
1441 }
1442 }
1443
1444 it = xcalloc(1, sizeof(*it));
1445 it->next = pbase_tree;
1446 pbase_tree = it;
1447
188960b4 1448 oidcpy(&it->pcache.oid, &tree_oid);
5379a5c5
JH
1449 it->pcache.tree_data = data;
1450 it->pcache.tree_size = size;
3f9ac8d2
JH
1451}
1452
0ef95f72
NP
1453static void cleanup_preferred_base(void)
1454{
1455 struct pbase_tree *it;
1456 unsigned i;
1457
1458 it = pbase_tree;
1459 pbase_tree = NULL;
1460 while (it) {
095b3b2c
BW
1461 struct pbase_tree *tmp = it;
1462 it = tmp->next;
1463 free(tmp->pcache.tree_data);
1464 free(tmp);
0ef95f72
NP
1465 }
1466
1467 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1468 if (!pbase_tree_cache[i])
1469 continue;
1470 free(pbase_tree_cache[i]->tree_data);
6a83d902 1471 FREE_AND_NULL(pbase_tree_cache[i]);
0ef95f72
NP
1472 }
1473
6a83d902 1474 FREE_AND_NULL(done_pbase_paths);
0ef95f72
NP
1475 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1476}
1477
2fa233a5
JK
1478/*
1479 * Return 1 iff the object specified by "delta" can be sent
1480 * literally as a delta against the base in "base_sha1". If
1481 * so, then *base_out will point to the entry in our packing
1482 * list, or NULL if we must use the external-base list.
1483 *
1484 * Depth value does not matter - find_deltas() will
1485 * never consider reused delta as the base object to
1486 * deltify other objects against, in order to avoid
1487 * circular deltas.
1488 */
1489static int can_reuse_delta(const unsigned char *base_sha1,
1490 struct object_entry *delta,
1491 struct object_entry **base_out)
1492{
1493 struct object_entry *base;
3c771448 1494 struct object_id base_oid;
2fa233a5
JK
1495
1496 if (!base_sha1)
1497 return 0;
1498
3df28cae
JK
1499 oidread(&base_oid, base_sha1);
1500
2fa233a5
JK
1501 /*
1502 * First see if we're already sending the base (or it's explicitly in
1503 * our "excluded" list).
1504 */
3a37876b 1505 base = packlist_find(&to_pack, &base_oid);
2fa233a5
JK
1506 if (base) {
1507 if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1508 return 0;
1509 *base_out = base;
1510 return 1;
1511 }
1512
1513 /*
1514 * Otherwise, reachability bitmaps may tell us if the receiver has it,
1515 * even if it was buried too deep in history to make it into the
1516 * packing list.
1517 */
3c771448 1518 if (thin && bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {
2fa233a5 1519 if (use_delta_islands) {
2fa233a5
JK
1520 if (!in_same_island(&delta->idx.oid, &base_oid))
1521 return 0;
1522 }
1523 *base_out = NULL;
1524 return 1;
1525 }
1526
1527 return 0;
1528}
1529
c323ac7d
LT
1530static void check_object(struct object_entry *entry)
1531{
ac77d0c3
NTND
1532 unsigned long canonical_size;
1533
43fa44fa
NTND
1534 if (IN_PACK(entry)) {
1535 struct packed_git *p = IN_PACK(entry);
03e79c88 1536 struct pack_window *w_curs = NULL;
5c49c116
NP
1537 const unsigned char *base_ref = NULL;
1538 struct object_entry *base_entry;
1539 unsigned long used, used_0;
ef49a7a0 1540 unsigned long avail;
5c49c116
NP
1541 off_t ofs;
1542 unsigned char *buf, c;
fd9b1bae 1543 enum object_type type;
27a7d067 1544 unsigned long in_pack_size;
780e6e73 1545
6777a59f 1546 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1547
5c49c116 1548 /*
fa736f72
NP
1549 * We want in_pack_type even if we do not reuse delta
1550 * since non-delta representations could still be reused.
ab7cd7bb 1551 */
09ded04b 1552 used = unpack_object_header_buffer(buf, avail,
fd9b1bae 1553 &type,
27a7d067 1554 &in_pack_size);
03d66015
NP
1555 if (used == 0)
1556 goto give_up;
ab7cd7bb 1557
fd9b1bae
NTND
1558 if (type < 0)
1559 BUG("invalid type %d", type);
1560 entry->in_pack_type = type;
1561
5c49c116
NP
1562 /*
1563 * Determine if this is a delta and if so whether we can
1564 * reuse it or not. Otherwise let's find out as cheaply as
1565 * possible what the actual type and size for this object is.
3f9ac8d2 1566 */
5c49c116
NP
1567 switch (entry->in_pack_type) {
1568 default:
1569 /* Not a delta hence we've already got all we need. */
fd9b1bae 1570 oe_set_type(entry, entry->in_pack_type);
ac77d0c3 1571 SET_SIZE(entry, in_pack_size);
5c49c116 1572 entry->in_pack_header_size = used;
fd9b1bae 1573 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
03d66015 1574 goto give_up;
5c49c116
NP
1575 unuse_pack(&w_curs);
1576 return;
1577 case OBJ_REF_DELTA:
a7de7130 1578 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1579 base_ref = use_pack(p, &w_curs,
1580 entry->in_pack_offset + used, NULL);
41179100 1581 entry->in_pack_header_size = used + the_hash_algo->rawsz;
5c49c116
NP
1582 break;
1583 case OBJ_OFS_DELTA:
1584 buf = use_pack(p, &w_curs,
1585 entry->in_pack_offset + used, NULL);
1586 used_0 = 0;
1587 c = buf[used_0++];
1588 ofs = c & 127;
1589 while (c & 128) {
1590 ofs += 1;
03d66015 1591 if (!ofs || MSB(ofs, 7)) {
f616db6a 1592 error(_("delta base offset overflow in pack for %s"),
e6a492b7 1593 oid_to_hex(&entry->idx.oid));
03d66015
NP
1594 goto give_up;
1595 }
5c49c116
NP
1596 c = buf[used_0++];
1597 ofs = (ofs << 7) + (c & 127);
780e6e73 1598 }
5c49c116 1599 ofs = entry->in_pack_offset - ofs;
03d66015 1600 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
f616db6a 1601 error(_("delta base offset out of bound for %s"),
e6a492b7 1602 oid_to_hex(&entry->idx.oid));
03d66015
NP
1603 goto give_up;
1604 }
a7de7130 1605 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1606 struct revindex_entry *revidx;
1607 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1608 if (!revidx)
1609 goto give_up;
3449f8c4
NP
1610 base_ref = nth_packed_object_sha1(p, revidx->nr);
1611 }
5c49c116
NP
1612 entry->in_pack_header_size = used + used_0;
1613 break;
780e6e73 1614 }
780e6e73 1615
2fa233a5 1616 if (can_reuse_delta(base_ref, entry, &base_entry)) {
fd9b1bae 1617 oe_set_type(entry, entry->in_pack_type);
ac77d0c3 1618 SET_SIZE(entry, in_pack_size); /* delta size */
0aca34e8 1619 SET_DELTA_SIZE(entry, in_pack_size);
6a1e32d5
JK
1620
1621 if (base_entry) {
1622 SET_DELTA(entry, base_entry);
1623 entry->delta_sibling_idx = base_entry->delta_child_idx;
1624 SET_DELTA_CHILD(base_entry, entry);
1625 } else {
1626 SET_DELTA_EXT(entry, base_ref);
1627 }
1628
5c49c116
NP
1629 unuse_pack(&w_curs);
1630 return;
1631 }
ab7cd7bb 1632
fd9b1bae 1633 if (oe_type(entry)) {
27a7d067
NTND
1634 off_t delta_pos;
1635
5c49c116
NP
1636 /*
1637 * This must be a delta and we already know what the
1638 * final object type is. Let's extract the actual
1639 * object size from the delta header.
1640 */
27a7d067 1641 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
ac77d0c3
NTND
1642 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1643 if (canonical_size == 0)
03d66015 1644 goto give_up;
ac77d0c3 1645 SET_SIZE(entry, canonical_size);
5c49c116 1646 unuse_pack(&w_curs);
3f9ac8d2
JH
1647 return;
1648 }
5c49c116
NP
1649
1650 /*
1651 * No choice but to fall back to the recursive delta walk
c93206b4 1652 * with oid_object_info() to find about the object type
5c49c116
NP
1653 * at this point...
1654 */
03d66015 1655 give_up:
5c49c116 1656 unuse_pack(&w_curs);
36e4d74a 1657 }
3f9ac8d2 1658
ad635e82
JH
1659 oe_set_type(entry,
1660 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
ac77d0c3
NTND
1661 if (entry->type_valid) {
1662 SET_SIZE(entry, canonical_size);
1663 } else {
1664 /*
1665 * Bad object type is checked in prepare_pack(). This is
1666 * to permit a missing preferred base object to be ignored
1667 * as a preferred base. Doing so can result in a larger
1668 * pack file, but the transfer will still take place.
1669 */
1670 }
3f9ac8d2
JH
1671}
1672
5c49c116
NP
1673static int pack_offset_sort(const void *_a, const void *_b)
1674{
1675 const struct object_entry *a = *(struct object_entry **)_a;
1676 const struct object_entry *b = *(struct object_entry **)_b;
43fa44fa
NTND
1677 const struct packed_git *a_in_pack = IN_PACK(a);
1678 const struct packed_git *b_in_pack = IN_PACK(b);
5c49c116
NP
1679
1680 /* avoid filesystem trashing with loose objects */
43fa44fa 1681 if (!a_in_pack && !b_in_pack)
e6a492b7 1682 return oidcmp(&a->idx.oid, &b->idx.oid);
5c49c116 1683
43fa44fa 1684 if (a_in_pack < b_in_pack)
5c49c116 1685 return -1;
43fa44fa 1686 if (a_in_pack > b_in_pack)
5c49c116
NP
1687 return 1;
1688 return a->in_pack_offset < b->in_pack_offset ? -1 :
1689 (a->in_pack_offset > b->in_pack_offset);
1690}
1691
4cf2143e
JK
1692/*
1693 * Drop an on-disk delta we were planning to reuse. Naively, this would
1694 * just involve blanking out the "delta" field, but we have to deal
1695 * with some extra book-keeping:
1696 *
1697 * 1. Removing ourselves from the delta_sibling linked list.
1698 *
1699 * 2. Updating our size/type to the non-delta representation. These were
1700 * either not recorded initially (size) or overwritten with the delta type
1701 * (type) when check_object() decided to reuse the delta.
7dbabbbe
JK
1702 *
1703 * 3. Resetting our delta depth, as we are now a base object.
4cf2143e
JK
1704 */
1705static void drop_reused_delta(struct object_entry *entry)
1706{
898eba5e 1707 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
4cf2143e 1708 struct object_info oi = OBJECT_INFO_INIT;
fd9b1bae 1709 enum object_type type;
ac77d0c3 1710 unsigned long size;
4cf2143e 1711
898eba5e
NTND
1712 while (*idx) {
1713 struct object_entry *oe = &to_pack.objects[*idx - 1];
4cf2143e 1714
898eba5e
NTND
1715 if (oe == entry)
1716 *idx = oe->delta_sibling_idx;
4cf2143e 1717 else
898eba5e 1718 idx = &oe->delta_sibling_idx;
4cf2143e 1719 }
898eba5e 1720 SET_DELTA(entry, NULL);
7dbabbbe 1721 entry->depth = 0;
4cf2143e 1722
ac77d0c3 1723 oi.sizep = &size;
fd9b1bae 1724 oi.typep = &type;
ad635e82 1725 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
4cf2143e
JK
1726 /*
1727 * We failed to get the info from this pack for some reason;
c93206b4 1728 * fall back to oid_object_info, which may find another copy.
fd9b1bae 1729 * And if that fails, the error will be recorded in oe_type(entry)
4cf2143e
JK
1730 * and dealt with in prepare_pack().
1731 */
ad635e82
JH
1732 oe_set_type(entry,
1733 oid_object_info(the_repository, &entry->idx.oid, &size));
fd9b1bae
NTND
1734 } else {
1735 oe_set_type(entry, type);
4cf2143e 1736 }
ac77d0c3 1737 SET_SIZE(entry, size);
4cf2143e
JK
1738}
1739
1740/*
1741 * Follow the chain of deltas from this entry onward, throwing away any links
1742 * that cause us to hit a cycle (as determined by the DFS state flags in
1743 * the entries).
7dbabbbe
JK
1744 *
1745 * We also detect too-long reused chains that would violate our --depth
1746 * limit.
4cf2143e
JK
1747 */
1748static void break_delta_chains(struct object_entry *entry)
1749{
42b766d7
JK
1750 /*
1751 * The actual depth of each object we will write is stored as an int,
1752 * as it cannot exceed our int "depth" limit. But before we break
1753 * changes based no that limit, we may potentially go as deep as the
1754 * number of objects, which is elsewhere bounded to a uint32_t.
1755 */
1756 uint32_t total_depth;
1757 struct object_entry *cur, *next;
1758
1759 for (cur = entry, total_depth = 0;
1760 cur;
898eba5e 1761 cur = DELTA(cur), total_depth++) {
42b766d7
JK
1762 if (cur->dfs_state == DFS_DONE) {
1763 /*
1764 * We've already seen this object and know it isn't
1765 * part of a cycle. We do need to append its depth
1766 * to our count.
1767 */
1768 total_depth += cur->depth;
1769 break;
1770 }
4cf2143e 1771
4cf2143e 1772 /*
42b766d7
JK
1773 * We break cycles before looping, so an ACTIVE state (or any
1774 * other cruft which made its way into the state variable)
1775 * is a bug.
4cf2143e 1776 */
42b766d7 1777 if (cur->dfs_state != DFS_NONE)
033abf97 1778 BUG("confusing delta dfs state in first pass: %d",
42b766d7 1779 cur->dfs_state);
4cf2143e 1780
4cf2143e 1781 /*
42b766d7
JK
1782 * Now we know this is the first time we've seen the object. If
1783 * it's not a delta, we're done traversing, but we'll mark it
1784 * done to save time on future traversals.
4cf2143e 1785 */
898eba5e 1786 if (!DELTA(cur)) {
42b766d7
JK
1787 cur->dfs_state = DFS_DONE;
1788 break;
1789 }
4cf2143e 1790
4cf2143e 1791 /*
42b766d7
JK
1792 * Mark ourselves as active and see if the next step causes
1793 * us to cycle to another active object. It's important to do
1794 * this _before_ we loop, because it impacts where we make the
1795 * cut, and thus how our total_depth counter works.
1796 * E.g., We may see a partial loop like:
1797 *
1798 * A -> B -> C -> D -> B
1799 *
1800 * Cutting B->C breaks the cycle. But now the depth of A is
1801 * only 1, and our total_depth counter is at 3. The size of the
1802 * error is always one less than the size of the cycle we
1803 * broke. Commits C and D were "lost" from A's chain.
1804 *
1805 * If we instead cut D->B, then the depth of A is correct at 3.
1806 * We keep all commits in the chain that we examined.
4cf2143e 1807 */
42b766d7 1808 cur->dfs_state = DFS_ACTIVE;
898eba5e 1809 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
42b766d7
JK
1810 drop_reused_delta(cur);
1811 cur->dfs_state = DFS_DONE;
1812 break;
7dbabbbe 1813 }
42b766d7 1814 }
7dbabbbe 1815
42b766d7
JK
1816 /*
1817 * And now that we've gone all the way to the bottom of the chain, we
1818 * need to clear the active flags and set the depth fields as
1819 * appropriate. Unlike the loop above, which can quit when it drops a
1820 * delta, we need to keep going to look for more depth cuts. So we need
1821 * an extra "next" pointer to keep going after we reset cur->delta.
1822 */
1823 for (cur = entry; cur; cur = next) {
898eba5e 1824 next = DELTA(cur);
4cf2143e 1825
42b766d7
JK
1826 /*
1827 * We should have a chain of zero or more ACTIVE states down to
1828 * a final DONE. We can quit after the DONE, because either it
1829 * has no bases, or we've already handled them in a previous
1830 * call.
1831 */
1832 if (cur->dfs_state == DFS_DONE)
1833 break;
1834 else if (cur->dfs_state != DFS_ACTIVE)
033abf97 1835 BUG("confusing delta dfs state in second pass: %d",
42b766d7 1836 cur->dfs_state);
4cf2143e 1837
4cf2143e 1838 /*
42b766d7
JK
1839 * If the total_depth is more than depth, then we need to snip
1840 * the chain into two or more smaller chains that don't exceed
1841 * the maximum depth. Most of the resulting chains will contain
1842 * (depth + 1) entries (i.e., depth deltas plus one base), and
1843 * the last chain (i.e., the one containing entry) will contain
1844 * whatever entries are left over, namely
1845 * (total_depth % (depth + 1)) of them.
1846 *
1847 * Since we are iterating towards decreasing depth, we need to
1848 * decrement total_depth as we go, and we need to write to the
1849 * entry what its final depth will be after all of the
1850 * snipping. Since we're snipping into chains of length (depth
1851 * + 1) entries, the final depth of an entry will be its
1852 * original depth modulo (depth + 1). Any time we encounter an
1853 * entry whose final depth is supposed to be zero, we snip it
1854 * from its delta base, thereby making it so.
4cf2143e 1855 */
42b766d7
JK
1856 cur->depth = (total_depth--) % (depth + 1);
1857 if (!cur->depth)
1858 drop_reused_delta(cur);
1859
1860 cur->dfs_state = DFS_DONE;
4cf2143e
JK
1861 }
1862}
1863
c323ac7d
LT
1864static void get_object_details(void)
1865{
7cadf491 1866 uint32_t i;
5c49c116
NP
1867 struct object_entry **sorted_by_offset;
1868
5af05043
NTND
1869 if (progress)
1870 progress_state = start_progress(_("Counting objects"),
1871 to_pack.nr_objects);
1872
2834bc27
VM
1873 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1874 for (i = 0; i < to_pack.nr_objects; i++)
1875 sorted_by_offset[i] = to_pack.objects + i;
9ed0d8d6 1876 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
c323ac7d 1877
2834bc27 1878 for (i = 0; i < to_pack.nr_objects; i++) {
15366280
JH
1879 struct object_entry *entry = sorted_by_offset[i];
1880 check_object(entry);
ac77d0c3
NTND
1881 if (entry->type_valid &&
1882 oe_size_greater_than(&to_pack, entry, big_file_threshold))
15366280 1883 entry->no_try_delta = 1;
5af05043 1884 display_progress(progress_state, i + 1);
15366280 1885 }
5af05043 1886 stop_progress(&progress_state);
3449f8c4 1887
4cf2143e
JK
1888 /*
1889 * This must happen in a second pass, since we rely on the delta
1890 * information for the whole list being completed.
1891 */
1892 for (i = 0; i < to_pack.nr_objects; i++)
1893 break_delta_chains(&to_pack.objects[i]);
1894
5c49c116 1895 free(sorted_by_offset);
c323ac7d
LT
1896}
1897
b904166c
NP
1898/*
1899 * We search for deltas in a list sorted by type, by filename hash, and then
1900 * by size, so that we see progressively smaller and smaller files.
1901 * That's because we prefer deltas to be from the bigger file
1902 * to the smaller -- deletes are potentially cheaper, but perhaps
1903 * more importantly, the bigger file is likely the more recent
1904 * one. The deepest deltas are therefore the oldest objects which are
1905 * less susceptible to be accessed often.
1906 */
9668cf59 1907static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1908{
9668cf59
NP
1909 const struct object_entry *a = *(struct object_entry **)_a;
1910 const struct object_entry *b = *(struct object_entry **)_b;
33de80b1
SL
1911 const enum object_type a_type = oe_type(a);
1912 const enum object_type b_type = oe_type(b);
1913 const unsigned long a_size = SIZE(a);
1914 const unsigned long b_size = SIZE(b);
9668cf59 1915
fd9b1bae 1916 if (a_type > b_type)
27225f2e 1917 return -1;
fd9b1bae 1918 if (a_type < b_type)
27225f2e 1919 return 1;
b904166c 1920 if (a->hash > b->hash)
7a979d99 1921 return -1;
b904166c 1922 if (a->hash < b->hash)
7a979d99 1923 return 1;
b904166c 1924 if (a->preferred_base > b->preferred_base)
c323ac7d 1925 return -1;
b904166c
NP
1926 if (a->preferred_base < b->preferred_base)
1927 return 1;
28b8a730 1928 if (use_delta_islands) {
33de80b1 1929 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
28b8a730
JK
1930 if (island_cmp)
1931 return island_cmp;
1932 }
ac77d0c3 1933 if (a_size > b_size)
b904166c 1934 return -1;
ac77d0c3 1935 if (a_size < b_size)
c323ac7d 1936 return 1;
b904166c 1937 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1938}
1939
1940struct unpacked {
1941 struct object_entry *entry;
1942 void *data;
f6c7081a 1943 struct delta_index *index;
5a235b5e 1944 unsigned depth;
c323ac7d
LT
1945};
1946
d250626c
NP
1947static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1948 unsigned long delta_size)
074b2eea
MK
1949{
1950 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1951 return 0;
1952
e3dfddb3
MK
1953 if (delta_size < cache_max_small_delta_size)
1954 return 1;
1955
074b2eea
MK
1956 /* cache delta, if objects are large enough compared to delta size */
1957 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1958 return 1;
1959
1960 return 0;
1961}
1962
ffbd51cc 1963/* Protect delta_cache_size */
44626dc7 1964static pthread_mutex_t cache_mutex;
3c701839
NP
1965#define cache_lock() pthread_mutex_lock(&cache_mutex)
1966#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1967
ffbd51cc
NTND
1968/*
1969 * Protect object list partitioning (e.g. struct thread_param) and
1970 * progress_state
1971 */
44626dc7 1972static pthread_mutex_t progress_mutex;
8ecce684
NP
1973#define progress_lock() pthread_mutex_lock(&progress_mutex)
1974#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1975
ffbd51cc
NTND
1976/*
1977 * Access to struct object_entry is unprotected since each thread owns
1978 * a portion of the main object list. Just don't access object entries
1979 * ahead in the list because they can be stolen and would need
1980 * progress_mutex for protection.
1981 */
8ecce684 1982
ac77d0c3
NTND
1983/*
1984 * Return the size of the object without doing any delta
1985 * reconstruction (so non-deltas are true object sizes, but deltas
1986 * return the size of the delta data).
1987 */
1988unsigned long oe_get_size_slow(struct packing_data *pack,
1989 const struct object_entry *e)
1990{
1991 struct packed_git *p;
1992 struct pack_window *w_curs;
1993 unsigned char *buf;
1994 enum object_type type;
1995 unsigned long used, avail, size;
1996
1997 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
edb673cf 1998 packing_data_lock(&to_pack);
ad635e82 1999 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
ac77d0c3
NTND
2000 die(_("unable to get size of %s"),
2001 oid_to_hex(&e->idx.oid));
edb673cf 2002 packing_data_unlock(&to_pack);
ac77d0c3
NTND
2003 return size;
2004 }
2005
2006 p = oe_in_pack(pack, e);
2007 if (!p)
2008 BUG("when e->type is a delta, it must belong to a pack");
2009
edb673cf 2010 packing_data_lock(&to_pack);
ac77d0c3
NTND
2011 w_curs = NULL;
2012 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2013 used = unpack_object_header_buffer(buf, avail, &type, &size);
2014 if (used == 0)
2015 die(_("unable to parse object header of %s"),
2016 oid_to_hex(&e->idx.oid));
2017
2018 unuse_pack(&w_curs);
edb673cf 2019 packing_data_unlock(&to_pack);
ac77d0c3
NTND
2020 return size;
2021}
2022
f6c7081a 2023static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 2024 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 2025{
f6c7081a
NP
2026 struct object_entry *trg_entry = trg->entry;
2027 struct object_entry *src_entry = src->entry;
560b25a8 2028 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 2029 unsigned ref_depth;
21666f1a 2030 enum object_type type;
c323ac7d
LT
2031 void *delta_buf;
2032
2033 /* Don't bother doing diffs between different types */
fd9b1bae 2034 if (oe_type(trg_entry) != oe_type(src_entry))
c323ac7d
LT
2035 return -1;
2036
51d1e83f 2037 /*
15f07e06
JK
2038 * We do not bother to try a delta that we discarded on an
2039 * earlier try, but only when reusing delta data. Note that
2040 * src_entry that is marked as the preferred_base should always
2041 * be considered, as even if we produce a suboptimal delta against
2042 * it, we will still save the transfer cost, as we already know
2043 * the other side has it and we won't send src_entry at all.
51d1e83f 2044 */
43fa44fa
NTND
2045 if (reuse_delta && IN_PACK(trg_entry) &&
2046 IN_PACK(trg_entry) == IN_PACK(src_entry) &&
15f07e06 2047 !src_entry->preferred_base &&
e9195b58
JH
2048 trg_entry->in_pack_type != OBJ_REF_DELTA &&
2049 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
2050 return 0;
2051
898b14ce 2052 /* Let's not bust the allowed depth. */
5a235b5e 2053 if (src->depth >= max_depth)
d116a45a 2054 return 0;
c323ac7d 2055
c3b06a69 2056 /* Now some size filtering heuristics. */
ac77d0c3 2057 trg_size = SIZE(trg_entry);
898eba5e 2058 if (!DELTA(trg_entry)) {
41179100 2059 max_size = trg_size/2 - the_hash_algo->rawsz;
c83f032e
NP
2060 ref_depth = 1;
2061 } else {
0aca34e8 2062 max_size = DELTA_SIZE(trg_entry);
5a235b5e 2063 ref_depth = trg->depth;
c83f032e 2064 }
720fe22d 2065 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 2066 (max_depth - ref_depth + 1);
c3b06a69
NP
2067 if (max_size == 0)
2068 return 0;
ac77d0c3 2069 src_size = SIZE(src_entry);
560b25a8 2070 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 2071 if (sizediff >= max_size)
f527cb8c 2072 return 0;
a1dab41a
BD
2073 if (trg_size < src_size / 32)
2074 return 0;
f6c7081a 2075
28b8a730
JK
2076 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2077 return 0;
2078
560b25a8
NP
2079 /* Load data if not already done */
2080 if (!trg->data) {
edb673cf 2081 packing_data_lock(&to_pack);
b4f5aca4 2082 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
edb673cf 2083 packing_data_unlock(&to_pack);
2e3404c3 2084 if (!trg->data)
f616db6a 2085 die(_("object %s cannot be read"),
e6a492b7 2086 oid_to_hex(&trg_entry->idx.oid));
560b25a8 2087 if (sz != trg_size)
ca473cef
TB
2088 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2089 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,
2090 (uintmax_t)trg_size);
ef0316fc 2091 *mem_usage += sz;
560b25a8
NP
2092 }
2093 if (!src->data) {
edb673cf 2094 packing_data_lock(&to_pack);
b4f5aca4 2095 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
edb673cf 2096 packing_data_unlock(&to_pack);
71064a95
NP
2097 if (!src->data) {
2098 if (src_entry->preferred_base) {
2099 static int warned = 0;
2100 if (!warned++)
f616db6a 2101 warning(_("object %s cannot be read"),
e6a492b7 2102 oid_to_hex(&src_entry->idx.oid));
71064a95
NP
2103 /*
2104 * Those objects are not included in the
2105 * resulting pack. Be resilient and ignore
2106 * them if they can't be read, in case the
2107 * pack could be created nevertheless.
2108 */
2109 return 0;
2110 }
f616db6a 2111 die(_("object %s cannot be read"),
e6a492b7 2112 oid_to_hex(&src_entry->idx.oid));
71064a95 2113 }
560b25a8 2114 if (sz != src_size)
ca473cef
TB
2115 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),
2116 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,
2117 (uintmax_t)src_size);
ef0316fc 2118 *mem_usage += sz;
560b25a8
NP
2119 }
2120 if (!src->index) {
2121 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
2122 if (!src->index) {
2123 static int warned = 0;
2124 if (!warned++)
f616db6a 2125 warning(_("suboptimal pack - out of memory"));
a588d88a
MK
2126 return 0;
2127 }
ef0316fc 2128 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
2129 }
2130
2131 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 2132 if (!delta_buf)
75c42d8c 2133 return 0;
f6c7081a 2134
898eba5e 2135 if (DELTA(trg_entry)) {
848d732c 2136 /* Prefer only shallower same-sized deltas. */
0aca34e8 2137 if (delta_size == DELTA_SIZE(trg_entry) &&
5a235b5e 2138 src->depth + 1 >= trg->depth) {
848d732c
BD
2139 free(delta_buf);
2140 return 0;
2141 }
074b2eea 2142 }
9e2d57a0 2143
3c701839
NP
2144 /*
2145 * Handle memory allocation outside of the cache
2146 * accounting lock. Compiler will optimize the strangeness
7eb151d6 2147 * away when NO_PTHREADS is defined.
3c701839 2148 */
8e0f7003 2149 free(trg_entry->delta_data);
3c701839 2150 cache_lock();
9e2d57a0 2151 if (trg_entry->delta_data) {
0aca34e8 2152 delta_cache_size -= DELTA_SIZE(trg_entry);
9e2d57a0
NP
2153 trg_entry->delta_data = NULL;
2154 }
d250626c 2155 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 2156 delta_cache_size += delta_size;
3c701839
NP
2157 cache_unlock();
2158 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2159 } else {
2160 cache_unlock();
074b2eea 2161 free(delta_buf);
3c701839
NP
2162 }
2163
898eba5e 2164 SET_DELTA(trg_entry, src_entry);
0aca34e8 2165 SET_DELTA_SIZE(trg_entry, delta_size);
b7a28f78
NP
2166 trg->depth = src->depth + 1;
2167
f6c7081a 2168 return 1;
c323ac7d
LT
2169}
2170
898b14ce 2171static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 2172{
898eba5e 2173 struct object_entry *child = DELTA_CHILD(me);
898b14ce
NP
2174 unsigned int m = n;
2175 while (child) {
33de80b1 2176 const unsigned int c = check_delta_limit(child, n + 1);
898b14ce
NP
2177 if (m < c)
2178 m = c;
898eba5e 2179 child = DELTA_SIBLING(child);
898b14ce
NP
2180 }
2181 return m;
b2504a0d
NP
2182}
2183
75ad235c 2184static unsigned long free_unpacked(struct unpacked *n)
a97773ce 2185{
ef0316fc 2186 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
2187 free_delta_index(n->index);
2188 n->index = NULL;
2189 if (n->data) {
ac77d0c3 2190 freed_mem += SIZE(n->entry);
6a83d902 2191 FREE_AND_NULL(n->data);
a97773ce
BD
2192 }
2193 n->entry = NULL;
7d7baa5e 2194 n->depth = 0;
ef0316fc 2195 return freed_mem;
a97773ce
BD
2196}
2197
384b32c0 2198static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 2199 int window, int depth, unsigned *processed)
c323ac7d 2200{
384b32c0 2201 uint32_t i, idx = 0, count = 0;
7cadf491 2202 struct unpacked *array;
ef0316fc 2203 unsigned long mem_usage = 0;
c323ac7d 2204
19d4b416 2205 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 2206
384b32c0 2207 for (;;) {
421b488a 2208 struct object_entry *entry;
c323ac7d 2209 struct unpacked *n = array + idx;
ef0316fc 2210 int j, max_depth, best_base = -1;
c323ac7d 2211
384b32c0
NP
2212 progress_lock();
2213 if (!*list_size) {
2214 progress_unlock();
2215 break;
2216 }
421b488a 2217 entry = *list++;
384b32c0
NP
2218 (*list_size)--;
2219 if (!entry->preferred_base) {
2220 (*processed)++;
2221 display_progress(progress_state, *processed);
2222 }
2223 progress_unlock();
2224
ef0316fc 2225 mem_usage -= free_unpacked(n);
c323ac7d 2226 n->entry = entry;
ab7cd7bb 2227
a97773ce 2228 while (window_memory_limit &&
ef0316fc 2229 mem_usage > window_memory_limit &&
a97773ce 2230 count > 1) {
33de80b1 2231 const uint32_t tail = (idx + window - count) % window;
75ad235c 2232 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
2233 count--;
2234 }
2235
75d39853
NP
2236 /* We do not compute delta to *create* objects we are not
2237 * going to pack.
2238 */
2239 if (entry->preferred_base)
2240 goto next;
2241
898b14ce
NP
2242 /*
2243 * If the current object is at pack edge, take the depth the
2244 * objects that depend on the current object into account
2245 * otherwise they would become too deep.
2246 */
2247 max_depth = depth;
898eba5e 2248 if (DELTA_CHILD(entry)) {
898b14ce
NP
2249 max_depth -= check_delta_limit(entry, 0);
2250 if (max_depth <= 0)
2251 goto next;
2252 }
2253
78817c15
LT
2254 j = window;
2255 while (--j > 0) {
77639870 2256 int ret;
7cadf491 2257 uint32_t other_idx = idx + j;
c323ac7d 2258 struct unpacked *m;
78817c15
LT
2259 if (other_idx >= window)
2260 other_idx -= window;
c323ac7d
LT
2261 m = array + other_idx;
2262 if (!m->entry)
2263 break;
ef0316fc 2264 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 2265 if (ret < 0)
c323ac7d 2266 break;
77639870
JH
2267 else if (ret > 0)
2268 best_base = other_idx;
c323ac7d 2269 }
898b14ce 2270
ed4a9031
NP
2271 /*
2272 * If we decided to cache the delta data, then it is best
2273 * to compress it right away. First because we have to do
2274 * it anyway, and doing it here while we're threaded will
2275 * save a lot of time in the non threaded write phase,
2276 * as well as allow for caching more deltas within
2277 * the same cache size limit.
2278 * ...
2279 * But only if not writing to stdout, since in that case
2280 * the network is most likely throttling writes anyway,
2281 * and therefore it is best to go to the write phase ASAP
2282 * instead, as we can afford spending more time compressing
2283 * between writes at that moment.
2284 */
2285 if (entry->delta_data && !pack_to_stdout) {
0cb3c142
NTND
2286 unsigned long size;
2287
0aca34e8 2288 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
0cb3c142
NTND
2289 if (size < (1U << OE_Z_DELTA_BITS)) {
2290 entry->z_delta_size = size;
2291 cache_lock();
0aca34e8 2292 delta_cache_size -= DELTA_SIZE(entry);
0cb3c142
NTND
2293 delta_cache_size += entry->z_delta_size;
2294 cache_unlock();
2295 } else {
2296 FREE_AND_NULL(entry->delta_data);
2297 entry->z_delta_size = 0;
2298 }
ed4a9031
NP
2299 }
2300
70ca1a3f
JH
2301 /* if we made n a delta, and if n is already at max
2302 * depth, leaving it in the window is pointless. we
2303 * should evict it first.
70ca1a3f 2304 */
898eba5e 2305 if (DELTA(entry) && max_depth <= n->depth)
70ca1a3f 2306 continue;
ff45715c 2307
77639870
JH
2308 /*
2309 * Move the best delta base up in the window, after the
2310 * currently deltified object, to keep it longer. It will
2311 * be the first base object to be attempted next.
2312 */
898eba5e 2313 if (DELTA(entry)) {
77639870
JH
2314 struct unpacked swap = array[best_base];
2315 int dist = (window + idx - best_base) % window;
2316 int dst = best_base;
2317 while (dist--) {
2318 int src = (dst + 1) % window;
2319 array[dst] = array[src];
2320 dst = src;
2321 }
2322 array[dst] = swap;
2323 }
2324
898b14ce 2325 next:
521a4f4c 2326 idx++;
a97773ce
BD
2327 if (count + 1 < window)
2328 count++;
521a4f4c
LT
2329 if (idx >= window)
2330 idx = 0;
384b32c0 2331 }
adee7bdf 2332
f6c7081a 2333 for (i = 0; i < window; ++i) {
ff45715c 2334 free_delta_index(array[i].index);
adee7bdf 2335 free(array[i].data);
f6c7081a 2336 }
adee7bdf 2337 free(array);
c323ac7d
LT
2338}
2339
a9a74636
NP
2340static void try_to_free_from_threads(size_t size)
2341{
edb673cf 2342 packing_data_lock(&to_pack);
7c3ecb32 2343 release_pack_memory(size);
edb673cf 2344 packing_data_unlock(&to_pack);
a9a74636
NP
2345}
2346
bc9b2175 2347static try_to_free_t old_try_to_free_routine;
851c34b0 2348
50f22ada 2349/*
ffbd51cc
NTND
2350 * The main object list is split into smaller lists, each is handed to
2351 * one worker.
2352 *
50f22ada
JS
2353 * The main thread waits on the condition that (at least) one of the workers
2354 * has stopped working (which is indicated in the .working member of
2355 * struct thread_params).
ffbd51cc 2356 *
50f22ada
JS
2357 * When a work thread has completed its work, it sets .working to 0 and
2358 * signals the main thread and waits on the condition that .data_ready
2359 * becomes 1.
ffbd51cc
NTND
2360 *
2361 * The main thread steals half of the work from the worker that has
2362 * most work left to hand it to the idle worker.
50f22ada
JS
2363 */
2364
8ecce684
NP
2365struct thread_params {
2366 pthread_t thread;
2367 struct object_entry **list;
2368 unsigned list_size;
384b32c0 2369 unsigned remaining;
8ecce684
NP
2370 int window;
2371 int depth;
50f22ada
JS
2372 int working;
2373 int data_ready;
2374 pthread_mutex_t mutex;
2375 pthread_cond_t cond;
8ecce684
NP
2376 unsigned *processed;
2377};
2378
44626dc7
AH
2379static pthread_cond_t progress_cond;
2380
2381/*
2382 * Mutex and conditional variable can't be statically-initialized on Windows.
2383 */
2384static void init_threaded_search(void)
2385{
44626dc7
AH
2386 pthread_mutex_init(&cache_mutex, NULL);
2387 pthread_mutex_init(&progress_mutex, NULL);
2388 pthread_cond_init(&progress_cond, NULL);
851c34b0 2389 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
2390}
2391
2392static void cleanup_threaded_search(void)
2393{
851c34b0 2394 set_try_to_free_routine(old_try_to_free_routine);
44626dc7 2395 pthread_cond_destroy(&progress_cond);
44626dc7
AH
2396 pthread_mutex_destroy(&cache_mutex);
2397 pthread_mutex_destroy(&progress_mutex);
2398}
c2a33679 2399
8ecce684
NP
2400static void *threaded_find_deltas(void *arg)
2401{
c2a33679
NP
2402 struct thread_params *me = arg;
2403
0c2ad00b 2404 progress_lock();
50f22ada 2405 while (me->remaining) {
0c2ad00b
2406 progress_unlock();
2407
384b32c0 2408 find_deltas(me->list, &me->remaining,
c2a33679 2409 me->window, me->depth, me->processed);
50f22ada
JS
2410
2411 progress_lock();
2412 me->working = 0;
2413 pthread_cond_signal(&progress_cond);
2414 progress_unlock();
2415
2416 /*
2417 * We must not set ->data_ready before we wait on the
2418 * condition because the main thread may have set it to 1
2419 * before we get here. In order to be sure that new
2420 * work is available if we see 1 in ->data_ready, it
2421 * was initialized to 0 before this thread was spawned
2422 * and we reset it to 0 right away.
2423 */
2424 pthread_mutex_lock(&me->mutex);
2425 while (!me->data_ready)
2426 pthread_cond_wait(&me->cond, &me->mutex);
2427 me->data_ready = 0;
2428 pthread_mutex_unlock(&me->mutex);
0c2ad00b
2429
2430 progress_lock();
c2a33679 2431 }
0c2ad00b 2432 progress_unlock();
50f22ada
JS
2433 /* leave ->working 1 so that this doesn't get more work assigned */
2434 return NULL;
8ecce684
NP
2435}
2436
8ecce684
NP
2437static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2438 int window, int depth, unsigned *processed)
2439{
dcda3614 2440 struct thread_params *p;
384b32c0 2441 int i, ret, active_threads = 0;
c2a33679 2442
44626dc7
AH
2443 init_threaded_search();
2444
367f4a43 2445 if (delta_search_threads <= 1) {
384b32c0 2446 find_deltas(list, &list_size, window, depth, processed);
44626dc7 2447 cleanup_threaded_search();
367f4a43
NP
2448 return;
2449 }
43cc2b42 2450 if (progress > pack_to_stdout)
f616db6a 2451 fprintf_ln(stderr, _("Delta compression using up to %d threads"),
1a07e59c 2452 delta_search_threads);
dcda3614 2453 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 2454
50f22ada 2455 /* Partition the work amongst work threads. */
367f4a43 2456 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
2457 unsigned sub_size = list_size / (delta_search_threads - i);
2458
bf874896
NP
2459 /* don't use too small segments or no deltas will be found */
2460 if (sub_size < 2*window && i+1 < delta_search_threads)
2461 sub_size = 0;
2462
8ecce684
NP
2463 p[i].window = window;
2464 p[i].depth = depth;
2465 p[i].processed = processed;
50f22ada
JS
2466 p[i].working = 1;
2467 p[i].data_ready = 0;
c2a33679 2468
59921b4b 2469 /* try to split chunks on "path" boundaries */
6fc74703
NP
2470 while (sub_size && sub_size < list_size &&
2471 list[sub_size]->hash &&
384b32c0
NP
2472 list[sub_size]->hash == list[sub_size-1]->hash)
2473 sub_size++;
2474
50f22ada
JS
2475 p[i].list = list;
2476 p[i].list_size = sub_size;
2477 p[i].remaining = sub_size;
59921b4b 2478
384b32c0
NP
2479 list += sub_size;
2480 list_size -= sub_size;
2481 }
2482
50f22ada
JS
2483 /* Start work threads. */
2484 for (i = 0; i < delta_search_threads; i++) {
2485 if (!p[i].list_size)
2486 continue;
68e6a4f8
JS
2487 pthread_mutex_init(&p[i].mutex, NULL);
2488 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
2489 ret = pthread_create(&p[i].thread, NULL,
2490 threaded_find_deltas, &p[i]);
2491 if (ret)
f616db6a 2492 die(_("unable to create thread: %s"), strerror(ret));
50f22ada
JS
2493 active_threads++;
2494 }
2495
384b32c0
NP
2496 /*
2497 * Now let's wait for work completion. Each time a thread is done
2498 * with its work, we steal half of the remaining work from the
2499 * thread with the largest number of unprocessed objects and give
2500 * it to that newly idle thread. This ensure good load balancing
2501 * until the remaining object list segments are simply too short
2502 * to be worth splitting anymore.
2503 */
50f22ada
JS
2504 while (active_threads) {
2505 struct thread_params *target = NULL;
384b32c0
NP
2506 struct thread_params *victim = NULL;
2507 unsigned sub_size = 0;
384b32c0
NP
2508
2509 progress_lock();
50f22ada
JS
2510 for (;;) {
2511 for (i = 0; !target && i < delta_search_threads; i++)
2512 if (!p[i].working)
2513 target = &p[i];
2514 if (target)
2515 break;
2516 pthread_cond_wait(&progress_cond, &progress_mutex);
2517 }
2518
384b32c0
NP
2519 for (i = 0; i < delta_search_threads; i++)
2520 if (p[i].remaining > 2*window &&
2521 (!victim || victim->remaining < p[i].remaining))
2522 victim = &p[i];
2523 if (victim) {
2524 sub_size = victim->remaining / 2;
2525 list = victim->list + victim->list_size - sub_size;
2526 while (sub_size && list[0]->hash &&
2527 list[0]->hash == list[-1]->hash) {
2528 list++;
2529 sub_size--;
2530 }
eb9688ff
NP
2531 if (!sub_size) {
2532 /*
2533 * It is possible for some "paths" to have
2534 * so many objects that no hash boundary
2535 * might be found. Let's just steal the
2536 * exact half in that case.
2537 */
2538 sub_size = victim->remaining / 2;
2539 list -= sub_size;
2540 }
384b32c0
NP
2541 target->list = list;
2542 victim->list_size -= sub_size;
2543 victim->remaining -= sub_size;
2544 }
384b32c0
NP
2545 target->list_size = sub_size;
2546 target->remaining = sub_size;
50f22ada
JS
2547 target->working = 1;
2548 progress_unlock();
2549
2550 pthread_mutex_lock(&target->mutex);
2551 target->data_ready = 1;
2552 pthread_cond_signal(&target->cond);
2553 pthread_mutex_unlock(&target->mutex);
c2a33679 2554
384b32c0 2555 if (!sub_size) {
b81d9af7 2556 pthread_join(target->thread, NULL);
50f22ada
JS
2557 pthread_cond_destroy(&target->cond);
2558 pthread_mutex_destroy(&target->mutex);
384b32c0 2559 active_threads--;
c2a33679 2560 }
50f22ada 2561 }
44626dc7 2562 cleanup_threaded_search();
dcda3614 2563 free(p);
8ecce684
NP
2564}
2565
b773ddea
JK
2566static void add_tag_chain(const struct object_id *oid)
2567{
2568 struct tag *tag;
2569
2570 /*
2571 * We catch duplicates already in add_object_entry(), but we'd
2572 * prefer to do this extra check to avoid having to parse the
2573 * tag at all if we already know that it's being packed (e.g., if
2574 * it was included via bitmaps, we would not have parsed it
2575 * previously).
2576 */
3a37876b 2577 if (packlist_find(&to_pack, oid))
b773ddea
JK
2578 return;
2579
ce71efb7 2580 tag = lookup_tag(the_repository, oid);
b773ddea
JK
2581 while (1) {
2582 if (!tag || parse_tag(tag) || !tag->tagged)
f616db6a 2583 die(_("unable to pack objects reachable from tag %s"),
b773ddea
JK
2584 oid_to_hex(oid));
2585
188960b4 2586 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
b773ddea
JK
2587
2588 if (tag->tagged->type != OBJ_TAG)
2589 return;
2590
2591 tag = (struct tag *)tag->tagged;
2592 }
2593}
2594
d155254c 2595static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
f0a24aa5 2596{
d155254c 2597 struct object_id peeled;
f0a24aa5 2598
59556548 2599 if (starts_with(path, "refs/tags/") && /* is a tag? */
b420d909 2600 !peel_ref(path, &peeled) && /* peelable? */
3a37876b 2601 packlist_find(&to_pack, &peeled)) /* object packed? */
b773ddea 2602 add_tag_chain(oid);
f0a24aa5
SP
2603 return 0;
2604}
2605
f3123c4a
JH
2606static void prepare_pack(int window, int depth)
2607{
9668cf59 2608 struct object_entry **delta_list;
6e1c2344
RJ
2609 uint32_t i, nr_deltas;
2610 unsigned n;
9668cf59 2611
28b8a730 2612 if (use_delta_islands)
385cb64f 2613 resolve_tree_islands(the_repository, progress, &to_pack);
28b8a730 2614
3f9ac8d2 2615 get_object_details();
9668cf59 2616
0e8189e2
NP
2617 /*
2618 * If we're locally repacking then we need to be doubly careful
2619 * from now on in order to make sure no stealth corruption gets
2620 * propagated to the new pack. Clients receiving streamed packs
2621 * should validate everything they get anyway so no need to incur
2622 * the additional cost here in that case.
2623 */
2624 if (!pack_to_stdout)
2625 do_check_packed_object_crc = 1;
2626
2834bc27 2627 if (!to_pack.nr_objects || !window || !depth)
9668cf59
NP
2628 return;
2629
b32fa95f 2630 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
75d39853
NP
2631 nr_deltas = n = 0;
2632
2834bc27
VM
2633 for (i = 0; i < to_pack.nr_objects; i++) {
2634 struct object_entry *entry = to_pack.objects + i;
75d39853 2635
898eba5e 2636 if (DELTA(entry))
75d39853 2637 /* This happens if we decided to reuse existing
a7de7130 2638 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2639 */
2640 continue;
2641
ac77d0c3
NTND
2642 if (!entry->type_valid ||
2643 oe_size_less_than(&to_pack, entry, 50))
75d39853
NP
2644 continue;
2645
2646 if (entry->no_try_delta)
2647 continue;
2648
6d6f9cdd 2649 if (!entry->preferred_base) {
75d39853 2650 nr_deltas++;
fd9b1bae 2651 if (oe_type(entry) < 0)
f616db6a 2652 die(_("unable to get type of object %s"),
e6a492b7 2653 oid_to_hex(&entry->idx.oid));
eede9f42 2654 } else {
fd9b1bae 2655 if (oe_type(entry) < 0) {
eede9f42
NP
2656 /*
2657 * This object is not found, but we
2658 * don't have to include it anyway.
2659 */
2660 continue;
2661 }
6d6f9cdd 2662 }
75d39853
NP
2663
2664 delta_list[n++] = entry;
2665 }
2666
2f8b8947 2667 if (nr_deltas && n > 1) {
e334977d
NP
2668 unsigned nr_done = 0;
2669 if (progress)
754dbc43 2670 progress_state = start_progress(_("Compressing objects"),
dc6a0757 2671 nr_deltas);
9ed0d8d6 2672 QSORT(delta_list, n, type_size_sort);
8ecce684 2673 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2674 stop_progress(&progress_state);
e334977d 2675 if (nr_done != nr_deltas)
f616db6a 2676 die(_("inconsistency with delta count"));
75d39853 2677 }
9668cf59 2678 free(delta_list);
f3123c4a
JH
2679}
2680
ef90d6d4 2681static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2682{
eeefa7c9 2683 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2684 window = git_config_int(k, v);
2685 return 0;
2686 }
a97773ce
BD
2687 if (!strcmp(k, "pack.windowmemory")) {
2688 window_memory_limit = git_config_ulong(k, v);
2689 return 0;
2690 }
2691 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2692 depth = git_config_int(k, v);
2693 return 0;
2694 }
074b2eea
MK
2695 if (!strcmp(k, "pack.deltacachesize")) {
2696 max_delta_cache_size = git_config_int(k, v);
2697 return 0;
2698 }
e3dfddb3
MK
2699 if (!strcmp(k, "pack.deltacachelimit")) {
2700 cache_max_small_delta_size = git_config_int(k, v);
2701 return 0;
2702 }
ae4f07fb
VM
2703 if (!strcmp(k, "pack.writebitmaphashcache")) {
2704 if (git_config_bool(k, v))
2705 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2706 else
2707 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2708 }
6b8fda2d 2709 if (!strcmp(k, "pack.usebitmaps")) {
645c432d 2710 use_bitmap_index_default = git_config_bool(k, v);
6b8fda2d
VM
2711 return 0;
2712 }
3d036eb0
DS
2713 if (!strcmp(k, "pack.usesparse")) {
2714 sparse = git_config_bool(k, v);
2715 return 0;
2716 }
693b86ff
NP
2717 if (!strcmp(k, "pack.threads")) {
2718 delta_search_threads = git_config_int(k, v);
833e3df1 2719 if (delta_search_threads < 0)
f616db6a 2720 die(_("invalid number of threads specified (%d)"),
693b86ff 2721 delta_search_threads);
9c897c5c 2722 if (!HAVE_THREADS && delta_search_threads != 1) {
f616db6a 2723 warning(_("no threads support, ignoring %s"), k);
2e96d815
ÆAB
2724 delta_search_threads = 0;
2725 }
693b86ff
NP
2726 return 0;
2727 }
4d00bda2 2728 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2729 pack_idx_opts.version = git_config_int(k, v);
2730 if (pack_idx_opts.version > 2)
f616db6a 2731 die(_("bad pack.indexversion=%"PRIu32),
ebcfb379 2732 pack_idx_opts.version);
4d00bda2
NP
2733 return 0;
2734 }
ef90d6d4 2735 return git_default_config(k, v, cb);
4812a93a
JK
2736}
2737
b5d97e6b 2738static void read_object_list_from_stdin(void)
c323ac7d 2739{
188960b4 2740 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2741 struct object_id oid;
2742 const char *p;
b2504a0d 2743
da93d12b 2744 for (;;) {
da93d12b
LT
2745 if (!fgets(line, sizeof(line), stdin)) {
2746 if (feof(stdin))
2747 break;
2748 if (!ferror(stdin))
1a07e59c 2749 die("BUG: fgets returned NULL, not EOF, not error!");
687dd75c 2750 if (errno != EINTR)
d824cbba 2751 die_errno("fgets");
687dd75c
JH
2752 clearerr(stdin);
2753 continue;
da93d12b 2754 }
7a979d99 2755 if (line[0] == '-') {
188960b4 2756 if (get_oid_hex(line+1, &oid))
f616db6a 2757 die(_("expected edge object ID, got garbage:\n %s"),
b5d97e6b 2758 line);
188960b4 2759 add_preferred_base(&oid);
7a979d99 2760 continue;
21fcd1bd 2761 }
188960b4 2762 if (parse_oid_hex(line, &oid, &p))
f616db6a 2763 die(_("expected object ID, got garbage:\n %s"), line);
b5d97e6b 2764
188960b4 2765 add_preferred_base_object(p + 1);
fd9b1bae 2766 add_object_entry(&oid, OBJ_NONE, p + 1, 0);
c323ac7d 2767 }
b5d97e6b
JH
2768}
2769
95308d64 2770/* Remember to update object flag allocation in object.h */
08cdfb13
JH
2771#define OBJECT_ADDED (1u<<20)
2772
11c211fa 2773static void show_commit(struct commit *commit, void *data)
b5d97e6b 2774{
188960b4 2775 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
08cdfb13 2776 commit->object.flags |= OBJECT_ADDED;
7cc8f971
VM
2777
2778 if (write_bitmap_index)
2779 index_commit_for_bitmap(commit);
28b8a730
JK
2780
2781 if (use_delta_islands)
2782 propagate_island_marks(commit);
b5d97e6b
JH
2783}
2784
de1e67d0 2785static void show_object(struct object *obj, const char *name, void *data)
b5d97e6b 2786{
8d2dfc49 2787 add_preferred_base_object(name);
188960b4 2788 add_object_entry(&obj->oid, obj->type, name, 0);
8d2dfc49 2789 obj->flags |= OBJECT_ADDED;
28b8a730
JK
2790
2791 if (use_delta_islands) {
2792 const char *p;
39490536 2793 unsigned depth;
28b8a730
JK
2794 struct object_entry *ent;
2795
39490536
JK
2796 /* the empty string is a root tree, which is depth 0 */
2797 depth = *name ? 1 : 0;
28b8a730
JK
2798 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2799 depth++;
2800
3a37876b 2801 ent = packlist_find(&to_pack, &obj->oid);
108f5303
CC
2802 if (ent && depth > oe_tree_depth(&to_pack, ent))
2803 oe_set_tree_depth(&to_pack, ent, depth);
28b8a730 2804 }
b5d97e6b
JH
2805}
2806
9535ce73
JH
2807static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2808{
2809 assert(arg_missing_action == MA_ALLOW_ANY);
2810
2811 /*
2812 * Quietly ignore ALL missing objects. This avoids problems with
2813 * staging them now and getting an odd error later.
2814 */
2815 if (!has_object_file(&obj->oid))
2816 return;
2817
2818 show_object(obj, name, data);
2819}
2820
0c16cd49
JT
2821static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2822{
2823 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2824
2825 /*
2826 * Quietly ignore EXPECTED missing objects. This avoids problems with
2827 * staging them now and getting an odd error later.
2828 */
2829 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2830 return;
2831
2832 show_object(obj, name, data);
2833}
2834
9535ce73
JH
2835static int option_parse_missing_action(const struct option *opt,
2836 const char *arg, int unset)
2837{
2838 assert(arg);
2839 assert(!unset);
2840
2841 if (!strcmp(arg, "error")) {
2842 arg_missing_action = MA_ERROR;
2843 fn_show_object = show_object;
2844 return 0;
2845 }
2846
2847 if (!strcmp(arg, "allow-any")) {
2848 arg_missing_action = MA_ALLOW_ANY;
0c16cd49 2849 fetch_if_missing = 0;
9535ce73
JH
2850 fn_show_object = show_object__ma_allow_any;
2851 return 0;
2852 }
2853
0c16cd49
JT
2854 if (!strcmp(arg, "allow-promisor")) {
2855 arg_missing_action = MA_ALLOW_PROMISOR;
2856 fetch_if_missing = 0;
2857 fn_show_object = show_object__ma_allow_promisor;
2858 return 0;
2859 }
2860
9535ce73
JH
2861 die(_("invalid value for --missing"));
2862 return 0;
2863}
2864
8d1d8f83
JH
2865static void show_edge(struct commit *commit)
2866{
188960b4 2867 add_preferred_base(&commit->object.oid);
8d1d8f83
JH
2868}
2869
08cdfb13
JH
2870struct in_pack_object {
2871 off_t offset;
2872 struct object *object;
2873};
2874
2875struct in_pack {
071bcaab
RJ
2876 unsigned int alloc;
2877 unsigned int nr;
08cdfb13
JH
2878 struct in_pack_object *array;
2879};
2880
2881static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2882{
ed1c9977 2883 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
08cdfb13
JH
2884 in_pack->array[in_pack->nr].object = object;
2885 in_pack->nr++;
2886}
2887
2888/*
2889 * Compare the objects in the offset order, in order to emulate the
f18d244a 2890 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2891 */
2892static int ofscmp(const void *a_, const void *b_)
2893{
2894 struct in_pack_object *a = (struct in_pack_object *)a_;
2895 struct in_pack_object *b = (struct in_pack_object *)b_;
2896
2897 if (a->offset < b->offset)
2898 return -1;
2899 else if (a->offset > b->offset)
2900 return 1;
2901 else
f2fd0760 2902 return oidcmp(&a->object->oid, &b->object->oid);
08cdfb13
JH
2903}
2904
7a2a7216 2905static void add_objects_in_unpacked_packs(void)
08cdfb13
JH
2906{
2907 struct packed_git *p;
2908 struct in_pack in_pack;
2909 uint32_t i;
2910
2911 memset(&in_pack, 0, sizeof(in_pack));
2912
454ea2e4 2913 for (p = get_all_packs(the_repository); p; p = p->next) {
188960b4 2914 struct object_id oid;
08cdfb13
JH
2915 struct object *o;
2916
ed7e5fc3 2917 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
08cdfb13
JH
2918 continue;
2919 if (open_pack_index(p))
f616db6a 2920 die(_("cannot open pack index"));
08cdfb13
JH
2921
2922 ALLOC_GROW(in_pack.array,
2923 in_pack.nr + p->num_objects,
2924 in_pack.alloc);
2925
2926 for (i = 0; i < p->num_objects; i++) {
188960b4 2927 nth_packed_object_oid(&oid, p, i);
0ebbcf70 2928 o = lookup_unknown_object(&oid);
08cdfb13
JH
2929 if (!(o->flags & OBJECT_ADDED))
2930 mark_in_pack_object(o, p, &in_pack);
2931 o->flags |= OBJECT_ADDED;
2932 }
2933 }
2934
2935 if (in_pack.nr) {
9ed0d8d6 2936 QSORT(in_pack.array, in_pack.nr, ofscmp);
08cdfb13
JH
2937 for (i = 0; i < in_pack.nr; i++) {
2938 struct object *o = in_pack.array[i].object;
188960b4 2939 add_object_entry(&o->oid, o->type, "", 0);
08cdfb13
JH
2940 }
2941 }
2942 free(in_pack.array);
2943}
2944
76c1d9a0 2945static int add_loose_object(const struct object_id *oid, const char *path,
e26a8c47
JK
2946 void *data)
2947{
0df8e965 2948 enum object_type type = oid_object_info(the_repository, oid, NULL);
e26a8c47
JK
2949
2950 if (type < 0) {
f616db6a 2951 warning(_("loose object at %s could not be examined"), path);
e26a8c47
JK
2952 return 0;
2953 }
2954
188960b4 2955 add_object_entry(oid, type, "", 0);
e26a8c47
JK
2956 return 0;
2957}
2958
2959/*
2960 * We actually don't even have to worry about reachability here.
2961 * add_object_entry will weed out duplicates, so we just add every
2962 * loose object we find.
2963 */
2964static void add_unreachable_loose_objects(void)
2965{
2966 for_each_loose_file_in_objdir(get_object_directory(),
2967 add_loose_object,
2968 NULL, NULL, NULL);
2969}
2970
188960b4 2971static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
094085e3
BC
2972{
2973 static struct packed_git *last_found = (void *)1;
2974 struct packed_git *p;
2975
a80d72db 2976 p = (last_found != (void *)1) ? last_found :
454ea2e4 2977 get_all_packs(the_repository);
094085e3
BC
2978
2979 while (p) {
ed7e5fc3
NTND
2980 if ((!p->pack_local || p->pack_keep ||
2981 p->pack_keep_in_core) &&
188960b4 2982 find_pack_entry_one(oid->hash, p)) {
094085e3
BC
2983 last_found = p;
2984 return 1;
2985 }
2986 if (p == last_found)
454ea2e4 2987 p = get_all_packs(the_repository);
094085e3
BC
2988 else
2989 p = p->next;
2990 if (p == last_found)
2991 p = p->next;
2992 }
2993 return 0;
2994}
2995
abcb8655
JK
2996/*
2997 * Store a list of sha1s that are should not be discarded
2998 * because they are either written too recently, or are
2999 * reachable from another object that was.
3000 *
3001 * This is filled by get_object_list.
3002 */
910650d2 3003static struct oid_array recent_objects;
abcb8655 3004
4ce3621a 3005static int loosened_object_can_be_discarded(const struct object_id *oid,
dddbad72 3006 timestamp_t mtime)
d0d46abc
JK
3007{
3008 if (!unpack_unreachable_expiration)
3009 return 0;
3010 if (mtime > unpack_unreachable_expiration)
3011 return 0;
910650d2 3012 if (oid_array_lookup(&recent_objects, oid) >= 0)
abcb8655 3013 return 0;
d0d46abc
JK
3014 return 1;
3015}
3016
7a2a7216 3017static void loosen_unused_packed_objects(void)
ca11b212
NP
3018{
3019 struct packed_git *p;
3020 uint32_t i;
4ce3621a 3021 struct object_id oid;
ca11b212 3022
454ea2e4 3023 for (p = get_all_packs(the_repository); p; p = p->next) {
ed7e5fc3 3024 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
ca11b212
NP
3025 continue;
3026
3027 if (open_pack_index(p))
f616db6a 3028 die(_("cannot open pack index"));
ca11b212
NP
3029
3030 for (i = 0; i < p->num_objects; i++) {
4ce3621a 3031 nth_packed_object_oid(&oid, p, i);
3a37876b 3032 if (!packlist_find(&to_pack, &oid) &&
188960b4 3033 !has_sha1_pack_kept_or_nonlocal(&oid) &&
4ce3621a 3034 !loosened_object_can_be_discarded(&oid, p->mtime))
4bdb70a4 3035 if (force_object_loose(&oid, p->mtime))
f616db6a 3036 die(_("unable to force loose object"));
ca11b212
NP
3037 }
3038 }
3039}
3040
69e4b342 3041/*
645c432d
KS
3042 * This tracks any options which pack-reuse code expects to be on, or which a
3043 * reader of the pack might not understand, and which would therefore prevent
3044 * blind reuse of what we have on disk.
69e4b342
JK
3045 */
3046static int pack_options_allow_reuse(void)
3047{
9df4a607
JK
3048 return pack_to_stdout &&
3049 allow_ofs_delta &&
ed7e5fc3
NTND
3050 !ignore_packed_keep_on_disk &&
3051 !ignore_packed_keep_in_core &&
9df4a607
JK
3052 (!local || !have_non_local_packs) &&
3053 !incremental;
69e4b342
JK
3054}
3055
6b8fda2d
VM
3056static int get_object_list_from_bitmap(struct rev_info *revs)
3057{
3ae5fa07 3058 if (!(bitmap_git = prepare_bitmap_walk(revs)))
6b8fda2d
VM
3059 return -1;
3060
69e4b342
JK
3061 if (pack_options_allow_reuse() &&
3062 !reuse_partial_packfile_from_bitmap(
3ae5fa07 3063 bitmap_git,
6b8fda2d
VM
3064 &reuse_packfile,
3065 &reuse_packfile_objects,
3066 &reuse_packfile_offset)) {
3067 assert(reuse_packfile_objects);
3068 nr_result += reuse_packfile_objects;
78d2214e 3069 display_progress(progress_state, nr_result);
6b8fda2d
VM
3070 }
3071
3ae5fa07 3072 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
6b8fda2d
VM
3073 return 0;
3074}
3075
abcb8655 3076static void record_recent_object(struct object *obj,
de1e67d0 3077 const char *name,
abcb8655
JK
3078 void *data)
3079{
910650d2 3080 oid_array_append(&recent_objects, &obj->oid);
abcb8655
JK
3081}
3082
3083static void record_recent_commit(struct commit *commit, void *data)
3084{
910650d2 3085 oid_array_append(&recent_objects, &commit->object.oid);
abcb8655
JK
3086}
3087
8d1d8f83 3088static void get_object_list(int ac, const char **av)
b5d97e6b
JH
3089{
3090 struct rev_info revs;
bbcde41a
MD
3091 struct setup_revision_opt s_r_opt = {
3092 .allow_exclude_promisor_objects = 1,
3093 };
b5d97e6b 3094 char line[1000];
b5d97e6b 3095 int flags = 0;
a4544b31 3096 int save_warning;
b5d97e6b 3097
2abf3503 3098 repo_init_revisions(the_repository, &revs, NULL);
b5d97e6b 3099 save_commit_buffer = 0;
bbcde41a 3100 setup_revisions(ac, av, &revs, &s_r_opt);
b5d97e6b 3101
b790e0f6 3102 /* make sure shallows are read */
c8813487 3103 is_repository_shallow(the_repository);
b790e0f6 3104
a4544b31
DS
3105 save_warning = warn_on_object_refname_ambiguity;
3106 warn_on_object_refname_ambiguity = 0;
3107
b5d97e6b
JH
3108 while (fgets(line, sizeof(line), stdin) != NULL) {
3109 int len = strlen(line);
872c930d 3110 if (len && line[len - 1] == '\n')
b5d97e6b
JH
3111 line[--len] = 0;
3112 if (!len)
3113 break;
3114 if (*line == '-') {
3115 if (!strcmp(line, "--not")) {
3116 flags ^= UNINTERESTING;
7cc8f971 3117 write_bitmap_index = 0;
b5d97e6b
JH
3118 continue;
3119 }
b790e0f6 3120 if (starts_with(line, "--shallow ")) {
e92b848c 3121 struct object_id oid;
3122 if (get_oid_hex(line + 10, &oid))
b790e0f6 3123 die("not an SHA-1 '%s'", line + 10);
19143f13 3124 register_shallow(the_repository, &oid);
f7f91086 3125 use_bitmap_index = 0;
b790e0f6
NTND
3126 continue;
3127 }
f616db6a 3128 die(_("not a rev '%s'"), line);
b5d97e6b 3129 }
8e676e8b 3130 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
f616db6a 3131 die(_("bad revision '%s'"), line);
b5d97e6b
JH
3132 }
3133
a4544b31
DS
3134 warn_on_object_refname_ambiguity = save_warning;
3135
6b8fda2d
VM
3136 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3137 return;
3138
28b8a730 3139 if (use_delta_islands)
bdbdf42f 3140 load_delta_islands(the_repository, progress);
28b8a730 3141
3d51e1b5 3142 if (prepare_revision_walk(&revs))
f616db6a 3143 die(_("revision walk setup failed"));
4f6d26b1 3144 mark_edges_uninteresting(&revs, show_edge, sparse);
9535ce73
JH
3145
3146 if (!fn_show_object)
3147 fn_show_object = show_object;
3148 traverse_commit_list_filtered(&filter_options, &revs,
3149 show_commit, fn_show_object, NULL,
3150 NULL);
08cdfb13 3151
abcb8655
JK
3152 if (unpack_unreachable_expiration) {
3153 revs.ignore_missing_links = 1;
3154 if (add_unseen_recent_objects_to_traversal(&revs,
3155 unpack_unreachable_expiration))
f616db6a 3156 die(_("unable to add recent objects"));
abcb8655 3157 if (prepare_revision_walk(&revs))
f616db6a 3158 die(_("revision walk setup failed"));
abcb8655
JK
3159 traverse_commit_list(&revs, record_recent_commit,
3160 record_recent_object, NULL);
3161 }
3162
08cdfb13 3163 if (keep_unreachable)
7a2a7216 3164 add_objects_in_unpacked_packs();
e26a8c47
JK
3165 if (pack_loose_unreachable)
3166 add_unreachable_loose_objects();
ca11b212 3167 if (unpack_unreachable)
7a2a7216 3168 loosen_unused_packed_objects();
abcb8655 3169
910650d2 3170 oid_array_clear(&recent_objects);
b5d97e6b
JH
3171}