object-store: move packed_git and packed_git_mru to object store
[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"
abcb8655
JK
27#include "reachable.h"
28#include "sha1-array.h"
edfbb2aa 29#include "argv-array.h"
ec2dd32c 30#include "list.h"
0317f455 31#include "packfile.h"
a80d72db 32#include "object-store.h"
8ecce684 33
99fb6e04 34static const char *pack_usage[] = {
b8c1d275
AH
35 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
36 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
99fb6e04
NTND
37 NULL
38};
c323ac7d 39
3f9ac8d2 40/*
2834bc27
VM
41 * Objects we are going to pack are collected in the `to_pack` structure.
42 * It contains an array (dynamically expanded) of the object data, and a map
43 * that can resolve SHA1s to their position in the array.
3f9ac8d2 44 */
2834bc27
VM
45static struct packing_data to_pack;
46
79814f42 47static struct pack_idx_entry **written_list;
2834bc27 48static uint32_t nr_result, nr_written;
3f9ac8d2 49
96f1e58f 50static int non_empty;
a7de7130 51static int reuse_delta = 1, reuse_object = 1;
e5e9714a 52static int keep_unreachable, unpack_unreachable, include_tag;
dddbad72 53static timestamp_t unpack_unreachable_expiration;
e26a8c47 54static int pack_loose_unreachable;
96f1e58f 55static int local;
56dfeb62 56static int have_non_local_packs;
96f1e58f 57static int incremental;
e96fb9b8 58static int ignore_packed_keep;
be6b1914 59static int allow_ofs_delta;
ebcfb379 60static struct pack_idx_option pack_idx_opts;
d01fb92f 61static const char *base_name;
024701f1 62static int progress = 1;
4812a93a 63static int window = 10;
568508e7 64static unsigned long pack_size_limit;
618e613a 65static int depth = 50;
43cc2b42 66static int delta_search_threads;
df6d6101 67static int pack_to_stdout;
8d1d8f83 68static int num_preferred_base;
dc6a0757 69static struct progress *progress_state;
c323ac7d 70
6b8fda2d
VM
71static struct packed_git *reuse_packfile;
72static uint32_t reuse_packfile_objects;
73static off_t reuse_packfile_offset;
74
645c432d
KS
75static int use_bitmap_index_default = 1;
76static int use_bitmap_index = -1;
7cc8f971 77static int write_bitmap_index;
ae4f07fb 78static uint16_t write_bitmap_options;
6b8fda2d 79
0c16cd49
JT
80static int exclude_promisor_objects;
81
074b2eea 82static unsigned long delta_cache_size = 0;
5749b0b2 83static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
e3dfddb3 84static unsigned long cache_max_small_delta_size = 1000;
074b2eea 85
a97773ce
BD
86static unsigned long window_memory_limit = 0;
87
9535ce73
JH
88static struct list_objects_filter_options filter_options;
89
90enum missing_action {
0c16cd49
JT
91 MA_ERROR = 0, /* fail if any missing objects are encountered */
92 MA_ALLOW_ANY, /* silently allow ALL missing objects */
93 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
9535ce73
JH
94};
95static enum missing_action arg_missing_action;
96static show_object_fn fn_show_object;
97
3f9ac8d2
JH
98/*
99 * stats
100 */
7cadf491
SP
101static uint32_t written, written_delta;
102static uint32_t reused, reused_delta;
3f9ac8d2 103
7cc8f971
VM
104/*
105 * Indexed commits
106 */
107static struct commit **indexed_commits;
108static unsigned int indexed_commits_nr;
109static unsigned int indexed_commits_alloc;
110
111static void index_commit_for_bitmap(struct commit *commit)
112{
113 if (indexed_commits_nr >= indexed_commits_alloc) {
114 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
2756ca43 115 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
7cc8f971
VM
116 }
117
118 indexed_commits[indexed_commits_nr++] = commit;
119}
780e6e73 120
3613f9b4 121static void *get_delta(struct object_entry *entry)
c323ac7d 122{
3613f9b4
NP
123 unsigned long size, base_size, delta_size;
124 void *buf, *base_buf, *delta_buf;
21666f1a 125 enum object_type type;
c323ac7d 126
e6a492b7 127 buf = read_sha1_file(entry->idx.oid.hash, &type, &size);
3613f9b4 128 if (!buf)
e6a492b7 129 die("unable to read %s", oid_to_hex(&entry->idx.oid));
130 base_buf = read_sha1_file(entry->delta->idx.oid.hash, &type,
131 &base_size);
3613f9b4 132 if (!base_buf)
e6a492b7 133 die("unable to read %s",
134 oid_to_hex(&entry->delta->idx.oid));
3613f9b4 135 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 136 buf, size, &delta_size, 0);
3613f9b4 137 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 138 die("delta size changed");
3613f9b4
NP
139 free(buf);
140 free(base_buf);
c323ac7d
LT
141 return delta_buf;
142}
143
30ebb40a
NP
144static unsigned long do_compress(void **pptr, unsigned long size)
145{
ef49a7a0 146 git_zstream stream;
30ebb40a
NP
147 void *in, *out;
148 unsigned long maxsize;
149
55bb5c91 150 git_deflate_init(&stream, pack_compression_level);
225a6f10 151 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
152
153 in = *pptr;
154 out = xmalloc(maxsize);
155 *pptr = out;
156
157 stream.next_in = in;
158 stream.avail_in = size;
159 stream.next_out = out;
160 stream.avail_out = maxsize;
55bb5c91 161 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 162 ; /* nothing */
55bb5c91 163 git_deflate_end(&stream);
30ebb40a
NP
164
165 free(in);
166 return stream.total_out;
167}
168
cf2ba13a 169static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
188960b4 170 const struct object_id *oid)
cf2ba13a
NTND
171{
172 git_zstream stream;
173 unsigned char ibuf[1024 * 16];
174 unsigned char obuf[1024 * 16];
175 unsigned long olen = 0;
176
cf2ba13a
NTND
177 git_deflate_init(&stream, pack_compression_level);
178
179 for (;;) {
180 ssize_t readlen;
181 int zret = Z_OK;
182 readlen = read_istream(st, ibuf, sizeof(ibuf));
183 if (readlen == -1)
188960b4 184 die(_("unable to read %s"), oid_to_hex(oid));
cf2ba13a
NTND
185
186 stream.next_in = ibuf;
187 stream.avail_in = readlen;
188 while ((stream.avail_in || readlen == 0) &&
189 (zret == Z_OK || zret == Z_BUF_ERROR)) {
190 stream.next_out = obuf;
191 stream.avail_out = sizeof(obuf);
192 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
193 sha1write(f, obuf, stream.next_out - obuf);
194 olen += stream.next_out - obuf;
195 }
196 if (stream.avail_in)
197 die(_("deflate error (%d)"), zret);
198 if (readlen == 0) {
199 if (zret != Z_STREAM_END)
200 die(_("deflate error (%d)"), zret);
201 break;
202 }
203 }
204 git_deflate_end(&stream);
205 return olen;
206}
207
780e6e73
NP
208/*
209 * we are going to reuse the existing object data as is. make
210 * sure it is not corrupt.
211 */
079afb18
SP
212static int check_pack_inflate(struct packed_git *p,
213 struct pack_window **w_curs,
6777a59f
SP
214 off_t offset,
215 off_t len,
079afb18
SP
216 unsigned long expect)
217{
ef49a7a0 218 git_zstream stream;
079afb18
SP
219 unsigned char fakebuf[4096], *in;
220 int st;
221
222 memset(&stream, 0, sizeof(stream));
39c68542 223 git_inflate_init(&stream);
079afb18
SP
224 do {
225 in = use_pack(p, w_curs, offset, &stream.avail_in);
226 stream.next_in = in;
227 stream.next_out = fakebuf;
228 stream.avail_out = sizeof(fakebuf);
39c68542 229 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
230 offset += stream.next_in - in;
231 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 232 git_inflate_end(&stream);
079afb18
SP
233 return (st == Z_STREAM_END &&
234 stream.total_out == expect &&
235 stream.total_in == len) ? 0 : -1;
236}
237
238static void copy_pack_data(struct sha1file *f,
239 struct packed_git *p,
240 struct pack_window **w_curs,
6777a59f
SP
241 off_t offset,
242 off_t len)
079afb18
SP
243{
244 unsigned char *in;
ef49a7a0 245 unsigned long avail;
079afb18
SP
246
247 while (len) {
248 in = use_pack(p, w_curs, offset, &avail);
249 if (avail > len)
ef49a7a0 250 avail = (unsigned long)len;
079afb18
SP
251 sha1write(f, in, avail);
252 offset += avail;
253 len -= avail;
254 }
255}
256
1b4bb16b 257/* Return 0 if we will bust the pack-size limit */
c9018b03
NTND
258static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
259 unsigned long limit, int usable_delta)
c323ac7d 260{
c9018b03 261 unsigned long size, datalen;
2c5e2865
JK
262 unsigned char header[MAX_PACK_OBJECT_HEADER],
263 dheader[MAX_PACK_OBJECT_HEADER];
6777a59f 264 unsigned hdrlen;
2c5ef824 265 enum object_type type;
c9018b03 266 void *buf;
cf2ba13a 267 struct git_istream *st = NULL;
c9018b03
NTND
268
269 if (!usable_delta) {
cf2ba13a
NTND
270 if (entry->type == OBJ_BLOB &&
271 entry->size > big_file_threshold &&
e6a492b7 272 (st = open_istream(entry->idx.oid.hash, &type, &size, NULL)) != NULL)
cf2ba13a
NTND
273 buf = NULL;
274 else {
e6a492b7 275 buf = read_sha1_file(entry->idx.oid.hash, &type,
276 &size);
cf2ba13a 277 if (!buf)
e6a492b7 278 die(_("unable to read %s"),
279 oid_to_hex(&entry->idx.oid));
cf2ba13a 280 }
c9018b03
NTND
281 /*
282 * make sure no cached delta data remains from a
283 * previous attempt before a pack split occurred.
284 */
6a83d902 285 FREE_AND_NULL(entry->delta_data);
c9018b03
NTND
286 entry->z_delta_size = 0;
287 } else if (entry->delta_data) {
288 size = entry->delta_size;
289 buf = entry->delta_data;
290 entry->delta_data = NULL;
291 type = (allow_ofs_delta && entry->delta->idx.offset) ?
292 OBJ_OFS_DELTA : OBJ_REF_DELTA;
293 } else {
294 buf = get_delta(entry);
295 size = entry->delta_size;
296 type = (allow_ofs_delta && entry->delta->idx.offset) ?
297 OBJ_OFS_DELTA : OBJ_REF_DELTA;
298 }
299
cf2ba13a
NTND
300 if (st) /* large blob case, just assume we don't compress well */
301 datalen = size;
302 else if (entry->z_delta_size)
c9018b03
NTND
303 datalen = entry->z_delta_size;
304 else
305 datalen = do_compress(&buf, size);
306
307 /*
308 * The object header is a byte of 'type' followed by zero or
309 * more bytes of length.
310 */
7202a6fa
JK
311 hdrlen = encode_in_pack_object_header(header, sizeof(header),
312 type, size);
c9018b03
NTND
313
314 if (type == OBJ_OFS_DELTA) {
315 /*
316 * Deltas with relative base contain an additional
317 * encoding of the relative offset for the delta
318 * base from this object's position in the pack.
319 */
320 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
321 unsigned pos = sizeof(dheader) - 1;
322 dheader[pos] = ofs & 127;
323 while (ofs >>= 7)
324 dheader[--pos] = 128 | (--ofs & 127);
325 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
cf2ba13a
NTND
326 if (st)
327 close_istream(st);
c9018b03
NTND
328 free(buf);
329 return 0;
330 }
331 sha1write(f, header, hdrlen);
332 sha1write(f, dheader + pos, sizeof(dheader) - pos);
333 hdrlen += sizeof(dheader) - pos;
334 } else if (type == OBJ_REF_DELTA) {
335 /*
336 * Deltas with a base reference contain
337 * an additional 20 bytes for the base sha1.
338 */
339 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
cf2ba13a
NTND
340 if (st)
341 close_istream(st);
c9018b03
NTND
342 free(buf);
343 return 0;
344 }
345 sha1write(f, header, hdrlen);
e6a492b7 346 sha1write(f, entry->delta->idx.oid.hash, 20);
c9018b03
NTND
347 hdrlen += 20;
348 } else {
349 if (limit && hdrlen + datalen + 20 >= limit) {
cf2ba13a
NTND
350 if (st)
351 close_istream(st);
c9018b03
NTND
352 free(buf);
353 return 0;
354 }
355 sha1write(f, header, hdrlen);
356 }
cf2ba13a 357 if (st) {
188960b4 358 datalen = write_large_blob_data(st, f, &entry->idx.oid);
cf2ba13a
NTND
359 close_istream(st);
360 } else {
361 sha1write(f, buf, datalen);
362 free(buf);
363 }
c9018b03
NTND
364
365 return hdrlen + datalen;
366}
367
368/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
369static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
370 unsigned long limit, int usable_delta)
c9018b03
NTND
371{
372 struct packed_git *p = entry->in_pack;
373 struct pack_window *w_curs = NULL;
374 struct revindex_entry *revidx;
375 off_t offset;
376 enum object_type type = entry->type;
211c61c6 377 off_t datalen;
2c5e2865
JK
378 unsigned char header[MAX_PACK_OBJECT_HEADER],
379 dheader[MAX_PACK_OBJECT_HEADER];
c9018b03
NTND
380 unsigned hdrlen;
381
382 if (entry->delta)
383 type = (allow_ofs_delta && entry->delta->idx.offset) ?
384 OBJ_OFS_DELTA : OBJ_REF_DELTA;
7202a6fa
JK
385 hdrlen = encode_in_pack_object_header(header, sizeof(header),
386 type, entry->size);
c9018b03
NTND
387
388 offset = entry->in_pack_offset;
389 revidx = find_pack_revindex(p, offset);
390 datalen = revidx[1].offset - offset;
391 if (!pack_to_stdout && p->index_version > 1 &&
392 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
e6a492b7 393 error("bad packed object CRC for %s",
394 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
395 unuse_pack(&w_curs);
396 return write_no_reuse_object(f, entry, limit, usable_delta);
397 }
398
399 offset += entry->in_pack_header_size;
400 datalen -= entry->in_pack_header_size;
401
402 if (!pack_to_stdout && p->index_version == 1 &&
403 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
e6a492b7 404 error("corrupt packed object for %s",
405 oid_to_hex(&entry->idx.oid));
c9018b03
NTND
406 unuse_pack(&w_curs);
407 return write_no_reuse_object(f, entry, limit, usable_delta);
408 }
409
410 if (type == OBJ_OFS_DELTA) {
411 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
412 unsigned pos = sizeof(dheader) - 1;
413 dheader[pos] = ofs & 127;
414 while (ofs >>= 7)
415 dheader[--pos] = 128 | (--ofs & 127);
416 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
417 unuse_pack(&w_curs);
418 return 0;
419 }
420 sha1write(f, header, hdrlen);
421 sha1write(f, dheader + pos, sizeof(dheader) - pos);
422 hdrlen += sizeof(dheader) - pos;
423 reused_delta++;
424 } else if (type == OBJ_REF_DELTA) {
425 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
426 unuse_pack(&w_curs);
427 return 0;
428 }
429 sha1write(f, header, hdrlen);
e6a492b7 430 sha1write(f, entry->delta->idx.oid.hash, 20);
c9018b03
NTND
431 hdrlen += 20;
432 reused_delta++;
433 } else {
434 if (limit && hdrlen + datalen + 20 >= limit) {
435 unuse_pack(&w_curs);
436 return 0;
437 }
438 sha1write(f, header, hdrlen);
439 }
440 copy_pack_data(f, p, &w_curs, offset, datalen);
441 unuse_pack(&w_curs);
442 reused++;
443 return hdrlen + datalen;
444}
445
446/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
447static off_t write_object(struct sha1file *f,
448 struct object_entry *entry,
449 off_t write_offset)
c9018b03 450{
af92a645
NTND
451 unsigned long limit;
452 off_t len;
2c5ef824 453 int usable_delta, to_reuse;
c323ac7d 454
78d1e84f
NP
455 if (!pack_to_stdout)
456 crc32_begin(f);
457
a2430dde 458 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
459 if (!pack_size_limit || !nr_written)
460 limit = 0;
461 else if (pack_size_limit <= write_offset)
462 /*
463 * the earlier object did not fit the limit; avoid
464 * mistaking this with unlimited (i.e. limit = 0).
465 */
466 limit = 1;
467 else
468 limit = pack_size_limit - write_offset;
2c5ef824
NP
469
470 if (!entry->delta)
471 usable_delta = 0; /* no delta */
472 else if (!pack_size_limit)
473 usable_delta = 1; /* unlimited packfile */
474 else if (entry->delta->idx.offset == (off_t)-1)
475 usable_delta = 0; /* base was written to another pack */
476 else if (entry->delta->idx.offset)
477 usable_delta = 1; /* base already exists in this pack */
478 else
479 usable_delta = 0; /* base could end up in another pack */
480
a7de7130 481 if (!reuse_object)
fa736f72
NP
482 to_reuse = 0; /* explicit */
483 else if (!entry->in_pack)
ab7cd7bb 484 to_reuse = 0; /* can't reuse what we don't have */
c9018b03 485 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
17b08f2c
DH
486 /* check_object() decided it for us ... */
487 to_reuse = usable_delta;
488 /* ... but pack split may override that */
c9018b03 489 else if (entry->type != entry->in_pack_type)
ab7cd7bb
JH
490 to_reuse = 0; /* pack has delta which is unusable */
491 else if (entry->delta)
492 to_reuse = 0; /* we want to pack afresh */
493 else
494 to_reuse = 1; /* we have it in-pack undeltified,
495 * and we do not need to deltify it.
496 */
497
c9018b03
NTND
498 if (!to_reuse)
499 len = write_no_reuse_object(f, entry, limit, usable_delta);
500 else
501 len = write_reuse_object(f, entry, limit, usable_delta);
502 if (!len)
503 return 0;
64bd76b1 504
17b08f2c 505 if (usable_delta)
ab7cd7bb 506 written_delta++;
3f9ac8d2 507 written++;
78d1e84f 508 if (!pack_to_stdout)
aa7e44bf 509 entry->idx.crc32 = crc32_end(f);
c9018b03 510 return len;
c323ac7d
LT
511}
512
f63c79db
JH
513enum write_one_status {
514 WRITE_ONE_SKIP = -1, /* already written */
515 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
516 WRITE_ONE_WRITTEN = 1, /* normal */
517 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
518};
519
520static enum write_one_status write_one(struct sha1file *f,
521 struct object_entry *e,
522 off_t *offset)
9d5ab962 523{
af92a645 524 off_t size;
f63c79db 525 int recursing;
d7dd0223 526
f63c79db
JH
527 /*
528 * we set offset to 1 (which is an impossible value) to mark
529 * the fact that this object is involved in "write its base
530 * first before writing a deltified object" recursion.
531 */
532 recursing = (e->idx.offset == 1);
533 if (recursing) {
534 warning("recursive delta detected for object %s",
e6a492b7 535 oid_to_hex(&e->idx.oid));
f63c79db
JH
536 return WRITE_ONE_RECURSIVE;
537 } else if (e->idx.offset || e->preferred_base) {
538 /* offset is non zero if object is written already. */
539 return WRITE_ONE_SKIP;
540 }
d7dd0223 541
720c9f7b 542 /* if we are deltified, write out base object first. */
f63c79db
JH
543 if (e->delta) {
544 e->idx.offset = 1; /* now recurse */
545 switch (write_one(f, e->delta, offset)) {
546 case WRITE_ONE_RECURSIVE:
547 /* we cannot depend on this one */
548 e->delta = NULL;
549 break;
550 default:
551 break;
552 case WRITE_ONE_BREAK:
553 e->idx.offset = recursing;
554 return WRITE_ONE_BREAK;
555 }
556 }
d7dd0223 557
6ed7f25e
NP
558 e->idx.offset = *offset;
559 size = write_object(f, e, *offset);
17b08f2c 560 if (!size) {
f63c79db
JH
561 e->idx.offset = recursing;
562 return WRITE_ONE_BREAK;
17b08f2c 563 }
79814f42 564 written_list[nr_written++] = &e->idx;
d7dd0223
NP
565
566 /* make sure off_t is sufficiently large not to wrap */
c03c8315 567 if (signed_add_overflows(*offset, size))
d7dd0223 568 die("pack too large for current definition of off_t");
6ed7f25e 569 *offset += size;
f63c79db 570 return WRITE_ONE_WRITTEN;
9d5ab962
JH
571}
572
d155254c 573static int mark_tagged(const char *path, const struct object_id *oid, int flag,
1b4bb16b
JH
574 void *cb_data)
575{
188960b4 576 struct object_id peeled;
d155254c 577 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
1b4bb16b
JH
578
579 if (entry)
580 entry->tagged = 1;
b420d909 581 if (!peel_ref(path, &peeled)) {
188960b4 582 entry = packlist_find(&to_pack, peeled.hash, NULL);
1b4bb16b
JH
583 if (entry)
584 entry->tagged = 1;
585 }
586 return 0;
587}
588
be126818 589static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 590 unsigned int *endp,
1b4bb16b
JH
591 struct object_entry *e)
592{
593 if (e->filled)
594 return;
595 wo[(*endp)++] = e;
596 e->filled = 1;
597}
598
599static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 600 unsigned int *endp,
1b4bb16b
JH
601 struct object_entry *e)
602{
f380872f
DM
603 int add_to_order = 1;
604 while (e) {
605 if (add_to_order) {
606 struct object_entry *s;
607 /* add this node... */
608 add_to_write_order(wo, endp, e);
609 /* all its siblings... */
610 for (s = e->delta_sibling; s; s = s->delta_sibling) {
611 add_to_write_order(wo, endp, s);
612 }
613 }
614 /* drop down a level to add left subtree nodes if possible */
615 if (e->delta_child) {
616 add_to_order = 1;
617 e = e->delta_child;
618 } else {
619 add_to_order = 0;
620 /* our sibling might have some children, it is next */
621 if (e->delta_sibling) {
622 e = e->delta_sibling;
623 continue;
624 }
625 /* go back to our parent node */
626 e = e->delta;
627 while (e && !e->delta_sibling) {
628 /* we're on the right side of a subtree, keep
629 * going up until we can go right again */
630 e = e->delta;
631 }
632 if (!e) {
633 /* done- we hit our original root node */
634 return;
635 }
636 /* pass it off to sibling at this level */
637 e = e->delta_sibling;
638 }
639 };
1b4bb16b
JH
640}
641
642static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 643 unsigned int *endp,
1b4bb16b
JH
644 struct object_entry *e)
645{
646 struct object_entry *root;
647
648 for (root = e; root->delta; root = root->delta)
649 ; /* nothing */
1b4bb16b
JH
650 add_descendants_to_write_order(wo, endp, root);
651}
652
653static struct object_entry **compute_write_order(void)
654{
38d4debb 655 unsigned int i, wo_end, last_untagged;
1b4bb16b 656
b32fa95f 657 struct object_entry **wo;
2834bc27 658 struct object_entry *objects = to_pack.objects;
1b4bb16b 659
2834bc27 660 for (i = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
661 objects[i].tagged = 0;
662 objects[i].filled = 0;
663 objects[i].delta_child = NULL;
664 objects[i].delta_sibling = NULL;
665 }
666
667 /*
668 * Fully connect delta_child/delta_sibling network.
669 * Make sure delta_sibling is sorted in the original
670 * recency order.
671 */
2834bc27 672 for (i = to_pack.nr_objects; i > 0;) {
92bef1a1 673 struct object_entry *e = &objects[--i];
1b4bb16b
JH
674 if (!e->delta)
675 continue;
676 /* Mark me as the first child */
677 e->delta_sibling = e->delta->delta_child;
678 e->delta->delta_child = e;
679 }
680
681 /*
682 * Mark objects that are at the tip of tags.
683 */
d155254c 684 for_each_tag_ref(mark_tagged, NULL);
1b4bb16b
JH
685
686 /*
38d4debb 687 * Give the objects in the original recency order until
1b4bb16b
JH
688 * we see a tagged tip.
689 */
b32fa95f 690 ALLOC_ARRAY(wo, to_pack.nr_objects);
2834bc27 691 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
692 if (objects[i].tagged)
693 break;
694 add_to_write_order(wo, &wo_end, &objects[i]);
695 }
38d4debb 696 last_untagged = i;
1b4bb16b
JH
697
698 /*
699 * Then fill all the tagged tips.
700 */
2834bc27 701 for (; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
702 if (objects[i].tagged)
703 add_to_write_order(wo, &wo_end, &objects[i]);
704 }
705
706 /*
707 * And then all remaining commits and tags.
708 */
2834bc27 709 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
710 if (objects[i].type != OBJ_COMMIT &&
711 objects[i].type != OBJ_TAG)
712 continue;
713 add_to_write_order(wo, &wo_end, &objects[i]);
714 }
715
716 /*
717 * And then all the trees.
718 */
2834bc27 719 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
720 if (objects[i].type != OBJ_TREE)
721 continue;
722 add_to_write_order(wo, &wo_end, &objects[i]);
723 }
724
725 /*
726 * Finally all the rest in really tight order
727 */
2834bc27 728 for (i = last_untagged; i < to_pack.nr_objects; i++) {
38d4debb
DM
729 if (!objects[i].filled)
730 add_family_to_write_order(wo, &wo_end, &objects[i]);
731 }
732
2834bc27
VM
733 if (wo_end != to_pack.nr_objects)
734 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
1b4bb16b
JH
735
736 return wo;
737}
738
6b8fda2d
VM
739static off_t write_reused_pack(struct sha1file *f)
740{
741 unsigned char buffer[8192];
657673f1 742 off_t to_write, total;
6b8fda2d
VM
743 int fd;
744
745 if (!is_pack_valid(reuse_packfile))
746 die("packfile is invalid: %s", reuse_packfile->pack_name);
747
a5436b57 748 fd = git_open(reuse_packfile->pack_name);
6b8fda2d
VM
749 if (fd < 0)
750 die_errno("unable to open packfile for reuse: %s",
751 reuse_packfile->pack_name);
752
753 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
754 die_errno("unable to seek in reused packfile");
755
756 if (reuse_packfile_offset < 0)
757 reuse_packfile_offset = reuse_packfile->pack_size - 20;
758
657673f1 759 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
6b8fda2d
VM
760
761 while (to_write) {
762 int read_pack = xread(fd, buffer, sizeof(buffer));
763
764 if (read_pack <= 0)
765 die_errno("unable to read from reused packfile");
766
767 if (read_pack > to_write)
768 read_pack = to_write;
769
770 sha1write(f, buffer, read_pack);
771 to_write -= read_pack;
657673f1
JK
772
773 /*
774 * We don't know the actual number of objects written,
775 * only how many bytes written, how many bytes total, and
776 * how many objects total. So we can fake it by pretending all
777 * objects we are writing are the same size. This gives us a
778 * smooth progress meter, and at the end it matches the true
779 * answer.
780 */
781 written = reuse_packfile_objects *
782 (((double)(total - to_write)) / total);
783 display_progress(progress_state, written);
6b8fda2d
VM
784 }
785
786 close(fd);
657673f1
JK
787 written = reuse_packfile_objects;
788 display_progress(progress_state, written);
6b8fda2d
VM
789 return reuse_packfile_offset - sizeof(struct pack_header);
790}
791
9cea46cd
EW
792static const char no_split_warning[] = N_(
793"disabling bitmap writing, packs are split due to pack.packSizeLimit"
794);
795
d01fb92f 796static void write_pack_file(void)
c323ac7d 797{
ebe27b13 798 uint32_t i = 0, j;
d22b9290 799 struct sha1file *f;
6ed7f25e 800 off_t offset;
ebe27b13 801 uint32_t nr_remaining = nr_result;
f746bae8 802 time_t last_mtime = 0;
1b4bb16b 803 struct object_entry **write_order;
81a216a5 804
bcd7954e 805 if (progress > pack_to_stdout)
754dbc43 806 progress_state = start_progress(_("Writing objects"), nr_result);
b32fa95f 807 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1b4bb16b 808 write_order = compute_write_order();
183bdb2c 809
ebe27b13 810 do {
188960b4 811 struct object_id oid;
7ba502c4 812 char *pack_tmp_name = NULL;
aa7e44bf 813
cdf9db3c 814 if (pack_to_stdout)
2a128d63 815 f = sha1fd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
816 else
817 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 818
c0ad4657 819 offset = write_pack_header(f, nr_remaining);
6b8fda2d
VM
820
821 if (reuse_packfile) {
822 off_t packfile_size;
823 assert(pack_to_stdout);
824
825 packfile_size = write_reused_pack(f);
826 offset += packfile_size;
827 }
828
ebe27b13 829 nr_written = 0;
2834bc27 830 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 831 struct object_entry *e = write_order[i];
cddec4f8 832 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
833 break;
834 display_progress(progress_state, written);
835 }
c553ca25 836
ebe27b13
DH
837 /*
838 * Did we write the wrong # entries in the header?
839 * If so, rewrite it like in fast-import
840 */
54352bb2 841 if (pack_to_stdout) {
188960b4 842 sha1close(f, oid.hash, CSUM_CLOSE);
54352bb2 843 } else if (nr_written == nr_remaining) {
188960b4 844 sha1close(f, oid.hash, CSUM_FSYNC);
ebe27b13 845 } else {
188960b4 846 int fd = sha1close(f, oid.hash, 0);
847 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
848 nr_written, oid.hash, offset);
7ba502c4 849 close(fd);
9cea46cd
EW
850 if (write_bitmap_index) {
851 warning(_(no_split_warning));
852 write_bitmap_index = 0;
853 }
ebe27b13
DH
854 }
855
856 if (!pack_to_stdout) {
f746bae8 857 struct stat st;
58892711 858 struct strbuf tmpname = STRBUF_INIT;
d01fb92f 859
f746bae8
NP
860 /*
861 * Packs are runtime accessed in their mtime
862 * order since newer packs are more likely to contain
863 * younger objects. So if we are creating multiple
864 * packs then we should modify the mtime of later ones
865 * to preserve this property.
866 */
0e990530 867 if (stat(pack_tmp_name, &st) < 0) {
54d47394 868 warning_errno("failed to stat %s", pack_tmp_name);
f746bae8
NP
869 } else if (!last_mtime) {
870 last_mtime = st.st_mtime;
871 } else {
872 struct utimbuf utb;
873 utb.actime = st.st_atime;
874 utb.modtime = --last_mtime;
0e990530 875 if (utime(pack_tmp_name, &utb) < 0)
54d47394 876 warning_errno("failed utime() on %s", pack_tmp_name);
f746bae8
NP
877 }
878
58892711 879 strbuf_addf(&tmpname, "%s-", base_name);
7cc8f971
VM
880
881 if (write_bitmap_index) {
188960b4 882 bitmap_writer_set_checksum(oid.hash);
7cc8f971
VM
883 bitmap_writer_build_type_index(written_list, nr_written);
884 }
885
58892711 886 finish_tmp_packfile(&tmpname, pack_tmp_name,
0e990530 887 written_list, nr_written,
188960b4 888 &pack_idx_opts, oid.hash);
7cc8f971
VM
889
890 if (write_bitmap_index) {
188960b4 891 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
7cc8f971
VM
892
893 stop_progress(&progress_state);
894
895 bitmap_writer_show_progress(progress);
896 bitmap_writer_reuse_bitmaps(&to_pack);
897 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
898 bitmap_writer_build(&to_pack);
ae4f07fb 899 bitmap_writer_finish(written_list, nr_written,
58892711 900 tmpname.buf, write_bitmap_options);
7cc8f971
VM
901 write_bitmap_index = 0;
902 }
903
58892711 904 strbuf_release(&tmpname);
7ba502c4 905 free(pack_tmp_name);
188960b4 906 puts(oid_to_hex(&oid));
ebe27b13
DH
907 }
908
909 /* mark written objects as written to previous pack */
910 for (j = 0; j < nr_written; j++) {
79814f42 911 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
912 }
913 nr_remaining -= nr_written;
2834bc27 914 } while (nr_remaining && i < to_pack.nr_objects);
ebe27b13
DH
915
916 free(written_list);
1b4bb16b 917 free(write_order);
4d4fcc54 918 stop_progress(&progress_state);
67c08ce1 919 if (written != nr_result)
6e1c2344
RJ
920 die("wrote %"PRIu32" objects while expecting %"PRIu32,
921 written, nr_result);
c323ac7d
LT
922}
923
a74db82e
JH
924static int no_try_delta(const char *path)
925{
2aef63d3 926 static struct attr_check *check;
a74db82e 927
2aef63d3
JH
928 if (!check)
929 check = attr_check_initl("delta", NULL);
930 if (git_check_attr(path, check))
a74db82e 931 return 0;
2aef63d3 932 if (ATTR_FALSE(check->items[0].value))
a74db82e
JH
933 return 1;
934 return 0;
935}
936
ce2bc424
JK
937/*
938 * When adding an object, check whether we have already added it
939 * to our packing list. If so, we can skip. However, if we are
940 * being asked to excludei t, but the previous mention was to include
941 * it, make sure to adjust its flags and tweak our numbers accordingly.
942 *
943 * As an optimization, we pass out the index position where we would have
944 * found the item, since that saves us from having to look it up again a
945 * few lines later when we want to add the new entry.
946 */
188960b4 947static int have_duplicate_entry(const struct object_id *oid,
ce2bc424
JK
948 int exclude,
949 uint32_t *index_pos)
c323ac7d 950{
c323ac7d 951 struct object_entry *entry;
29b734e4 952
188960b4 953 entry = packlist_find(&to_pack, oid->hash, index_pos);
ce2bc424 954 if (!entry)
29b734e4 955 return 0;
ce2bc424
JK
956
957 if (exclude) {
958 if (!entry->preferred_base)
959 nr_result--;
960 entry->preferred_base = 1;
29b734e4 961 }
c323ac7d 962
ce2bc424
JK
963 return 1;
964}
965
702d1b95
KS
966static int want_found_object(int exclude, struct packed_git *p)
967{
968 if (exclude)
969 return 1;
970 if (incremental)
971 return 0;
972
973 /*
974 * When asked to do --local (do not include an object that appears in a
975 * pack we borrow from elsewhere) or --honor-pack-keep (do not include
976 * an object that appears in a pack marked with .keep), finding a pack
977 * that matches the criteria is sufficient for us to decide to omit it.
978 * However, even if this pack does not satisfy the criteria, we need to
979 * make sure no copy of this object appears in _any_ pack that makes us
980 * to omit the object, so we need to check all the packs.
981 *
982 * We can however first check whether these options can possible matter;
983 * if they do not matter we know we want the object in generated pack.
984 * Otherwise, we signal "-1" at the end to tell the caller that we do
985 * not know either way, and it needs to check more packs.
986 */
987 if (!ignore_packed_keep &&
988 (!local || !have_non_local_packs))
989 return 1;
990
991 if (local && !p->pack_local)
992 return 0;
993 if (ignore_packed_keep && p->pack_local && p->pack_keep)
994 return 0;
995
996 /* we don't know yet; keep looking for more packs */
997 return -1;
998}
999
ce2bc424
JK
1000/*
1001 * Check whether we want the object in the pack (e.g., we do not want
1002 * objects found in non-local stores if the "--local" option was used).
1003 *
702d1b95
KS
1004 * If the caller already knows an existing pack it wants to take the object
1005 * from, that is passed in *found_pack and *found_offset; otherwise this
1006 * function finds if there is any pack that has the object and returns the pack
1007 * and its offset in these variables.
ce2bc424 1008 */
188960b4 1009static int want_object_in_pack(const struct object_id *oid,
ce2bc424
JK
1010 int exclude,
1011 struct packed_git **found_pack,
1012 off_t *found_offset)
1013{
702d1b95 1014 int want;
8865859d 1015 struct list_head *pos;
ce2bc424 1016
188960b4 1017 if (!exclude && local && has_loose_object_nonlocal(oid->hash))
daae0625
BC
1018 return 0;
1019
702d1b95
KS
1020 /*
1021 * If we already know the pack object lives in, start checks from that
1022 * pack - in the usual case when neither --local was given nor .keep files
1023 * are present we will determine the answer right now.
1024 */
1025 if (*found_pack) {
1026 want = want_found_object(exclude, *found_pack);
1027 if (want != -1)
1028 return want;
1029 }
a80d72db 1030 list_for_each(pos, get_packed_git_mru(the_repository)) {
ec2dd32c 1031 struct packed_git *p = list_entry(pos, struct packed_git, mru);
702d1b95
KS
1032 off_t offset;
1033
1034 if (p == *found_pack)
1035 offset = *found_offset;
1036 else
188960b4 1037 offset = find_pack_entry_one(oid->hash, p);
702d1b95 1038
5c49c116 1039 if (offset) {
ce2bc424 1040 if (!*found_pack) {
319b678a 1041 if (!is_pack_valid(p))
4c080182 1042 continue;
ce2bc424
JK
1043 *found_offset = offset;
1044 *found_pack = p;
64560374 1045 }
702d1b95 1046 want = want_found_object(exclude, p);
e6e24c94 1047 if (!exclude && want > 0)
a80d72db
SB
1048 list_move(&p->mru,
1049 get_packed_git_mru(the_repository));
702d1b95
KS
1050 if (want != -1)
1051 return want;
64560374
LT
1052 }
1053 }
eb019375 1054
ce2bc424
JK
1055 return 1;
1056}
1057
188960b4 1058static void create_object_entry(const struct object_id *oid,
ce2bc424
JK
1059 enum object_type type,
1060 uint32_t hash,
1061 int exclude,
1062 int no_try_delta,
1063 uint32_t index_pos,
1064 struct packed_git *found_pack,
1065 off_t found_offset)
1066{
1067 struct object_entry *entry;
29b734e4 1068
188960b4 1069 entry = packlist_alloc(&to_pack, oid->hash, index_pos);
27225f2e 1070 entry->hash = hash;
5c49c116
NP
1071 if (type)
1072 entry->type = type;
29b734e4
NP
1073 if (exclude)
1074 entry->preferred_base = 1;
81a216a5
NP
1075 else
1076 nr_result++;
29b734e4
NP
1077 if (found_pack) {
1078 entry->in_pack = found_pack;
1079 entry->in_pack_offset = found_offset;
1080 }
7a979d99 1081
ce2bc424
JK
1082 entry->no_try_delta = no_try_delta;
1083}
29b734e4 1084
373c67da
JK
1085static const char no_closure_warning[] = N_(
1086"disabling bitmap writing, as some objects are not being packed"
1087);
1088
188960b4 1089static int add_object_entry(const struct object_id *oid, enum object_type type,
ce2bc424
JK
1090 const char *name, int exclude)
1091{
702d1b95
KS
1092 struct packed_git *found_pack = NULL;
1093 off_t found_offset = 0;
ce2bc424 1094 uint32_t index_pos;
7a979d99 1095
188960b4 1096 if (have_duplicate_entry(oid, exclude, &index_pos))
ce2bc424 1097 return 0;
a74db82e 1098
188960b4 1099 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
373c67da
JK
1100 /* The pack is missing an object, so it will not have closure */
1101 if (write_bitmap_index) {
1102 warning(_(no_closure_warning));
1103 write_bitmap_index = 0;
1104 }
ce2bc424 1105 return 0;
373c67da 1106 }
29b734e4 1107
188960b4 1108 create_object_entry(oid, type, pack_name_hash(name),
ce2bc424
JK
1109 exclude, name && no_try_delta(name),
1110 index_pos, found_pack, found_offset);
a74db82e 1111
78d2214e 1112 display_progress(progress_state, nr_result);
29b734e4 1113 return 1;
c323ac7d
LT
1114}
1115
20664967 1116static int add_object_entry_from_bitmap(const struct object_id *oid,
6b8fda2d
VM
1117 enum object_type type,
1118 int flags, uint32_t name_hash,
1119 struct packed_git *pack, off_t offset)
1120{
1121 uint32_t index_pos;
1122
188960b4 1123 if (have_duplicate_entry(oid, 0, &index_pos))
6b8fda2d
VM
1124 return 0;
1125
188960b4 1126 if (!want_object_in_pack(oid, 0, &pack, &offset))
702d1b95
KS
1127 return 0;
1128
188960b4 1129 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
6b8fda2d 1130
78d2214e 1131 display_progress(progress_state, nr_result);
29b734e4 1132 return 1;
c323ac7d
LT
1133}
1134
5379a5c5 1135struct pbase_tree_cache {
188960b4 1136 struct object_id oid;
5379a5c5
JH
1137 int ref;
1138 int temporary;
1139 void *tree_data;
1140 unsigned long tree_size;
1141};
1142
1143static struct pbase_tree_cache *(pbase_tree_cache[256]);
188960b4 1144static int pbase_tree_cache_ix(const struct object_id *oid)
5379a5c5 1145{
188960b4 1146 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
5379a5c5
JH
1147}
1148static int pbase_tree_cache_ix_incr(int ix)
1149{
1150 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1151}
1152
1153static struct pbase_tree {
1154 struct pbase_tree *next;
1155 /* This is a phony "cache" entry; we are not
01689909 1156 * going to evict it or find it through _get()
5379a5c5
JH
1157 * mechanism -- this is for the toplevel node that
1158 * would almost always change with any commit.
1159 */
1160 struct pbase_tree_cache pcache;
1161} *pbase_tree;
1162
188960b4 1163static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
5379a5c5
JH
1164{
1165 struct pbase_tree_cache *ent, *nent;
1166 void *data;
1167 unsigned long size;
21666f1a 1168 enum object_type type;
5379a5c5 1169 int neigh;
188960b4 1170 int my_ix = pbase_tree_cache_ix(oid);
5379a5c5
JH
1171 int available_ix = -1;
1172
1173 /* pbase-tree-cache acts as a limited hashtable.
1174 * your object will be found at your index or within a few
1175 * slots after that slot if it is cached.
1176 */
1177 for (neigh = 0; neigh < 8; neigh++) {
1178 ent = pbase_tree_cache[my_ix];
188960b4 1179 if (ent && !oidcmp(&ent->oid, oid)) {
5379a5c5
JH
1180 ent->ref++;
1181 return ent;
1182 }
1183 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1184 ((0 <= available_ix) &&
1185 (!ent && pbase_tree_cache[available_ix])))
1186 available_ix = my_ix;
1187 if (!ent)
1188 break;
1189 my_ix = pbase_tree_cache_ix_incr(my_ix);
1190 }
1191
1192 /* Did not find one. Either we got a bogus request or
1193 * we need to read and perhaps cache.
1194 */
188960b4 1195 data = read_sha1_file(oid->hash, &type, &size);
5379a5c5
JH
1196 if (!data)
1197 return NULL;
21666f1a 1198 if (type != OBJ_TREE) {
5379a5c5
JH
1199 free(data);
1200 return NULL;
1201 }
1202
1203 /* We need to either cache or return a throwaway copy */
1204
1205 if (available_ix < 0)
1206 ent = NULL;
1207 else {
1208 ent = pbase_tree_cache[available_ix];
1209 my_ix = available_ix;
1210 }
1211
1212 if (!ent) {
1213 nent = xmalloc(sizeof(*nent));
1214 nent->temporary = (available_ix < 0);
1215 }
1216 else {
1217 /* evict and reuse */
1218 free(ent->tree_data);
1219 nent = ent;
1220 }
188960b4 1221 oidcpy(&nent->oid, oid);
5379a5c5
JH
1222 nent->tree_data = data;
1223 nent->tree_size = size;
1224 nent->ref = 1;
1225 if (!nent->temporary)
1226 pbase_tree_cache[my_ix] = nent;
1227 return nent;
1228}
1229
1230static void pbase_tree_put(struct pbase_tree_cache *cache)
1231{
1232 if (!cache->temporary) {
1233 cache->ref--;
1234 return;
1235 }
1236 free(cache->tree_data);
1237 free(cache);
1238}
1239
1240static int name_cmp_len(const char *name)
1241{
1242 int i;
1243 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1244 ;
1245 return i;
1246}
1247
1248static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1249 const char *name,
ce0bd642
LT
1250 int cmplen,
1251 const char *fullname)
3f9ac8d2 1252{
4c068a98 1253 struct name_entry entry;
8a5a8d6c 1254 int cmp;
4c068a98
LT
1255
1256 while (tree_entry(tree,&entry)) {
1211be6b
LT
1257 if (S_ISGITLINK(entry.mode))
1258 continue;
0de16337 1259 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1260 memcmp(name, entry.path, cmplen);
1261 if (cmp > 0)
7a979d99 1262 continue;
8a5a8d6c
NP
1263 if (cmp < 0)
1264 return;
5379a5c5 1265 if (name[cmplen] != '/') {
188960b4 1266 add_object_entry(entry.oid,
4d1012c3 1267 object_type(entry.mode),
bc32fed5 1268 fullname, 1);
5379a5c5
JH
1269 return;
1270 }
8a5a8d6c 1271 if (S_ISDIR(entry.mode)) {
7a979d99 1272 struct tree_desc sub;
5379a5c5
JH
1273 struct pbase_tree_cache *tree;
1274 const char *down = name+cmplen+1;
1275 int downlen = name_cmp_len(down);
1276
188960b4 1277 tree = pbase_tree_get(entry.oid);
5379a5c5
JH
1278 if (!tree)
1279 return;
6fda5e51 1280 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1281
ce0bd642 1282 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1283 pbase_tree_put(tree);
1284 }
1285 }
1286}
1d6b38cc 1287
5379a5c5
JH
1288static unsigned *done_pbase_paths;
1289static int done_pbase_paths_num;
1290static int done_pbase_paths_alloc;
1291static int done_pbase_path_pos(unsigned hash)
1292{
1293 int lo = 0;
1294 int hi = done_pbase_paths_num;
1295 while (lo < hi) {
19716b21 1296 int mi = lo + (hi - lo) / 2;
5379a5c5
JH
1297 if (done_pbase_paths[mi] == hash)
1298 return mi;
1299 if (done_pbase_paths[mi] < hash)
1300 hi = mi;
1301 else
1302 lo = mi + 1;
1303 }
1304 return -lo-1;
1305}
1306
1307static int check_pbase_path(unsigned hash)
1308{
c7b07805 1309 int pos = done_pbase_path_pos(hash);
5379a5c5
JH
1310 if (0 <= pos)
1311 return 1;
1312 pos = -pos - 1;
25e19407
DD
1313 ALLOC_GROW(done_pbase_paths,
1314 done_pbase_paths_num + 1,
1315 done_pbase_paths_alloc);
5379a5c5
JH
1316 done_pbase_paths_num++;
1317 if (pos < done_pbase_paths_num)
f331ab9d
RS
1318 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1319 done_pbase_paths_num - pos - 1);
5379a5c5
JH
1320 done_pbase_paths[pos] = hash;
1321 return 0;
1322}
1323
bc32fed5 1324static void add_preferred_base_object(const char *name)
5379a5c5
JH
1325{
1326 struct pbase_tree *it;
8a5a8d6c 1327 int cmplen;
68fb36eb 1328 unsigned hash = pack_name_hash(name);
5379a5c5 1329
8a5a8d6c 1330 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1331 return;
1332
8a5a8d6c 1333 cmplen = name_cmp_len(name);
5379a5c5
JH
1334 for (it = pbase_tree; it; it = it->next) {
1335 if (cmplen == 0) {
188960b4 1336 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
5379a5c5
JH
1337 }
1338 else {
1339 struct tree_desc tree;
6fda5e51 1340 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1341 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1342 }
3f9ac8d2 1343 }
3f9ac8d2
JH
1344}
1345
188960b4 1346static void add_preferred_base(struct object_id *oid)
3f9ac8d2 1347{
5379a5c5
JH
1348 struct pbase_tree *it;
1349 void *data;
1350 unsigned long size;
188960b4 1351 struct object_id tree_oid;
1d6b38cc 1352
8d1d8f83
JH
1353 if (window <= num_preferred_base++)
1354 return;
1355
188960b4 1356 data = read_object_with_reference(oid->hash, tree_type, &size, tree_oid.hash);
5379a5c5 1357 if (!data)
7a979d99 1358 return;
5379a5c5
JH
1359
1360 for (it = pbase_tree; it; it = it->next) {
188960b4 1361 if (!oidcmp(&it->pcache.oid, &tree_oid)) {
5379a5c5
JH
1362 free(data);
1363 return;
1364 }
1365 }
1366
1367 it = xcalloc(1, sizeof(*it));
1368 it->next = pbase_tree;
1369 pbase_tree = it;
1370
188960b4 1371 oidcpy(&it->pcache.oid, &tree_oid);
5379a5c5
JH
1372 it->pcache.tree_data = data;
1373 it->pcache.tree_size = size;
3f9ac8d2
JH
1374}
1375
0ef95f72
NP
1376static void cleanup_preferred_base(void)
1377{
1378 struct pbase_tree *it;
1379 unsigned i;
1380
1381 it = pbase_tree;
1382 pbase_tree = NULL;
1383 while (it) {
1384 struct pbase_tree *this = it;
1385 it = this->next;
1386 free(this->pcache.tree_data);
1387 free(this);
1388 }
1389
1390 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1391 if (!pbase_tree_cache[i])
1392 continue;
1393 free(pbase_tree_cache[i]->tree_data);
6a83d902 1394 FREE_AND_NULL(pbase_tree_cache[i]);
0ef95f72
NP
1395 }
1396
6a83d902 1397 FREE_AND_NULL(done_pbase_paths);
0ef95f72
NP
1398 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1399}
1400
c323ac7d
LT
1401static void check_object(struct object_entry *entry)
1402{
5c49c116 1403 if (entry->in_pack) {
780e6e73 1404 struct packed_git *p = entry->in_pack;
03e79c88 1405 struct pack_window *w_curs = NULL;
5c49c116
NP
1406 const unsigned char *base_ref = NULL;
1407 struct object_entry *base_entry;
1408 unsigned long used, used_0;
ef49a7a0 1409 unsigned long avail;
5c49c116
NP
1410 off_t ofs;
1411 unsigned char *buf, c;
780e6e73 1412
6777a59f 1413 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1414
5c49c116 1415 /*
fa736f72
NP
1416 * We want in_pack_type even if we do not reuse delta
1417 * since non-delta representations could still be reused.
ab7cd7bb 1418 */
09ded04b 1419 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1420 &entry->in_pack_type,
1421 &entry->size);
03d66015
NP
1422 if (used == 0)
1423 goto give_up;
ab7cd7bb 1424
5c49c116
NP
1425 /*
1426 * Determine if this is a delta and if so whether we can
1427 * reuse it or not. Otherwise let's find out as cheaply as
1428 * possible what the actual type and size for this object is.
3f9ac8d2 1429 */
5c49c116
NP
1430 switch (entry->in_pack_type) {
1431 default:
1432 /* Not a delta hence we've already got all we need. */
1433 entry->type = entry->in_pack_type;
1434 entry->in_pack_header_size = used;
03d66015
NP
1435 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1436 goto give_up;
5c49c116
NP
1437 unuse_pack(&w_curs);
1438 return;
1439 case OBJ_REF_DELTA:
a7de7130 1440 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1441 base_ref = use_pack(p, &w_curs,
1442 entry->in_pack_offset + used, NULL);
1443 entry->in_pack_header_size = used + 20;
1444 break;
1445 case OBJ_OFS_DELTA:
1446 buf = use_pack(p, &w_curs,
1447 entry->in_pack_offset + used, NULL);
1448 used_0 = 0;
1449 c = buf[used_0++];
1450 ofs = c & 127;
1451 while (c & 128) {
1452 ofs += 1;
03d66015
NP
1453 if (!ofs || MSB(ofs, 7)) {
1454 error("delta base offset overflow in pack for %s",
e6a492b7 1455 oid_to_hex(&entry->idx.oid));
03d66015
NP
1456 goto give_up;
1457 }
5c49c116
NP
1458 c = buf[used_0++];
1459 ofs = (ofs << 7) + (c & 127);
780e6e73 1460 }
5c49c116 1461 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1462 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1463 error("delta base offset out of bound for %s",
e6a492b7 1464 oid_to_hex(&entry->idx.oid));
03d66015
NP
1465 goto give_up;
1466 }
a7de7130 1467 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1468 struct revindex_entry *revidx;
1469 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1470 if (!revidx)
1471 goto give_up;
3449f8c4
NP
1472 base_ref = nth_packed_object_sha1(p, revidx->nr);
1473 }
5c49c116
NP
1474 entry->in_pack_header_size = used + used_0;
1475 break;
780e6e73 1476 }
780e6e73 1477
2834bc27 1478 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
5c49c116
NP
1479 /*
1480 * If base_ref was set above that means we wish to
1481 * reuse delta data, and we even found that base
1482 * in the list of objects we want to pack. Goodie!
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.
3f9ac8d2 1488 */
780e6e73 1489 entry->type = entry->in_pack_type;
5c49c116 1490 entry->delta = base_entry;
64bd76b1 1491 entry->delta_size = entry->size;
15b4d577
JH
1492 entry->delta_sibling = base_entry->delta_child;
1493 base_entry->delta_child = entry;
5c49c116
NP
1494 unuse_pack(&w_curs);
1495 return;
1496 }
ab7cd7bb 1497
5c49c116
NP
1498 if (entry->type) {
1499 /*
1500 * This must be a delta and we already know what the
1501 * final object type is. Let's extract the actual
1502 * object size from the delta header.
1503 */
1504 entry->size = get_size_from_delta(p, &w_curs,
1505 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1506 if (entry->size == 0)
1507 goto give_up;
5c49c116 1508 unuse_pack(&w_curs);
3f9ac8d2
JH
1509 return;
1510 }
5c49c116
NP
1511
1512 /*
1513 * No choice but to fall back to the recursive delta walk
1514 * with sha1_object_info() to find about the object type
1515 * at this point...
1516 */
03d66015 1517 give_up:
5c49c116 1518 unuse_pack(&w_curs);
36e4d74a 1519 }
3f9ac8d2 1520
e6a492b7 1521 entry->type = sha1_object_info(entry->idx.oid.hash, &entry->size);
6d6f9cdd
SP
1522 /*
1523 * The error condition is checked in prepare_pack(). This is
1524 * to permit a missing preferred base object to be ignored
1525 * as a preferred base. Doing so can result in a larger
1526 * pack file, but the transfer will still take place.
1527 */
3f9ac8d2
JH
1528}
1529
5c49c116
NP
1530static int pack_offset_sort(const void *_a, const void *_b)
1531{
1532 const struct object_entry *a = *(struct object_entry **)_a;
1533 const struct object_entry *b = *(struct object_entry **)_b;
1534
1535 /* avoid filesystem trashing with loose objects */
1536 if (!a->in_pack && !b->in_pack)
e6a492b7 1537 return oidcmp(&a->idx.oid, &b->idx.oid);
5c49c116
NP
1538
1539 if (a->in_pack < b->in_pack)
1540 return -1;
1541 if (a->in_pack > b->in_pack)
1542 return 1;
1543 return a->in_pack_offset < b->in_pack_offset ? -1 :
1544 (a->in_pack_offset > b->in_pack_offset);
1545}
1546
4cf2143e
JK
1547/*
1548 * Drop an on-disk delta we were planning to reuse. Naively, this would
1549 * just involve blanking out the "delta" field, but we have to deal
1550 * with some extra book-keeping:
1551 *
1552 * 1. Removing ourselves from the delta_sibling linked list.
1553 *
1554 * 2. Updating our size/type to the non-delta representation. These were
1555 * either not recorded initially (size) or overwritten with the delta type
1556 * (type) when check_object() decided to reuse the delta.
7dbabbbe
JK
1557 *
1558 * 3. Resetting our delta depth, as we are now a base object.
4cf2143e
JK
1559 */
1560static void drop_reused_delta(struct object_entry *entry)
1561{
1562 struct object_entry **p = &entry->delta->delta_child;
1563 struct object_info oi = OBJECT_INFO_INIT;
1564
1565 while (*p) {
1566 if (*p == entry)
1567 *p = (*p)->delta_sibling;
1568 else
1569 p = &(*p)->delta_sibling;
1570 }
1571 entry->delta = NULL;
7dbabbbe 1572 entry->depth = 0;
4cf2143e
JK
1573
1574 oi.sizep = &entry->size;
1575 oi.typep = &entry->type;
1576 if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1577 /*
1578 * We failed to get the info from this pack for some reason;
1579 * fall back to sha1_object_info, which may find another copy.
1580 * And if that fails, the error will be recorded in entry->type
1581 * and dealt with in prepare_pack().
1582 */
e6a492b7 1583 entry->type = sha1_object_info(entry->idx.oid.hash,
1584 &entry->size);
4cf2143e
JK
1585 }
1586}
1587
1588/*
1589 * Follow the chain of deltas from this entry onward, throwing away any links
1590 * that cause us to hit a cycle (as determined by the DFS state flags in
1591 * the entries).
7dbabbbe
JK
1592 *
1593 * We also detect too-long reused chains that would violate our --depth
1594 * limit.
4cf2143e
JK
1595 */
1596static void break_delta_chains(struct object_entry *entry)
1597{
42b766d7
JK
1598 /*
1599 * The actual depth of each object we will write is stored as an int,
1600 * as it cannot exceed our int "depth" limit. But before we break
1601 * changes based no that limit, we may potentially go as deep as the
1602 * number of objects, which is elsewhere bounded to a uint32_t.
1603 */
1604 uint32_t total_depth;
1605 struct object_entry *cur, *next;
1606
1607 for (cur = entry, total_depth = 0;
1608 cur;
1609 cur = cur->delta, total_depth++) {
1610 if (cur->dfs_state == DFS_DONE) {
1611 /*
1612 * We've already seen this object and know it isn't
1613 * part of a cycle. We do need to append its depth
1614 * to our count.
1615 */
1616 total_depth += cur->depth;
1617 break;
1618 }
4cf2143e 1619
4cf2143e 1620 /*
42b766d7
JK
1621 * We break cycles before looping, so an ACTIVE state (or any
1622 * other cruft which made its way into the state variable)
1623 * is a bug.
4cf2143e 1624 */
42b766d7
JK
1625 if (cur->dfs_state != DFS_NONE)
1626 die("BUG: confusing delta dfs state in first pass: %d",
1627 cur->dfs_state);
4cf2143e 1628
4cf2143e 1629 /*
42b766d7
JK
1630 * Now we know this is the first time we've seen the object. If
1631 * it's not a delta, we're done traversing, but we'll mark it
1632 * done to save time on future traversals.
4cf2143e 1633 */
42b766d7
JK
1634 if (!cur->delta) {
1635 cur->dfs_state = DFS_DONE;
1636 break;
1637 }
4cf2143e 1638
4cf2143e 1639 /*
42b766d7
JK
1640 * Mark ourselves as active and see if the next step causes
1641 * us to cycle to another active object. It's important to do
1642 * this _before_ we loop, because it impacts where we make the
1643 * cut, and thus how our total_depth counter works.
1644 * E.g., We may see a partial loop like:
1645 *
1646 * A -> B -> C -> D -> B
1647 *
1648 * Cutting B->C breaks the cycle. But now the depth of A is
1649 * only 1, and our total_depth counter is at 3. The size of the
1650 * error is always one less than the size of the cycle we
1651 * broke. Commits C and D were "lost" from A's chain.
1652 *
1653 * If we instead cut D->B, then the depth of A is correct at 3.
1654 * We keep all commits in the chain that we examined.
4cf2143e 1655 */
42b766d7
JK
1656 cur->dfs_state = DFS_ACTIVE;
1657 if (cur->delta->dfs_state == DFS_ACTIVE) {
1658 drop_reused_delta(cur);
1659 cur->dfs_state = DFS_DONE;
1660 break;
7dbabbbe 1661 }
42b766d7 1662 }
7dbabbbe 1663
42b766d7
JK
1664 /*
1665 * And now that we've gone all the way to the bottom of the chain, we
1666 * need to clear the active flags and set the depth fields as
1667 * appropriate. Unlike the loop above, which can quit when it drops a
1668 * delta, we need to keep going to look for more depth cuts. So we need
1669 * an extra "next" pointer to keep going after we reset cur->delta.
1670 */
1671 for (cur = entry; cur; cur = next) {
1672 next = cur->delta;
4cf2143e 1673
42b766d7
JK
1674 /*
1675 * We should have a chain of zero or more ACTIVE states down to
1676 * a final DONE. We can quit after the DONE, because either it
1677 * has no bases, or we've already handled them in a previous
1678 * call.
1679 */
1680 if (cur->dfs_state == DFS_DONE)
1681 break;
1682 else if (cur->dfs_state != DFS_ACTIVE)
1683 die("BUG: confusing delta dfs state in second pass: %d",
1684 cur->dfs_state);
4cf2143e 1685
4cf2143e 1686 /*
42b766d7
JK
1687 * If the total_depth is more than depth, then we need to snip
1688 * the chain into two or more smaller chains that don't exceed
1689 * the maximum depth. Most of the resulting chains will contain
1690 * (depth + 1) entries (i.e., depth deltas plus one base), and
1691 * the last chain (i.e., the one containing entry) will contain
1692 * whatever entries are left over, namely
1693 * (total_depth % (depth + 1)) of them.
1694 *
1695 * Since we are iterating towards decreasing depth, we need to
1696 * decrement total_depth as we go, and we need to write to the
1697 * entry what its final depth will be after all of the
1698 * snipping. Since we're snipping into chains of length (depth
1699 * + 1) entries, the final depth of an entry will be its
1700 * original depth modulo (depth + 1). Any time we encounter an
1701 * entry whose final depth is supposed to be zero, we snip it
1702 * from its delta base, thereby making it so.
4cf2143e 1703 */
42b766d7
JK
1704 cur->depth = (total_depth--) % (depth + 1);
1705 if (!cur->depth)
1706 drop_reused_delta(cur);
1707
1708 cur->dfs_state = DFS_DONE;
4cf2143e
JK
1709 }
1710}
1711
c323ac7d
LT
1712static void get_object_details(void)
1713{
7cadf491 1714 uint32_t i;
5c49c116
NP
1715 struct object_entry **sorted_by_offset;
1716
2834bc27
VM
1717 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1718 for (i = 0; i < to_pack.nr_objects; i++)
1719 sorted_by_offset[i] = to_pack.objects + i;
9ed0d8d6 1720 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
c323ac7d 1721
2834bc27 1722 for (i = 0; i < to_pack.nr_objects; i++) {
15366280
JH
1723 struct object_entry *entry = sorted_by_offset[i];
1724 check_object(entry);
754980d0 1725 if (big_file_threshold < entry->size)
15366280
JH
1726 entry->no_try_delta = 1;
1727 }
3449f8c4 1728
4cf2143e
JK
1729 /*
1730 * This must happen in a second pass, since we rely on the delta
1731 * information for the whole list being completed.
1732 */
1733 for (i = 0; i < to_pack.nr_objects; i++)
1734 break_delta_chains(&to_pack.objects[i]);
1735
5c49c116 1736 free(sorted_by_offset);
c323ac7d
LT
1737}
1738
b904166c
NP
1739/*
1740 * We search for deltas in a list sorted by type, by filename hash, and then
1741 * by size, so that we see progressively smaller and smaller files.
1742 * That's because we prefer deltas to be from the bigger file
1743 * to the smaller -- deletes are potentially cheaper, but perhaps
1744 * more importantly, the bigger file is likely the more recent
1745 * one. The deepest deltas are therefore the oldest objects which are
1746 * less susceptible to be accessed often.
1747 */
9668cf59 1748static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1749{
9668cf59
NP
1750 const struct object_entry *a = *(struct object_entry **)_a;
1751 const struct object_entry *b = *(struct object_entry **)_b;
1752
c323ac7d 1753 if (a->type > b->type)
27225f2e 1754 return -1;
b904166c 1755 if (a->type < b->type)
27225f2e 1756 return 1;
b904166c 1757 if (a->hash > b->hash)
7a979d99 1758 return -1;
b904166c 1759 if (a->hash < b->hash)
7a979d99 1760 return 1;
b904166c 1761 if (a->preferred_base > b->preferred_base)
c323ac7d 1762 return -1;
b904166c
NP
1763 if (a->preferred_base < b->preferred_base)
1764 return 1;
c323ac7d 1765 if (a->size > b->size)
b904166c
NP
1766 return -1;
1767 if (a->size < b->size)
c323ac7d 1768 return 1;
b904166c 1769 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1770}
1771
1772struct unpacked {
1773 struct object_entry *entry;
1774 void *data;
f6c7081a 1775 struct delta_index *index;
5a235b5e 1776 unsigned depth;
c323ac7d
LT
1777};
1778
d250626c
NP
1779static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1780 unsigned long delta_size)
074b2eea
MK
1781{
1782 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1783 return 0;
1784
e3dfddb3
MK
1785 if (delta_size < cache_max_small_delta_size)
1786 return 1;
1787
074b2eea
MK
1788 /* cache delta, if objects are large enough compared to delta size */
1789 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1790 return 1;
1791
1792 return 0;
1793}
1794
7eb151d6 1795#ifndef NO_PTHREADS
8ecce684 1796
44626dc7 1797static pthread_mutex_t read_mutex;
8ecce684
NP
1798#define read_lock() pthread_mutex_lock(&read_mutex)
1799#define read_unlock() pthread_mutex_unlock(&read_mutex)
1800
44626dc7 1801static pthread_mutex_t cache_mutex;
3c701839
NP
1802#define cache_lock() pthread_mutex_lock(&cache_mutex)
1803#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1804
44626dc7 1805static pthread_mutex_t progress_mutex;
8ecce684
NP
1806#define progress_lock() pthread_mutex_lock(&progress_mutex)
1807#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1808
1809#else
1810
e1ef8673
JH
1811#define read_lock() (void)0
1812#define read_unlock() (void)0
1813#define cache_lock() (void)0
1814#define cache_unlock() (void)0
1815#define progress_lock() (void)0
1816#define progress_unlock() (void)0
8ecce684
NP
1817
1818#endif
1819
f6c7081a 1820static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1821 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1822{
f6c7081a
NP
1823 struct object_entry *trg_entry = trg->entry;
1824 struct object_entry *src_entry = src->entry;
560b25a8 1825 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1826 unsigned ref_depth;
21666f1a 1827 enum object_type type;
c323ac7d
LT
1828 void *delta_buf;
1829
1830 /* Don't bother doing diffs between different types */
f6c7081a 1831 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1832 return -1;
1833
51d1e83f 1834 /*
15f07e06
JK
1835 * We do not bother to try a delta that we discarded on an
1836 * earlier try, but only when reusing delta data. Note that
1837 * src_entry that is marked as the preferred_base should always
1838 * be considered, as even if we produce a suboptimal delta against
1839 * it, we will still save the transfer cost, as we already know
1840 * the other side has it and we won't send src_entry at all.
51d1e83f 1841 */
a7de7130 1842 if (reuse_delta && trg_entry->in_pack &&
e9195b58 1843 trg_entry->in_pack == src_entry->in_pack &&
15f07e06 1844 !src_entry->preferred_base &&
e9195b58
JH
1845 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1846 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1847 return 0;
1848
898b14ce 1849 /* Let's not bust the allowed depth. */
5a235b5e 1850 if (src->depth >= max_depth)
d116a45a 1851 return 0;
c323ac7d 1852
c3b06a69 1853 /* Now some size filtering heuristics. */
560b25a8 1854 trg_size = trg_entry->size;
c83f032e
NP
1855 if (!trg_entry->delta) {
1856 max_size = trg_size/2 - 20;
1857 ref_depth = 1;
1858 } else {
1859 max_size = trg_entry->delta_size;
5a235b5e 1860 ref_depth = trg->depth;
c83f032e 1861 }
720fe22d 1862 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1863 (max_depth - ref_depth + 1);
c3b06a69
NP
1864 if (max_size == 0)
1865 return 0;
f6c7081a 1866 src_size = src_entry->size;
560b25a8 1867 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1868 if (sizediff >= max_size)
f527cb8c 1869 return 0;
a1dab41a
BD
1870 if (trg_size < src_size / 32)
1871 return 0;
f6c7081a 1872
560b25a8
NP
1873 /* Load data if not already done */
1874 if (!trg->data) {
8ecce684 1875 read_lock();
e6a492b7 1876 trg->data = read_sha1_file(trg_entry->idx.oid.hash, &type,
1877 &sz);
8ecce684 1878 read_unlock();
2e3404c3
JH
1879 if (!trg->data)
1880 die("object %s cannot be read",
e6a492b7 1881 oid_to_hex(&trg_entry->idx.oid));
560b25a8
NP
1882 if (sz != trg_size)
1883 die("object %s inconsistent object length (%lu vs %lu)",
e6a492b7 1884 oid_to_hex(&trg_entry->idx.oid), sz,
1885 trg_size);
ef0316fc 1886 *mem_usage += sz;
560b25a8
NP
1887 }
1888 if (!src->data) {
8ecce684 1889 read_lock();
e6a492b7 1890 src->data = read_sha1_file(src_entry->idx.oid.hash, &type,
1891 &sz);
8ecce684 1892 read_unlock();
71064a95
NP
1893 if (!src->data) {
1894 if (src_entry->preferred_base) {
1895 static int warned = 0;
1896 if (!warned++)
1897 warning("object %s cannot be read",
e6a492b7 1898 oid_to_hex(&src_entry->idx.oid));
71064a95
NP
1899 /*
1900 * Those objects are not included in the
1901 * resulting pack. Be resilient and ignore
1902 * them if they can't be read, in case the
1903 * pack could be created nevertheless.
1904 */
1905 return 0;
1906 }
2e3404c3 1907 die("object %s cannot be read",
e6a492b7 1908 oid_to_hex(&src_entry->idx.oid));
71064a95 1909 }
560b25a8
NP
1910 if (sz != src_size)
1911 die("object %s inconsistent object length (%lu vs %lu)",
e6a492b7 1912 oid_to_hex(&src_entry->idx.oid), sz,
1913 src_size);
ef0316fc 1914 *mem_usage += sz;
560b25a8
NP
1915 }
1916 if (!src->index) {
1917 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1918 if (!src->index) {
1919 static int warned = 0;
1920 if (!warned++)
1921 warning("suboptimal pack - out of memory");
1922 return 0;
1923 }
ef0316fc 1924 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1925 }
1926
1927 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1928 if (!delta_buf)
75c42d8c 1929 return 0;
f6c7081a 1930
9e2d57a0 1931 if (trg_entry->delta) {
848d732c
BD
1932 /* Prefer only shallower same-sized deltas. */
1933 if (delta_size == trg_entry->delta_size &&
5a235b5e 1934 src->depth + 1 >= trg->depth) {
848d732c
BD
1935 free(delta_buf);
1936 return 0;
1937 }
074b2eea 1938 }
9e2d57a0 1939
3c701839
NP
1940 /*
1941 * Handle memory allocation outside of the cache
1942 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1943 * away when NO_PTHREADS is defined.
3c701839 1944 */
8e0f7003 1945 free(trg_entry->delta_data);
3c701839 1946 cache_lock();
9e2d57a0
NP
1947 if (trg_entry->delta_data) {
1948 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1949 trg_entry->delta_data = NULL;
1950 }
d250626c 1951 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1952 delta_cache_size += delta_size;
3c701839
NP
1953 cache_unlock();
1954 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1955 } else {
1956 cache_unlock();
074b2eea 1957 free(delta_buf);
3c701839
NP
1958 }
1959
b7a28f78
NP
1960 trg_entry->delta = src_entry;
1961 trg_entry->delta_size = delta_size;
1962 trg->depth = src->depth + 1;
1963
f6c7081a 1964 return 1;
c323ac7d
LT
1965}
1966
898b14ce 1967static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1968{
898b14ce
NP
1969 struct object_entry *child = me->delta_child;
1970 unsigned int m = n;
1971 while (child) {
1972 unsigned int c = check_delta_limit(child, n + 1);
1973 if (m < c)
1974 m = c;
1975 child = child->delta_sibling;
1976 }
1977 return m;
b2504a0d
NP
1978}
1979
75ad235c 1980static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1981{
ef0316fc 1982 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1983 free_delta_index(n->index);
1984 n->index = NULL;
1985 if (n->data) {
ef0316fc 1986 freed_mem += n->entry->size;
6a83d902 1987 FREE_AND_NULL(n->data);
a97773ce
BD
1988 }
1989 n->entry = NULL;
7d7baa5e 1990 n->depth = 0;
ef0316fc 1991 return freed_mem;
a97773ce
BD
1992}
1993
384b32c0 1994static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1995 int window, int depth, unsigned *processed)
c323ac7d 1996{
384b32c0 1997 uint32_t i, idx = 0, count = 0;
7cadf491 1998 struct unpacked *array;
ef0316fc 1999 unsigned long mem_usage = 0;
c323ac7d 2000
19d4b416 2001 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 2002
384b32c0 2003 for (;;) {
421b488a 2004 struct object_entry *entry;
c323ac7d 2005 struct unpacked *n = array + idx;
ef0316fc 2006 int j, max_depth, best_base = -1;
c323ac7d 2007
384b32c0
NP
2008 progress_lock();
2009 if (!*list_size) {
2010 progress_unlock();
2011 break;
2012 }
421b488a 2013 entry = *list++;
384b32c0
NP
2014 (*list_size)--;
2015 if (!entry->preferred_base) {
2016 (*processed)++;
2017 display_progress(progress_state, *processed);
2018 }
2019 progress_unlock();
2020
ef0316fc 2021 mem_usage -= free_unpacked(n);
c323ac7d 2022 n->entry = entry;
ab7cd7bb 2023
a97773ce 2024 while (window_memory_limit &&
ef0316fc 2025 mem_usage > window_memory_limit &&
a97773ce
BD
2026 count > 1) {
2027 uint32_t tail = (idx + window - count) % window;
75ad235c 2028 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
2029 count--;
2030 }
2031
75d39853
NP
2032 /* We do not compute delta to *create* objects we are not
2033 * going to pack.
2034 */
2035 if (entry->preferred_base)
2036 goto next;
2037
898b14ce
NP
2038 /*
2039 * If the current object is at pack edge, take the depth the
2040 * objects that depend on the current object into account
2041 * otherwise they would become too deep.
2042 */
2043 max_depth = depth;
2044 if (entry->delta_child) {
2045 max_depth -= check_delta_limit(entry, 0);
2046 if (max_depth <= 0)
2047 goto next;
2048 }
2049
78817c15
LT
2050 j = window;
2051 while (--j > 0) {
77639870 2052 int ret;
7cadf491 2053 uint32_t other_idx = idx + j;
c323ac7d 2054 struct unpacked *m;
78817c15
LT
2055 if (other_idx >= window)
2056 other_idx -= window;
c323ac7d
LT
2057 m = array + other_idx;
2058 if (!m->entry)
2059 break;
ef0316fc 2060 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 2061 if (ret < 0)
c323ac7d 2062 break;
77639870
JH
2063 else if (ret > 0)
2064 best_base = other_idx;
c323ac7d 2065 }
898b14ce 2066
ed4a9031
NP
2067 /*
2068 * If we decided to cache the delta data, then it is best
2069 * to compress it right away. First because we have to do
2070 * it anyway, and doing it here while we're threaded will
2071 * save a lot of time in the non threaded write phase,
2072 * as well as allow for caching more deltas within
2073 * the same cache size limit.
2074 * ...
2075 * But only if not writing to stdout, since in that case
2076 * the network is most likely throttling writes anyway,
2077 * and therefore it is best to go to the write phase ASAP
2078 * instead, as we can afford spending more time compressing
2079 * between writes at that moment.
2080 */
2081 if (entry->delta_data && !pack_to_stdout) {
2082 entry->z_delta_size = do_compress(&entry->delta_data,
2083 entry->delta_size);
2084 cache_lock();
2085 delta_cache_size -= entry->delta_size;
2086 delta_cache_size += entry->z_delta_size;
2087 cache_unlock();
2088 }
2089
70ca1a3f
JH
2090 /* if we made n a delta, and if n is already at max
2091 * depth, leaving it in the window is pointless. we
2092 * should evict it first.
70ca1a3f 2093 */
70baf5d4 2094 if (entry->delta && max_depth <= n->depth)
70ca1a3f 2095 continue;
ff45715c 2096
77639870
JH
2097 /*
2098 * Move the best delta base up in the window, after the
2099 * currently deltified object, to keep it longer. It will
2100 * be the first base object to be attempted next.
2101 */
2102 if (entry->delta) {
2103 struct unpacked swap = array[best_base];
2104 int dist = (window + idx - best_base) % window;
2105 int dst = best_base;
2106 while (dist--) {
2107 int src = (dst + 1) % window;
2108 array[dst] = array[src];
2109 dst = src;
2110 }
2111 array[dst] = swap;
2112 }
2113
898b14ce 2114 next:
521a4f4c 2115 idx++;
a97773ce
BD
2116 if (count + 1 < window)
2117 count++;
521a4f4c
LT
2118 if (idx >= window)
2119 idx = 0;
384b32c0 2120 }
adee7bdf 2121
f6c7081a 2122 for (i = 0; i < window; ++i) {
ff45715c 2123 free_delta_index(array[i].index);
adee7bdf 2124 free(array[i].data);
f6c7081a 2125 }
adee7bdf 2126 free(array);
c323ac7d
LT
2127}
2128
7eb151d6 2129#ifndef NO_PTHREADS
8ecce684 2130
a9a74636
NP
2131static void try_to_free_from_threads(size_t size)
2132{
2133 read_lock();
7c3ecb32 2134 release_pack_memory(size);
a9a74636
NP
2135 read_unlock();
2136}
2137
bc9b2175 2138static try_to_free_t old_try_to_free_routine;
851c34b0 2139
50f22ada
JS
2140/*
2141 * The main thread waits on the condition that (at least) one of the workers
2142 * has stopped working (which is indicated in the .working member of
2143 * struct thread_params).
2144 * When a work thread has completed its work, it sets .working to 0 and
2145 * signals the main thread and waits on the condition that .data_ready
2146 * becomes 1.
2147 */
2148
8ecce684
NP
2149struct thread_params {
2150 pthread_t thread;
2151 struct object_entry **list;
2152 unsigned list_size;
384b32c0 2153 unsigned remaining;
8ecce684
NP
2154 int window;
2155 int depth;
50f22ada
JS
2156 int working;
2157 int data_ready;
2158 pthread_mutex_t mutex;
2159 pthread_cond_t cond;
8ecce684
NP
2160 unsigned *processed;
2161};
2162
44626dc7
AH
2163static pthread_cond_t progress_cond;
2164
2165/*
2166 * Mutex and conditional variable can't be statically-initialized on Windows.
2167 */
2168static void init_threaded_search(void)
2169{
93749194 2170 init_recursive_mutex(&read_mutex);
44626dc7
AH
2171 pthread_mutex_init(&cache_mutex, NULL);
2172 pthread_mutex_init(&progress_mutex, NULL);
2173 pthread_cond_init(&progress_cond, NULL);
851c34b0 2174 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
2175}
2176
2177static void cleanup_threaded_search(void)
2178{
851c34b0 2179 set_try_to_free_routine(old_try_to_free_routine);
44626dc7
AH
2180 pthread_cond_destroy(&progress_cond);
2181 pthread_mutex_destroy(&read_mutex);
2182 pthread_mutex_destroy(&cache_mutex);
2183 pthread_mutex_destroy(&progress_mutex);
2184}
c2a33679 2185
8ecce684
NP
2186static void *threaded_find_deltas(void *arg)
2187{
c2a33679
NP
2188 struct thread_params *me = arg;
2189
0c2ad00b 2190 progress_lock();
50f22ada 2191 while (me->remaining) {
0c2ad00b
2192 progress_unlock();
2193
384b32c0 2194 find_deltas(me->list, &me->remaining,
c2a33679 2195 me->window, me->depth, me->processed);
50f22ada
JS
2196
2197 progress_lock();
2198 me->working = 0;
2199 pthread_cond_signal(&progress_cond);
2200 progress_unlock();
2201
2202 /*
2203 * We must not set ->data_ready before we wait on the
2204 * condition because the main thread may have set it to 1
2205 * before we get here. In order to be sure that new
2206 * work is available if we see 1 in ->data_ready, it
2207 * was initialized to 0 before this thread was spawned
2208 * and we reset it to 0 right away.
2209 */
2210 pthread_mutex_lock(&me->mutex);
2211 while (!me->data_ready)
2212 pthread_cond_wait(&me->cond, &me->mutex);
2213 me->data_ready = 0;
2214 pthread_mutex_unlock(&me->mutex);
0c2ad00b
2215
2216 progress_lock();
c2a33679 2217 }
0c2ad00b 2218 progress_unlock();
50f22ada
JS
2219 /* leave ->working 1 so that this doesn't get more work assigned */
2220 return NULL;
8ecce684
NP
2221}
2222
8ecce684
NP
2223static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2224 int window, int depth, unsigned *processed)
2225{
dcda3614 2226 struct thread_params *p;
384b32c0 2227 int i, ret, active_threads = 0;
c2a33679 2228
44626dc7
AH
2229 init_threaded_search();
2230
367f4a43 2231 if (delta_search_threads <= 1) {
384b32c0 2232 find_deltas(list, &list_size, window, depth, processed);
44626dc7 2233 cleanup_threaded_search();
367f4a43
NP
2234 return;
2235 }
43cc2b42 2236 if (progress > pack_to_stdout)
b6c29915 2237 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 2238 delta_search_threads);
dcda3614 2239 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 2240
50f22ada 2241 /* Partition the work amongst work threads. */
367f4a43 2242 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
2243 unsigned sub_size = list_size / (delta_search_threads - i);
2244
bf874896
NP
2245 /* don't use too small segments or no deltas will be found */
2246 if (sub_size < 2*window && i+1 < delta_search_threads)
2247 sub_size = 0;
2248
8ecce684
NP
2249 p[i].window = window;
2250 p[i].depth = depth;
2251 p[i].processed = processed;
50f22ada
JS
2252 p[i].working = 1;
2253 p[i].data_ready = 0;
c2a33679 2254
59921b4b 2255 /* try to split chunks on "path" boundaries */
6fc74703
NP
2256 while (sub_size && sub_size < list_size &&
2257 list[sub_size]->hash &&
384b32c0
NP
2258 list[sub_size]->hash == list[sub_size-1]->hash)
2259 sub_size++;
2260
50f22ada
JS
2261 p[i].list = list;
2262 p[i].list_size = sub_size;
2263 p[i].remaining = sub_size;
59921b4b 2264
384b32c0
NP
2265 list += sub_size;
2266 list_size -= sub_size;
2267 }
2268
50f22ada
JS
2269 /* Start work threads. */
2270 for (i = 0; i < delta_search_threads; i++) {
2271 if (!p[i].list_size)
2272 continue;
68e6a4f8
JS
2273 pthread_mutex_init(&p[i].mutex, NULL);
2274 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
2275 ret = pthread_create(&p[i].thread, NULL,
2276 threaded_find_deltas, &p[i]);
2277 if (ret)
2278 die("unable to create thread: %s", strerror(ret));
2279 active_threads++;
2280 }
2281
384b32c0
NP
2282 /*
2283 * Now let's wait for work completion. Each time a thread is done
2284 * with its work, we steal half of the remaining work from the
2285 * thread with the largest number of unprocessed objects and give
2286 * it to that newly idle thread. This ensure good load balancing
2287 * until the remaining object list segments are simply too short
2288 * to be worth splitting anymore.
2289 */
50f22ada
JS
2290 while (active_threads) {
2291 struct thread_params *target = NULL;
384b32c0
NP
2292 struct thread_params *victim = NULL;
2293 unsigned sub_size = 0;
384b32c0
NP
2294
2295 progress_lock();
50f22ada
JS
2296 for (;;) {
2297 for (i = 0; !target && i < delta_search_threads; i++)
2298 if (!p[i].working)
2299 target = &p[i];
2300 if (target)
2301 break;
2302 pthread_cond_wait(&progress_cond, &progress_mutex);
2303 }
2304
384b32c0
NP
2305 for (i = 0; i < delta_search_threads; i++)
2306 if (p[i].remaining > 2*window &&
2307 (!victim || victim->remaining < p[i].remaining))
2308 victim = &p[i];
2309 if (victim) {
2310 sub_size = victim->remaining / 2;
2311 list = victim->list + victim->list_size - sub_size;
2312 while (sub_size && list[0]->hash &&
2313 list[0]->hash == list[-1]->hash) {
2314 list++;
2315 sub_size--;
2316 }
eb9688ff
NP
2317 if (!sub_size) {
2318 /*
2319 * It is possible for some "paths" to have
2320 * so many objects that no hash boundary
2321 * might be found. Let's just steal the
2322 * exact half in that case.
2323 */
2324 sub_size = victim->remaining / 2;
2325 list -= sub_size;
2326 }
384b32c0
NP
2327 target->list = list;
2328 victim->list_size -= sub_size;
2329 victim->remaining -= sub_size;
2330 }
384b32c0
NP
2331 target->list_size = sub_size;
2332 target->remaining = sub_size;
50f22ada
JS
2333 target->working = 1;
2334 progress_unlock();
2335
2336 pthread_mutex_lock(&target->mutex);
2337 target->data_ready = 1;
2338 pthread_cond_signal(&target->cond);
2339 pthread_mutex_unlock(&target->mutex);
c2a33679 2340
384b32c0 2341 if (!sub_size) {
b81d9af7 2342 pthread_join(target->thread, NULL);
50f22ada
JS
2343 pthread_cond_destroy(&target->cond);
2344 pthread_mutex_destroy(&target->mutex);
384b32c0 2345 active_threads--;
c2a33679 2346 }
50f22ada 2347 }
44626dc7 2348 cleanup_threaded_search();
dcda3614 2349 free(p);
8ecce684
NP
2350}
2351
2352#else
384b32c0 2353#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
2354#endif
2355
b773ddea
JK
2356static void add_tag_chain(const struct object_id *oid)
2357{
2358 struct tag *tag;
2359
2360 /*
2361 * We catch duplicates already in add_object_entry(), but we'd
2362 * prefer to do this extra check to avoid having to parse the
2363 * tag at all if we already know that it's being packed (e.g., if
2364 * it was included via bitmaps, we would not have parsed it
2365 * previously).
2366 */
2367 if (packlist_find(&to_pack, oid->hash, NULL))
2368 return;
2369
d3101b53 2370 tag = lookup_tag(oid);
b773ddea
JK
2371 while (1) {
2372 if (!tag || parse_tag(tag) || !tag->tagged)
2373 die("unable to pack objects reachable from tag %s",
2374 oid_to_hex(oid));
2375
188960b4 2376 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
b773ddea
JK
2377
2378 if (tag->tagged->type != OBJ_TAG)
2379 return;
2380
2381 tag = (struct tag *)tag->tagged;
2382 }
2383}
2384
d155254c 2385static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
f0a24aa5 2386{
d155254c 2387 struct object_id peeled;
f0a24aa5 2388
59556548 2389 if (starts_with(path, "refs/tags/") && /* is a tag? */
b420d909 2390 !peel_ref(path, &peeled) && /* peelable? */
d155254c 2391 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
b773ddea 2392 add_tag_chain(oid);
f0a24aa5
SP
2393 return 0;
2394}
2395
f3123c4a
JH
2396static void prepare_pack(int window, int depth)
2397{
9668cf59 2398 struct object_entry **delta_list;
6e1c2344
RJ
2399 uint32_t i, nr_deltas;
2400 unsigned n;
9668cf59 2401
3f9ac8d2 2402 get_object_details();
9668cf59 2403
0e8189e2
NP
2404 /*
2405 * If we're locally repacking then we need to be doubly careful
2406 * from now on in order to make sure no stealth corruption gets
2407 * propagated to the new pack. Clients receiving streamed packs
2408 * should validate everything they get anyway so no need to incur
2409 * the additional cost here in that case.
2410 */
2411 if (!pack_to_stdout)
2412 do_check_packed_object_crc = 1;
2413
2834bc27 2414 if (!to_pack.nr_objects || !window || !depth)
9668cf59
NP
2415 return;
2416
b32fa95f 2417 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
75d39853
NP
2418 nr_deltas = n = 0;
2419
2834bc27
VM
2420 for (i = 0; i < to_pack.nr_objects; i++) {
2421 struct object_entry *entry = to_pack.objects + i;
75d39853
NP
2422
2423 if (entry->delta)
2424 /* This happens if we decided to reuse existing
a7de7130 2425 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2426 */
2427 continue;
2428
2429 if (entry->size < 50)
2430 continue;
2431
2432 if (entry->no_try_delta)
2433 continue;
2434
6d6f9cdd 2435 if (!entry->preferred_base) {
75d39853 2436 nr_deltas++;
6d6f9cdd
SP
2437 if (entry->type < 0)
2438 die("unable to get type of object %s",
e6a492b7 2439 oid_to_hex(&entry->idx.oid));
eede9f42
NP
2440 } else {
2441 if (entry->type < 0) {
2442 /*
2443 * This object is not found, but we
2444 * don't have to include it anyway.
2445 */
2446 continue;
2447 }
6d6f9cdd 2448 }
75d39853
NP
2449
2450 delta_list[n++] = entry;
2451 }
2452
2f8b8947 2453 if (nr_deltas && n > 1) {
e334977d
NP
2454 unsigned nr_done = 0;
2455 if (progress)
754dbc43 2456 progress_state = start_progress(_("Compressing objects"),
dc6a0757 2457 nr_deltas);
9ed0d8d6 2458 QSORT(delta_list, n, type_size_sort);
8ecce684 2459 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2460 stop_progress(&progress_state);
e334977d
NP
2461 if (nr_done != nr_deltas)
2462 die("inconsistency with delta count");
75d39853 2463 }
9668cf59 2464 free(delta_list);
f3123c4a
JH
2465}
2466
ef90d6d4 2467static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2468{
eeefa7c9 2469 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2470 window = git_config_int(k, v);
2471 return 0;
2472 }
a97773ce
BD
2473 if (!strcmp(k, "pack.windowmemory")) {
2474 window_memory_limit = git_config_ulong(k, v);
2475 return 0;
2476 }
2477 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2478 depth = git_config_int(k, v);
2479 return 0;
2480 }
074b2eea
MK
2481 if (!strcmp(k, "pack.deltacachesize")) {
2482 max_delta_cache_size = git_config_int(k, v);
2483 return 0;
2484 }
e3dfddb3
MK
2485 if (!strcmp(k, "pack.deltacachelimit")) {
2486 cache_max_small_delta_size = git_config_int(k, v);
2487 return 0;
2488 }
ae4f07fb
VM
2489 if (!strcmp(k, "pack.writebitmaphashcache")) {
2490 if (git_config_bool(k, v))
2491 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2492 else
2493 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2494 }
6b8fda2d 2495 if (!strcmp(k, "pack.usebitmaps")) {
645c432d 2496 use_bitmap_index_default = git_config_bool(k, v);
6b8fda2d
VM
2497 return 0;
2498 }
693b86ff
NP
2499 if (!strcmp(k, "pack.threads")) {
2500 delta_search_threads = git_config_int(k, v);
833e3df1 2501 if (delta_search_threads < 0)
693b86ff
NP
2502 die("invalid number of threads specified (%d)",
2503 delta_search_threads);
7eb151d6 2504#ifdef NO_PTHREADS
2e96d815 2505 if (delta_search_threads != 1) {
693b86ff 2506 warning("no threads support, ignoring %s", k);
2e96d815
ÆAB
2507 delta_search_threads = 0;
2508 }
693b86ff
NP
2509#endif
2510 return 0;
2511 }
4d00bda2 2512 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2513 pack_idx_opts.version = git_config_int(k, v);
2514 if (pack_idx_opts.version > 2)
6e1c2344 2515 die("bad pack.indexversion=%"PRIu32,
ebcfb379 2516 pack_idx_opts.version);
4d00bda2
NP
2517 return 0;
2518 }
ef90d6d4 2519 return git_default_config(k, v, cb);
4812a93a
JK
2520}
2521
b5d97e6b 2522static void read_object_list_from_stdin(void)
c323ac7d 2523{
188960b4 2524 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2525 struct object_id oid;
2526 const char *p;
b2504a0d 2527
da93d12b 2528 for (;;) {
da93d12b
LT
2529 if (!fgets(line, sizeof(line), stdin)) {
2530 if (feof(stdin))
2531 break;
2532 if (!ferror(stdin))
2533 die("fgets returned NULL, not EOF, not error!");
687dd75c 2534 if (errno != EINTR)
d824cbba 2535 die_errno("fgets");
687dd75c
JH
2536 clearerr(stdin);
2537 continue;
da93d12b 2538 }
7a979d99 2539 if (line[0] == '-') {
188960b4 2540 if (get_oid_hex(line+1, &oid))
2541 die("expected edge object ID, got garbage:\n %s",
b5d97e6b 2542 line);
188960b4 2543 add_preferred_base(&oid);
7a979d99 2544 continue;
21fcd1bd 2545 }
188960b4 2546 if (parse_oid_hex(line, &oid, &p))
2547 die("expected object ID, got garbage:\n %s", line);
b5d97e6b 2548
188960b4 2549 add_preferred_base_object(p + 1);
2550 add_object_entry(&oid, 0, p + 1, 0);
c323ac7d 2551 }
b5d97e6b
JH
2552}
2553
08cdfb13
JH
2554#define OBJECT_ADDED (1u<<20)
2555
11c211fa 2556static void show_commit(struct commit *commit, void *data)
b5d97e6b 2557{
188960b4 2558 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
08cdfb13 2559 commit->object.flags |= OBJECT_ADDED;
7cc8f971
VM
2560
2561 if (write_bitmap_index)
2562 index_commit_for_bitmap(commit);
b5d97e6b
JH
2563}
2564
de1e67d0 2565static void show_object(struct object *obj, const char *name, void *data)
b5d97e6b 2566{
8d2dfc49 2567 add_preferred_base_object(name);
188960b4 2568 add_object_entry(&obj->oid, obj->type, name, 0);
8d2dfc49 2569 obj->flags |= OBJECT_ADDED;
b5d97e6b
JH
2570}
2571
9535ce73
JH
2572static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2573{
2574 assert(arg_missing_action == MA_ALLOW_ANY);
2575
2576 /*
2577 * Quietly ignore ALL missing objects. This avoids problems with
2578 * staging them now and getting an odd error later.
2579 */
2580 if (!has_object_file(&obj->oid))
2581 return;
2582
2583 show_object(obj, name, data);
2584}
2585
0c16cd49
JT
2586static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2587{
2588 assert(arg_missing_action == MA_ALLOW_PROMISOR);
2589
2590 /*
2591 * Quietly ignore EXPECTED missing objects. This avoids problems with
2592 * staging them now and getting an odd error later.
2593 */
2594 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2595 return;
2596
2597 show_object(obj, name, data);
2598}
2599
9535ce73
JH
2600static int option_parse_missing_action(const struct option *opt,
2601 const char *arg, int unset)
2602{
2603 assert(arg);
2604 assert(!unset);
2605
2606 if (!strcmp(arg, "error")) {
2607 arg_missing_action = MA_ERROR;
2608 fn_show_object = show_object;
2609 return 0;
2610 }
2611
2612 if (!strcmp(arg, "allow-any")) {
2613 arg_missing_action = MA_ALLOW_ANY;
0c16cd49 2614 fetch_if_missing = 0;
9535ce73
JH
2615 fn_show_object = show_object__ma_allow_any;
2616 return 0;
2617 }
2618
0c16cd49
JT
2619 if (!strcmp(arg, "allow-promisor")) {
2620 arg_missing_action = MA_ALLOW_PROMISOR;
2621 fetch_if_missing = 0;
2622 fn_show_object = show_object__ma_allow_promisor;
2623 return 0;
2624 }
2625
9535ce73
JH
2626 die(_("invalid value for --missing"));
2627 return 0;
2628}
2629
8d1d8f83
JH
2630static void show_edge(struct commit *commit)
2631{
188960b4 2632 add_preferred_base(&commit->object.oid);
8d1d8f83
JH
2633}
2634
08cdfb13
JH
2635struct in_pack_object {
2636 off_t offset;
2637 struct object *object;
2638};
2639
2640struct in_pack {
071bcaab
RJ
2641 unsigned int alloc;
2642 unsigned int nr;
08cdfb13
JH
2643 struct in_pack_object *array;
2644};
2645
2646static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2647{
ed1c9977 2648 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
08cdfb13
JH
2649 in_pack->array[in_pack->nr].object = object;
2650 in_pack->nr++;
2651}
2652
2653/*
2654 * Compare the objects in the offset order, in order to emulate the
f18d244a 2655 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2656 */
2657static int ofscmp(const void *a_, const void *b_)
2658{
2659 struct in_pack_object *a = (struct in_pack_object *)a_;
2660 struct in_pack_object *b = (struct in_pack_object *)b_;
2661
2662 if (a->offset < b->offset)
2663 return -1;
2664 else if (a->offset > b->offset)
2665 return 1;
2666 else
f2fd0760 2667 return oidcmp(&a->object->oid, &b->object->oid);
08cdfb13
JH
2668}
2669
2670static void add_objects_in_unpacked_packs(struct rev_info *revs)
2671{
2672 struct packed_git *p;
2673 struct in_pack in_pack;
2674 uint32_t i;
2675
2676 memset(&in_pack, 0, sizeof(in_pack));
2677
a80d72db 2678 for (p = get_packed_git(the_repository); p; p = p->next) {
188960b4 2679 struct object_id oid;
08cdfb13
JH
2680 struct object *o;
2681
79bc4c71 2682 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2683 continue;
2684 if (open_pack_index(p))
2685 die("cannot open pack index");
2686
2687 ALLOC_GROW(in_pack.array,
2688 in_pack.nr + p->num_objects,
2689 in_pack.alloc);
2690
2691 for (i = 0; i < p->num_objects; i++) {
188960b4 2692 nth_packed_object_oid(&oid, p, i);
2693 o = lookup_unknown_object(oid.hash);
08cdfb13
JH
2694 if (!(o->flags & OBJECT_ADDED))
2695 mark_in_pack_object(o, p, &in_pack);
2696 o->flags |= OBJECT_ADDED;
2697 }
2698 }
2699
2700 if (in_pack.nr) {
9ed0d8d6 2701 QSORT(in_pack.array, in_pack.nr, ofscmp);
08cdfb13
JH
2702 for (i = 0; i < in_pack.nr; i++) {
2703 struct object *o = in_pack.array[i].object;
188960b4 2704 add_object_entry(&o->oid, o->type, "", 0);
08cdfb13
JH
2705 }
2706 }
2707 free(in_pack.array);
2708}
2709
76c1d9a0 2710static int add_loose_object(const struct object_id *oid, const char *path,
e26a8c47
JK
2711 void *data)
2712{
76c1d9a0 2713 enum object_type type = sha1_object_info(oid->hash, NULL);
e26a8c47
JK
2714
2715 if (type < 0) {
2716 warning("loose object at %s could not be examined", path);
2717 return 0;
2718 }
2719
188960b4 2720 add_object_entry(oid, type, "", 0);
e26a8c47
JK
2721 return 0;
2722}
2723
2724/*
2725 * We actually don't even have to worry about reachability here.
2726 * add_object_entry will weed out duplicates, so we just add every
2727 * loose object we find.
2728 */
2729static void add_unreachable_loose_objects(void)
2730{
2731 for_each_loose_file_in_objdir(get_object_directory(),
2732 add_loose_object,
2733 NULL, NULL, NULL);
2734}
2735
188960b4 2736static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
094085e3
BC
2737{
2738 static struct packed_git *last_found = (void *)1;
2739 struct packed_git *p;
2740
a80d72db
SB
2741 p = (last_found != (void *)1) ? last_found :
2742 get_packed_git(the_repository);
094085e3
BC
2743
2744 while (p) {
2745 if ((!p->pack_local || p->pack_keep) &&
188960b4 2746 find_pack_entry_one(oid->hash, p)) {
094085e3
BC
2747 last_found = p;
2748 return 1;
2749 }
2750 if (p == last_found)
a80d72db 2751 p = get_packed_git(the_repository);
094085e3
BC
2752 else
2753 p = p->next;
2754 if (p == last_found)
2755 p = p->next;
2756 }
2757 return 0;
2758}
2759
abcb8655
JK
2760/*
2761 * Store a list of sha1s that are should not be discarded
2762 * because they are either written too recently, or are
2763 * reachable from another object that was.
2764 *
2765 * This is filled by get_object_list.
2766 */
910650d2 2767static struct oid_array recent_objects;
abcb8655 2768
4ce3621a 2769static int loosened_object_can_be_discarded(const struct object_id *oid,
dddbad72 2770 timestamp_t mtime)
d0d46abc
JK
2771{
2772 if (!unpack_unreachable_expiration)
2773 return 0;
2774 if (mtime > unpack_unreachable_expiration)
2775 return 0;
910650d2 2776 if (oid_array_lookup(&recent_objects, oid) >= 0)
abcb8655 2777 return 0;
d0d46abc
JK
2778 return 1;
2779}
2780
ca11b212
NP
2781static void loosen_unused_packed_objects(struct rev_info *revs)
2782{
2783 struct packed_git *p;
2784 uint32_t i;
4ce3621a 2785 struct object_id oid;
ca11b212 2786
a80d72db 2787 for (p = get_packed_git(the_repository); p; p = p->next) {
79bc4c71 2788 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2789 continue;
2790
2791 if (open_pack_index(p))
2792 die("cannot open pack index");
2793
2794 for (i = 0; i < p->num_objects; i++) {
4ce3621a 2795 nth_packed_object_oid(&oid, p, i);
2796 if (!packlist_find(&to_pack, oid.hash, NULL) &&
188960b4 2797 !has_sha1_pack_kept_or_nonlocal(&oid) &&
4ce3621a 2798 !loosened_object_can_be_discarded(&oid, p->mtime))
2799 if (force_object_loose(oid.hash, p->mtime))
ca11b212
NP
2800 die("unable to force loose object");
2801 }
2802 }
2803}
2804
69e4b342 2805/*
645c432d
KS
2806 * This tracks any options which pack-reuse code expects to be on, or which a
2807 * reader of the pack might not understand, and which would therefore prevent
2808 * blind reuse of what we have on disk.
69e4b342
JK
2809 */
2810static int pack_options_allow_reuse(void)
2811{
9df4a607
JK
2812 return pack_to_stdout &&
2813 allow_ofs_delta &&
2814 !ignore_packed_keep &&
2815 (!local || !have_non_local_packs) &&
2816 !incremental;
69e4b342
JK
2817}
2818
6b8fda2d
VM
2819static int get_object_list_from_bitmap(struct rev_info *revs)
2820{
2821 if (prepare_bitmap_walk(revs) < 0)
2822 return -1;
2823
69e4b342
JK
2824 if (pack_options_allow_reuse() &&
2825 !reuse_partial_packfile_from_bitmap(
6b8fda2d
VM
2826 &reuse_packfile,
2827 &reuse_packfile_objects,
2828 &reuse_packfile_offset)) {
2829 assert(reuse_packfile_objects);
2830 nr_result += reuse_packfile_objects;
78d2214e 2831 display_progress(progress_state, nr_result);
6b8fda2d
VM
2832 }
2833
2834 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2835 return 0;
2836}
2837
abcb8655 2838static void record_recent_object(struct object *obj,
de1e67d0 2839 const char *name,
abcb8655
JK
2840 void *data)
2841{
910650d2 2842 oid_array_append(&recent_objects, &obj->oid);
abcb8655
JK
2843}
2844
2845static void record_recent_commit(struct commit *commit, void *data)
2846{
910650d2 2847 oid_array_append(&recent_objects, &commit->object.oid);
abcb8655
JK
2848}
2849
8d1d8f83 2850static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2851{
2852 struct rev_info revs;
2853 char line[1000];
b5d97e6b
JH
2854 int flags = 0;
2855
b5d97e6b
JH
2856 init_revisions(&revs, NULL);
2857 save_commit_buffer = 0;
b5d97e6b
JH
2858 setup_revisions(ac, av, &revs, NULL);
2859
b790e0f6
NTND
2860 /* make sure shallows are read */
2861 is_repository_shallow();
2862
b5d97e6b
JH
2863 while (fgets(line, sizeof(line), stdin) != NULL) {
2864 int len = strlen(line);
872c930d 2865 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2866 line[--len] = 0;
2867 if (!len)
2868 break;
2869 if (*line == '-') {
2870 if (!strcmp(line, "--not")) {
2871 flags ^= UNINTERESTING;
7cc8f971 2872 write_bitmap_index = 0;
b5d97e6b
JH
2873 continue;
2874 }
b790e0f6 2875 if (starts_with(line, "--shallow ")) {
e92b848c 2876 struct object_id oid;
2877 if (get_oid_hex(line + 10, &oid))
b790e0f6 2878 die("not an SHA-1 '%s'", line + 10);
e92b848c 2879 register_shallow(&oid);
f7f91086 2880 use_bitmap_index = 0;
b790e0f6
NTND
2881 continue;
2882 }
b5d97e6b
JH
2883 die("not a rev '%s'", line);
2884 }
8e676e8b 2885 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
b5d97e6b
JH
2886 die("bad revision '%s'", line);
2887 }
2888
6b8fda2d
VM
2889 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2890 return;
2891
3d51e1b5
MK
2892 if (prepare_revision_walk(&revs))
2893 die("revision walk setup failed");
e76a5fb4 2894 mark_edges_uninteresting(&revs, show_edge);
9535ce73
JH
2895
2896 if (!fn_show_object)
2897 fn_show_object = show_object;
2898 traverse_commit_list_filtered(&filter_options, &revs,
2899 show_commit, fn_show_object, NULL,
2900 NULL);
08cdfb13 2901
abcb8655
JK
2902 if (unpack_unreachable_expiration) {
2903 revs.ignore_missing_links = 1;
2904 if (add_unseen_recent_objects_to_traversal(&revs,
2905 unpack_unreachable_expiration))
2906 die("unable to add recent objects");
2907 if (prepare_revision_walk(&revs))
2908 die("revision walk setup failed");
2909 traverse_commit_list(&revs, record_recent_commit,
2910 record_recent_object, NULL);
2911 }
2912
08cdfb13
JH
2913 if (keep_unreachable)
2914 add_objects_in_unpacked_packs(&revs);
e26a8c47
JK
2915 if (pack_loose_unreachable)
2916 add_unreachable_loose_objects();
ca11b212
NP
2917 if (unpack_unreachable)
2918 loosen_unused_packed_objects(&revs);
abcb8655 2919
910650d2 2920 oid_array_clear(&recent_objects);
b5d97e6b
JH
2921}
2922
99fb6e04
NTND
2923static int option_parse_index_version(const struct option *opt,
2924 const char *arg, int unset)
2925{
2926 char *c;
2927 const char *val = arg;
2928 pack_idx_opts.version = strtoul(val, &c, 10);
2929 if (pack_idx_opts.version > 2)
2930 die(_("unsupported index version %s"), val);
2931 if (*c == ',' && c[1])
2932 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2933 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2934 die(_("bad index version '%s'"), val);
2935 return 0;
2936}
2937
7e52f566
JK
2938static int option_parse_unpack_unreachable(const struct option *opt,
2939 const char *arg, int unset)
2940{
2941 if (unset) {
2942 unpack_unreachable = 0;
2943 unpack_unreachable_expiration = 0;
2944 }
2945 else {
2946 unpack_unreachable = 1;
2947 if (arg)
2948 unpack_unreachable_expiration = approxidate(arg);
2949 }
2950 return 0;
2951}
2952
b5d97e6b
JH
2953int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2954{
b5d97e6b 2955 int use_internal_rev_list = 0;
8d1d8f83 2956 int thin = 0;
2dacf26d 2957 int shallow = 0;
4f366275 2958 int all_progress_implied = 0;
edfbb2aa 2959 struct argv_array rp = ARGV_ARRAY_INIT;
99fb6e04 2960 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
c90f9e13 2961 int rev_list_index = 0;
99fb6e04
NTND
2962 struct option pack_objects_options[] = {
2963 OPT_SET_INT('q', "quiet", &progress,
4c688120 2964 N_("do not show progress meter"), 0),
99fb6e04 2965 OPT_SET_INT(0, "progress", &progress,
4c688120 2966 N_("show progress meter"), 1),
99fb6e04 2967 OPT_SET_INT(0, "all-progress", &progress,
4c688120 2968 N_("show progress meter during object writing phase"), 2),
99fb6e04
NTND
2969 OPT_BOOL(0, "all-progress-implied",
2970 &all_progress_implied,
4c688120
NTND
2971 N_("similar to --all-progress when progress meter is shown")),
2972 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2973 N_("write the pack index file in the specified idx format version"),
99fb6e04 2974 0, option_parse_index_version },
2a514ed8
CB
2975 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2976 N_("maximum size of each output pack file")),
99fb6e04 2977 OPT_BOOL(0, "local", &local,
4c688120 2978 N_("ignore borrowed objects from alternate object store")),
99fb6e04 2979 OPT_BOOL(0, "incremental", &incremental,
4c688120 2980 N_("ignore packed objects")),
99fb6e04 2981 OPT_INTEGER(0, "window", &window,
4c688120 2982 N_("limit pack window by objects")),
2a514ed8
CB
2983 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2984 N_("limit pack window by memory in addition to object limit")),
99fb6e04 2985 OPT_INTEGER(0, "depth", &depth,
4c688120 2986 N_("maximum length of delta chain allowed in the resulting pack")),
99fb6e04 2987 OPT_BOOL(0, "reuse-delta", &reuse_delta,
4c688120 2988 N_("reuse existing deltas")),
99fb6e04 2989 OPT_BOOL(0, "reuse-object", &reuse_object,
4c688120 2990 N_("reuse existing objects")),
99fb6e04 2991 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
4c688120 2992 N_("use OFS_DELTA objects")),
99fb6e04 2993 OPT_INTEGER(0, "threads", &delta_search_threads,
4c688120 2994 N_("use threads when searching for best delta matches")),
99fb6e04 2995 OPT_BOOL(0, "non-empty", &non_empty,
4c688120 2996 N_("do not create an empty pack output")),
99fb6e04 2997 OPT_BOOL(0, "revs", &use_internal_rev_list,
4c688120 2998 N_("read revision arguments from standard input")),
99fb6e04 2999 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
4c688120 3000 N_("limit the objects to those that are not yet packed"),
99fb6e04
NTND
3001 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3002 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
4c688120 3003 N_("include objects reachable from any reference"),
99fb6e04
NTND
3004 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3005 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
4c688120 3006 N_("include objects referred by reflog entries"),
99fb6e04 3007 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
c90f9e13
JK
3008 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
3009 N_("include objects referred to by the index"),
3010 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
99fb6e04 3011 OPT_BOOL(0, "stdout", &pack_to_stdout,
4c688120 3012 N_("output pack to stdout")),
99fb6e04 3013 OPT_BOOL(0, "include-tag", &include_tag,
4c688120 3014 N_("include tag objects that refer to objects to be packed")),
99fb6e04 3015 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
4c688120 3016 N_("keep unreachable objects")),
e26a8c47
JK
3017 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3018 N_("pack loose unreachable objects")),
4c688120
NTND
3019 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3020 N_("unpack unreachable objects newer than <time>"),
7e52f566 3021 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
99fb6e04 3022 OPT_BOOL(0, "thin", &thin,
4c688120 3023 N_("create thin packs")),
2dacf26d 3024 OPT_BOOL(0, "shallow", &shallow,
3025 N_("create packs suitable for shallow fetches")),
99fb6e04 3026 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
4c688120 3027 N_("ignore packs that have companion .keep file")),
99fb6e04 3028 OPT_INTEGER(0, "compression", &pack_compression_level,
4c688120 3029 N_("pack compression level")),
99fb6e04 3030 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
4c688120 3031 N_("do not hide commits by grafts"), 0),
6b8fda2d
VM
3032 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3033 N_("use a bitmap index if available to speed up counting objects")),
7cc8f971
VM
3034 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3035 N_("write a bitmap index together with the pack index")),
9535ce73
JH
3036 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3037 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3038 N_("handling for missing objects"), PARSE_OPT_NONEG,
3039 option_parse_missing_action },
0c16cd49
JT
3040 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3041 N_("do not pack objects in promisor packfiles")),
99fb6e04
NTND
3042 OPT_END(),
3043 };
8d1d8f83 3044
afc711b8 3045 check_replace_refs = 0;
dae556bd 3046
ebcfb379 3047 reset_pack_idx_option(&pack_idx_opts);
ef90d6d4 3048 git_config(git_pack_config, NULL);
b5d97e6b
JH
3049
3050 progress = isatty(2);
99fb6e04
NTND
3051 argc = parse_options(argc, argv, prefix, pack_objects_options,
3052 pack_usage, 0);
b5d97e6b 3053
99fb6e04
NTND
3054 if (argc) {
3055 base_name = argv[0];
3056 argc--;
b5d97e6b 3057 }
99fb6e04
NTND
3058 if (pack_to_stdout != !base_name || argc)
3059 usage_with_options(pack_usage, pack_objects_options);
b5d97e6b 3060
edfbb2aa 3061 argv_array_push(&rp, "pack-objects");
99fb6e04
NTND
3062 if (thin) {
3063 use_internal_rev_list = 1;
2dacf26d 3064 argv_array_push(&rp, shallow
3065 ? "--objects-edge-aggressive"
3066 : "--objects-edge");
99fb6e04 3067 } else
edfbb2aa 3068 argv_array_push(&rp, "--objects");
b5d97e6b 3069
99fb6e04
NTND
3070 if (rev_list_all) {
3071 use_internal_rev_list = 1;
edfbb2aa 3072 argv_array_push(&rp, "--all");
99fb6e04
NTND
3073 }
3074 if (rev_list_reflog) {
3075 use_internal_rev_list = 1;
edfbb2aa 3076 argv_array_push(&rp, "--reflog");
99fb6e04 3077 }
c90f9e13
JK
3078 if (rev_list_index) {
3079 use_internal_rev_list = 1;
3080 argv_array_push(&rp, "--indexed-objects");
99fb6e04
NTND
3081 }
3082 if (rev_list_unpacked) {
3083 use_internal_rev_list = 1;
edfbb2aa 3084 argv_array_push(&rp, "--unpacked");
99fb6e04 3085 }
b5d97e6b 3086
0c16cd49
JT
3087 if (exclude_promisor_objects) {
3088 use_internal_rev_list = 1;
3089 fetch_if_missing = 0;
3090 argv_array_push(&rp, "--exclude-promisor-objects");
3091 }
3092
99fb6e04
NTND
3093 if (!reuse_object)
3094 reuse_delta = 0;
3095 if (pack_compression_level == -1)
3096 pack_compression_level = Z_DEFAULT_COMPRESSION;
3097 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3098 die("bad pack compression level %d", pack_compression_level);
0c45d258
JH
3099
3100 if (!delta_search_threads) /* --threads=0 means autodetect */
3101 delta_search_threads = online_cpus();
3102
99fb6e04
NTND
3103#ifdef NO_PTHREADS
3104 if (delta_search_threads != 1)
2b34e486 3105 warning("no threads support, ignoring --threads");
99fb6e04 3106#endif
2b84b5a8
JS
3107 if (!pack_to_stdout && !pack_size_limit)
3108 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
3109 if (pack_to_stdout && pack_size_limit)
3110 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
3111 if (pack_size_limit && pack_size_limit < 1024*1024) {
3112 warning("minimum pack size limit is 1 MiB");
3113 pack_size_limit = 1024*1024;
3114 }
01c12a23 3115
8d1d8f83
JH
3116 if (!pack_to_stdout && thin)
3117 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 3118
ca11b212
NP
3119 if (keep_unreachable && unpack_unreachable)
3120 die("--keep-unreachable and --unpack-unreachable are incompatible.");
b1e757f3
JK
3121 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3122 unpack_unreachable_expiration = 0;
ca11b212 3123
9535ce73
JH
3124 if (filter_options.choice) {
3125 if (!pack_to_stdout)
3126 die("cannot use --filter without --stdout.");
3127 use_bitmap_index = 0;
3128 }
3129
645c432d
KS
3130 /*
3131 * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3132 *
3133 * - to produce good pack (with bitmap index not-yet-packed objects are
3134 * packed in suboptimal order).
3135 *
3136 * - to use more robust pack-generation codepath (avoiding possible
3137 * bugs in bitmap code and possible bitmap index corruption).
3138 */
3139 if (!pack_to_stdout)
3140 use_bitmap_index_default = 0;
3141
3142 if (use_bitmap_index < 0)
3143 use_bitmap_index = use_bitmap_index_default;
3144
3145 /* "hard" reasons not to use bitmaps; these just won't work at all */
3146 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
6b8fda2d
VM
3147 use_bitmap_index = 0;
3148
7cc8f971
VM
3149 if (pack_to_stdout || !rev_list_all)
3150 write_bitmap_index = 0;
3151
4f366275
NP
3152 if (progress && all_progress_implied)
3153 progress = 2;
3154
b5d97e6b 3155 prepare_packed_git();
56dfeb62
JK
3156 if (ignore_packed_keep) {
3157 struct packed_git *p;
a80d72db 3158 for (p = get_packed_git(the_repository); p; p = p->next)
56dfeb62
JK
3159 if (p->pack_local && p->pack_keep)
3160 break;
3161 if (!p) /* no keep-able packs found */
3162 ignore_packed_keep = 0;
3163 }
3164 if (local) {
3165 /*
3166 * unlike ignore_packed_keep above, we do not want to
3167 * unset "local" based on looking at packs, as it
3168 * also covers non-local objects
3169 */
3170 struct packed_git *p;
a80d72db 3171 for (p = get_packed_git(the_repository); p; p = p->next) {
56dfeb62
JK
3172 if (!p->pack_local) {
3173 have_non_local_packs = 1;
3174 break;
3175 }
3176 }
3177 }
b5d97e6b 3178
13aaf148 3179 if (progress)
754dbc43 3180 progress_state = start_progress(_("Counting objects"), 0);
b5d97e6b
JH
3181 if (!use_internal_rev_list)
3182 read_object_list_from_stdin();
8d1d8f83 3183 else {
edfbb2aa
JK
3184 get_object_list(rp.argc, rp.argv);
3185 argv_array_clear(&rp);
8d1d8f83 3186 }
0ef95f72 3187 cleanup_preferred_base();
d155254c
MH
3188 if (include_tag && nr_result)
3189 for_each_ref(add_ref_tag, NULL);
4d4fcc54 3190 stop_progress(&progress_state);
96a02f8f 3191
f0b0af1b 3192 if (non_empty && !nr_result)
1c4a2912 3193 return 0;
f7ae6a93
NP
3194 if (nr_result)
3195 prepare_pack(window, depth);
d01fb92f 3196 write_pack_file();
ab7cd7bb 3197 if (progress)
6e1c2344
RJ
3198 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3199 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 3200 written, written_delta, reused, reused_delta);
c323ac7d
LT
3201 return 0;
3202}