t5305: simplify packname handling
[git/git.git] / builtin / pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d 2#include "cache.h"
a74db82e 3#include "attr.h"
c323ac7d 4#include "object.h"
8e440259
PE
5#include "blob.h"
6#include "commit.h"
7#include "tag.h"
8#include "tree.h"
c323ac7d 9#include "delta.h"
a733cb60 10#include "pack.h"
3449f8c4 11#include "pack-revindex.h"
c38138cd 12#include "csum-file.h"
1b0c7174 13#include "tree-walk.h"
b5d97e6b
JH
14#include "diff.h"
15#include "revision.h"
16#include "list-objects.h"
2834bc27 17#include "pack-objects.h"
96a02f8f 18#include "progress.h"
f0a24aa5 19#include "refs.h"
cf2ba13a 20#include "streaming.h"
93749194 21#include "thread-utils.h"
6b8fda2d 22#include "pack-bitmap.h"
abcb8655
JK
23#include "reachable.h"
24#include "sha1-array.h"
edfbb2aa 25#include "argv-array.h"
8ecce684 26
99fb6e04 27static const char *pack_usage[] = {
b8c1d275
AH
28 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
29 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
99fb6e04
NTND
30 NULL
31};
c323ac7d 32
3f9ac8d2 33/*
2834bc27
VM
34 * Objects we are going to pack are collected in the `to_pack` structure.
35 * It contains an array (dynamically expanded) of the object data, and a map
36 * that can resolve SHA1s to their position in the array.
3f9ac8d2 37 */
2834bc27
VM
38static struct packing_data to_pack;
39
79814f42 40static struct pack_idx_entry **written_list;
2834bc27 41static uint32_t nr_result, nr_written;
3f9ac8d2 42
96f1e58f 43static int non_empty;
a7de7130 44static int reuse_delta = 1, reuse_object = 1;
e5e9714a 45static int keep_unreachable, unpack_unreachable, include_tag;
7e52f566 46static unsigned long unpack_unreachable_expiration;
96f1e58f
DR
47static int local;
48static int incremental;
e96fb9b8 49static int ignore_packed_keep;
be6b1914 50static int allow_ofs_delta;
ebcfb379 51static struct pack_idx_option pack_idx_opts;
d01fb92f 52static const char *base_name;
024701f1 53static int progress = 1;
4812a93a 54static int window = 10;
568508e7 55static unsigned long pack_size_limit;
618e613a 56static int depth = 50;
43cc2b42 57static int delta_search_threads;
df6d6101 58static int pack_to_stdout;
8d1d8f83 59static int num_preferred_base;
dc6a0757 60static struct progress *progress_state;
960ccca6
DH
61static int pack_compression_level = Z_DEFAULT_COMPRESSION;
62static int pack_compression_seen;
c323ac7d 63
6b8fda2d
VM
64static struct packed_git *reuse_packfile;
65static uint32_t reuse_packfile_objects;
66static off_t reuse_packfile_offset;
67
68static int use_bitmap_index = 1;
7cc8f971 69static int write_bitmap_index;
ae4f07fb 70static uint16_t write_bitmap_options;
6b8fda2d 71
074b2eea 72static unsigned long delta_cache_size = 0;
5749b0b2 73static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
e3dfddb3 74static unsigned long cache_max_small_delta_size = 1000;
074b2eea 75
a97773ce
BD
76static unsigned long window_memory_limit = 0;
77
3f9ac8d2
JH
78/*
79 * stats
80 */
7cadf491
SP
81static uint32_t written, written_delta;
82static uint32_t reused, reused_delta;
3f9ac8d2 83
7cc8f971
VM
84/*
85 * Indexed commits
86 */
87static struct commit **indexed_commits;
88static unsigned int indexed_commits_nr;
89static unsigned int indexed_commits_alloc;
90
91static void index_commit_for_bitmap(struct commit *commit)
92{
93 if (indexed_commits_nr >= indexed_commits_alloc) {
94 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
2756ca43 95 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
7cc8f971
VM
96 }
97
98 indexed_commits[indexed_commits_nr++] = commit;
99}
780e6e73 100
3613f9b4 101static void *get_delta(struct object_entry *entry)
c323ac7d 102{
3613f9b4
NP
103 unsigned long size, base_size, delta_size;
104 void *buf, *base_buf, *delta_buf;
21666f1a 105 enum object_type type;
c323ac7d 106
3613f9b4
NP
107 buf = read_sha1_file(entry->idx.sha1, &type, &size);
108 if (!buf)
109 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
110 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
111 if (!base_buf)
aa7e44bf 112 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
3613f9b4 113 delta_buf = diff_delta(base_buf, base_size,
dcde55bc 114 buf, size, &delta_size, 0);
3613f9b4 115 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 116 die("delta size changed");
3613f9b4
NP
117 free(buf);
118 free(base_buf);
c323ac7d
LT
119 return delta_buf;
120}
121
30ebb40a
NP
122static unsigned long do_compress(void **pptr, unsigned long size)
123{
ef49a7a0 124 git_zstream stream;
30ebb40a
NP
125 void *in, *out;
126 unsigned long maxsize;
127
55bb5c91 128 git_deflate_init(&stream, pack_compression_level);
225a6f10 129 maxsize = git_deflate_bound(&stream, size);
30ebb40a
NP
130
131 in = *pptr;
132 out = xmalloc(maxsize);
133 *pptr = out;
134
135 stream.next_in = in;
136 stream.avail_in = size;
137 stream.next_out = out;
138 stream.avail_out = maxsize;
55bb5c91 139 while (git_deflate(&stream, Z_FINISH) == Z_OK)
30ebb40a 140 ; /* nothing */
55bb5c91 141 git_deflate_end(&stream);
30ebb40a
NP
142
143 free(in);
144 return stream.total_out;
145}
146
cf2ba13a
NTND
147static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
148 const unsigned char *sha1)
149{
150 git_zstream stream;
151 unsigned char ibuf[1024 * 16];
152 unsigned char obuf[1024 * 16];
153 unsigned long olen = 0;
154
cf2ba13a
NTND
155 git_deflate_init(&stream, pack_compression_level);
156
157 for (;;) {
158 ssize_t readlen;
159 int zret = Z_OK;
160 readlen = read_istream(st, ibuf, sizeof(ibuf));
161 if (readlen == -1)
162 die(_("unable to read %s"), sha1_to_hex(sha1));
163
164 stream.next_in = ibuf;
165 stream.avail_in = readlen;
166 while ((stream.avail_in || readlen == 0) &&
167 (zret == Z_OK || zret == Z_BUF_ERROR)) {
168 stream.next_out = obuf;
169 stream.avail_out = sizeof(obuf);
170 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
171 sha1write(f, obuf, stream.next_out - obuf);
172 olen += stream.next_out - obuf;
173 }
174 if (stream.avail_in)
175 die(_("deflate error (%d)"), zret);
176 if (readlen == 0) {
177 if (zret != Z_STREAM_END)
178 die(_("deflate error (%d)"), zret);
179 break;
180 }
181 }
182 git_deflate_end(&stream);
183 return olen;
184}
185
780e6e73
NP
186/*
187 * we are going to reuse the existing object data as is. make
188 * sure it is not corrupt.
189 */
079afb18
SP
190static int check_pack_inflate(struct packed_git *p,
191 struct pack_window **w_curs,
6777a59f
SP
192 off_t offset,
193 off_t len,
079afb18
SP
194 unsigned long expect)
195{
ef49a7a0 196 git_zstream stream;
079afb18
SP
197 unsigned char fakebuf[4096], *in;
198 int st;
199
200 memset(&stream, 0, sizeof(stream));
39c68542 201 git_inflate_init(&stream);
079afb18
SP
202 do {
203 in = use_pack(p, w_curs, offset, &stream.avail_in);
204 stream.next_in = in;
205 stream.next_out = fakebuf;
206 stream.avail_out = sizeof(fakebuf);
39c68542 207 st = git_inflate(&stream, Z_FINISH);
079afb18
SP
208 offset += stream.next_in - in;
209 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 210 git_inflate_end(&stream);
079afb18
SP
211 return (st == Z_STREAM_END &&
212 stream.total_out == expect &&
213 stream.total_in == len) ? 0 : -1;
214}
215
216static void copy_pack_data(struct sha1file *f,
217 struct packed_git *p,
218 struct pack_window **w_curs,
6777a59f
SP
219 off_t offset,
220 off_t len)
079afb18
SP
221{
222 unsigned char *in;
ef49a7a0 223 unsigned long avail;
079afb18
SP
224
225 while (len) {
226 in = use_pack(p, w_curs, offset, &avail);
227 if (avail > len)
ef49a7a0 228 avail = (unsigned long)len;
079afb18
SP
229 sha1write(f, in, avail);
230 offset += avail;
231 len -= avail;
232 }
233}
234
1b4bb16b 235/* Return 0 if we will bust the pack-size limit */
c9018b03
NTND
236static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
237 unsigned long limit, int usable_delta)
c323ac7d 238{
c9018b03 239 unsigned long size, datalen;
2c5ef824 240 unsigned char header[10], dheader[10];
6777a59f 241 unsigned hdrlen;
2c5ef824 242 enum object_type type;
c9018b03 243 void *buf;
cf2ba13a 244 struct git_istream *st = NULL;
c9018b03
NTND
245
246 if (!usable_delta) {
cf2ba13a
NTND
247 if (entry->type == OBJ_BLOB &&
248 entry->size > big_file_threshold &&
249 (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
250 buf = NULL;
251 else {
252 buf = read_sha1_file(entry->idx.sha1, &type, &size);
253 if (!buf)
254 die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
255 }
c9018b03
NTND
256 /*
257 * make sure no cached delta data remains from a
258 * previous attempt before a pack split occurred.
259 */
260 free(entry->delta_data);
261 entry->delta_data = NULL;
262 entry->z_delta_size = 0;
263 } else if (entry->delta_data) {
264 size = entry->delta_size;
265 buf = entry->delta_data;
266 entry->delta_data = NULL;
267 type = (allow_ofs_delta && entry->delta->idx.offset) ?
268 OBJ_OFS_DELTA : OBJ_REF_DELTA;
269 } else {
270 buf = get_delta(entry);
271 size = entry->delta_size;
272 type = (allow_ofs_delta && entry->delta->idx.offset) ?
273 OBJ_OFS_DELTA : OBJ_REF_DELTA;
274 }
275
cf2ba13a
NTND
276 if (st) /* large blob case, just assume we don't compress well */
277 datalen = size;
278 else if (entry->z_delta_size)
c9018b03
NTND
279 datalen = entry->z_delta_size;
280 else
281 datalen = do_compress(&buf, size);
282
283 /*
284 * The object header is a byte of 'type' followed by zero or
285 * more bytes of length.
286 */
287 hdrlen = encode_in_pack_object_header(type, size, header);
288
289 if (type == OBJ_OFS_DELTA) {
290 /*
291 * Deltas with relative base contain an additional
292 * encoding of the relative offset for the delta
293 * base from this object's position in the pack.
294 */
295 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
296 unsigned pos = sizeof(dheader) - 1;
297 dheader[pos] = ofs & 127;
298 while (ofs >>= 7)
299 dheader[--pos] = 128 | (--ofs & 127);
300 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
cf2ba13a
NTND
301 if (st)
302 close_istream(st);
c9018b03
NTND
303 free(buf);
304 return 0;
305 }
306 sha1write(f, header, hdrlen);
307 sha1write(f, dheader + pos, sizeof(dheader) - pos);
308 hdrlen += sizeof(dheader) - pos;
309 } else if (type == OBJ_REF_DELTA) {
310 /*
311 * Deltas with a base reference contain
312 * an additional 20 bytes for the base sha1.
313 */
314 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
cf2ba13a
NTND
315 if (st)
316 close_istream(st);
c9018b03
NTND
317 free(buf);
318 return 0;
319 }
320 sha1write(f, header, hdrlen);
321 sha1write(f, entry->delta->idx.sha1, 20);
322 hdrlen += 20;
323 } else {
324 if (limit && hdrlen + datalen + 20 >= limit) {
cf2ba13a
NTND
325 if (st)
326 close_istream(st);
c9018b03
NTND
327 free(buf);
328 return 0;
329 }
330 sha1write(f, header, hdrlen);
331 }
cf2ba13a
NTND
332 if (st) {
333 datalen = write_large_blob_data(st, f, entry->idx.sha1);
334 close_istream(st);
335 } else {
336 sha1write(f, buf, datalen);
337 free(buf);
338 }
c9018b03
NTND
339
340 return hdrlen + datalen;
341}
342
343/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
344static off_t write_reuse_object(struct sha1file *f, struct object_entry *entry,
345 unsigned long limit, int usable_delta)
c9018b03
NTND
346{
347 struct packed_git *p = entry->in_pack;
348 struct pack_window *w_curs = NULL;
349 struct revindex_entry *revidx;
350 off_t offset;
351 enum object_type type = entry->type;
211c61c6 352 off_t datalen;
c9018b03
NTND
353 unsigned char header[10], dheader[10];
354 unsigned hdrlen;
355
356 if (entry->delta)
357 type = (allow_ofs_delta && entry->delta->idx.offset) ?
358 OBJ_OFS_DELTA : OBJ_REF_DELTA;
359 hdrlen = encode_in_pack_object_header(type, entry->size, header);
360
361 offset = entry->in_pack_offset;
362 revidx = find_pack_revindex(p, offset);
363 datalen = revidx[1].offset - offset;
364 if (!pack_to_stdout && p->index_version > 1 &&
365 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
366 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
367 unuse_pack(&w_curs);
368 return write_no_reuse_object(f, entry, limit, usable_delta);
369 }
370
371 offset += entry->in_pack_header_size;
372 datalen -= entry->in_pack_header_size;
373
374 if (!pack_to_stdout && p->index_version == 1 &&
375 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
376 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
377 unuse_pack(&w_curs);
378 return write_no_reuse_object(f, entry, limit, usable_delta);
379 }
380
381 if (type == OBJ_OFS_DELTA) {
382 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
383 unsigned pos = sizeof(dheader) - 1;
384 dheader[pos] = ofs & 127;
385 while (ofs >>= 7)
386 dheader[--pos] = 128 | (--ofs & 127);
387 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
388 unuse_pack(&w_curs);
389 return 0;
390 }
391 sha1write(f, header, hdrlen);
392 sha1write(f, dheader + pos, sizeof(dheader) - pos);
393 hdrlen += sizeof(dheader) - pos;
394 reused_delta++;
395 } else if (type == OBJ_REF_DELTA) {
396 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
397 unuse_pack(&w_curs);
398 return 0;
399 }
400 sha1write(f, header, hdrlen);
401 sha1write(f, entry->delta->idx.sha1, 20);
402 hdrlen += 20;
403 reused_delta++;
404 } else {
405 if (limit && hdrlen + datalen + 20 >= limit) {
406 unuse_pack(&w_curs);
407 return 0;
408 }
409 sha1write(f, header, hdrlen);
410 }
411 copy_pack_data(f, p, &w_curs, offset, datalen);
412 unuse_pack(&w_curs);
413 reused++;
414 return hdrlen + datalen;
415}
416
417/* Return 0 if we will bust the pack-size limit */
af92a645
NTND
418static off_t write_object(struct sha1file *f,
419 struct object_entry *entry,
420 off_t write_offset)
c9018b03 421{
af92a645
NTND
422 unsigned long limit;
423 off_t len;
2c5ef824 424 int usable_delta, to_reuse;
c323ac7d 425
78d1e84f
NP
426 if (!pack_to_stdout)
427 crc32_begin(f);
428
a2430dde 429 /* apply size limit if limited packsize and not first object */
a1e4760f
NP
430 if (!pack_size_limit || !nr_written)
431 limit = 0;
432 else if (pack_size_limit <= write_offset)
433 /*
434 * the earlier object did not fit the limit; avoid
435 * mistaking this with unlimited (i.e. limit = 0).
436 */
437 limit = 1;
438 else
439 limit = pack_size_limit - write_offset;
2c5ef824
NP
440
441 if (!entry->delta)
442 usable_delta = 0; /* no delta */
443 else if (!pack_size_limit)
444 usable_delta = 1; /* unlimited packfile */
445 else if (entry->delta->idx.offset == (off_t)-1)
446 usable_delta = 0; /* base was written to another pack */
447 else if (entry->delta->idx.offset)
448 usable_delta = 1; /* base already exists in this pack */
449 else
450 usable_delta = 0; /* base could end up in another pack */
451
a7de7130 452 if (!reuse_object)
fa736f72
NP
453 to_reuse = 0; /* explicit */
454 else if (!entry->in_pack)
ab7cd7bb 455 to_reuse = 0; /* can't reuse what we don't have */
c9018b03 456 else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
17b08f2c
DH
457 /* check_object() decided it for us ... */
458 to_reuse = usable_delta;
459 /* ... but pack split may override that */
c9018b03 460 else if (entry->type != entry->in_pack_type)
ab7cd7bb
JH
461 to_reuse = 0; /* pack has delta which is unusable */
462 else if (entry->delta)
463 to_reuse = 0; /* we want to pack afresh */
464 else
465 to_reuse = 1; /* we have it in-pack undeltified,
466 * and we do not need to deltify it.
467 */
468
c9018b03
NTND
469 if (!to_reuse)
470 len = write_no_reuse_object(f, entry, limit, usable_delta);
471 else
472 len = write_reuse_object(f, entry, limit, usable_delta);
473 if (!len)
474 return 0;
64bd76b1 475
17b08f2c 476 if (usable_delta)
ab7cd7bb 477 written_delta++;
3f9ac8d2 478 written++;
78d1e84f 479 if (!pack_to_stdout)
aa7e44bf 480 entry->idx.crc32 = crc32_end(f);
c9018b03 481 return len;
c323ac7d
LT
482}
483
f63c79db
JH
484enum write_one_status {
485 WRITE_ONE_SKIP = -1, /* already written */
486 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
487 WRITE_ONE_WRITTEN = 1, /* normal */
488 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
489};
490
491static enum write_one_status write_one(struct sha1file *f,
492 struct object_entry *e,
493 off_t *offset)
9d5ab962 494{
af92a645 495 off_t size;
f63c79db 496 int recursing;
d7dd0223 497
f63c79db
JH
498 /*
499 * we set offset to 1 (which is an impossible value) to mark
500 * the fact that this object is involved in "write its base
501 * first before writing a deltified object" recursion.
502 */
503 recursing = (e->idx.offset == 1);
504 if (recursing) {
505 warning("recursive delta detected for object %s",
506 sha1_to_hex(e->idx.sha1));
507 return WRITE_ONE_RECURSIVE;
508 } else if (e->idx.offset || e->preferred_base) {
509 /* offset is non zero if object is written already. */
510 return WRITE_ONE_SKIP;
511 }
d7dd0223 512
720c9f7b 513 /* if we are deltified, write out base object first. */
f63c79db
JH
514 if (e->delta) {
515 e->idx.offset = 1; /* now recurse */
516 switch (write_one(f, e->delta, offset)) {
517 case WRITE_ONE_RECURSIVE:
518 /* we cannot depend on this one */
519 e->delta = NULL;
520 break;
521 default:
522 break;
523 case WRITE_ONE_BREAK:
524 e->idx.offset = recursing;
525 return WRITE_ONE_BREAK;
526 }
527 }
d7dd0223 528
6ed7f25e
NP
529 e->idx.offset = *offset;
530 size = write_object(f, e, *offset);
17b08f2c 531 if (!size) {
f63c79db
JH
532 e->idx.offset = recursing;
533 return WRITE_ONE_BREAK;
17b08f2c 534 }
79814f42 535 written_list[nr_written++] = &e->idx;
d7dd0223
NP
536
537 /* make sure off_t is sufficiently large not to wrap */
c03c8315 538 if (signed_add_overflows(*offset, size))
d7dd0223 539 die("pack too large for current definition of off_t");
6ed7f25e 540 *offset += size;
f63c79db 541 return WRITE_ONE_WRITTEN;
9d5ab962
JH
542}
543
d155254c 544static int mark_tagged(const char *path, const struct object_id *oid, int flag,
1b4bb16b
JH
545 void *cb_data)
546{
547 unsigned char peeled[20];
d155254c 548 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
1b4bb16b
JH
549
550 if (entry)
551 entry->tagged = 1;
552 if (!peel_ref(path, peeled)) {
2834bc27 553 entry = packlist_find(&to_pack, peeled, NULL);
1b4bb16b
JH
554 if (entry)
555 entry->tagged = 1;
556 }
557 return 0;
558}
559
be126818 560static inline void add_to_write_order(struct object_entry **wo,
92bef1a1 561 unsigned int *endp,
1b4bb16b
JH
562 struct object_entry *e)
563{
564 if (e->filled)
565 return;
566 wo[(*endp)++] = e;
567 e->filled = 1;
568}
569
570static void add_descendants_to_write_order(struct object_entry **wo,
92bef1a1 571 unsigned int *endp,
1b4bb16b
JH
572 struct object_entry *e)
573{
f380872f
DM
574 int add_to_order = 1;
575 while (e) {
576 if (add_to_order) {
577 struct object_entry *s;
578 /* add this node... */
579 add_to_write_order(wo, endp, e);
580 /* all its siblings... */
581 for (s = e->delta_sibling; s; s = s->delta_sibling) {
582 add_to_write_order(wo, endp, s);
583 }
584 }
585 /* drop down a level to add left subtree nodes if possible */
586 if (e->delta_child) {
587 add_to_order = 1;
588 e = e->delta_child;
589 } else {
590 add_to_order = 0;
591 /* our sibling might have some children, it is next */
592 if (e->delta_sibling) {
593 e = e->delta_sibling;
594 continue;
595 }
596 /* go back to our parent node */
597 e = e->delta;
598 while (e && !e->delta_sibling) {
599 /* we're on the right side of a subtree, keep
600 * going up until we can go right again */
601 e = e->delta;
602 }
603 if (!e) {
604 /* done- we hit our original root node */
605 return;
606 }
607 /* pass it off to sibling at this level */
608 e = e->delta_sibling;
609 }
610 };
1b4bb16b
JH
611}
612
613static void add_family_to_write_order(struct object_entry **wo,
92bef1a1 614 unsigned int *endp,
1b4bb16b
JH
615 struct object_entry *e)
616{
617 struct object_entry *root;
618
619 for (root = e; root->delta; root = root->delta)
620 ; /* nothing */
1b4bb16b
JH
621 add_descendants_to_write_order(wo, endp, root);
622}
623
624static struct object_entry **compute_write_order(void)
625{
38d4debb 626 unsigned int i, wo_end, last_untagged;
1b4bb16b 627
b32fa95f 628 struct object_entry **wo;
2834bc27 629 struct object_entry *objects = to_pack.objects;
1b4bb16b 630
2834bc27 631 for (i = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
632 objects[i].tagged = 0;
633 objects[i].filled = 0;
634 objects[i].delta_child = NULL;
635 objects[i].delta_sibling = NULL;
636 }
637
638 /*
639 * Fully connect delta_child/delta_sibling network.
640 * Make sure delta_sibling is sorted in the original
641 * recency order.
642 */
2834bc27 643 for (i = to_pack.nr_objects; i > 0;) {
92bef1a1 644 struct object_entry *e = &objects[--i];
1b4bb16b
JH
645 if (!e->delta)
646 continue;
647 /* Mark me as the first child */
648 e->delta_sibling = e->delta->delta_child;
649 e->delta->delta_child = e;
650 }
651
652 /*
653 * Mark objects that are at the tip of tags.
654 */
d155254c 655 for_each_tag_ref(mark_tagged, NULL);
1b4bb16b
JH
656
657 /*
38d4debb 658 * Give the objects in the original recency order until
1b4bb16b
JH
659 * we see a tagged tip.
660 */
b32fa95f 661 ALLOC_ARRAY(wo, to_pack.nr_objects);
2834bc27 662 for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
663 if (objects[i].tagged)
664 break;
665 add_to_write_order(wo, &wo_end, &objects[i]);
666 }
38d4debb 667 last_untagged = i;
1b4bb16b
JH
668
669 /*
670 * Then fill all the tagged tips.
671 */
2834bc27 672 for (; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
673 if (objects[i].tagged)
674 add_to_write_order(wo, &wo_end, &objects[i]);
675 }
676
677 /*
678 * And then all remaining commits and tags.
679 */
2834bc27 680 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
681 if (objects[i].type != OBJ_COMMIT &&
682 objects[i].type != OBJ_TAG)
683 continue;
684 add_to_write_order(wo, &wo_end, &objects[i]);
685 }
686
687 /*
688 * And then all the trees.
689 */
2834bc27 690 for (i = last_untagged; i < to_pack.nr_objects; i++) {
1b4bb16b
JH
691 if (objects[i].type != OBJ_TREE)
692 continue;
693 add_to_write_order(wo, &wo_end, &objects[i]);
694 }
695
696 /*
697 * Finally all the rest in really tight order
698 */
2834bc27 699 for (i = last_untagged; i < to_pack.nr_objects; i++) {
38d4debb
DM
700 if (!objects[i].filled)
701 add_family_to_write_order(wo, &wo_end, &objects[i]);
702 }
703
2834bc27
VM
704 if (wo_end != to_pack.nr_objects)
705 die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
1b4bb16b
JH
706
707 return wo;
708}
709
6b8fda2d
VM
710static off_t write_reused_pack(struct sha1file *f)
711{
712 unsigned char buffer[8192];
657673f1 713 off_t to_write, total;
6b8fda2d
VM
714 int fd;
715
716 if (!is_pack_valid(reuse_packfile))
717 die("packfile is invalid: %s", reuse_packfile->pack_name);
718
719 fd = git_open_noatime(reuse_packfile->pack_name);
720 if (fd < 0)
721 die_errno("unable to open packfile for reuse: %s",
722 reuse_packfile->pack_name);
723
724 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
725 die_errno("unable to seek in reused packfile");
726
727 if (reuse_packfile_offset < 0)
728 reuse_packfile_offset = reuse_packfile->pack_size - 20;
729
657673f1 730 total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
6b8fda2d
VM
731
732 while (to_write) {
733 int read_pack = xread(fd, buffer, sizeof(buffer));
734
735 if (read_pack <= 0)
736 die_errno("unable to read from reused packfile");
737
738 if (read_pack > to_write)
739 read_pack = to_write;
740
741 sha1write(f, buffer, read_pack);
742 to_write -= read_pack;
657673f1
JK
743
744 /*
745 * We don't know the actual number of objects written,
746 * only how many bytes written, how many bytes total, and
747 * how many objects total. So we can fake it by pretending all
748 * objects we are writing are the same size. This gives us a
749 * smooth progress meter, and at the end it matches the true
750 * answer.
751 */
752 written = reuse_packfile_objects *
753 (((double)(total - to_write)) / total);
754 display_progress(progress_state, written);
6b8fda2d
VM
755 }
756
757 close(fd);
657673f1
JK
758 written = reuse_packfile_objects;
759 display_progress(progress_state, written);
6b8fda2d
VM
760 return reuse_packfile_offset - sizeof(struct pack_header);
761}
762
9cea46cd
EW
763static const char no_split_warning[] = N_(
764"disabling bitmap writing, packs are split due to pack.packSizeLimit"
765);
766
d01fb92f 767static void write_pack_file(void)
c323ac7d 768{
ebe27b13 769 uint32_t i = 0, j;
d22b9290 770 struct sha1file *f;
6ed7f25e 771 off_t offset;
ebe27b13 772 uint32_t nr_remaining = nr_result;
f746bae8 773 time_t last_mtime = 0;
1b4bb16b 774 struct object_entry **write_order;
81a216a5 775
bcd7954e 776 if (progress > pack_to_stdout)
754dbc43 777 progress_state = start_progress(_("Writing objects"), nr_result);
b32fa95f 778 ALLOC_ARRAY(written_list, to_pack.nr_objects);
1b4bb16b 779 write_order = compute_write_order();
183bdb2c 780
ebe27b13 781 do {
aa7e44bf 782 unsigned char sha1[20];
7ba502c4 783 char *pack_tmp_name = NULL;
aa7e44bf 784
cdf9db3c 785 if (pack_to_stdout)
2a128d63 786 f = sha1fd_throughput(1, "<stdout>", progress_state);
cdf9db3c
JH
787 else
788 f = create_tmp_packfile(&pack_tmp_name);
ebe27b13 789
c0ad4657 790 offset = write_pack_header(f, nr_remaining);
6b8fda2d
VM
791
792 if (reuse_packfile) {
793 off_t packfile_size;
794 assert(pack_to_stdout);
795
796 packfile_size = write_reused_pack(f);
797 offset += packfile_size;
798 }
799
ebe27b13 800 nr_written = 0;
2834bc27 801 for (; i < to_pack.nr_objects; i++) {
1b4bb16b 802 struct object_entry *e = write_order[i];
cddec4f8 803 if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
720c9f7b
NP
804 break;
805 display_progress(progress_state, written);
806 }
c553ca25 807
ebe27b13
DH
808 /*
809 * Did we write the wrong # entries in the header?
810 * If so, rewrite it like in fast-import
811 */
54352bb2
LT
812 if (pack_to_stdout) {
813 sha1close(f, sha1, CSUM_CLOSE);
814 } else if (nr_written == nr_remaining) {
815 sha1close(f, sha1, CSUM_FSYNC);
ebe27b13 816 } else {
ac0463ed 817 int fd = sha1close(f, sha1, 0);
abeb40e5 818 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
ac0463ed 819 nr_written, sha1, offset);
7ba502c4 820 close(fd);
9cea46cd
EW
821 if (write_bitmap_index) {
822 warning(_(no_split_warning));
823 write_bitmap_index = 0;
824 }
ebe27b13
DH
825 }
826
827 if (!pack_to_stdout) {
f746bae8 828 struct stat st;
58892711 829 struct strbuf tmpname = STRBUF_INIT;
d01fb92f 830
f746bae8
NP
831 /*
832 * Packs are runtime accessed in their mtime
833 * order since newer packs are more likely to contain
834 * younger objects. So if we are creating multiple
835 * packs then we should modify the mtime of later ones
836 * to preserve this property.
837 */
0e990530 838 if (stat(pack_tmp_name, &st) < 0) {
54d47394 839 warning_errno("failed to stat %s", pack_tmp_name);
f746bae8
NP
840 } else if (!last_mtime) {
841 last_mtime = st.st_mtime;
842 } else {
843 struct utimbuf utb;
844 utb.actime = st.st_atime;
845 utb.modtime = --last_mtime;
0e990530 846 if (utime(pack_tmp_name, &utb) < 0)
54d47394 847 warning_errno("failed utime() on %s", pack_tmp_name);
f746bae8
NP
848 }
849
58892711 850 strbuf_addf(&tmpname, "%s-", base_name);
7cc8f971
VM
851
852 if (write_bitmap_index) {
853 bitmap_writer_set_checksum(sha1);
854 bitmap_writer_build_type_index(written_list, nr_written);
855 }
856
58892711 857 finish_tmp_packfile(&tmpname, pack_tmp_name,
0e990530
JH
858 written_list, nr_written,
859 &pack_idx_opts, sha1);
7cc8f971
VM
860
861 if (write_bitmap_index) {
58892711 862 strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
7cc8f971
VM
863
864 stop_progress(&progress_state);
865
866 bitmap_writer_show_progress(progress);
867 bitmap_writer_reuse_bitmaps(&to_pack);
868 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
869 bitmap_writer_build(&to_pack);
ae4f07fb 870 bitmap_writer_finish(written_list, nr_written,
58892711 871 tmpname.buf, write_bitmap_options);
7cc8f971
VM
872 write_bitmap_index = 0;
873 }
874
58892711 875 strbuf_release(&tmpname);
7ba502c4 876 free(pack_tmp_name);
aa7e44bf 877 puts(sha1_to_hex(sha1));
ebe27b13
DH
878 }
879
880 /* mark written objects as written to previous pack */
881 for (j = 0; j < nr_written; j++) {
79814f42 882 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
883 }
884 nr_remaining -= nr_written;
2834bc27 885 } while (nr_remaining && i < to_pack.nr_objects);
ebe27b13
DH
886
887 free(written_list);
1b4bb16b 888 free(write_order);
4d4fcc54 889 stop_progress(&progress_state);
67c08ce1 890 if (written != nr_result)
6e1c2344
RJ
891 die("wrote %"PRIu32" objects while expecting %"PRIu32,
892 written, nr_result);
c323ac7d
LT
893}
894
a74db82e
JH
895static void setup_delta_attr_check(struct git_attr_check *check)
896{
897 static struct git_attr *attr_delta;
898
899 if (!attr_delta)
7fb0eaa2 900 attr_delta = git_attr("delta");
a74db82e
JH
901
902 check[0].attr = attr_delta;
903}
904
905static int no_try_delta(const char *path)
906{
907 struct git_attr_check check[1];
908
909 setup_delta_attr_check(check);
d932f4eb 910 if (git_check_attr(path, ARRAY_SIZE(check), check))
a74db82e
JH
911 return 0;
912 if (ATTR_FALSE(check->value))
913 return 1;
914 return 0;
915}
916
ce2bc424
JK
917/*
918 * When adding an object, check whether we have already added it
919 * to our packing list. If so, we can skip. However, if we are
920 * being asked to excludei t, but the previous mention was to include
921 * it, make sure to adjust its flags and tweak our numbers accordingly.
922 *
923 * As an optimization, we pass out the index position where we would have
924 * found the item, since that saves us from having to look it up again a
925 * few lines later when we want to add the new entry.
926 */
927static int have_duplicate_entry(const unsigned char *sha1,
928 int exclude,
929 uint32_t *index_pos)
c323ac7d 930{
c323ac7d 931 struct object_entry *entry;
29b734e4 932
ce2bc424
JK
933 entry = packlist_find(&to_pack, sha1, index_pos);
934 if (!entry)
29b734e4 935 return 0;
ce2bc424
JK
936
937 if (exclude) {
938 if (!entry->preferred_base)
939 nr_result--;
940 entry->preferred_base = 1;
29b734e4 941 }
c323ac7d 942
ce2bc424
JK
943 return 1;
944}
945
946/*
947 * Check whether we want the object in the pack (e.g., we do not want
948 * objects found in non-local stores if the "--local" option was used).
949 *
950 * As a side effect of this check, we will find the packed version of this
951 * object, if any. We therefore pass out the pack information to avoid having
952 * to look it up again later.
953 */
954static int want_object_in_pack(const unsigned char *sha1,
955 int exclude,
956 struct packed_git **found_pack,
957 off_t *found_offset)
958{
959 struct packed_git *p;
960
daae0625
BC
961 if (!exclude && local && has_loose_object_nonlocal(sha1))
962 return 0;
963
ce2bc424
JK
964 *found_pack = NULL;
965 *found_offset = 0;
966
5c49c116
NP
967 for (p = packed_git; p; p = p->next) {
968 off_t offset = find_pack_entry_one(sha1, p);
969 if (offset) {
ce2bc424 970 if (!*found_pack) {
319b678a 971 if (!is_pack_valid(p))
4c080182 972 continue;
ce2bc424
JK
973 *found_offset = offset;
974 *found_pack = p;
64560374 975 }
5c49c116 976 if (exclude)
ce2bc424 977 return 1;
5c49c116
NP
978 if (incremental)
979 return 0;
980 if (local && !p->pack_local)
981 return 0;
e96fb9b8
BC
982 if (ignore_packed_keep && p->pack_local && p->pack_keep)
983 return 0;
64560374
LT
984 }
985 }
eb019375 986
ce2bc424
JK
987 return 1;
988}
989
990static void create_object_entry(const unsigned char *sha1,
991 enum object_type type,
992 uint32_t hash,
993 int exclude,
994 int no_try_delta,
995 uint32_t index_pos,
996 struct packed_git *found_pack,
997 off_t found_offset)
998{
999 struct object_entry *entry;
29b734e4 1000
2834bc27 1001 entry = packlist_alloc(&to_pack, sha1, index_pos);
27225f2e 1002 entry->hash = hash;
5c49c116
NP
1003 if (type)
1004 entry->type = type;
29b734e4
NP
1005 if (exclude)
1006 entry->preferred_base = 1;
81a216a5
NP
1007 else
1008 nr_result++;
29b734e4
NP
1009 if (found_pack) {
1010 entry->in_pack = found_pack;
1011 entry->in_pack_offset = found_offset;
1012 }
7a979d99 1013
ce2bc424
JK
1014 entry->no_try_delta = no_try_delta;
1015}
29b734e4 1016
373c67da
JK
1017static const char no_closure_warning[] = N_(
1018"disabling bitmap writing, as some objects are not being packed"
1019);
1020
ce2bc424
JK
1021static int add_object_entry(const unsigned char *sha1, enum object_type type,
1022 const char *name, int exclude)
1023{
1024 struct packed_git *found_pack;
1025 off_t found_offset;
1026 uint32_t index_pos;
7a979d99 1027
ce2bc424
JK
1028 if (have_duplicate_entry(sha1, exclude, &index_pos))
1029 return 0;
a74db82e 1030
373c67da
JK
1031 if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1032 /* The pack is missing an object, so it will not have closure */
1033 if (write_bitmap_index) {
1034 warning(_(no_closure_warning));
1035 write_bitmap_index = 0;
1036 }
ce2bc424 1037 return 0;
373c67da 1038 }
29b734e4 1039
ce2bc424
JK
1040 create_object_entry(sha1, type, pack_name_hash(name),
1041 exclude, name && no_try_delta(name),
1042 index_pos, found_pack, found_offset);
a74db82e 1043
78d2214e 1044 display_progress(progress_state, nr_result);
29b734e4 1045 return 1;
c323ac7d
LT
1046}
1047
6b8fda2d
VM
1048static int add_object_entry_from_bitmap(const unsigned char *sha1,
1049 enum object_type type,
1050 int flags, uint32_t name_hash,
1051 struct packed_git *pack, off_t offset)
1052{
1053 uint32_t index_pos;
1054
1055 if (have_duplicate_entry(sha1, 0, &index_pos))
1056 return 0;
1057
1058 create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1059
78d2214e 1060 display_progress(progress_state, nr_result);
29b734e4 1061 return 1;
c323ac7d
LT
1062}
1063
5379a5c5
JH
1064struct pbase_tree_cache {
1065 unsigned char sha1[20];
1066 int ref;
1067 int temporary;
1068 void *tree_data;
1069 unsigned long tree_size;
1070};
1071
1072static struct pbase_tree_cache *(pbase_tree_cache[256]);
1073static int pbase_tree_cache_ix(const unsigned char *sha1)
1074{
1075 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1076}
1077static int pbase_tree_cache_ix_incr(int ix)
1078{
1079 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1080}
1081
1082static struct pbase_tree {
1083 struct pbase_tree *next;
1084 /* This is a phony "cache" entry; we are not
01689909 1085 * going to evict it or find it through _get()
5379a5c5
JH
1086 * mechanism -- this is for the toplevel node that
1087 * would almost always change with any commit.
1088 */
1089 struct pbase_tree_cache pcache;
1090} *pbase_tree;
1091
1092static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1093{
1094 struct pbase_tree_cache *ent, *nent;
1095 void *data;
1096 unsigned long size;
21666f1a 1097 enum object_type type;
5379a5c5
JH
1098 int neigh;
1099 int my_ix = pbase_tree_cache_ix(sha1);
1100 int available_ix = -1;
1101
1102 /* pbase-tree-cache acts as a limited hashtable.
1103 * your object will be found at your index or within a few
1104 * slots after that slot if it is cached.
1105 */
1106 for (neigh = 0; neigh < 8; neigh++) {
1107 ent = pbase_tree_cache[my_ix];
a89fccd2 1108 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
1109 ent->ref++;
1110 return ent;
1111 }
1112 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1113 ((0 <= available_ix) &&
1114 (!ent && pbase_tree_cache[available_ix])))
1115 available_ix = my_ix;
1116 if (!ent)
1117 break;
1118 my_ix = pbase_tree_cache_ix_incr(my_ix);
1119 }
1120
1121 /* Did not find one. Either we got a bogus request or
1122 * we need to read and perhaps cache.
1123 */
21666f1a 1124 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
1125 if (!data)
1126 return NULL;
21666f1a 1127 if (type != OBJ_TREE) {
5379a5c5
JH
1128 free(data);
1129 return NULL;
1130 }
1131
1132 /* We need to either cache or return a throwaway copy */
1133
1134 if (available_ix < 0)
1135 ent = NULL;
1136 else {
1137 ent = pbase_tree_cache[available_ix];
1138 my_ix = available_ix;
1139 }
1140
1141 if (!ent) {
1142 nent = xmalloc(sizeof(*nent));
1143 nent->temporary = (available_ix < 0);
1144 }
1145 else {
1146 /* evict and reuse */
1147 free(ent->tree_data);
1148 nent = ent;
1149 }
e702496e 1150 hashcpy(nent->sha1, sha1);
5379a5c5
JH
1151 nent->tree_data = data;
1152 nent->tree_size = size;
1153 nent->ref = 1;
1154 if (!nent->temporary)
1155 pbase_tree_cache[my_ix] = nent;
1156 return nent;
1157}
1158
1159static void pbase_tree_put(struct pbase_tree_cache *cache)
1160{
1161 if (!cache->temporary) {
1162 cache->ref--;
1163 return;
1164 }
1165 free(cache->tree_data);
1166 free(cache);
1167}
1168
1169static int name_cmp_len(const char *name)
1170{
1171 int i;
1172 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1173 ;
1174 return i;
1175}
1176
1177static void add_pbase_object(struct tree_desc *tree,
5379a5c5 1178 const char *name,
ce0bd642
LT
1179 int cmplen,
1180 const char *fullname)
3f9ac8d2 1181{
4c068a98 1182 struct name_entry entry;
8a5a8d6c 1183 int cmp;
4c068a98
LT
1184
1185 while (tree_entry(tree,&entry)) {
1211be6b
LT
1186 if (S_ISGITLINK(entry.mode))
1187 continue;
0de16337 1188 cmp = tree_entry_len(&entry) != cmplen ? 1 :
8a5a8d6c
NP
1189 memcmp(name, entry.path, cmplen);
1190 if (cmp > 0)
7a979d99 1191 continue;
8a5a8d6c
NP
1192 if (cmp < 0)
1193 return;
5379a5c5 1194 if (name[cmplen] != '/') {
7d924c91 1195 add_object_entry(entry.oid->hash,
4d1012c3 1196 object_type(entry.mode),
bc32fed5 1197 fullname, 1);
5379a5c5
JH
1198 return;
1199 }
8a5a8d6c 1200 if (S_ISDIR(entry.mode)) {
7a979d99 1201 struct tree_desc sub;
5379a5c5
JH
1202 struct pbase_tree_cache *tree;
1203 const char *down = name+cmplen+1;
1204 int downlen = name_cmp_len(down);
1205
7d924c91 1206 tree = pbase_tree_get(entry.oid->hash);
5379a5c5
JH
1207 if (!tree)
1208 return;
6fda5e51 1209 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1210
ce0bd642 1211 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1212 pbase_tree_put(tree);
1213 }
1214 }
1215}
1d6b38cc 1216
5379a5c5
JH
1217static unsigned *done_pbase_paths;
1218static int done_pbase_paths_num;
1219static int done_pbase_paths_alloc;
1220static int done_pbase_path_pos(unsigned hash)
1221{
1222 int lo = 0;
1223 int hi = done_pbase_paths_num;
1224 while (lo < hi) {
1225 int mi = (hi + lo) / 2;
1226 if (done_pbase_paths[mi] == hash)
1227 return mi;
1228 if (done_pbase_paths[mi] < hash)
1229 hi = mi;
1230 else
1231 lo = mi + 1;
1232 }
1233 return -lo-1;
1234}
1235
1236static int check_pbase_path(unsigned hash)
1237{
1238 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1239 if (0 <= pos)
1240 return 1;
1241 pos = -pos - 1;
25e19407
DD
1242 ALLOC_GROW(done_pbase_paths,
1243 done_pbase_paths_num + 1,
1244 done_pbase_paths_alloc);
5379a5c5
JH
1245 done_pbase_paths_num++;
1246 if (pos < done_pbase_paths_num)
1247 memmove(done_pbase_paths + pos + 1,
1248 done_pbase_paths + pos,
1249 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1250 done_pbase_paths[pos] = hash;
1251 return 0;
1252}
1253
bc32fed5 1254static void add_preferred_base_object(const char *name)
5379a5c5
JH
1255{
1256 struct pbase_tree *it;
8a5a8d6c 1257 int cmplen;
68fb36eb 1258 unsigned hash = pack_name_hash(name);
5379a5c5 1259
8a5a8d6c 1260 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1261 return;
1262
8a5a8d6c 1263 cmplen = name_cmp_len(name);
5379a5c5
JH
1264 for (it = pbase_tree; it; it = it->next) {
1265 if (cmplen == 0) {
bc32fed5 1266 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
1267 }
1268 else {
1269 struct tree_desc tree;
6fda5e51 1270 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1271 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1272 }
3f9ac8d2 1273 }
3f9ac8d2
JH
1274}
1275
7a979d99 1276static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 1277{
5379a5c5
JH
1278 struct pbase_tree *it;
1279 void *data;
1280 unsigned long size;
1281 unsigned char tree_sha1[20];
1d6b38cc 1282
8d1d8f83
JH
1283 if (window <= num_preferred_base++)
1284 return;
1285
5379a5c5
JH
1286 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1287 if (!data)
7a979d99 1288 return;
5379a5c5
JH
1289
1290 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1291 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1292 free(data);
1293 return;
1294 }
1295 }
1296
1297 it = xcalloc(1, sizeof(*it));
1298 it->next = pbase_tree;
1299 pbase_tree = it;
1300
e702496e 1301 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1302 it->pcache.tree_data = data;
1303 it->pcache.tree_size = size;
3f9ac8d2
JH
1304}
1305
0ef95f72
NP
1306static void cleanup_preferred_base(void)
1307{
1308 struct pbase_tree *it;
1309 unsigned i;
1310
1311 it = pbase_tree;
1312 pbase_tree = NULL;
1313 while (it) {
1314 struct pbase_tree *this = it;
1315 it = this->next;
1316 free(this->pcache.tree_data);
1317 free(this);
1318 }
1319
1320 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1321 if (!pbase_tree_cache[i])
1322 continue;
1323 free(pbase_tree_cache[i]->tree_data);
1324 free(pbase_tree_cache[i]);
1325 pbase_tree_cache[i] = NULL;
1326 }
1327
1328 free(done_pbase_paths);
1329 done_pbase_paths = NULL;
1330 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1331}
1332
c323ac7d
LT
1333static void check_object(struct object_entry *entry)
1334{
5c49c116 1335 if (entry->in_pack) {
780e6e73 1336 struct packed_git *p = entry->in_pack;
03e79c88 1337 struct pack_window *w_curs = NULL;
5c49c116
NP
1338 const unsigned char *base_ref = NULL;
1339 struct object_entry *base_entry;
1340 unsigned long used, used_0;
ef49a7a0 1341 unsigned long avail;
5c49c116
NP
1342 off_t ofs;
1343 unsigned char *buf, c;
780e6e73 1344
6777a59f 1345 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1346
5c49c116 1347 /*
fa736f72
NP
1348 * We want in_pack_type even if we do not reuse delta
1349 * since non-delta representations could still be reused.
ab7cd7bb 1350 */
09ded04b 1351 used = unpack_object_header_buffer(buf, avail,
5c49c116
NP
1352 &entry->in_pack_type,
1353 &entry->size);
03d66015
NP
1354 if (used == 0)
1355 goto give_up;
ab7cd7bb 1356
5c49c116
NP
1357 /*
1358 * Determine if this is a delta and if so whether we can
1359 * reuse it or not. Otherwise let's find out as cheaply as
1360 * possible what the actual type and size for this object is.
3f9ac8d2 1361 */
5c49c116
NP
1362 switch (entry->in_pack_type) {
1363 default:
1364 /* Not a delta hence we've already got all we need. */
1365 entry->type = entry->in_pack_type;
1366 entry->in_pack_header_size = used;
03d66015
NP
1367 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1368 goto give_up;
5c49c116
NP
1369 unuse_pack(&w_curs);
1370 return;
1371 case OBJ_REF_DELTA:
a7de7130 1372 if (reuse_delta && !entry->preferred_base)
5c49c116
NP
1373 base_ref = use_pack(p, &w_curs,
1374 entry->in_pack_offset + used, NULL);
1375 entry->in_pack_header_size = used + 20;
1376 break;
1377 case OBJ_OFS_DELTA:
1378 buf = use_pack(p, &w_curs,
1379 entry->in_pack_offset + used, NULL);
1380 used_0 = 0;
1381 c = buf[used_0++];
1382 ofs = c & 127;
1383 while (c & 128) {
1384 ofs += 1;
03d66015
NP
1385 if (!ofs || MSB(ofs, 7)) {
1386 error("delta base offset overflow in pack for %s",
1387 sha1_to_hex(entry->idx.sha1));
1388 goto give_up;
1389 }
5c49c116
NP
1390 c = buf[used_0++];
1391 ofs = (ofs << 7) + (c & 127);
780e6e73 1392 }
5c49c116 1393 ofs = entry->in_pack_offset - ofs;
03d66015
NP
1394 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1395 error("delta base offset out of bound for %s",
1396 sha1_to_hex(entry->idx.sha1));
1397 goto give_up;
1398 }
a7de7130 1399 if (reuse_delta && !entry->preferred_base) {
3449f8c4
NP
1400 struct revindex_entry *revidx;
1401 revidx = find_pack_revindex(p, ofs);
08698b1e
NP
1402 if (!revidx)
1403 goto give_up;
3449f8c4
NP
1404 base_ref = nth_packed_object_sha1(p, revidx->nr);
1405 }
5c49c116
NP
1406 entry->in_pack_header_size = used + used_0;
1407 break;
780e6e73 1408 }
780e6e73 1409
2834bc27 1410 if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
5c49c116
NP
1411 /*
1412 * If base_ref was set above that means we wish to
1413 * reuse delta data, and we even found that base
1414 * in the list of objects we want to pack. Goodie!
1415 *
1416 * Depth value does not matter - find_deltas() will
1417 * never consider reused delta as the base object to
1418 * deltify other objects against, in order to avoid
1419 * circular deltas.
3f9ac8d2 1420 */
780e6e73 1421 entry->type = entry->in_pack_type;
5c49c116 1422 entry->delta = base_entry;
64bd76b1 1423 entry->delta_size = entry->size;
15b4d577
JH
1424 entry->delta_sibling = base_entry->delta_child;
1425 base_entry->delta_child = entry;
5c49c116
NP
1426 unuse_pack(&w_curs);
1427 return;
1428 }
ab7cd7bb 1429
5c49c116
NP
1430 if (entry->type) {
1431 /*
1432 * This must be a delta and we already know what the
1433 * final object type is. Let's extract the actual
1434 * object size from the delta header.
1435 */
1436 entry->size = get_size_from_delta(p, &w_curs,
1437 entry->in_pack_offset + entry->in_pack_header_size);
03d66015
NP
1438 if (entry->size == 0)
1439 goto give_up;
5c49c116 1440 unuse_pack(&w_curs);
3f9ac8d2
JH
1441 return;
1442 }
5c49c116
NP
1443
1444 /*
1445 * No choice but to fall back to the recursive delta walk
1446 * with sha1_object_info() to find about the object type
1447 * at this point...
1448 */
03d66015 1449 give_up:
5c49c116 1450 unuse_pack(&w_curs);
36e4d74a 1451 }
3f9ac8d2 1452
aa7e44bf 1453 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
6d6f9cdd
SP
1454 /*
1455 * The error condition is checked in prepare_pack(). This is
1456 * to permit a missing preferred base object to be ignored
1457 * as a preferred base. Doing so can result in a larger
1458 * pack file, but the transfer will still take place.
1459 */
3f9ac8d2
JH
1460}
1461
5c49c116
NP
1462static int pack_offset_sort(const void *_a, const void *_b)
1463{
1464 const struct object_entry *a = *(struct object_entry **)_a;
1465 const struct object_entry *b = *(struct object_entry **)_b;
1466
1467 /* avoid filesystem trashing with loose objects */
1468 if (!a->in_pack && !b->in_pack)
aa7e44bf 1469 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1470
1471 if (a->in_pack < b->in_pack)
1472 return -1;
1473 if (a->in_pack > b->in_pack)
1474 return 1;
1475 return a->in_pack_offset < b->in_pack_offset ? -1 :
1476 (a->in_pack_offset > b->in_pack_offset);
1477}
1478
c323ac7d
LT
1479static void get_object_details(void)
1480{
7cadf491 1481 uint32_t i;
5c49c116
NP
1482 struct object_entry **sorted_by_offset;
1483
2834bc27
VM
1484 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1485 for (i = 0; i < to_pack.nr_objects; i++)
1486 sorted_by_offset[i] = to_pack.objects + i;
1487 qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1488
2834bc27 1489 for (i = 0; i < to_pack.nr_objects; i++) {
15366280
JH
1490 struct object_entry *entry = sorted_by_offset[i];
1491 check_object(entry);
754980d0 1492 if (big_file_threshold < entry->size)
15366280
JH
1493 entry->no_try_delta = 1;
1494 }
3449f8c4 1495
5c49c116 1496 free(sorted_by_offset);
c323ac7d
LT
1497}
1498
b904166c
NP
1499/*
1500 * We search for deltas in a list sorted by type, by filename hash, and then
1501 * by size, so that we see progressively smaller and smaller files.
1502 * That's because we prefer deltas to be from the bigger file
1503 * to the smaller -- deletes are potentially cheaper, but perhaps
1504 * more importantly, the bigger file is likely the more recent
1505 * one. The deepest deltas are therefore the oldest objects which are
1506 * less susceptible to be accessed often.
1507 */
9668cf59 1508static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1509{
9668cf59
NP
1510 const struct object_entry *a = *(struct object_entry **)_a;
1511 const struct object_entry *b = *(struct object_entry **)_b;
1512
c323ac7d 1513 if (a->type > b->type)
27225f2e 1514 return -1;
b904166c 1515 if (a->type < b->type)
27225f2e 1516 return 1;
b904166c 1517 if (a->hash > b->hash)
7a979d99 1518 return -1;
b904166c 1519 if (a->hash < b->hash)
7a979d99 1520 return 1;
b904166c 1521 if (a->preferred_base > b->preferred_base)
c323ac7d 1522 return -1;
b904166c
NP
1523 if (a->preferred_base < b->preferred_base)
1524 return 1;
c323ac7d 1525 if (a->size > b->size)
b904166c
NP
1526 return -1;
1527 if (a->size < b->size)
c323ac7d 1528 return 1;
b904166c 1529 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1530}
1531
1532struct unpacked {
1533 struct object_entry *entry;
1534 void *data;
f6c7081a 1535 struct delta_index *index;
5a235b5e 1536 unsigned depth;
c323ac7d
LT
1537};
1538
d250626c
NP
1539static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1540 unsigned long delta_size)
074b2eea
MK
1541{
1542 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1543 return 0;
1544
e3dfddb3
MK
1545 if (delta_size < cache_max_small_delta_size)
1546 return 1;
1547
074b2eea
MK
1548 /* cache delta, if objects are large enough compared to delta size */
1549 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1550 return 1;
1551
1552 return 0;
1553}
1554
7eb151d6 1555#ifndef NO_PTHREADS
8ecce684 1556
44626dc7 1557static pthread_mutex_t read_mutex;
8ecce684
NP
1558#define read_lock() pthread_mutex_lock(&read_mutex)
1559#define read_unlock() pthread_mutex_unlock(&read_mutex)
1560
44626dc7 1561static pthread_mutex_t cache_mutex;
3c701839
NP
1562#define cache_lock() pthread_mutex_lock(&cache_mutex)
1563#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1564
44626dc7 1565static pthread_mutex_t progress_mutex;
8ecce684
NP
1566#define progress_lock() pthread_mutex_lock(&progress_mutex)
1567#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1568
1569#else
1570
e1ef8673
JH
1571#define read_lock() (void)0
1572#define read_unlock() (void)0
1573#define cache_lock() (void)0
1574#define cache_unlock() (void)0
1575#define progress_lock() (void)0
1576#define progress_unlock() (void)0
8ecce684
NP
1577
1578#endif
1579
f6c7081a 1580static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1581 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1582{
f6c7081a
NP
1583 struct object_entry *trg_entry = trg->entry;
1584 struct object_entry *src_entry = src->entry;
560b25a8 1585 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1586 unsigned ref_depth;
21666f1a 1587 enum object_type type;
c323ac7d
LT
1588 void *delta_buf;
1589
1590 /* Don't bother doing diffs between different types */
f6c7081a 1591 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1592 return -1;
1593
51d1e83f 1594 /*
15f07e06
JK
1595 * We do not bother to try a delta that we discarded on an
1596 * earlier try, but only when reusing delta data. Note that
1597 * src_entry that is marked as the preferred_base should always
1598 * be considered, as even if we produce a suboptimal delta against
1599 * it, we will still save the transfer cost, as we already know
1600 * the other side has it and we won't send src_entry at all.
51d1e83f 1601 */
a7de7130 1602 if (reuse_delta && trg_entry->in_pack &&
e9195b58 1603 trg_entry->in_pack == src_entry->in_pack &&
15f07e06 1604 !src_entry->preferred_base &&
e9195b58
JH
1605 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1606 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1607 return 0;
1608
898b14ce 1609 /* Let's not bust the allowed depth. */
5a235b5e 1610 if (src->depth >= max_depth)
d116a45a 1611 return 0;
c323ac7d 1612
c3b06a69 1613 /* Now some size filtering heuristics. */
560b25a8 1614 trg_size = trg_entry->size;
c83f032e
NP
1615 if (!trg_entry->delta) {
1616 max_size = trg_size/2 - 20;
1617 ref_depth = 1;
1618 } else {
1619 max_size = trg_entry->delta_size;
5a235b5e 1620 ref_depth = trg->depth;
c83f032e 1621 }
720fe22d 1622 max_size = (uint64_t)max_size * (max_depth - src->depth) /
c83f032e 1623 (max_depth - ref_depth + 1);
c3b06a69
NP
1624 if (max_size == 0)
1625 return 0;
f6c7081a 1626 src_size = src_entry->size;
560b25a8 1627 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1628 if (sizediff >= max_size)
f527cb8c 1629 return 0;
a1dab41a
BD
1630 if (trg_size < src_size / 32)
1631 return 0;
f6c7081a 1632
560b25a8
NP
1633 /* Load data if not already done */
1634 if (!trg->data) {
8ecce684 1635 read_lock();
aa7e44bf 1636 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1637 read_unlock();
2e3404c3
JH
1638 if (!trg->data)
1639 die("object %s cannot be read",
1640 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1641 if (sz != trg_size)
1642 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1643 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1644 *mem_usage += sz;
560b25a8
NP
1645 }
1646 if (!src->data) {
8ecce684 1647 read_lock();
aa7e44bf 1648 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1649 read_unlock();
71064a95
NP
1650 if (!src->data) {
1651 if (src_entry->preferred_base) {
1652 static int warned = 0;
1653 if (!warned++)
1654 warning("object %s cannot be read",
1655 sha1_to_hex(src_entry->idx.sha1));
1656 /*
1657 * Those objects are not included in the
1658 * resulting pack. Be resilient and ignore
1659 * them if they can't be read, in case the
1660 * pack could be created nevertheless.
1661 */
1662 return 0;
1663 }
2e3404c3
JH
1664 die("object %s cannot be read",
1665 sha1_to_hex(src_entry->idx.sha1));
71064a95 1666 }
560b25a8
NP
1667 if (sz != src_size)
1668 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1669 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1670 *mem_usage += sz;
560b25a8
NP
1671 }
1672 if (!src->index) {
1673 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1674 if (!src->index) {
1675 static int warned = 0;
1676 if (!warned++)
1677 warning("suboptimal pack - out of memory");
1678 return 0;
1679 }
ef0316fc 1680 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1681 }
1682
1683 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1684 if (!delta_buf)
75c42d8c 1685 return 0;
f6c7081a 1686
9e2d57a0 1687 if (trg_entry->delta) {
848d732c
BD
1688 /* Prefer only shallower same-sized deltas. */
1689 if (delta_size == trg_entry->delta_size &&
5a235b5e 1690 src->depth + 1 >= trg->depth) {
848d732c
BD
1691 free(delta_buf);
1692 return 0;
1693 }
074b2eea 1694 }
9e2d57a0 1695
3c701839
NP
1696 /*
1697 * Handle memory allocation outside of the cache
1698 * accounting lock. Compiler will optimize the strangeness
7eb151d6 1699 * away when NO_PTHREADS is defined.
3c701839 1700 */
8e0f7003 1701 free(trg_entry->delta_data);
3c701839 1702 cache_lock();
9e2d57a0
NP
1703 if (trg_entry->delta_data) {
1704 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1705 trg_entry->delta_data = NULL;
1706 }
d250626c 1707 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1708 delta_cache_size += delta_size;
3c701839
NP
1709 cache_unlock();
1710 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1711 } else {
1712 cache_unlock();
074b2eea 1713 free(delta_buf);
3c701839
NP
1714 }
1715
b7a28f78
NP
1716 trg_entry->delta = src_entry;
1717 trg_entry->delta_size = delta_size;
1718 trg->depth = src->depth + 1;
1719
f6c7081a 1720 return 1;
c323ac7d
LT
1721}
1722
898b14ce 1723static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1724{
898b14ce
NP
1725 struct object_entry *child = me->delta_child;
1726 unsigned int m = n;
1727 while (child) {
1728 unsigned int c = check_delta_limit(child, n + 1);
1729 if (m < c)
1730 m = c;
1731 child = child->delta_sibling;
1732 }
1733 return m;
b2504a0d
NP
1734}
1735
75ad235c 1736static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1737{
ef0316fc 1738 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1739 free_delta_index(n->index);
1740 n->index = NULL;
1741 if (n->data) {
ef0316fc 1742 freed_mem += n->entry->size;
a97773ce
BD
1743 free(n->data);
1744 n->data = NULL;
a97773ce
BD
1745 }
1746 n->entry = NULL;
7d7baa5e 1747 n->depth = 0;
ef0316fc 1748 return freed_mem;
a97773ce
BD
1749}
1750
384b32c0 1751static void find_deltas(struct object_entry **list, unsigned *list_size,
e334977d 1752 int window, int depth, unsigned *processed)
c323ac7d 1753{
384b32c0 1754 uint32_t i, idx = 0, count = 0;
7cadf491 1755 struct unpacked *array;
ef0316fc 1756 unsigned long mem_usage = 0;
c323ac7d 1757
19d4b416 1758 array = xcalloc(window, sizeof(struct unpacked));
21fcd1bd 1759
384b32c0 1760 for (;;) {
421b488a 1761 struct object_entry *entry;
c323ac7d 1762 struct unpacked *n = array + idx;
ef0316fc 1763 int j, max_depth, best_base = -1;
c323ac7d 1764
384b32c0
NP
1765 progress_lock();
1766 if (!*list_size) {
1767 progress_unlock();
1768 break;
1769 }
421b488a 1770 entry = *list++;
384b32c0
NP
1771 (*list_size)--;
1772 if (!entry->preferred_base) {
1773 (*processed)++;
1774 display_progress(progress_state, *processed);
1775 }
1776 progress_unlock();
1777
ef0316fc 1778 mem_usage -= free_unpacked(n);
c323ac7d 1779 n->entry = entry;
ab7cd7bb 1780
a97773ce 1781 while (window_memory_limit &&
ef0316fc 1782 mem_usage > window_memory_limit &&
a97773ce
BD
1783 count > 1) {
1784 uint32_t tail = (idx + window - count) % window;
75ad235c 1785 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1786 count--;
1787 }
1788
75d39853
NP
1789 /* We do not compute delta to *create* objects we are not
1790 * going to pack.
1791 */
1792 if (entry->preferred_base)
1793 goto next;
1794
898b14ce
NP
1795 /*
1796 * If the current object is at pack edge, take the depth the
1797 * objects that depend on the current object into account
1798 * otherwise they would become too deep.
1799 */
1800 max_depth = depth;
1801 if (entry->delta_child) {
1802 max_depth -= check_delta_limit(entry, 0);
1803 if (max_depth <= 0)
1804 goto next;
1805 }
1806
78817c15
LT
1807 j = window;
1808 while (--j > 0) {
77639870 1809 int ret;
7cadf491 1810 uint32_t other_idx = idx + j;
c323ac7d 1811 struct unpacked *m;
78817c15
LT
1812 if (other_idx >= window)
1813 other_idx -= window;
c323ac7d
LT
1814 m = array + other_idx;
1815 if (!m->entry)
1816 break;
ef0316fc 1817 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1818 if (ret < 0)
c323ac7d 1819 break;
77639870
JH
1820 else if (ret > 0)
1821 best_base = other_idx;
c323ac7d 1822 }
898b14ce 1823
ed4a9031
NP
1824 /*
1825 * If we decided to cache the delta data, then it is best
1826 * to compress it right away. First because we have to do
1827 * it anyway, and doing it here while we're threaded will
1828 * save a lot of time in the non threaded write phase,
1829 * as well as allow for caching more deltas within
1830 * the same cache size limit.
1831 * ...
1832 * But only if not writing to stdout, since in that case
1833 * the network is most likely throttling writes anyway,
1834 * and therefore it is best to go to the write phase ASAP
1835 * instead, as we can afford spending more time compressing
1836 * between writes at that moment.
1837 */
1838 if (entry->delta_data && !pack_to_stdout) {
1839 entry->z_delta_size = do_compress(&entry->delta_data,
1840 entry->delta_size);
1841 cache_lock();
1842 delta_cache_size -= entry->delta_size;
1843 delta_cache_size += entry->z_delta_size;
1844 cache_unlock();
1845 }
1846
70ca1a3f
JH
1847 /* if we made n a delta, and if n is already at max
1848 * depth, leaving it in the window is pointless. we
1849 * should evict it first.
70ca1a3f 1850 */
70baf5d4 1851 if (entry->delta && max_depth <= n->depth)
70ca1a3f 1852 continue;
ff45715c 1853
77639870
JH
1854 /*
1855 * Move the best delta base up in the window, after the
1856 * currently deltified object, to keep it longer. It will
1857 * be the first base object to be attempted next.
1858 */
1859 if (entry->delta) {
1860 struct unpacked swap = array[best_base];
1861 int dist = (window + idx - best_base) % window;
1862 int dst = best_base;
1863 while (dist--) {
1864 int src = (dst + 1) % window;
1865 array[dst] = array[src];
1866 dst = src;
1867 }
1868 array[dst] = swap;
1869 }
1870
898b14ce 1871 next:
521a4f4c 1872 idx++;
a97773ce
BD
1873 if (count + 1 < window)
1874 count++;
521a4f4c
LT
1875 if (idx >= window)
1876 idx = 0;
384b32c0 1877 }
adee7bdf 1878
f6c7081a 1879 for (i = 0; i < window; ++i) {
ff45715c 1880 free_delta_index(array[i].index);
adee7bdf 1881 free(array[i].data);
f6c7081a 1882 }
adee7bdf 1883 free(array);
c323ac7d
LT
1884}
1885
7eb151d6 1886#ifndef NO_PTHREADS
8ecce684 1887
a9a74636
NP
1888static void try_to_free_from_threads(size_t size)
1889{
1890 read_lock();
7c3ecb32 1891 release_pack_memory(size);
a9a74636
NP
1892 read_unlock();
1893}
1894
bc9b2175 1895static try_to_free_t old_try_to_free_routine;
851c34b0 1896
50f22ada
JS
1897/*
1898 * The main thread waits on the condition that (at least) one of the workers
1899 * has stopped working (which is indicated in the .working member of
1900 * struct thread_params).
1901 * When a work thread has completed its work, it sets .working to 0 and
1902 * signals the main thread and waits on the condition that .data_ready
1903 * becomes 1.
1904 */
1905
8ecce684
NP
1906struct thread_params {
1907 pthread_t thread;
1908 struct object_entry **list;
1909 unsigned list_size;
384b32c0 1910 unsigned remaining;
8ecce684
NP
1911 int window;
1912 int depth;
50f22ada
JS
1913 int working;
1914 int data_ready;
1915 pthread_mutex_t mutex;
1916 pthread_cond_t cond;
8ecce684
NP
1917 unsigned *processed;
1918};
1919
44626dc7
AH
1920static pthread_cond_t progress_cond;
1921
1922/*
1923 * Mutex and conditional variable can't be statically-initialized on Windows.
1924 */
1925static void init_threaded_search(void)
1926{
93749194 1927 init_recursive_mutex(&read_mutex);
44626dc7
AH
1928 pthread_mutex_init(&cache_mutex, NULL);
1929 pthread_mutex_init(&progress_mutex, NULL);
1930 pthread_cond_init(&progress_cond, NULL);
851c34b0 1931 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
44626dc7
AH
1932}
1933
1934static void cleanup_threaded_search(void)
1935{
851c34b0 1936 set_try_to_free_routine(old_try_to_free_routine);
44626dc7
AH
1937 pthread_cond_destroy(&progress_cond);
1938 pthread_mutex_destroy(&read_mutex);
1939 pthread_mutex_destroy(&cache_mutex);
1940 pthread_mutex_destroy(&progress_mutex);
1941}
c2a33679 1942
8ecce684
NP
1943static void *threaded_find_deltas(void *arg)
1944{
c2a33679
NP
1945 struct thread_params *me = arg;
1946
50f22ada 1947 while (me->remaining) {
384b32c0 1948 find_deltas(me->list, &me->remaining,
c2a33679 1949 me->window, me->depth, me->processed);
50f22ada
JS
1950
1951 progress_lock();
1952 me->working = 0;
1953 pthread_cond_signal(&progress_cond);
1954 progress_unlock();
1955
1956 /*
1957 * We must not set ->data_ready before we wait on the
1958 * condition because the main thread may have set it to 1
1959 * before we get here. In order to be sure that new
1960 * work is available if we see 1 in ->data_ready, it
1961 * was initialized to 0 before this thread was spawned
1962 * and we reset it to 0 right away.
1963 */
1964 pthread_mutex_lock(&me->mutex);
1965 while (!me->data_ready)
1966 pthread_cond_wait(&me->cond, &me->mutex);
1967 me->data_ready = 0;
1968 pthread_mutex_unlock(&me->mutex);
c2a33679 1969 }
50f22ada
JS
1970 /* leave ->working 1 so that this doesn't get more work assigned */
1971 return NULL;
8ecce684
NP
1972}
1973
8ecce684
NP
1974static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1975 int window, int depth, unsigned *processed)
1976{
dcda3614 1977 struct thread_params *p;
384b32c0 1978 int i, ret, active_threads = 0;
c2a33679 1979
44626dc7
AH
1980 init_threaded_search();
1981
367f4a43 1982 if (delta_search_threads <= 1) {
384b32c0 1983 find_deltas(list, &list_size, window, depth, processed);
44626dc7 1984 cleanup_threaded_search();
367f4a43
NP
1985 return;
1986 }
43cc2b42 1987 if (progress > pack_to_stdout)
b6c29915 1988 fprintf(stderr, "Delta compression using up to %d threads.\n",
43cc2b42 1989 delta_search_threads);
dcda3614 1990 p = xcalloc(delta_search_threads, sizeof(*p));
367f4a43 1991
50f22ada 1992 /* Partition the work amongst work threads. */
367f4a43 1993 for (i = 0; i < delta_search_threads; i++) {
50f22ada
JS
1994 unsigned sub_size = list_size / (delta_search_threads - i);
1995
bf874896
NP
1996 /* don't use too small segments or no deltas will be found */
1997 if (sub_size < 2*window && i+1 < delta_search_threads)
1998 sub_size = 0;
1999
8ecce684
NP
2000 p[i].window = window;
2001 p[i].depth = depth;
2002 p[i].processed = processed;
50f22ada
JS
2003 p[i].working = 1;
2004 p[i].data_ready = 0;
c2a33679 2005
59921b4b 2006 /* try to split chunks on "path" boundaries */
6fc74703
NP
2007 while (sub_size && sub_size < list_size &&
2008 list[sub_size]->hash &&
384b32c0
NP
2009 list[sub_size]->hash == list[sub_size-1]->hash)
2010 sub_size++;
2011
50f22ada
JS
2012 p[i].list = list;
2013 p[i].list_size = sub_size;
2014 p[i].remaining = sub_size;
59921b4b 2015
384b32c0
NP
2016 list += sub_size;
2017 list_size -= sub_size;
2018 }
2019
50f22ada
JS
2020 /* Start work threads. */
2021 for (i = 0; i < delta_search_threads; i++) {
2022 if (!p[i].list_size)
2023 continue;
68e6a4f8
JS
2024 pthread_mutex_init(&p[i].mutex, NULL);
2025 pthread_cond_init(&p[i].cond, NULL);
50f22ada
JS
2026 ret = pthread_create(&p[i].thread, NULL,
2027 threaded_find_deltas, &p[i]);
2028 if (ret)
2029 die("unable to create thread: %s", strerror(ret));
2030 active_threads++;
2031 }
2032
384b32c0
NP
2033 /*
2034 * Now let's wait for work completion. Each time a thread is done
2035 * with its work, we steal half of the remaining work from the
2036 * thread with the largest number of unprocessed objects and give
2037 * it to that newly idle thread. This ensure good load balancing
2038 * until the remaining object list segments are simply too short
2039 * to be worth splitting anymore.
2040 */
50f22ada
JS
2041 while (active_threads) {
2042 struct thread_params *target = NULL;
384b32c0
NP
2043 struct thread_params *victim = NULL;
2044 unsigned sub_size = 0;
384b32c0
NP
2045
2046 progress_lock();
50f22ada
JS
2047 for (;;) {
2048 for (i = 0; !target && i < delta_search_threads; i++)
2049 if (!p[i].working)
2050 target = &p[i];
2051 if (target)
2052 break;
2053 pthread_cond_wait(&progress_cond, &progress_mutex);
2054 }
2055
384b32c0
NP
2056 for (i = 0; i < delta_search_threads; i++)
2057 if (p[i].remaining > 2*window &&
2058 (!victim || victim->remaining < p[i].remaining))
2059 victim = &p[i];
2060 if (victim) {
2061 sub_size = victim->remaining / 2;
2062 list = victim->list + victim->list_size - sub_size;
2063 while (sub_size && list[0]->hash &&
2064 list[0]->hash == list[-1]->hash) {
2065 list++;
2066 sub_size--;
2067 }
eb9688ff
NP
2068 if (!sub_size) {
2069 /*
2070 * It is possible for some "paths" to have
2071 * so many objects that no hash boundary
2072 * might be found. Let's just steal the
2073 * exact half in that case.
2074 */
2075 sub_size = victim->remaining / 2;
2076 list -= sub_size;
2077 }
384b32c0
NP
2078 target->list = list;
2079 victim->list_size -= sub_size;
2080 victim->remaining -= sub_size;
2081 }
384b32c0
NP
2082 target->list_size = sub_size;
2083 target->remaining = sub_size;
50f22ada
JS
2084 target->working = 1;
2085 progress_unlock();
2086
2087 pthread_mutex_lock(&target->mutex);
2088 target->data_ready = 1;
2089 pthread_cond_signal(&target->cond);
2090 pthread_mutex_unlock(&target->mutex);
c2a33679 2091
384b32c0 2092 if (!sub_size) {
b81d9af7 2093 pthread_join(target->thread, NULL);
50f22ada
JS
2094 pthread_cond_destroy(&target->cond);
2095 pthread_mutex_destroy(&target->mutex);
384b32c0 2096 active_threads--;
c2a33679 2097 }
50f22ada 2098 }
44626dc7 2099 cleanup_threaded_search();
dcda3614 2100 free(p);
8ecce684
NP
2101}
2102
2103#else
384b32c0 2104#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
8ecce684
NP
2105#endif
2106
d155254c 2107static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
f0a24aa5 2108{
d155254c 2109 struct object_id peeled;
f0a24aa5 2110
59556548 2111 if (starts_with(path, "refs/tags/") && /* is a tag? */
d155254c
MH
2112 !peel_ref(path, peeled.hash) && /* peelable? */
2113 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */
2114 add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
f0a24aa5
SP
2115 return 0;
2116}
2117
f3123c4a
JH
2118static void prepare_pack(int window, int depth)
2119{
9668cf59 2120 struct object_entry **delta_list;
6e1c2344
RJ
2121 uint32_t i, nr_deltas;
2122 unsigned n;
9668cf59 2123
3f9ac8d2 2124 get_object_details();
9668cf59 2125
0e8189e2
NP
2126 /*
2127 * If we're locally repacking then we need to be doubly careful
2128 * from now on in order to make sure no stealth corruption gets
2129 * propagated to the new pack. Clients receiving streamed packs
2130 * should validate everything they get anyway so no need to incur
2131 * the additional cost here in that case.
2132 */
2133 if (!pack_to_stdout)
2134 do_check_packed_object_crc = 1;
2135
2834bc27 2136 if (!to_pack.nr_objects || !window || !depth)
9668cf59
NP
2137 return;
2138
b32fa95f 2139 ALLOC_ARRAY(delta_list, to_pack.nr_objects);
75d39853
NP
2140 nr_deltas = n = 0;
2141
2834bc27
VM
2142 for (i = 0; i < to_pack.nr_objects; i++) {
2143 struct object_entry *entry = to_pack.objects + i;
75d39853
NP
2144
2145 if (entry->delta)
2146 /* This happens if we decided to reuse existing
a7de7130 2147 * delta from a pack. "reuse_delta &&" is implied.
75d39853
NP
2148 */
2149 continue;
2150
2151 if (entry->size < 50)
2152 continue;
2153
2154 if (entry->no_try_delta)
2155 continue;
2156
6d6f9cdd 2157 if (!entry->preferred_base) {
75d39853 2158 nr_deltas++;
6d6f9cdd
SP
2159 if (entry->type < 0)
2160 die("unable to get type of object %s",
2161 sha1_to_hex(entry->idx.sha1));
eede9f42
NP
2162 } else {
2163 if (entry->type < 0) {
2164 /*
2165 * This object is not found, but we
2166 * don't have to include it anyway.
2167 */
2168 continue;
2169 }
6d6f9cdd 2170 }
75d39853
NP
2171
2172 delta_list[n++] = entry;
2173 }
2174
2f8b8947 2175 if (nr_deltas && n > 1) {
e334977d
NP
2176 unsigned nr_done = 0;
2177 if (progress)
754dbc43 2178 progress_state = start_progress(_("Compressing objects"),
dc6a0757 2179 nr_deltas);
75d39853 2180 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 2181 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 2182 stop_progress(&progress_state);
e334977d
NP
2183 if (nr_done != nr_deltas)
2184 die("inconsistency with delta count");
75d39853 2185 }
9668cf59 2186 free(delta_list);
f3123c4a
JH
2187}
2188
ef90d6d4 2189static int git_pack_config(const char *k, const char *v, void *cb)
4812a93a 2190{
eeefa7c9 2191 if (!strcmp(k, "pack.window")) {
4812a93a
JK
2192 window = git_config_int(k, v);
2193 return 0;
2194 }
a97773ce
BD
2195 if (!strcmp(k, "pack.windowmemory")) {
2196 window_memory_limit = git_config_ulong(k, v);
2197 return 0;
2198 }
2199 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
2200 depth = git_config_int(k, v);
2201 return 0;
2202 }
960ccca6
DH
2203 if (!strcmp(k, "pack.compression")) {
2204 int level = git_config_int(k, v);
2205 if (level == -1)
2206 level = Z_DEFAULT_COMPRESSION;
2207 else if (level < 0 || level > Z_BEST_COMPRESSION)
2208 die("bad pack compression level %d", level);
2209 pack_compression_level = level;
2210 pack_compression_seen = 1;
2211 return 0;
2212 }
074b2eea
MK
2213 if (!strcmp(k, "pack.deltacachesize")) {
2214 max_delta_cache_size = git_config_int(k, v);
2215 return 0;
2216 }
e3dfddb3
MK
2217 if (!strcmp(k, "pack.deltacachelimit")) {
2218 cache_max_small_delta_size = git_config_int(k, v);
2219 return 0;
2220 }
ae4f07fb
VM
2221 if (!strcmp(k, "pack.writebitmaphashcache")) {
2222 if (git_config_bool(k, v))
2223 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2224 else
2225 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2226 }
6b8fda2d
VM
2227 if (!strcmp(k, "pack.usebitmaps")) {
2228 use_bitmap_index = git_config_bool(k, v);
2229 return 0;
2230 }
693b86ff
NP
2231 if (!strcmp(k, "pack.threads")) {
2232 delta_search_threads = git_config_int(k, v);
833e3df1 2233 if (delta_search_threads < 0)
693b86ff
NP
2234 die("invalid number of threads specified (%d)",
2235 delta_search_threads);
7eb151d6 2236#ifdef NO_PTHREADS
833e3df1 2237 if (delta_search_threads != 1)
693b86ff
NP
2238 warning("no threads support, ignoring %s", k);
2239#endif
2240 return 0;
2241 }
4d00bda2 2242 if (!strcmp(k, "pack.indexversion")) {
ebcfb379
JH
2243 pack_idx_opts.version = git_config_int(k, v);
2244 if (pack_idx_opts.version > 2)
6e1c2344 2245 die("bad pack.indexversion=%"PRIu32,
ebcfb379 2246 pack_idx_opts.version);
4d00bda2
NP
2247 return 0;
2248 }
ef90d6d4 2249 return git_default_config(k, v, cb);
4812a93a
JK
2250}
2251
b5d97e6b 2252static void read_object_list_from_stdin(void)
c323ac7d 2253{
b5d97e6b
JH
2254 char line[40 + 1 + PATH_MAX + 2];
2255 unsigned char sha1[20];
b2504a0d 2256
da93d12b 2257 for (;;) {
da93d12b
LT
2258 if (!fgets(line, sizeof(line), stdin)) {
2259 if (feof(stdin))
2260 break;
2261 if (!ferror(stdin))
2262 die("fgets returned NULL, not EOF, not error!");
687dd75c 2263 if (errno != EINTR)
d824cbba 2264 die_errno("fgets");
687dd75c
JH
2265 clearerr(stdin);
2266 continue;
da93d12b 2267 }
7a979d99
JH
2268 if (line[0] == '-') {
2269 if (get_sha1_hex(line+1, sha1))
2270 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 2271 line);
8d1d8f83 2272 add_preferred_base(sha1);
7a979d99 2273 continue;
21fcd1bd 2274 }
c323ac7d 2275 if (get_sha1_hex(line, sha1))
ef07618f 2276 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 2277
bc32fed5
JH
2278 add_preferred_base_object(line+41);
2279 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 2280 }
b5d97e6b
JH
2281}
2282
08cdfb13
JH
2283#define OBJECT_ADDED (1u<<20)
2284
11c211fa 2285static void show_commit(struct commit *commit, void *data)
b5d97e6b 2286{
ed1c9977 2287 add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
08cdfb13 2288 commit->object.flags |= OBJECT_ADDED;
7cc8f971
VM
2289
2290 if (write_bitmap_index)
2291 index_commit_for_bitmap(commit);
b5d97e6b
JH
2292}
2293
de1e67d0 2294static void show_object(struct object *obj, const char *name, void *data)
b5d97e6b 2295{
8d2dfc49 2296 add_preferred_base_object(name);
ed1c9977 2297 add_object_entry(obj->oid.hash, obj->type, name, 0);
8d2dfc49 2298 obj->flags |= OBJECT_ADDED;
b5d97e6b
JH
2299}
2300
8d1d8f83
JH
2301static void show_edge(struct commit *commit)
2302{
ed1c9977 2303 add_preferred_base(commit->object.oid.hash);
8d1d8f83
JH
2304}
2305
08cdfb13
JH
2306struct in_pack_object {
2307 off_t offset;
2308 struct object *object;
2309};
2310
2311struct in_pack {
2312 int alloc;
2313 int nr;
2314 struct in_pack_object *array;
2315};
2316
2317static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2318{
ed1c9977 2319 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
08cdfb13
JH
2320 in_pack->array[in_pack->nr].object = object;
2321 in_pack->nr++;
2322}
2323
2324/*
2325 * Compare the objects in the offset order, in order to emulate the
f18d244a 2326 * "git rev-list --objects" output that produced the pack originally.
08cdfb13
JH
2327 */
2328static int ofscmp(const void *a_, const void *b_)
2329{
2330 struct in_pack_object *a = (struct in_pack_object *)a_;
2331 struct in_pack_object *b = (struct in_pack_object *)b_;
2332
2333 if (a->offset < b->offset)
2334 return -1;
2335 else if (a->offset > b->offset)
2336 return 1;
2337 else
f2fd0760 2338 return oidcmp(&a->object->oid, &b->object->oid);
08cdfb13
JH
2339}
2340
2341static void add_objects_in_unpacked_packs(struct rev_info *revs)
2342{
2343 struct packed_git *p;
2344 struct in_pack in_pack;
2345 uint32_t i;
2346
2347 memset(&in_pack, 0, sizeof(in_pack));
2348
2349 for (p = packed_git; p; p = p->next) {
2350 const unsigned char *sha1;
2351 struct object *o;
2352
79bc4c71 2353 if (!p->pack_local || p->pack_keep)
08cdfb13
JH
2354 continue;
2355 if (open_pack_index(p))
2356 die("cannot open pack index");
2357
2358 ALLOC_GROW(in_pack.array,
2359 in_pack.nr + p->num_objects,
2360 in_pack.alloc);
2361
2362 for (i = 0; i < p->num_objects; i++) {
2363 sha1 = nth_packed_object_sha1(p, i);
2364 o = lookup_unknown_object(sha1);
2365 if (!(o->flags & OBJECT_ADDED))
2366 mark_in_pack_object(o, p, &in_pack);
2367 o->flags |= OBJECT_ADDED;
2368 }
2369 }
2370
2371 if (in_pack.nr) {
2372 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2373 ofscmp);
2374 for (i = 0; i < in_pack.nr; i++) {
2375 struct object *o = in_pack.array[i].object;
ed1c9977 2376 add_object_entry(o->oid.hash, o->type, "", 0);
08cdfb13
JH
2377 }
2378 }
2379 free(in_pack.array);
2380}
2381
094085e3
BC
2382static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2383{
2384 static struct packed_git *last_found = (void *)1;
2385 struct packed_git *p;
2386
2387 p = (last_found != (void *)1) ? last_found : packed_git;
2388
2389 while (p) {
2390 if ((!p->pack_local || p->pack_keep) &&
2391 find_pack_entry_one(sha1, p)) {
2392 last_found = p;
2393 return 1;
2394 }
2395 if (p == last_found)
2396 p = packed_git;
2397 else
2398 p = p->next;
2399 if (p == last_found)
2400 p = p->next;
2401 }
2402 return 0;
2403}
2404
abcb8655
JK
2405/*
2406 * Store a list of sha1s that are should not be discarded
2407 * because they are either written too recently, or are
2408 * reachable from another object that was.
2409 *
2410 * This is filled by get_object_list.
2411 */
2412static struct sha1_array recent_objects;
2413
d0d46abc
JK
2414static int loosened_object_can_be_discarded(const unsigned char *sha1,
2415 unsigned long mtime)
2416{
2417 if (!unpack_unreachable_expiration)
2418 return 0;
2419 if (mtime > unpack_unreachable_expiration)
2420 return 0;
abcb8655
JK
2421 if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2422 return 0;
d0d46abc
JK
2423 return 1;
2424}
2425
ca11b212
NP
2426static void loosen_unused_packed_objects(struct rev_info *revs)
2427{
2428 struct packed_git *p;
2429 uint32_t i;
2430 const unsigned char *sha1;
2431
2432 for (p = packed_git; p; p = p->next) {
79bc4c71 2433 if (!p->pack_local || p->pack_keep)
ca11b212
NP
2434 continue;
2435
2436 if (open_pack_index(p))
2437 die("cannot open pack index");
2438
2439 for (i = 0; i < p->num_objects; i++) {
2440 sha1 = nth_packed_object_sha1(p, i);
2834bc27 2441 if (!packlist_find(&to_pack, sha1, NULL) &&
d0d46abc
JK
2442 !has_sha1_pack_kept_or_nonlocal(sha1) &&
2443 !loosened_object_can_be_discarded(sha1, p->mtime))
ca11b212
NP
2444 if (force_object_loose(sha1, p->mtime))
2445 die("unable to force loose object");
2446 }
2447 }
2448}
2449
69e4b342
JK
2450/*
2451 * This tracks any options which a reader of the pack might
2452 * not understand, and which would therefore prevent blind reuse
2453 * of what we have on disk.
2454 */
2455static int pack_options_allow_reuse(void)
2456{
2457 return allow_ofs_delta;
2458}
2459
6b8fda2d
VM
2460static int get_object_list_from_bitmap(struct rev_info *revs)
2461{
2462 if (prepare_bitmap_walk(revs) < 0)
2463 return -1;
2464
69e4b342
JK
2465 if (pack_options_allow_reuse() &&
2466 !reuse_partial_packfile_from_bitmap(
6b8fda2d
VM
2467 &reuse_packfile,
2468 &reuse_packfile_objects,
2469 &reuse_packfile_offset)) {
2470 assert(reuse_packfile_objects);
2471 nr_result += reuse_packfile_objects;
78d2214e 2472 display_progress(progress_state, nr_result);
6b8fda2d
VM
2473 }
2474
2475 traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2476 return 0;
2477}
2478
abcb8655 2479static void record_recent_object(struct object *obj,
de1e67d0 2480 const char *name,
abcb8655
JK
2481 void *data)
2482{
ed1c9977 2483 sha1_array_append(&recent_objects, obj->oid.hash);
abcb8655
JK
2484}
2485
2486static void record_recent_commit(struct commit *commit, void *data)
2487{
ed1c9977 2488 sha1_array_append(&recent_objects, commit->object.oid.hash);
abcb8655
JK
2489}
2490
8d1d8f83 2491static void get_object_list(int ac, const char **av)
b5d97e6b
JH
2492{
2493 struct rev_info revs;
2494 char line[1000];
b5d97e6b
JH
2495 int flags = 0;
2496
b5d97e6b
JH
2497 init_revisions(&revs, NULL);
2498 save_commit_buffer = 0;
b5d97e6b
JH
2499 setup_revisions(ac, av, &revs, NULL);
2500
b790e0f6
NTND
2501 /* make sure shallows are read */
2502 is_repository_shallow();
2503
b5d97e6b
JH
2504 while (fgets(line, sizeof(line), stdin) != NULL) {
2505 int len = strlen(line);
872c930d 2506 if (len && line[len - 1] == '\n')
b5d97e6b
JH
2507 line[--len] = 0;
2508 if (!len)
2509 break;
2510 if (*line == '-') {
2511 if (!strcmp(line, "--not")) {
2512 flags ^= UNINTERESTING;
7cc8f971 2513 write_bitmap_index = 0;
b5d97e6b
JH
2514 continue;
2515 }
b790e0f6
NTND
2516 if (starts_with(line, "--shallow ")) {
2517 unsigned char sha1[20];
2518 if (get_sha1_hex(line + 10, sha1))
2519 die("not an SHA-1 '%s'", line + 10);
2520 register_shallow(sha1);
f7f91086 2521 use_bitmap_index = 0;
b790e0f6
NTND
2522 continue;
2523 }
b5d97e6b
JH
2524 die("not a rev '%s'", line);
2525 }
8e676e8b 2526 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
b5d97e6b
JH
2527 die("bad revision '%s'", line);
2528 }
2529
6b8fda2d
VM
2530 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2531 return;
2532
3d51e1b5
MK
2533 if (prepare_revision_walk(&revs))
2534 die("revision walk setup failed");
e76a5fb4 2535 mark_edges_uninteresting(&revs, show_edge);
11c211fa 2536 traverse_commit_list(&revs, show_commit, show_object, NULL);
08cdfb13 2537
abcb8655
JK
2538 if (unpack_unreachable_expiration) {
2539 revs.ignore_missing_links = 1;
2540 if (add_unseen_recent_objects_to_traversal(&revs,
2541 unpack_unreachable_expiration))
2542 die("unable to add recent objects");
2543 if (prepare_revision_walk(&revs))
2544 die("revision walk setup failed");
2545 traverse_commit_list(&revs, record_recent_commit,
2546 record_recent_object, NULL);
2547 }
2548
08cdfb13
JH
2549 if (keep_unreachable)
2550 add_objects_in_unpacked_packs(&revs);
ca11b212
NP
2551 if (unpack_unreachable)
2552 loosen_unused_packed_objects(&revs);
abcb8655
JK
2553
2554 sha1_array_clear(&recent_objects);
b5d97e6b
JH
2555}
2556
99fb6e04
NTND
2557static int option_parse_index_version(const struct option *opt,
2558 const char *arg, int unset)
2559{
2560 char *c;
2561 const char *val = arg;
2562 pack_idx_opts.version = strtoul(val, &c, 10);
2563 if (pack_idx_opts.version > 2)
2564 die(_("unsupported index version %s"), val);
2565 if (*c == ',' && c[1])
2566 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2567 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2568 die(_("bad index version '%s'"), val);
2569 return 0;
2570}
2571
7e52f566
JK
2572static int option_parse_unpack_unreachable(const struct option *opt,
2573 const char *arg, int unset)
2574{
2575 if (unset) {
2576 unpack_unreachable = 0;
2577 unpack_unreachable_expiration = 0;
2578 }
2579 else {
2580 unpack_unreachable = 1;
2581 if (arg)
2582 unpack_unreachable_expiration = approxidate(arg);
2583 }
2584 return 0;
2585}
2586
b5d97e6b
JH
2587int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2588{
b5d97e6b 2589 int use_internal_rev_list = 0;
8d1d8f83 2590 int thin = 0;
2dacf26d 2591 int shallow = 0;
4f366275 2592 int all_progress_implied = 0;
edfbb2aa 2593 struct argv_array rp = ARGV_ARRAY_INIT;
99fb6e04 2594 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
c90f9e13 2595 int rev_list_index = 0;
99fb6e04
NTND
2596 struct option pack_objects_options[] = {
2597 OPT_SET_INT('q', "quiet", &progress,
4c688120 2598 N_("do not show progress meter"), 0),
99fb6e04 2599 OPT_SET_INT(0, "progress", &progress,
4c688120 2600 N_("show progress meter"), 1),
99fb6e04 2601 OPT_SET_INT(0, "all-progress", &progress,
4c688120 2602 N_("show progress meter during object writing phase"), 2),
99fb6e04
NTND
2603 OPT_BOOL(0, "all-progress-implied",
2604 &all_progress_implied,
4c688120
NTND
2605 N_("similar to --all-progress when progress meter is shown")),
2606 { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2607 N_("write the pack index file in the specified idx format version"),
99fb6e04 2608 0, option_parse_index_version },
2a514ed8
CB
2609 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2610 N_("maximum size of each output pack file")),
99fb6e04 2611 OPT_BOOL(0, "local", &local,
4c688120 2612 N_("ignore borrowed objects from alternate object store")),
99fb6e04 2613 OPT_BOOL(0, "incremental", &incremental,
4c688120 2614 N_("ignore packed objects")),
99fb6e04 2615 OPT_INTEGER(0, "window", &window,
4c688120 2616 N_("limit pack window by objects")),
2a514ed8
CB
2617 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2618 N_("limit pack window by memory in addition to object limit")),
99fb6e04 2619 OPT_INTEGER(0, "depth", &depth,
4c688120 2620 N_("maximum length of delta chain allowed in the resulting pack")),
99fb6e04 2621 OPT_BOOL(0, "reuse-delta", &reuse_delta,
4c688120 2622 N_("reuse existing deltas")),
99fb6e04 2623 OPT_BOOL(0, "reuse-object", &reuse_object,
4c688120 2624 N_("reuse existing objects")),
99fb6e04 2625 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
4c688120 2626 N_("use OFS_DELTA objects")),
99fb6e04 2627 OPT_INTEGER(0, "threads", &delta_search_threads,
4c688120 2628 N_("use threads when searching for best delta matches")),
99fb6e04 2629 OPT_BOOL(0, "non-empty", &non_empty,
4c688120 2630 N_("do not create an empty pack output")),
99fb6e04 2631 OPT_BOOL(0, "revs", &use_internal_rev_list,
4c688120 2632 N_("read revision arguments from standard input")),
99fb6e04 2633 { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
4c688120 2634 N_("limit the objects to those that are not yet packed"),
99fb6e04
NTND
2635 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2636 { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
4c688120 2637 N_("include objects reachable from any reference"),
99fb6e04
NTND
2638 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2639 { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
4c688120 2640 N_("include objects referred by reflog entries"),
99fb6e04 2641 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
c90f9e13
JK
2642 { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2643 N_("include objects referred to by the index"),
2644 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
99fb6e04 2645 OPT_BOOL(0, "stdout", &pack_to_stdout,
4c688120 2646 N_("output pack to stdout")),
99fb6e04 2647 OPT_BOOL(0, "include-tag", &include_tag,
4c688120 2648 N_("include tag objects that refer to objects to be packed")),
99fb6e04 2649 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
4c688120
NTND
2650 N_("keep unreachable objects")),
2651 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2652 N_("unpack unreachable objects newer than <time>"),
7e52f566 2653 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
99fb6e04 2654 OPT_BOOL(0, "thin", &thin,
4c688120 2655 N_("create thin packs")),
2dacf26d 2656 OPT_BOOL(0, "shallow", &shallow,
2657 N_("create packs suitable for shallow fetches")),
99fb6e04 2658 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
4c688120 2659 N_("ignore packs that have companion .keep file")),
99fb6e04 2660 OPT_INTEGER(0, "compression", &pack_compression_level,
4c688120 2661 N_("pack compression level")),
99fb6e04 2662 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
4c688120 2663 N_("do not hide commits by grafts"), 0),
6b8fda2d
VM
2664 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2665 N_("use a bitmap index if available to speed up counting objects")),
7cc8f971
VM
2666 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2667 N_("write a bitmap index together with the pack index")),
99fb6e04
NTND
2668 OPT_END(),
2669 };
8d1d8f83 2670
afc711b8 2671 check_replace_refs = 0;
dae556bd 2672
ebcfb379 2673 reset_pack_idx_option(&pack_idx_opts);
ef90d6d4 2674 git_config(git_pack_config, NULL);
960ccca6
DH
2675 if (!pack_compression_seen && core_compression_seen)
2676 pack_compression_level = core_compression_level;
b5d97e6b
JH
2677
2678 progress = isatty(2);
99fb6e04
NTND
2679 argc = parse_options(argc, argv, prefix, pack_objects_options,
2680 pack_usage, 0);
b5d97e6b 2681
99fb6e04
NTND
2682 if (argc) {
2683 base_name = argv[0];
2684 argc--;
b5d97e6b 2685 }
99fb6e04
NTND
2686 if (pack_to_stdout != !base_name || argc)
2687 usage_with_options(pack_usage, pack_objects_options);
b5d97e6b 2688
edfbb2aa 2689 argv_array_push(&rp, "pack-objects");
99fb6e04
NTND
2690 if (thin) {
2691 use_internal_rev_list = 1;
2dacf26d 2692 argv_array_push(&rp, shallow
2693 ? "--objects-edge-aggressive"
2694 : "--objects-edge");
99fb6e04 2695 } else
edfbb2aa 2696 argv_array_push(&rp, "--objects");
b5d97e6b 2697
99fb6e04
NTND
2698 if (rev_list_all) {
2699 use_internal_rev_list = 1;
edfbb2aa 2700 argv_array_push(&rp, "--all");
99fb6e04
NTND
2701 }
2702 if (rev_list_reflog) {
2703 use_internal_rev_list = 1;
edfbb2aa 2704 argv_array_push(&rp, "--reflog");
99fb6e04 2705 }
c90f9e13
JK
2706 if (rev_list_index) {
2707 use_internal_rev_list = 1;
2708 argv_array_push(&rp, "--indexed-objects");
99fb6e04
NTND
2709 }
2710 if (rev_list_unpacked) {
2711 use_internal_rev_list = 1;
edfbb2aa 2712 argv_array_push(&rp, "--unpacked");
99fb6e04 2713 }
b5d97e6b 2714
99fb6e04
NTND
2715 if (!reuse_object)
2716 reuse_delta = 0;
2717 if (pack_compression_level == -1)
2718 pack_compression_level = Z_DEFAULT_COMPRESSION;
2719 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2720 die("bad pack compression level %d", pack_compression_level);
0c45d258
JH
2721
2722 if (!delta_search_threads) /* --threads=0 means autodetect */
2723 delta_search_threads = online_cpus();
2724
99fb6e04
NTND
2725#ifdef NO_PTHREADS
2726 if (delta_search_threads != 1)
2b34e486 2727 warning("no threads support, ignoring --threads");
99fb6e04 2728#endif
2b84b5a8
JS
2729 if (!pack_to_stdout && !pack_size_limit)
2730 pack_size_limit = pack_size_limit_cfg;
01c12a23
DH
2731 if (pack_to_stdout && pack_size_limit)
2732 die("--max-pack-size cannot be used to build a pack for transfer.");
07cf0f24
NP
2733 if (pack_size_limit && pack_size_limit < 1024*1024) {
2734 warning("minimum pack size limit is 1 MiB");
2735 pack_size_limit = 1024*1024;
2736 }
01c12a23 2737
8d1d8f83
JH
2738 if (!pack_to_stdout && thin)
2739 die("--thin cannot be used to build an indexable pack.");
b5d97e6b 2740
ca11b212
NP
2741 if (keep_unreachable && unpack_unreachable)
2742 die("--keep-unreachable and --unpack-unreachable are incompatible.");
b1e757f3
JK
2743 if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2744 unpack_unreachable_expiration = 0;
ca11b212 2745
6b8fda2d
VM
2746 if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2747 use_bitmap_index = 0;
2748
7cc8f971
VM
2749 if (pack_to_stdout || !rev_list_all)
2750 write_bitmap_index = 0;
2751
4f366275
NP
2752 if (progress && all_progress_implied)
2753 progress = 2;
2754
b5d97e6b
JH
2755 prepare_packed_git();
2756
13aaf148 2757 if (progress)
754dbc43 2758 progress_state = start_progress(_("Counting objects"), 0);
b5d97e6b
JH
2759 if (!use_internal_rev_list)
2760 read_object_list_from_stdin();
8d1d8f83 2761 else {
edfbb2aa
JK
2762 get_object_list(rp.argc, rp.argv);
2763 argv_array_clear(&rp);
8d1d8f83 2764 }
0ef95f72 2765 cleanup_preferred_base();
d155254c
MH
2766 if (include_tag && nr_result)
2767 for_each_ref(add_ref_tag, NULL);
4d4fcc54 2768 stop_progress(&progress_state);
96a02f8f 2769
f0b0af1b 2770 if (non_empty && !nr_result)
1c4a2912 2771 return 0;
f7ae6a93
NP
2772 if (nr_result)
2773 prepare_pack(window, depth);
d01fb92f 2774 write_pack_file();
ab7cd7bb 2775 if (progress)
6e1c2344
RJ
2776 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2777 " reused %"PRIu32" (delta %"PRIu32")\n",
67c08ce1 2778 written, written_delta, reused, reused_delta);
c323ac7d
LT
2779 return 0;
2780}