pack-objects: reverse the delta search sort list
[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"
c38138cd 11#include "csum-file.h"
1b0c7174 12#include "tree-walk.h"
b5d97e6b
JH
13#include "diff.h"
14#include "revision.h"
15#include "list-objects.h"
96a02f8f 16#include "progress.h"
c323ac7d 17
8ecce684
NP
18#ifdef THREADED_DELTA_SEARCH
19#include <pthread.h>
20#endif
21
231f240b 22static const char pack_usage[] = "\
a97773ce
BD
23git-pack-objects [{ -q | --progress | --all-progress }] \n\
24 [--max-pack-size=N] [--local] [--incremental] \n\
25 [--window=N] [--window-memory=N] [--depth=N] \n\
fa736f72 26 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
367f4a43 27 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
08cdfb13 28 [--stdout | base-name] [--keep-unreachable] [<ref-list | <object-list]";
c323ac7d 29
c323ac7d 30struct object_entry {
aa7e44bf 31 struct pack_idx_entry idx;
3f9ac8d2 32 unsigned long size; /* uncompressed size */
3f9ac8d2 33 struct packed_git *in_pack; /* already in pack */
6777a59f 34 off_t in_pack_offset;
a3fbf4df 35 struct object_entry *delta; /* delta base object */
82e5a82f 36 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
37 struct object_entry *delta_sibling; /* other deltified objects who
38 * uses the same base as me
39 */
074b2eea 40 void *delta_data; /* cached delta (uncompressed) */
a3fbf4df 41 unsigned long delta_size; /* delta data size (uncompressed) */
5a235b5e 42 unsigned int hash; /* name hint hash */
a3fbf4df
NP
43 enum object_type type;
44 enum object_type in_pack_type; /* could be delta */
45 unsigned char in_pack_header_size;
46 unsigned char preferred_base; /* we do not pack this, but is available
a74db82e 47 * to be used as the base object to delta
a3fbf4df
NP
48 * objects against.
49 */
a74db82e 50 unsigned char no_try_delta;
c323ac7d
LT
51};
52
3f9ac8d2 53/*
82e5a82f 54 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
55 * expanded). nr_objects & nr_alloc controls this array. They are stored
56 * in the order we see -- typically rev-list --objects order that gives us
57 * nice "minimum seek" order.
3f9ac8d2 58 */
e4d58311 59static struct object_entry *objects;
79814f42 60static struct pack_idx_entry **written_list;
d01fb92f 61static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
3f9ac8d2 62
96f1e58f 63static int non_empty;
08cdfb13 64static int no_reuse_delta, no_reuse_object, keep_unreachable;
96f1e58f
DR
65static int local;
66static int incremental;
be6b1914 67static int allow_ofs_delta;
d01fb92f 68static const char *base_name;
024701f1 69static int progress = 1;
4812a93a 70static int window = 10;
d01fb92f 71static uint32_t pack_size_limit;
618e613a 72static int depth = 50;
367f4a43 73static int delta_search_threads = 1;
df6d6101 74static int pack_to_stdout;
8d1d8f83 75static int num_preferred_base;
dc6a0757 76static struct progress *progress_state;
960ccca6
DH
77static int pack_compression_level = Z_DEFAULT_COMPRESSION;
78static int pack_compression_seen;
c323ac7d 79
074b2eea
MK
80static unsigned long delta_cache_size = 0;
81static unsigned long max_delta_cache_size = 0;
e3dfddb3 82static unsigned long cache_max_small_delta_size = 1000;
074b2eea 83
a97773ce
BD
84static unsigned long window_memory_limit = 0;
85
3f9ac8d2
JH
86/*
87 * The object names in objects array are hashed with this hashtable,
e4d58311 88 * to help looking up the entry by object name.
3f9ac8d2
JH
89 * This hashtable is built after all the objects are seen.
90 */
96f1e58f
DR
91static int *object_ix;
92static int object_ix_hashsz;
3f9ac8d2
JH
93
94/*
95 * Pack index for existing packs give us easy access to the offsets into
96 * corresponding pack file where each object's data starts, but the entries
97 * do not store the size of the compressed representation (uncompressed
780e6e73
NP
98 * size is easily available by examining the pack entry header). It is
99 * also rather expensive to find the sha1 for an object given its offset.
100 *
101 * We build a hashtable of existing packs (pack_revindex), and keep reverse
102 * index here -- pack index file is sorted by object name mapping to offset;
103 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
104 * ordered by offset, so if you know the offset of an object, next offset
105 * is where its packed representation ends and the index_nr can be used to
106 * get the object sha1 from the main index.
3f9ac8d2 107 */
780e6e73 108struct revindex_entry {
6777a59f 109 off_t offset;
780e6e73
NP
110 unsigned int nr;
111};
3f9ac8d2
JH
112struct pack_revindex {
113 struct packed_git *p;
780e6e73
NP
114 struct revindex_entry *revindex;
115};
116static struct pack_revindex *pack_revindex;
96f1e58f 117static int pack_revindex_hashsz;
3f9ac8d2
JH
118
119/*
120 * stats
121 */
7cadf491
SP
122static uint32_t written, written_delta;
123static uint32_t reused, reused_delta;
3f9ac8d2
JH
124
125static int pack_revindex_ix(struct packed_git *p)
126{
2b74cffa 127 unsigned long ui = (unsigned long)p;
3f9ac8d2
JH
128 int i;
129
130 ui = ui ^ (ui >> 16); /* defeat structure alignment */
131 i = (int)(ui % pack_revindex_hashsz);
132 while (pack_revindex[i].p) {
133 if (pack_revindex[i].p == p)
134 return i;
135 if (++i == pack_revindex_hashsz)
136 i = 0;
137 }
138 return -1 - i;
139}
140
141static void prepare_pack_ix(void)
142{
143 int num;
144 struct packed_git *p;
145 for (num = 0, p = packed_git; p; p = p->next)
146 num++;
147 if (!num)
148 return;
149 pack_revindex_hashsz = num * 11;
150 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
151 for (p = packed_git; p; p = p->next) {
152 num = pack_revindex_ix(p);
153 num = - 1 - num;
154 pack_revindex[num].p = p;
155 }
156 /* revindex elements are lazily initialized */
157}
158
159static int cmp_offset(const void *a_, const void *b_)
160{
780e6e73
NP
161 const struct revindex_entry *a = a_;
162 const struct revindex_entry *b = b_;
163 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
3f9ac8d2
JH
164}
165
166/*
167 * Ordered list of offsets of objects in the pack.
168 */
169static void prepare_pack_revindex(struct pack_revindex *rix)
170{
171 struct packed_git *p = rix->p;
57059091 172 int num_ent = p->num_objects;
3f9ac8d2 173 int i;
42873078 174 const char *index = p->index_data;
3f9ac8d2 175
780e6e73 176 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
c553ca25
NP
177 index += 4 * 256;
178
179 if (p->index_version > 1) {
180 const uint32_t *off_32 =
181 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
182 const uint32_t *off_64 = off_32 + p->num_objects;
183 for (i = 0; i < num_ent; i++) {
184 uint32_t off = ntohl(*off_32++);
185 if (!(off & 0x80000000)) {
186 rix->revindex[i].offset = off;
187 } else {
188 rix->revindex[i].offset =
189 ((uint64_t)ntohl(*off_64++)) << 32;
190 rix->revindex[i].offset |=
191 ntohl(*off_64++);
192 }
193 rix->revindex[i].nr = i;
194 }
195 } else {
196 for (i = 0; i < num_ent; i++) {
197 uint32_t hl = *((uint32_t *)(index + 24 * i));
198 rix->revindex[i].offset = ntohl(hl);
199 rix->revindex[i].nr = i;
200 }
3f9ac8d2 201 }
c553ca25 202
3f9ac8d2
JH
203 /* This knows the pack format -- the 20-byte trailer
204 * follows immediately after the last object data.
205 */
780e6e73
NP
206 rix->revindex[num_ent].offset = p->pack_size - 20;
207 rix->revindex[num_ent].nr = -1;
208 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
3f9ac8d2
JH
209}
210
780e6e73 211static struct revindex_entry * find_packed_object(struct packed_git *p,
6777a59f 212 off_t ofs)
3f9ac8d2
JH
213{
214 int num;
215 int lo, hi;
216 struct pack_revindex *rix;
780e6e73 217 struct revindex_entry *revindex;
3f9ac8d2
JH
218 num = pack_revindex_ix(p);
219 if (num < 0)
220 die("internal error: pack revindex uninitialized");
221 rix = &pack_revindex[num];
222 if (!rix->revindex)
223 prepare_pack_revindex(rix);
224 revindex = rix->revindex;
225 lo = 0;
57059091 226 hi = p->num_objects + 1;
3f9ac8d2
JH
227 do {
228 int mi = (lo + hi) / 2;
780e6e73
NP
229 if (revindex[mi].offset == ofs) {
230 return revindex + mi;
3f9ac8d2 231 }
780e6e73 232 else if (ofs < revindex[mi].offset)
3f9ac8d2
JH
233 hi = mi;
234 else
235 lo = mi + 1;
236 } while (lo < hi);
237 die("internal error: pack revindex corrupt");
238}
239
42873078
NP
240static const unsigned char *find_packed_object_name(struct packed_git *p,
241 off_t ofs)
780e6e73
NP
242{
243 struct revindex_entry *entry = find_packed_object(p, ofs);
d72308e0 244 return nth_packed_object_sha1(p, entry->nr);
780e6e73
NP
245}
246
c323ac7d
LT
247static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
248{
249 unsigned long othersize, delta_size;
21666f1a 250 enum object_type type;
aa7e44bf 251 void *otherbuf = read_sha1_file(entry->delta->idx.sha1, &type, &othersize);
c323ac7d
LT
252 void *delta_buf;
253
254 if (!otherbuf)
aa7e44bf 255 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
8ee378a0 256 delta_buf = diff_delta(otherbuf, othersize,
dcde55bc 257 buf, size, &delta_size, 0);
75c42d8c 258 if (!delta_buf || delta_size != entry->delta_size)
a6080a0a 259 die("delta size changed");
c323ac7d
LT
260 free(buf);
261 free(otherbuf);
262 return delta_buf;
263}
264
a733cb60
LT
265/*
266 * The per-object header is a pretty dense thing, which is
267 * - first byte: low four bits are "size", then three bits of "type",
268 * and the high bit is "size continues".
269 * - each byte afterwards: low seven bits are size continuation,
270 * with the high bit being "size continues"
271 */
272static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
273{
01247d87 274 int n = 1;
a733cb60
LT
275 unsigned char c;
276
eb32d236 277 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
278 die("bad type %d", type);
279
01247d87
LT
280 c = (type << 4) | (size & 15);
281 size >>= 4;
282 while (size) {
a733cb60 283 *hdr++ = c | 0x80;
01247d87
LT
284 c = size & 0x7f;
285 size >>= 7;
286 n++;
a733cb60
LT
287 }
288 *hdr = c;
289 return n;
290}
291
780e6e73
NP
292/*
293 * we are going to reuse the existing object data as is. make
294 * sure it is not corrupt.
295 */
079afb18
SP
296static int check_pack_inflate(struct packed_git *p,
297 struct pack_window **w_curs,
6777a59f
SP
298 off_t offset,
299 off_t len,
079afb18
SP
300 unsigned long expect)
301{
302 z_stream stream;
303 unsigned char fakebuf[4096], *in;
304 int st;
305
306 memset(&stream, 0, sizeof(stream));
307 inflateInit(&stream);
308 do {
309 in = use_pack(p, w_curs, offset, &stream.avail_in);
310 stream.next_in = in;
311 stream.next_out = fakebuf;
312 stream.avail_out = sizeof(fakebuf);
313 st = inflate(&stream, Z_FINISH);
314 offset += stream.next_in - in;
315 } while (st == Z_OK || st == Z_BUF_ERROR);
316 inflateEnd(&stream);
317 return (st == Z_STREAM_END &&
318 stream.total_out == expect &&
319 stream.total_in == len) ? 0 : -1;
320}
321
91ecbeca
NP
322static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
323 off_t offset, off_t len, unsigned int nr)
324{
325 const uint32_t *index_crc;
326 uint32_t data_crc = crc32(0, Z_NULL, 0);
327
328 do {
329 unsigned int avail;
330 void *data = use_pack(p, w_curs, offset, &avail);
331 if (avail > len)
332 avail = len;
333 data_crc = crc32(data_crc, data, avail);
334 offset += avail;
335 len -= avail;
336 } while (len);
337
338 index_crc = p->index_data;
339 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
340
341 return data_crc != ntohl(*index_crc);
342}
343
079afb18
SP
344static void copy_pack_data(struct sha1file *f,
345 struct packed_git *p,
346 struct pack_window **w_curs,
6777a59f
SP
347 off_t offset,
348 off_t len)
079afb18
SP
349{
350 unsigned char *in;
351 unsigned int avail;
352
353 while (len) {
354 in = use_pack(p, w_curs, offset, &avail);
355 if (avail > len)
6777a59f 356 avail = (unsigned int)len;
079afb18
SP
357 sha1write(f, in, avail);
358 offset += avail;
359 len -= avail;
360 }
361}
362
d7dd0223 363static unsigned long write_object(struct sha1file *f,
17b08f2c
DH
364 struct object_entry *entry,
365 off_t write_offset)
c323ac7d
LT
366{
367 unsigned long size;
21666f1a 368 enum object_type type;
3f9ac8d2 369 void *buf;
a733cb60 370 unsigned char header[10];
17b08f2c 371 unsigned char dheader[10];
6777a59f
SP
372 unsigned hdrlen;
373 off_t datalen;
a733cb60 374 enum object_type obj_type;
ab7cd7bb 375 int to_reuse = 0;
17b08f2c
DH
376 /* write limit if limited packsize and not first object */
377 unsigned long limit = pack_size_limit && nr_written ?
378 pack_size_limit - write_offset : 0;
379 /* no if no delta */
380 int usable_delta = !entry->delta ? 0 :
381 /* yes if unlimited packfile */
382 !pack_size_limit ? 1 :
383 /* no if base written to previous pack */
aa7e44bf 384 entry->delta->idx.offset == (off_t)-1 ? 0 :
17b08f2c
DH
385 /* otherwise double-check written to this
386 * pack, like we do below
387 */
aa7e44bf 388 entry->delta->idx.offset ? 1 : 0;
c323ac7d 389
78d1e84f
NP
390 if (!pack_to_stdout)
391 crc32_begin(f);
392
a733cb60 393 obj_type = entry->type;
fa736f72
NP
394 if (no_reuse_object)
395 to_reuse = 0; /* explicit */
396 else if (!entry->in_pack)
ab7cd7bb 397 to_reuse = 0; /* can't reuse what we don't have */
780e6e73 398 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
17b08f2c
DH
399 /* check_object() decided it for us ... */
400 to_reuse = usable_delta;
401 /* ... but pack split may override that */
ab7cd7bb
JH
402 else if (obj_type != entry->in_pack_type)
403 to_reuse = 0; /* pack has delta which is unusable */
404 else if (entry->delta)
405 to_reuse = 0; /* we want to pack afresh */
406 else
407 to_reuse = 1; /* we have it in-pack undeltified,
408 * and we do not need to deltify it.
409 */
410
df6d6101 411 if (!to_reuse) {
17b08f2c
DH
412 z_stream stream;
413 unsigned long maxsize;
414 void *out;
5476a8ad 415 if (!usable_delta) {
aa7e44bf 416 buf = read_sha1_file(entry->idx.sha1, &obj_type, &size);
5476a8ad 417 if (!buf)
aa7e44bf 418 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
5476a8ad 419 } else if (entry->delta_data) {
3f9ac8d2 420 size = entry->delta_size;
5476a8ad
NP
421 buf = entry->delta_data;
422 entry->delta_data = NULL;
aa7e44bf 423 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
be6b1914 424 OBJ_OFS_DELTA : OBJ_REF_DELTA;
17b08f2c 425 } else {
aa7e44bf 426 buf = read_sha1_file(entry->idx.sha1, &type, &size);
074b2eea 427 if (!buf)
aa7e44bf 428 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
5476a8ad
NP
429 buf = delta_against(buf, size, entry);
430 size = entry->delta_size;
aa7e44bf 431 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
5476a8ad 432 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2 433 }
17b08f2c
DH
434 /* compress the data to store and put compressed length in datalen */
435 memset(&stream, 0, sizeof(stream));
436 deflateInit(&stream, pack_compression_level);
437 maxsize = deflateBound(&stream, size);
438 out = xmalloc(maxsize);
439 /* Compress it */
440 stream.next_in = buf;
441 stream.avail_in = size;
442 stream.next_out = out;
443 stream.avail_out = maxsize;
444 while (deflate(&stream, Z_FINISH) == Z_OK)
445 /* nothing */;
446 deflateEnd(&stream);
447 datalen = stream.total_out;
448 deflateEnd(&stream);
3f9ac8d2
JH
449 /*
450 * The object header is a byte of 'type' followed by zero or
be6b1914 451 * more bytes of length.
3f9ac8d2
JH
452 */
453 hdrlen = encode_header(obj_type, size, header);
3f9ac8d2 454
be6b1914
NP
455 if (obj_type == OBJ_OFS_DELTA) {
456 /*
457 * Deltas with relative base contain an additional
458 * encoding of the relative offset for the delta
459 * base from this object's position in the pack.
460 */
aa7e44bf 461 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
462 unsigned pos = sizeof(dheader) - 1;
463 dheader[pos] = ofs & 127;
be6b1914 464 while (ofs >>= 7)
17b08f2c
DH
465 dheader[--pos] = 128 | (--ofs & 127);
466 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
467 free(out);
468 free(buf);
469 return 0;
470 }
471 sha1write(f, header, hdrlen);
472 sha1write(f, dheader + pos, sizeof(dheader) - pos);
473 hdrlen += sizeof(dheader) - pos;
be6b1914
NP
474 } else if (obj_type == OBJ_REF_DELTA) {
475 /*
476 * Deltas with a base reference contain
477 * an additional 20 bytes for the base sha1.
478 */
17b08f2c
DH
479 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
480 free(out);
481 free(buf);
482 return 0;
483 }
484 sha1write(f, header, hdrlen);
aa7e44bf 485 sha1write(f, entry->delta->idx.sha1, 20);
3f9ac8d2 486 hdrlen += 20;
17b08f2c
DH
487 } else {
488 if (limit && hdrlen + datalen + 20 >= limit) {
489 free(out);
490 free(buf);
491 return 0;
492 }
493 sha1write(f, header, hdrlen);
3f9ac8d2 494 }
17b08f2c
DH
495 sha1write(f, out, datalen);
496 free(out);
3f9ac8d2 497 free(buf);
c323ac7d 498 }
3f9ac8d2
JH
499 else {
500 struct packed_git *p = entry->in_pack;
03e79c88 501 struct pack_window *w_curs = NULL;
91ecbeca 502 struct revindex_entry *revidx;
6777a59f 503 off_t offset;
3f9ac8d2 504
780e6e73 505 if (entry->delta) {
aa7e44bf 506 obj_type = (allow_ofs_delta && entry->delta->idx.offset) ?
780e6e73
NP
507 OBJ_OFS_DELTA : OBJ_REF_DELTA;
508 reused_delta++;
509 }
510 hdrlen = encode_header(obj_type, entry->size, header);
91ecbeca
NP
511 offset = entry->in_pack_offset;
512 revidx = find_packed_object(p, offset);
513 datalen = revidx[1].offset - offset;
514 if (!pack_to_stdout && p->index_version > 1 &&
515 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
aa7e44bf 516 die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
91ecbeca
NP
517 offset += entry->in_pack_header_size;
518 datalen -= entry->in_pack_header_size;
17b08f2c 519 if (obj_type == OBJ_OFS_DELTA) {
aa7e44bf 520 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
17b08f2c
DH
521 unsigned pos = sizeof(dheader) - 1;
522 dheader[pos] = ofs & 127;
523 while (ofs >>= 7)
524 dheader[--pos] = 128 | (--ofs & 127);
525 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
526 return 0;
527 sha1write(f, header, hdrlen);
528 sha1write(f, dheader + pos, sizeof(dheader) - pos);
529 hdrlen += sizeof(dheader) - pos;
530 } else if (obj_type == OBJ_REF_DELTA) {
531 if (limit && hdrlen + 20 + datalen + 20 >= limit)
532 return 0;
533 sha1write(f, header, hdrlen);
aa7e44bf 534 sha1write(f, entry->delta->idx.sha1, 20);
17b08f2c
DH
535 hdrlen += 20;
536 } else {
537 if (limit && hdrlen + datalen + 20 >= limit)
538 return 0;
539 sha1write(f, header, hdrlen);
540 }
541
91ecbeca
NP
542 if (!pack_to_stdout && p->index_version == 1 &&
543 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
aa7e44bf 544 die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
079afb18 545 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 546 unuse_pack(&w_curs);
3f9ac8d2 547 reused++;
a733cb60 548 }
17b08f2c 549 if (usable_delta)
ab7cd7bb 550 written_delta++;
3f9ac8d2 551 written++;
78d1e84f 552 if (!pack_to_stdout)
aa7e44bf 553 entry->idx.crc32 = crc32_end(f);
c323ac7d
LT
554 return hdrlen + datalen;
555}
556
6777a59f 557static off_t write_one(struct sha1file *f,
9d5ab962 558 struct object_entry *e,
6777a59f 559 off_t offset)
9d5ab962 560{
d7dd0223
NP
561 unsigned long size;
562
563 /* offset is non zero if object is written already. */
aa7e44bf 564 if (e->idx.offset || e->preferred_base)
9d5ab962 565 return offset;
d7dd0223
NP
566
567 /* if we are deltified, write out base object first. */
17b08f2c 568 if (e->delta) {
9d5ab962 569 offset = write_one(f, e->delta, offset);
17b08f2c
DH
570 if (!offset)
571 return 0;
572 }
d7dd0223 573
aa7e44bf 574 e->idx.offset = offset;
17b08f2c
DH
575 size = write_object(f, e, offset);
576 if (!size) {
aa7e44bf 577 e->idx.offset = 0;
17b08f2c
DH
578 return 0;
579 }
79814f42 580 written_list[nr_written++] = &e->idx;
d7dd0223
NP
581
582 /* make sure off_t is sufficiently large not to wrap */
583 if (offset > offset + size)
584 die("pack too large for current definition of off_t");
585 return offset + size;
9d5ab962
JH
586}
587
aa7e44bf 588/* forward declaration for write_pack_file */
d01fb92f
DH
589static int adjust_perm(const char *path, mode_t mode);
590
591static void write_pack_file(void)
c323ac7d 592{
ebe27b13 593 uint32_t i = 0, j;
d22b9290 594 struct sha1file *f;
ebe27b13 595 off_t offset, offset_one, last_obj_offset = 0;
a733cb60 596 struct pack_header hdr;
ebe27b13
DH
597 int do_progress = progress >> pack_to_stdout;
598 uint32_t nr_remaining = nr_result;
81a216a5 599
13aaf148 600 if (do_progress)
dc6a0757 601 progress_state = start_progress("Writing objects", nr_result);
79814f42 602 written_list = xmalloc(nr_objects * sizeof(*written_list));
183bdb2c 603
ebe27b13 604 do {
aa7e44bf 605 unsigned char sha1[20];
7ba502c4 606 char *pack_tmp_name = NULL;
aa7e44bf 607
ebe27b13 608 if (pack_to_stdout) {
2a128d63 609 f = sha1fd_throughput(1, "<stdout>", progress_state);
ebe27b13 610 } else {
7ba502c4
NP
611 char tmpname[PATH_MAX];
612 int fd;
613 snprintf(tmpname, sizeof(tmpname),
614 "%s/tmp_pack_XXXXXX", get_object_directory());
615 fd = xmkstemp(tmpname);
ebe27b13
DH
616 pack_tmp_name = xstrdup(tmpname);
617 f = sha1fd(fd, pack_tmp_name);
618 }
619
620 hdr.hdr_signature = htonl(PACK_SIGNATURE);
621 hdr.hdr_version = htonl(PACK_VERSION);
622 hdr.hdr_entries = htonl(nr_remaining);
623 sha1write(f, &hdr, sizeof(hdr));
624 offset = sizeof(hdr);
625 nr_written = 0;
626 for (; i < nr_objects; i++) {
627 last_obj_offset = offset;
628 offset_one = write_one(f, objects + i, offset);
629 if (!offset_one)
630 break;
631 offset = offset_one;
4d4fcc54 632 display_progress(progress_state, written);
ebe27b13 633 }
c553ca25 634
ebe27b13
DH
635 /*
636 * Did we write the wrong # entries in the header?
637 * If so, rewrite it like in fast-import
638 */
639 if (pack_to_stdout || nr_written == nr_remaining) {
aa7e44bf 640 sha1close(f, sha1, 1);
ebe27b13 641 } else {
7ba502c4
NP
642 int fd = sha1close(f, NULL, 0);
643 fixup_pack_header_footer(fd, sha1, pack_tmp_name, nr_written);
644 close(fd);
ebe27b13
DH
645 }
646
647 if (!pack_to_stdout) {
d01fb92f 648 mode_t mode = umask(0);
7ba502c4 649 char *idx_tmp_name, tmpname[PATH_MAX];
d01fb92f
DH
650
651 umask(mode);
652 mode = 0444 & ~mode;
653
79814f42
NP
654 idx_tmp_name = write_idx_file(NULL, written_list,
655 nr_written, sha1);
d01fb92f 656 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
aa7e44bf 657 base_name, sha1_to_hex(sha1));
d01fb92f
DH
658 if (adjust_perm(pack_tmp_name, mode))
659 die("unable to make temporary pack file readable: %s",
660 strerror(errno));
661 if (rename(pack_tmp_name, tmpname))
662 die("unable to rename temporary pack file: %s",
663 strerror(errno));
664 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
aa7e44bf 665 base_name, sha1_to_hex(sha1));
d01fb92f
DH
666 if (adjust_perm(idx_tmp_name, mode))
667 die("unable to make temporary index file readable: %s",
668 strerror(errno));
669 if (rename(idx_tmp_name, tmpname))
670 die("unable to rename temporary index file: %s",
671 strerror(errno));
7ba502c4
NP
672 free(idx_tmp_name);
673 free(pack_tmp_name);
aa7e44bf 674 puts(sha1_to_hex(sha1));
ebe27b13
DH
675 }
676
677 /* mark written objects as written to previous pack */
678 for (j = 0; j < nr_written; j++) {
79814f42 679 written_list[j]->offset = (off_t)-1;
ebe27b13
DH
680 }
681 nr_remaining -= nr_written;
682 } while (nr_remaining && i < nr_objects);
683
684 free(written_list);
4d4fcc54 685 stop_progress(&progress_state);
67c08ce1 686 if (written != nr_result)
7cadf491 687 die("wrote %u objects while expecting %u", written, nr_result);
01c12a23
DH
688 /*
689 * We have scanned through [0 ... i). Since we have written
690 * the correct number of objects, the remaining [i ... nr_objects)
691 * items must be either already written (due to out-of-order delta base)
692 * or a preferred base. Count those which are neither and complain if any.
693 */
ebe27b13
DH
694 for (j = 0; i < nr_objects; i++) {
695 struct object_entry *e = objects + i;
aa7e44bf 696 j += !e->idx.offset && !e->preferred_base;
d01fb92f 697 }
ebe27b13
DH
698 if (j)
699 die("wrote %u objects as expected but %u unwritten", written, j);
c323ac7d
LT
700}
701
7a979d99
JH
702static int locate_object_entry_hash(const unsigned char *sha1)
703{
704 int i;
705 unsigned int ui;
706 memcpy(&ui, sha1, sizeof(unsigned int));
707 i = ui % object_ix_hashsz;
708 while (0 < object_ix[i]) {
aa7e44bf 709 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
7a979d99
JH
710 return i;
711 if (++i == object_ix_hashsz)
712 i = 0;
713 }
714 return -1 - i;
715}
716
717static struct object_entry *locate_object_entry(const unsigned char *sha1)
718{
719 int i;
720
721 if (!object_ix_hashsz)
722 return NULL;
723
724 i = locate_object_entry_hash(sha1);
725 if (0 <= i)
726 return &objects[object_ix[i]-1];
727 return NULL;
728}
729
730static void rehash_objects(void)
c323ac7d 731{
7cadf491 732 uint32_t i;
7a979d99
JH
733 struct object_entry *oe;
734
735 object_ix_hashsz = nr_objects * 3;
736 if (object_ix_hashsz < 1024)
737 object_ix_hashsz = 1024;
738 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 739 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99 740 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
aa7e44bf 741 int ix = locate_object_entry_hash(oe->idx.sha1);
7a979d99
JH
742 if (0 <= ix)
743 continue;
744 ix = -1 - ix;
745 object_ix[ix] = i + 1;
746 }
747}
748
ce0bd642 749static unsigned name_hash(const char *name)
1d6b38cc 750{
ce0bd642
LT
751 unsigned char c;
752 unsigned hash = 0;
753
bc32fed5
JH
754 if (!name)
755 return 0;
756
eeef7135 757 /*
ce0bd642
LT
758 * This effectively just creates a sortable number from the
759 * last sixteen non-whitespace characters. Last characters
760 * count "most", so things that end in ".c" sort together.
eeef7135 761 */
ce0bd642
LT
762 while ((c = *name++) != 0) {
763 if (isspace(c))
764 continue;
765 hash = (hash >> 2) + (c << 24);
766 }
1d6b38cc
JH
767 return hash;
768}
769
a74db82e
JH
770static void setup_delta_attr_check(struct git_attr_check *check)
771{
772 static struct git_attr *attr_delta;
773
774 if (!attr_delta)
775 attr_delta = git_attr("delta", 5);
776
777 check[0].attr = attr_delta;
778}
779
780static int no_try_delta(const char *path)
781{
782 struct git_attr_check check[1];
783
784 setup_delta_attr_check(check);
785 if (git_checkattr(path, ARRAY_SIZE(check), check))
786 return 0;
787 if (ATTR_FALSE(check->value))
788 return 1;
789 return 0;
790}
791
5c49c116 792static int add_object_entry(const unsigned char *sha1, enum object_type type,
bc32fed5 793 const char *name, int exclude)
c323ac7d 794{
c323ac7d 795 struct object_entry *entry;
29b734e4 796 struct packed_git *p, *found_pack = NULL;
6777a59f 797 off_t found_offset = 0;
29b734e4 798 int ix;
bc32fed5 799 unsigned hash = name_hash(name);
29b734e4
NP
800
801 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
802 if (ix >= 0) {
803 if (exclude) {
804 entry = objects + object_ix[ix] - 1;
81a216a5
NP
805 if (!entry->preferred_base)
806 nr_result--;
29b734e4
NP
807 entry->preferred_base = 1;
808 }
809 return 0;
810 }
c323ac7d 811
5c49c116
NP
812 for (p = packed_git; p; p = p->next) {
813 off_t offset = find_pack_entry_one(sha1, p);
814 if (offset) {
815 if (!found_pack) {
816 found_offset = offset;
817 found_pack = p;
64560374 818 }
5c49c116
NP
819 if (exclude)
820 break;
821 if (incremental)
822 return 0;
823 if (local && !p->pack_local)
824 return 0;
64560374
LT
825 }
826 }
eb019375 827
29b734e4
NP
828 if (nr_objects >= nr_alloc) {
829 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 830 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 831 }
29b734e4
NP
832
833 entry = objects + nr_objects++;
c323ac7d 834 memset(entry, 0, sizeof(*entry));
aa7e44bf 835 hashcpy(entry->idx.sha1, sha1);
27225f2e 836 entry->hash = hash;
5c49c116
NP
837 if (type)
838 entry->type = type;
29b734e4
NP
839 if (exclude)
840 entry->preferred_base = 1;
81a216a5
NP
841 else
842 nr_result++;
29b734e4
NP
843 if (found_pack) {
844 entry->in_pack = found_pack;
845 entry->in_pack_offset = found_offset;
846 }
7a979d99
JH
847
848 if (object_ix_hashsz * 3 <= nr_objects * 4)
849 rehash_objects();
29b734e4
NP
850 else
851 object_ix[-1 - ix] = nr_objects;
7a979d99 852
4d4fcc54 853 display_progress(progress_state, nr_objects);
29b734e4 854
a74db82e
JH
855 if (name && no_try_delta(name))
856 entry->no_try_delta = 1;
857
29b734e4 858 return 1;
c323ac7d
LT
859}
860
5379a5c5
JH
861struct pbase_tree_cache {
862 unsigned char sha1[20];
863 int ref;
864 int temporary;
865 void *tree_data;
866 unsigned long tree_size;
867};
868
869static struct pbase_tree_cache *(pbase_tree_cache[256]);
870static int pbase_tree_cache_ix(const unsigned char *sha1)
871{
872 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
873}
874static int pbase_tree_cache_ix_incr(int ix)
875{
876 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
877}
878
879static struct pbase_tree {
880 struct pbase_tree *next;
881 /* This is a phony "cache" entry; we are not
882 * going to evict it nor find it through _get()
883 * mechanism -- this is for the toplevel node that
884 * would almost always change with any commit.
885 */
886 struct pbase_tree_cache pcache;
887} *pbase_tree;
888
889static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
890{
891 struct pbase_tree_cache *ent, *nent;
892 void *data;
893 unsigned long size;
21666f1a 894 enum object_type type;
5379a5c5
JH
895 int neigh;
896 int my_ix = pbase_tree_cache_ix(sha1);
897 int available_ix = -1;
898
899 /* pbase-tree-cache acts as a limited hashtable.
900 * your object will be found at your index or within a few
901 * slots after that slot if it is cached.
902 */
903 for (neigh = 0; neigh < 8; neigh++) {
904 ent = pbase_tree_cache[my_ix];
a89fccd2 905 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
906 ent->ref++;
907 return ent;
908 }
909 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
910 ((0 <= available_ix) &&
911 (!ent && pbase_tree_cache[available_ix])))
912 available_ix = my_ix;
913 if (!ent)
914 break;
915 my_ix = pbase_tree_cache_ix_incr(my_ix);
916 }
917
918 /* Did not find one. Either we got a bogus request or
919 * we need to read and perhaps cache.
920 */
21666f1a 921 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
922 if (!data)
923 return NULL;
21666f1a 924 if (type != OBJ_TREE) {
5379a5c5
JH
925 free(data);
926 return NULL;
927 }
928
929 /* We need to either cache or return a throwaway copy */
930
931 if (available_ix < 0)
932 ent = NULL;
933 else {
934 ent = pbase_tree_cache[available_ix];
935 my_ix = available_ix;
936 }
937
938 if (!ent) {
939 nent = xmalloc(sizeof(*nent));
940 nent->temporary = (available_ix < 0);
941 }
942 else {
943 /* evict and reuse */
944 free(ent->tree_data);
945 nent = ent;
946 }
e702496e 947 hashcpy(nent->sha1, sha1);
5379a5c5
JH
948 nent->tree_data = data;
949 nent->tree_size = size;
950 nent->ref = 1;
951 if (!nent->temporary)
952 pbase_tree_cache[my_ix] = nent;
953 return nent;
954}
955
956static void pbase_tree_put(struct pbase_tree_cache *cache)
957{
958 if (!cache->temporary) {
959 cache->ref--;
960 return;
961 }
962 free(cache->tree_data);
963 free(cache);
964}
965
966static int name_cmp_len(const char *name)
967{
968 int i;
969 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
970 ;
971 return i;
972}
973
974static void add_pbase_object(struct tree_desc *tree,
5379a5c5 975 const char *name,
ce0bd642
LT
976 int cmplen,
977 const char *fullname)
3f9ac8d2 978{
4c068a98 979 struct name_entry entry;
8a5a8d6c 980 int cmp;
4c068a98
LT
981
982 while (tree_entry(tree,&entry)) {
1211be6b
LT
983 if (S_ISGITLINK(entry.mode))
984 continue;
8a5a8d6c
NP
985 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
986 memcmp(name, entry.path, cmplen);
987 if (cmp > 0)
7a979d99 988 continue;
8a5a8d6c
NP
989 if (cmp < 0)
990 return;
5379a5c5 991 if (name[cmplen] != '/') {
5c49c116 992 add_object_entry(entry.sha1,
4d1012c3 993 object_type(entry.mode),
bc32fed5 994 fullname, 1);
5379a5c5
JH
995 return;
996 }
8a5a8d6c 997 if (S_ISDIR(entry.mode)) {
7a979d99 998 struct tree_desc sub;
5379a5c5
JH
999 struct pbase_tree_cache *tree;
1000 const char *down = name+cmplen+1;
1001 int downlen = name_cmp_len(down);
1002
4c068a98 1003 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
1004 if (!tree)
1005 return;
6fda5e51 1006 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1007
ce0bd642 1008 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1009 pbase_tree_put(tree);
1010 }
1011 }
1012}
1d6b38cc 1013
5379a5c5
JH
1014static unsigned *done_pbase_paths;
1015static int done_pbase_paths_num;
1016static int done_pbase_paths_alloc;
1017static int done_pbase_path_pos(unsigned hash)
1018{
1019 int lo = 0;
1020 int hi = done_pbase_paths_num;
1021 while (lo < hi) {
1022 int mi = (hi + lo) / 2;
1023 if (done_pbase_paths[mi] == hash)
1024 return mi;
1025 if (done_pbase_paths[mi] < hash)
1026 hi = mi;
1027 else
1028 lo = mi + 1;
1029 }
1030 return -lo-1;
1031}
1032
1033static int check_pbase_path(unsigned hash)
1034{
1035 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1036 if (0 <= pos)
1037 return 1;
1038 pos = -pos - 1;
1039 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1040 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1041 done_pbase_paths = xrealloc(done_pbase_paths,
1042 done_pbase_paths_alloc *
1043 sizeof(unsigned));
1044 }
1045 done_pbase_paths_num++;
1046 if (pos < done_pbase_paths_num)
1047 memmove(done_pbase_paths + pos + 1,
1048 done_pbase_paths + pos,
1049 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1050 done_pbase_paths[pos] = hash;
1051 return 0;
1052}
1053
bc32fed5 1054static void add_preferred_base_object(const char *name)
5379a5c5
JH
1055{
1056 struct pbase_tree *it;
8a5a8d6c 1057 int cmplen;
bc32fed5 1058 unsigned hash = name_hash(name);
5379a5c5 1059
8a5a8d6c 1060 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1061 return;
1062
8a5a8d6c 1063 cmplen = name_cmp_len(name);
5379a5c5
JH
1064 for (it = pbase_tree; it; it = it->next) {
1065 if (cmplen == 0) {
bc32fed5 1066 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
5379a5c5
JH
1067 }
1068 else {
1069 struct tree_desc tree;
6fda5e51 1070 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1071 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1072 }
3f9ac8d2 1073 }
3f9ac8d2
JH
1074}
1075
7a979d99 1076static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 1077{
5379a5c5
JH
1078 struct pbase_tree *it;
1079 void *data;
1080 unsigned long size;
1081 unsigned char tree_sha1[20];
1d6b38cc 1082
8d1d8f83
JH
1083 if (window <= num_preferred_base++)
1084 return;
1085
5379a5c5
JH
1086 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1087 if (!data)
7a979d99 1088 return;
5379a5c5
JH
1089
1090 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1091 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1092 free(data);
1093 return;
1094 }
1095 }
1096
1097 it = xcalloc(1, sizeof(*it));
1098 it->next = pbase_tree;
1099 pbase_tree = it;
1100
e702496e 1101 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1102 it->pcache.tree_data = data;
1103 it->pcache.tree_size = size;
3f9ac8d2
JH
1104}
1105
c323ac7d
LT
1106static void check_object(struct object_entry *entry)
1107{
5c49c116 1108 if (entry->in_pack) {
780e6e73 1109 struct packed_git *p = entry->in_pack;
03e79c88 1110 struct pack_window *w_curs = NULL;
5c49c116
NP
1111 const unsigned char *base_ref = NULL;
1112 struct object_entry *base_entry;
1113 unsigned long used, used_0;
6777a59f 1114 unsigned int avail;
5c49c116
NP
1115 off_t ofs;
1116 unsigned char *buf, c;
780e6e73 1117
6777a59f 1118 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1119
5c49c116 1120 /*
fa736f72
NP
1121 * We want in_pack_type even if we do not reuse delta
1122 * since non-delta representations could still be reused.
ab7cd7bb 1123 */
6777a59f 1124 used = unpack_object_header_gently(buf, avail,
5c49c116
NP
1125 &entry->in_pack_type,
1126 &entry->size);
ab7cd7bb 1127
5c49c116
NP
1128 /*
1129 * Determine if this is a delta and if so whether we can
1130 * reuse it or not. Otherwise let's find out as cheaply as
1131 * possible what the actual type and size for this object is.
3f9ac8d2 1132 */
5c49c116
NP
1133 switch (entry->in_pack_type) {
1134 default:
1135 /* Not a delta hence we've already got all we need. */
1136 entry->type = entry->in_pack_type;
1137 entry->in_pack_header_size = used;
1138 unuse_pack(&w_curs);
1139 return;
1140 case OBJ_REF_DELTA:
1141 if (!no_reuse_delta && !entry->preferred_base)
1142 base_ref = use_pack(p, &w_curs,
1143 entry->in_pack_offset + used, NULL);
1144 entry->in_pack_header_size = used + 20;
1145 break;
1146 case OBJ_OFS_DELTA:
1147 buf = use_pack(p, &w_curs,
1148 entry->in_pack_offset + used, NULL);
1149 used_0 = 0;
1150 c = buf[used_0++];
1151 ofs = c & 127;
1152 while (c & 128) {
1153 ofs += 1;
1154 if (!ofs || MSB(ofs, 7))
1155 die("delta base offset overflow in pack for %s",
aa7e44bf 1156 sha1_to_hex(entry->idx.sha1));
5c49c116
NP
1157 c = buf[used_0++];
1158 ofs = (ofs << 7) + (c & 127);
780e6e73 1159 }
5c49c116
NP
1160 if (ofs >= entry->in_pack_offset)
1161 die("delta base offset out of bound for %s",
aa7e44bf 1162 sha1_to_hex(entry->idx.sha1));
5c49c116
NP
1163 ofs = entry->in_pack_offset - ofs;
1164 if (!no_reuse_delta && !entry->preferred_base)
1165 base_ref = find_packed_object_name(p, ofs);
1166 entry->in_pack_header_size = used + used_0;
1167 break;
780e6e73 1168 }
780e6e73 1169
5c49c116
NP
1170 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1171 /*
1172 * If base_ref was set above that means we wish to
1173 * reuse delta data, and we even found that base
1174 * in the list of objects we want to pack. Goodie!
1175 *
1176 * Depth value does not matter - find_deltas() will
1177 * never consider reused delta as the base object to
1178 * deltify other objects against, in order to avoid
1179 * circular deltas.
3f9ac8d2 1180 */
780e6e73 1181 entry->type = entry->in_pack_type;
5c49c116 1182 entry->delta = base_entry;
15b4d577
JH
1183 entry->delta_sibling = base_entry->delta_child;
1184 base_entry->delta_child = entry;
5c49c116
NP
1185 unuse_pack(&w_curs);
1186 return;
1187 }
ab7cd7bb 1188
5c49c116
NP
1189 if (entry->type) {
1190 /*
1191 * This must be a delta and we already know what the
1192 * final object type is. Let's extract the actual
1193 * object size from the delta header.
1194 */
1195 entry->size = get_size_from_delta(p, &w_curs,
1196 entry->in_pack_offset + entry->in_pack_header_size);
1197 unuse_pack(&w_curs);
3f9ac8d2
JH
1198 return;
1199 }
5c49c116
NP
1200
1201 /*
1202 * No choice but to fall back to the recursive delta walk
1203 * with sha1_object_info() to find about the object type
1204 * at this point...
1205 */
1206 unuse_pack(&w_curs);
36e4d74a 1207 }
3f9ac8d2 1208
aa7e44bf 1209 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
21666f1a 1210 if (entry->type < 0)
36e4d74a 1211 die("unable to get type of object %s",
aa7e44bf 1212 sha1_to_hex(entry->idx.sha1));
3f9ac8d2
JH
1213}
1214
5c49c116
NP
1215static int pack_offset_sort(const void *_a, const void *_b)
1216{
1217 const struct object_entry *a = *(struct object_entry **)_a;
1218 const struct object_entry *b = *(struct object_entry **)_b;
1219
1220 /* avoid filesystem trashing with loose objects */
1221 if (!a->in_pack && !b->in_pack)
aa7e44bf 1222 return hashcmp(a->idx.sha1, b->idx.sha1);
5c49c116
NP
1223
1224 if (a->in_pack < b->in_pack)
1225 return -1;
1226 if (a->in_pack > b->in_pack)
1227 return 1;
1228 return a->in_pack_offset < b->in_pack_offset ? -1 :
1229 (a->in_pack_offset > b->in_pack_offset);
1230}
1231
c323ac7d
LT
1232static void get_object_details(void)
1233{
7cadf491 1234 uint32_t i;
5c49c116
NP
1235 struct object_entry **sorted_by_offset;
1236
1237 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1238 for (i = 0; i < nr_objects; i++)
1239 sorted_by_offset[i] = objects + i;
1240 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1241
3f9ac8d2 1242 prepare_pack_ix();
5c49c116
NP
1243 for (i = 0; i < nr_objects; i++)
1244 check_object(sorted_by_offset[i]);
1245 free(sorted_by_offset);
c323ac7d
LT
1246}
1247
b904166c
NP
1248/*
1249 * We search for deltas in a list sorted by type, by filename hash, and then
1250 * by size, so that we see progressively smaller and smaller files.
1251 * That's because we prefer deltas to be from the bigger file
1252 * to the smaller -- deletes are potentially cheaper, but perhaps
1253 * more importantly, the bigger file is likely the more recent
1254 * one. The deepest deltas are therefore the oldest objects which are
1255 * less susceptible to be accessed often.
1256 */
9668cf59 1257static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1258{
9668cf59
NP
1259 const struct object_entry *a = *(struct object_entry **)_a;
1260 const struct object_entry *b = *(struct object_entry **)_b;
1261
c323ac7d 1262 if (a->type > b->type)
27225f2e 1263 return -1;
b904166c 1264 if (a->type < b->type)
27225f2e 1265 return 1;
b904166c 1266 if (a->hash > b->hash)
7a979d99 1267 return -1;
b904166c 1268 if (a->hash < b->hash)
7a979d99 1269 return 1;
b904166c 1270 if (a->preferred_base > b->preferred_base)
c323ac7d 1271 return -1;
b904166c
NP
1272 if (a->preferred_base < b->preferred_base)
1273 return 1;
c323ac7d 1274 if (a->size > b->size)
b904166c
NP
1275 return -1;
1276 if (a->size < b->size)
c323ac7d 1277 return 1;
b904166c 1278 return a < b ? -1 : (a > b); /* newest first */
c323ac7d
LT
1279}
1280
1281struct unpacked {
1282 struct object_entry *entry;
1283 void *data;
f6c7081a 1284 struct delta_index *index;
5a235b5e 1285 unsigned depth;
c323ac7d
LT
1286};
1287
d250626c
NP
1288static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1289 unsigned long delta_size)
074b2eea
MK
1290{
1291 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1292 return 0;
1293
e3dfddb3
MK
1294 if (delta_size < cache_max_small_delta_size)
1295 return 1;
1296
074b2eea
MK
1297 /* cache delta, if objects are large enough compared to delta size */
1298 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1299 return 1;
1300
1301 return 0;
1302}
1303
8ecce684
NP
1304#ifdef THREADED_DELTA_SEARCH
1305
1306static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1307#define read_lock() pthread_mutex_lock(&read_mutex)
1308#define read_unlock() pthread_mutex_unlock(&read_mutex)
1309
3c701839
NP
1310static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1311#define cache_lock() pthread_mutex_lock(&cache_mutex)
1312#define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1313
8ecce684
NP
1314static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1315#define progress_lock() pthread_mutex_lock(&progress_mutex)
1316#define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1317
1318#else
1319
e1ef8673
JH
1320#define read_lock() (void)0
1321#define read_unlock() (void)0
1322#define cache_lock() (void)0
1323#define cache_unlock() (void)0
1324#define progress_lock() (void)0
1325#define progress_unlock() (void)0
8ecce684
NP
1326
1327#endif
1328
f6c7081a 1329static int try_delta(struct unpacked *trg, struct unpacked *src,
ef0316fc 1330 unsigned max_depth, unsigned long *mem_usage)
c323ac7d 1331{
f6c7081a
NP
1332 struct object_entry *trg_entry = trg->entry;
1333 struct object_entry *src_entry = src->entry;
560b25a8 1334 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
c83f032e 1335 unsigned ref_depth;
21666f1a 1336 enum object_type type;
c323ac7d
LT
1337 void *delta_buf;
1338
1339 /* Don't bother doing diffs between different types */
f6c7081a 1340 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1341 return -1;
1342
51d1e83f
LT
1343 /*
1344 * We do not bother to try a delta that we discarded
8dbbd14e 1345 * on an earlier try, but only when reusing delta data.
51d1e83f 1346 */
8dbbd14e 1347 if (!no_reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1348 trg_entry->in_pack == src_entry->in_pack &&
1349 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1350 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1351 return 0;
1352
898b14ce 1353 /* Let's not bust the allowed depth. */
5a235b5e 1354 if (src->depth >= max_depth)
d116a45a 1355 return 0;
c323ac7d 1356
c3b06a69 1357 /* Now some size filtering heuristics. */
560b25a8 1358 trg_size = trg_entry->size;
c83f032e
NP
1359 if (!trg_entry->delta) {
1360 max_size = trg_size/2 - 20;
1361 ref_depth = 1;
1362 } else {
1363 max_size = trg_entry->delta_size;
5a235b5e 1364 ref_depth = trg->depth;
c83f032e 1365 }
5a235b5e 1366 max_size = max_size * (max_depth - src->depth) /
c83f032e 1367 (max_depth - ref_depth + 1);
c3b06a69
NP
1368 if (max_size == 0)
1369 return 0;
f6c7081a 1370 src_size = src_entry->size;
560b25a8 1371 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1372 if (sizediff >= max_size)
f527cb8c 1373 return 0;
a1dab41a
BD
1374 if (trg_size < src_size / 32)
1375 return 0;
f6c7081a 1376
560b25a8
NP
1377 /* Load data if not already done */
1378 if (!trg->data) {
8ecce684 1379 read_lock();
aa7e44bf 1380 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
8ecce684 1381 read_unlock();
2e3404c3
JH
1382 if (!trg->data)
1383 die("object %s cannot be read",
1384 sha1_to_hex(trg_entry->idx.sha1));
560b25a8
NP
1385 if (sz != trg_size)
1386 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1387 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
ef0316fc 1388 *mem_usage += sz;
560b25a8
NP
1389 }
1390 if (!src->data) {
8ecce684 1391 read_lock();
aa7e44bf 1392 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
8ecce684 1393 read_unlock();
2e3404c3
JH
1394 if (!src->data)
1395 die("object %s cannot be read",
1396 sha1_to_hex(src_entry->idx.sha1));
560b25a8
NP
1397 if (sz != src_size)
1398 die("object %s inconsistent object length (%lu vs %lu)",
aa7e44bf 1399 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
ef0316fc 1400 *mem_usage += sz;
560b25a8
NP
1401 }
1402 if (!src->index) {
1403 src->index = create_delta_index(src->data, src_size);
a588d88a
MK
1404 if (!src->index) {
1405 static int warned = 0;
1406 if (!warned++)
1407 warning("suboptimal pack - out of memory");
1408 return 0;
1409 }
ef0316fc 1410 *mem_usage += sizeof_delta_index(src->index);
560b25a8
NP
1411 }
1412
1413 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1414 if (!delta_buf)
75c42d8c 1415 return 0;
f6c7081a 1416
9e2d57a0 1417 if (trg_entry->delta) {
848d732c
BD
1418 /* Prefer only shallower same-sized deltas. */
1419 if (delta_size == trg_entry->delta_size &&
5a235b5e 1420 src->depth + 1 >= trg->depth) {
848d732c
BD
1421 free(delta_buf);
1422 return 0;
1423 }
074b2eea 1424 }
9e2d57a0 1425
3c701839
NP
1426 /*
1427 * Handle memory allocation outside of the cache
1428 * accounting lock. Compiler will optimize the strangeness
1429 * away when THREADED_DELTA_SEARCH is not defined.
1430 */
1431 if (trg_entry->delta_data)
1432 free(trg_entry->delta_data);
1433 cache_lock();
9e2d57a0
NP
1434 if (trg_entry->delta_data) {
1435 delta_cache_size -= trg_entry->delta_size;
9e2d57a0
NP
1436 trg_entry->delta_data = NULL;
1437 }
d250626c 1438 if (delta_cacheable(src_size, trg_size, delta_size)) {
b7a28f78 1439 delta_cache_size += delta_size;
3c701839
NP
1440 cache_unlock();
1441 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1442 } else {
1443 cache_unlock();
074b2eea 1444 free(delta_buf);
3c701839
NP
1445 }
1446
b7a28f78
NP
1447 trg_entry->delta = src_entry;
1448 trg_entry->delta_size = delta_size;
1449 trg->depth = src->depth + 1;
1450
f6c7081a 1451 return 1;
c323ac7d
LT
1452}
1453
898b14ce 1454static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1455{
898b14ce
NP
1456 struct object_entry *child = me->delta_child;
1457 unsigned int m = n;
1458 while (child) {
1459 unsigned int c = check_delta_limit(child, n + 1);
1460 if (m < c)
1461 m = c;
1462 child = child->delta_sibling;
1463 }
1464 return m;
b2504a0d
NP
1465}
1466
ef0316fc 1467static unsigned long free_unpacked(struct unpacked *n)
a97773ce 1468{
ef0316fc 1469 unsigned long freed_mem = sizeof_delta_index(n->index);
a97773ce
BD
1470 free_delta_index(n->index);
1471 n->index = NULL;
1472 if (n->data) {
ef0316fc 1473 freed_mem += n->entry->size;
a97773ce
BD
1474 free(n->data);
1475 n->data = NULL;
a97773ce
BD
1476 }
1477 n->entry = NULL;
7d7baa5e 1478 n->depth = 0;
ef0316fc 1479 return freed_mem;
a97773ce
BD
1480}
1481
75d39853 1482static void find_deltas(struct object_entry **list, unsigned list_size,
e334977d 1483 int window, int depth, unsigned *processed)
c323ac7d 1484{
b904166c 1485 uint32_t i = 0, idx = 0, count = 0;
c323ac7d 1486 unsigned int array_size = window * sizeof(struct unpacked);
7cadf491 1487 struct unpacked *array;
ef0316fc 1488 unsigned long mem_usage = 0;
c323ac7d 1489
7cadf491 1490 array = xmalloc(array_size);
c323ac7d 1491 memset(array, 0, array_size);
21fcd1bd 1492
7cadf491 1493 do {
b904166c 1494 struct object_entry *entry = list[i++];
c323ac7d 1495 struct unpacked *n = array + idx;
ef0316fc 1496 int j, max_depth, best_base = -1;
c323ac7d 1497
ef0316fc 1498 mem_usage -= free_unpacked(n);
c323ac7d 1499 n->entry = entry;
ab7cd7bb 1500
a97773ce 1501 while (window_memory_limit &&
ef0316fc 1502 mem_usage > window_memory_limit &&
a97773ce
BD
1503 count > 1) {
1504 uint32_t tail = (idx + window - count) % window;
ef0316fc 1505 mem_usage -= free_unpacked(array + tail);
a97773ce
BD
1506 count--;
1507 }
1508
75d39853
NP
1509 /* We do not compute delta to *create* objects we are not
1510 * going to pack.
1511 */
1512 if (entry->preferred_base)
1513 goto next;
1514
8ecce684 1515 progress_lock();
e334977d 1516 (*processed)++;
4d4fcc54 1517 display_progress(progress_state, *processed);
8ecce684 1518 progress_unlock();
75d39853 1519
898b14ce
NP
1520 /*
1521 * If the current object is at pack edge, take the depth the
1522 * objects that depend on the current object into account
1523 * otherwise they would become too deep.
1524 */
1525 max_depth = depth;
1526 if (entry->delta_child) {
1527 max_depth -= check_delta_limit(entry, 0);
1528 if (max_depth <= 0)
1529 goto next;
1530 }
1531
78817c15
LT
1532 j = window;
1533 while (--j > 0) {
77639870 1534 int ret;
7cadf491 1535 uint32_t other_idx = idx + j;
c323ac7d 1536 struct unpacked *m;
78817c15
LT
1537 if (other_idx >= window)
1538 other_idx -= window;
c323ac7d
LT
1539 m = array + other_idx;
1540 if (!m->entry)
1541 break;
ef0316fc 1542 ret = try_delta(n, m, max_depth, &mem_usage);
77639870 1543 if (ret < 0)
c323ac7d 1544 break;
77639870
JH
1545 else if (ret > 0)
1546 best_base = other_idx;
c323ac7d 1547 }
898b14ce 1548
70ca1a3f
JH
1549 /* if we made n a delta, and if n is already at max
1550 * depth, leaving it in the window is pointless. we
1551 * should evict it first.
70ca1a3f 1552 */
5a235b5e 1553 if (entry->delta && depth <= n->depth)
70ca1a3f 1554 continue;
ff45715c 1555
77639870
JH
1556 /*
1557 * Move the best delta base up in the window, after the
1558 * currently deltified object, to keep it longer. It will
1559 * be the first base object to be attempted next.
1560 */
1561 if (entry->delta) {
1562 struct unpacked swap = array[best_base];
1563 int dist = (window + idx - best_base) % window;
1564 int dst = best_base;
1565 while (dist--) {
1566 int src = (dst + 1) % window;
1567 array[dst] = array[src];
1568 dst = src;
1569 }
1570 array[dst] = swap;
1571 }
1572
898b14ce 1573 next:
521a4f4c 1574 idx++;
a97773ce
BD
1575 if (count + 1 < window)
1576 count++;
521a4f4c
LT
1577 if (idx >= window)
1578 idx = 0;
b904166c 1579 } while (i < list_size);
adee7bdf 1580
f6c7081a 1581 for (i = 0; i < window; ++i) {
ff45715c 1582 free_delta_index(array[i].index);
adee7bdf 1583 free(array[i].data);
f6c7081a 1584 }
adee7bdf 1585 free(array);
c323ac7d
LT
1586}
1587
8ecce684
NP
1588#ifdef THREADED_DELTA_SEARCH
1589
1590struct thread_params {
1591 pthread_t thread;
1592 struct object_entry **list;
1593 unsigned list_size;
1594 int window;
1595 int depth;
1596 unsigned *processed;
1597};
1598
c2a33679
NP
1599static pthread_mutex_t data_request = PTHREAD_MUTEX_INITIALIZER;
1600static pthread_mutex_t data_ready = PTHREAD_MUTEX_INITIALIZER;
1601static pthread_mutex_t data_provider = PTHREAD_MUTEX_INITIALIZER;
1602static struct thread_params *data_requester;
1603
8ecce684
NP
1604static void *threaded_find_deltas(void *arg)
1605{
c2a33679
NP
1606 struct thread_params *me = arg;
1607
1608 for (;;) {
1609 pthread_mutex_lock(&data_request);
1610 data_requester = me;
1611 pthread_mutex_unlock(&data_provider);
1612 pthread_mutex_lock(&data_ready);
b81d9af7 1613 pthread_mutex_unlock(&data_request);
c2a33679
NP
1614
1615 if (!me->list_size)
1616 return NULL;
1617
1618 find_deltas(me->list, me->list_size,
1619 me->window, me->depth, me->processed);
1620 }
8ecce684
NP
1621}
1622
8ecce684
NP
1623static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1624 int window, int depth, unsigned *processed)
1625{
b81d9af7 1626 struct thread_params *target, p[delta_search_threads];
8ecce684 1627 int i, ret;
c2a33679
NP
1628 unsigned chunk_size;
1629
367f4a43
NP
1630 if (delta_search_threads <= 1) {
1631 find_deltas(list, list_size, window, depth, processed);
1632 return;
1633 }
1634
c2a33679
NP
1635 pthread_mutex_lock(&data_provider);
1636 pthread_mutex_lock(&data_ready);
8ecce684 1637
367f4a43 1638 for (i = 0; i < delta_search_threads; i++) {
8ecce684
NP
1639 p[i].window = window;
1640 p[i].depth = depth;
1641 p[i].processed = processed;
1642 ret = pthread_create(&p[i].thread, NULL,
1643 threaded_find_deltas, &p[i]);
1644 if (ret)
1645 die("unable to create thread: %s", strerror(ret));
8ecce684
NP
1646 }
1647
c2a33679
NP
1648 /* this should be auto-tuned somehow */
1649 chunk_size = window * 1000;
1650
1651 do {
1652 unsigned sublist_size = chunk_size;
1653 if (sublist_size > list_size)
1654 sublist_size = list_size;
1655
59921b4b
NP
1656 /* try to split chunks on "path" boundaries */
1657 while (sublist_size < list_size && list[sublist_size]->hash &&
1658 list[sublist_size]->hash == list[sublist_size-1]->hash)
1659 sublist_size++;
1660
c2a33679 1661 pthread_mutex_lock(&data_provider);
b81d9af7
NP
1662 target = data_requester;
1663 target->list = list;
1664 target->list_size = sublist_size;
c2a33679
NP
1665 pthread_mutex_unlock(&data_ready);
1666
1667 list += sublist_size;
1668 list_size -= sublist_size;
1669 if (!sublist_size) {
b81d9af7 1670 pthread_join(target->thread, NULL);
c2a33679
NP
1671 i--;
1672 }
c2a33679 1673 } while (i);
8ecce684
NP
1674}
1675
1676#else
1677#define ll_find_deltas find_deltas
1678#endif
1679
f3123c4a
JH
1680static void prepare_pack(int window, int depth)
1681{
9668cf59 1682 struct object_entry **delta_list;
75d39853 1683 uint32_t i, n, nr_deltas;
9668cf59 1684
3f9ac8d2 1685 get_object_details();
9668cf59 1686
75d39853 1687 if (!nr_objects || !window || !depth)
9668cf59
NP
1688 return;
1689
1690 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
75d39853
NP
1691 nr_deltas = n = 0;
1692
1693 for (i = 0; i < nr_objects; i++) {
1694 struct object_entry *entry = objects + i;
1695
1696 if (entry->delta)
1697 /* This happens if we decided to reuse existing
1698 * delta from a pack. "!no_reuse_delta &&" is implied.
1699 */
1700 continue;
1701
1702 if (entry->size < 50)
1703 continue;
1704
1705 if (entry->no_try_delta)
1706 continue;
1707
1708 if (!entry->preferred_base)
1709 nr_deltas++;
1710
1711 delta_list[n++] = entry;
1712 }
1713
2f8b8947 1714 if (nr_deltas && n > 1) {
e334977d
NP
1715 unsigned nr_done = 0;
1716 if (progress)
dc6a0757
NP
1717 progress_state = start_progress("Compressing objects",
1718 nr_deltas);
75d39853 1719 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
8ecce684 1720 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
4d4fcc54 1721 stop_progress(&progress_state);
e334977d
NP
1722 if (nr_done != nr_deltas)
1723 die("inconsistency with delta count");
75d39853 1724 }
9668cf59 1725 free(delta_list);
f3123c4a
JH
1726}
1727
4812a93a
JK
1728static int git_pack_config(const char *k, const char *v)
1729{
1730 if(!strcmp(k, "pack.window")) {
1731 window = git_config_int(k, v);
1732 return 0;
1733 }
a97773ce
BD
1734 if (!strcmp(k, "pack.windowmemory")) {
1735 window_memory_limit = git_config_ulong(k, v);
1736 return 0;
1737 }
1738 if (!strcmp(k, "pack.depth")) {
842aaf93
TT
1739 depth = git_config_int(k, v);
1740 return 0;
1741 }
960ccca6
DH
1742 if (!strcmp(k, "pack.compression")) {
1743 int level = git_config_int(k, v);
1744 if (level == -1)
1745 level = Z_DEFAULT_COMPRESSION;
1746 else if (level < 0 || level > Z_BEST_COMPRESSION)
1747 die("bad pack compression level %d", level);
1748 pack_compression_level = level;
1749 pack_compression_seen = 1;
1750 return 0;
1751 }
074b2eea
MK
1752 if (!strcmp(k, "pack.deltacachesize")) {
1753 max_delta_cache_size = git_config_int(k, v);
1754 return 0;
1755 }
e3dfddb3
MK
1756 if (!strcmp(k, "pack.deltacachelimit")) {
1757 cache_max_small_delta_size = git_config_int(k, v);
1758 return 0;
1759 }
693b86ff
NP
1760 if (!strcmp(k, "pack.threads")) {
1761 delta_search_threads = git_config_int(k, v);
1762 if (delta_search_threads < 1)
1763 die("invalid number of threads specified (%d)",
1764 delta_search_threads);
1765#ifndef THREADED_DELTA_SEARCH
1766 if (delta_search_threads > 1)
1767 warning("no threads support, ignoring %s", k);
1768#endif
1769 return 0;
1770 }
4d00bda2
NP
1771 if (!strcmp(k, "pack.indexversion")) {
1772 pack_idx_default_version = git_config_int(k, v);
1773 if (pack_idx_default_version > 2)
1774 die("bad pack.indexversion=%d", pack_idx_default_version);
1775 return 0;
1776 }
4812a93a
JK
1777 return git_default_config(k, v);
1778}
1779
b5d97e6b 1780static void read_object_list_from_stdin(void)
c323ac7d 1781{
b5d97e6b
JH
1782 char line[40 + 1 + PATH_MAX + 2];
1783 unsigned char sha1[20];
b2504a0d 1784
da93d12b 1785 for (;;) {
da93d12b
LT
1786 if (!fgets(line, sizeof(line), stdin)) {
1787 if (feof(stdin))
1788 break;
1789 if (!ferror(stdin))
1790 die("fgets returned NULL, not EOF, not error!");
687dd75c
JH
1791 if (errno != EINTR)
1792 die("fgets: %s", strerror(errno));
1793 clearerr(stdin);
1794 continue;
da93d12b 1795 }
7a979d99
JH
1796 if (line[0] == '-') {
1797 if (get_sha1_hex(line+1, sha1))
1798 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1799 line);
8d1d8f83 1800 add_preferred_base(sha1);
7a979d99 1801 continue;
21fcd1bd 1802 }
c323ac7d 1803 if (get_sha1_hex(line, sha1))
ef07618f 1804 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1805
bc32fed5
JH
1806 add_preferred_base_object(line+41);
1807 add_object_entry(sha1, 0, line+41, 0);
c323ac7d 1808 }
b5d97e6b
JH
1809}
1810
08cdfb13
JH
1811#define OBJECT_ADDED (1u<<20)
1812
b5d97e6b
JH
1813static void show_commit(struct commit *commit)
1814{
bc32fed5 1815 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
08cdfb13 1816 commit->object.flags |= OBJECT_ADDED;
b5d97e6b
JH
1817}
1818
1819static void show_object(struct object_array_entry *p)
1820{
bc32fed5
JH
1821 add_preferred_base_object(p->name);
1822 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
08cdfb13 1823 p->item->flags |= OBJECT_ADDED;
b5d97e6b
JH
1824}
1825
8d1d8f83
JH
1826static void show_edge(struct commit *commit)
1827{
1828 add_preferred_base(commit->object.sha1);
1829}
1830
08cdfb13
JH
1831struct in_pack_object {
1832 off_t offset;
1833 struct object *object;
1834};
1835
1836struct in_pack {
1837 int alloc;
1838 int nr;
1839 struct in_pack_object *array;
1840};
1841
1842static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1843{
1844 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1845 in_pack->array[in_pack->nr].object = object;
1846 in_pack->nr++;
1847}
1848
1849/*
1850 * Compare the objects in the offset order, in order to emulate the
1851 * "git-rev-list --objects" output that produced the pack originally.
1852 */
1853static int ofscmp(const void *a_, const void *b_)
1854{
1855 struct in_pack_object *a = (struct in_pack_object *)a_;
1856 struct in_pack_object *b = (struct in_pack_object *)b_;
1857
1858 if (a->offset < b->offset)
1859 return -1;
1860 else if (a->offset > b->offset)
1861 return 1;
1862 else
1863 return hashcmp(a->object->sha1, b->object->sha1);
1864}
1865
1866static void add_objects_in_unpacked_packs(struct rev_info *revs)
1867{
1868 struct packed_git *p;
1869 struct in_pack in_pack;
1870 uint32_t i;
1871
1872 memset(&in_pack, 0, sizeof(in_pack));
1873
1874 for (p = packed_git; p; p = p->next) {
1875 const unsigned char *sha1;
1876 struct object *o;
1877
1878 for (i = 0; i < revs->num_ignore_packed; i++) {
1879 if (matches_pack_name(p, revs->ignore_packed[i]))
1880 break;
1881 }
1882 if (revs->num_ignore_packed <= i)
1883 continue;
1884 if (open_pack_index(p))
1885 die("cannot open pack index");
1886
1887 ALLOC_GROW(in_pack.array,
1888 in_pack.nr + p->num_objects,
1889 in_pack.alloc);
1890
1891 for (i = 0; i < p->num_objects; i++) {
1892 sha1 = nth_packed_object_sha1(p, i);
1893 o = lookup_unknown_object(sha1);
1894 if (!(o->flags & OBJECT_ADDED))
1895 mark_in_pack_object(o, p, &in_pack);
1896 o->flags |= OBJECT_ADDED;
1897 }
1898 }
1899
1900 if (in_pack.nr) {
1901 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1902 ofscmp);
1903 for (i = 0; i < in_pack.nr; i++) {
1904 struct object *o = in_pack.array[i].object;
1905 add_object_entry(o->sha1, o->type, "", 0);
1906 }
1907 }
1908 free(in_pack.array);
1909}
1910
8d1d8f83 1911static void get_object_list(int ac, const char **av)
b5d97e6b
JH
1912{
1913 struct rev_info revs;
1914 char line[1000];
b5d97e6b
JH
1915 int flags = 0;
1916
b5d97e6b
JH
1917 init_revisions(&revs, NULL);
1918 save_commit_buffer = 0;
1919 track_object_refs = 0;
1920 setup_revisions(ac, av, &revs, NULL);
1921
b5d97e6b
JH
1922 while (fgets(line, sizeof(line), stdin) != NULL) {
1923 int len = strlen(line);
1924 if (line[len - 1] == '\n')
1925 line[--len] = 0;
1926 if (!len)
1927 break;
1928 if (*line == '-') {
1929 if (!strcmp(line, "--not")) {
1930 flags ^= UNINTERESTING;
1931 continue;
1932 }
1933 die("not a rev '%s'", line);
1934 }
1935 if (handle_revision_arg(line, &revs, flags, 1))
1936 die("bad revision '%s'", line);
1937 }
1938
1939 prepare_revision_walk(&revs);
8d1d8f83 1940 mark_edges_uninteresting(revs.commits, &revs, show_edge);
b5d97e6b 1941 traverse_commit_list(&revs, show_commit, show_object);
08cdfb13
JH
1942
1943 if (keep_unreachable)
1944 add_objects_in_unpacked_packs(&revs);
b5d97e6b
JH
1945}
1946
b6b32ccb
JH
1947static int adjust_perm(const char *path, mode_t mode)
1948{
1949 if (chmod(path, mode))
1950 return -1;
1951 return adjust_shared_perm(path);
1952}
1953
b5d97e6b
JH
1954int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1955{
b5d97e6b 1956 int use_internal_rev_list = 0;
8d1d8f83 1957 int thin = 0;
7cadf491 1958 uint32_t i;
ffa84ffb
RD
1959 const char **rp_av;
1960 int rp_ac_alloc = 64;
8d1d8f83
JH
1961 int rp_ac;
1962
ffa84ffb
RD
1963 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1964
8d1d8f83
JH
1965 rp_av[0] = "pack-objects";
1966 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1967 rp_ac = 2;
b5d97e6b
JH
1968
1969 git_config(git_pack_config);
960ccca6
DH
1970 if (!pack_compression_seen && core_compression_seen)
1971 pack_compression_level = core_compression_level;
b5d97e6b
JH
1972
1973 progress = isatty(2);
1974 for (i = 1; i < argc; i++) {
1975 const char *arg = argv[i];
1976
1977 if (*arg != '-')
1978 break;
1979
1980 if (!strcmp("--non-empty", arg)) {
1981 non_empty = 1;
1982 continue;
1983 }
1984 if (!strcmp("--local", arg)) {
1985 local = 1;
1986 continue;
1987 }
b5d97e6b
JH
1988 if (!strcmp("--incremental", arg)) {
1989 incremental = 1;
1990 continue;
1991 }
960ccca6
DH
1992 if (!prefixcmp(arg, "--compression=")) {
1993 char *end;
1994 int level = strtoul(arg+14, &end, 0);
1995 if (!arg[14] || *end)
1996 usage(pack_usage);
1997 if (level == -1)
1998 level = Z_DEFAULT_COMPRESSION;
1999 else if (level < 0 || level > Z_BEST_COMPRESSION)
2000 die("bad pack compression level %d", level);
2001 pack_compression_level = level;
2002 continue;
2003 }
6b94b1a0
DH
2004 if (!prefixcmp(arg, "--max-pack-size=")) {
2005 char *end;
2006 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2007 if (!arg[16] || *end)
2008 usage(pack_usage);
2009 continue;
2010 }
599065a3 2011 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
2012 char *end;
2013 window = strtoul(arg+9, &end, 0);
2014 if (!arg[9] || *end)
2015 usage(pack_usage);
2016 continue;
2017 }
a97773ce
BD
2018 if (!prefixcmp(arg, "--window-memory=")) {
2019 if (!git_parse_ulong(arg+16, &window_memory_limit))
2020 usage(pack_usage);
2021 continue;
2022 }
367f4a43
NP
2023 if (!prefixcmp(arg, "--threads=")) {
2024 char *end;
2025 delta_search_threads = strtoul(arg+10, &end, 0);
2026 if (!arg[10] || *end || delta_search_threads < 1)
2027 usage(pack_usage);
2028#ifndef THREADED_DELTA_SEARCH
2029 if (delta_search_threads > 1)
2030 warning("no threads support, "
2031 "ignoring %s", arg);
2032#endif
2033 continue;
2034 }
599065a3 2035 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
2036 char *end;
2037 depth = strtoul(arg+8, &end, 0);
2038 if (!arg[8] || *end)
2039 usage(pack_usage);
2040 continue;
2041 }
2042 if (!strcmp("--progress", arg)) {
2043 progress = 1;
2044 continue;
2045 }
231f240b
NP
2046 if (!strcmp("--all-progress", arg)) {
2047 progress = 2;
2048 continue;
2049 }
b5d97e6b
JH
2050 if (!strcmp("-q", arg)) {
2051 progress = 0;
2052 continue;
2053 }
2054 if (!strcmp("--no-reuse-delta", arg)) {
2055 no_reuse_delta = 1;
2056 continue;
2057 }
fa736f72
NP
2058 if (!strcmp("--no-reuse-object", arg)) {
2059 no_reuse_object = no_reuse_delta = 1;
2060 continue;
2061 }
be6b1914 2062 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 2063 allow_ofs_delta = 1;
be6b1914
NP
2064 continue;
2065 }
b5d97e6b
JH
2066 if (!strcmp("--stdout", arg)) {
2067 pack_to_stdout = 1;
2068 continue;
2069 }
2070 if (!strcmp("--revs", arg)) {
2071 use_internal_rev_list = 1;
2072 continue;
2073 }
08cdfb13
JH
2074 if (!strcmp("--keep-unreachable", arg)) {
2075 keep_unreachable = 1;
2076 continue;
2077 }
8d1d8f83 2078 if (!strcmp("--unpacked", arg) ||
599065a3 2079 !prefixcmp(arg, "--unpacked=") ||
63049292 2080 !strcmp("--reflog", arg) ||
8d1d8f83
JH
2081 !strcmp("--all", arg)) {
2082 use_internal_rev_list = 1;
ffa84ffb
RD
2083 if (rp_ac >= rp_ac_alloc - 1) {
2084 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2085 rp_av = xrealloc(rp_av,
2086 rp_ac_alloc * sizeof(*rp_av));
2087 }
8d1d8f83 2088 rp_av[rp_ac++] = arg;
b5d97e6b
JH
2089 continue;
2090 }
8d1d8f83
JH
2091 if (!strcmp("--thin", arg)) {
2092 use_internal_rev_list = 1;
2093 thin = 1;
2094 rp_av[1] = "--objects-edge";
b5d97e6b 2095 continue;
4ba7d711
NP
2096 }
2097 if (!prefixcmp(arg, "--index-version=")) {
2098 char *c;
aa7e44bf
GB
2099 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2100 if (pack_idx_default_version > 2)
4ba7d711
NP
2101 die("bad %s", arg);
2102 if (*c == ',')
aa7e44bf
GB
2103 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2104 if (*c || pack_idx_off32_limit & 0x80000000)
4ba7d711
NP
2105 die("bad %s", arg);
2106 continue;
b5d97e6b
JH
2107 }
2108 usage(pack_usage);
2109 }
2110
2111 /* Traditionally "pack-objects [options] base extra" failed;
2112 * we would however want to take refs parameter that would
2113 * have been given to upstream rev-list ourselves, which means
2114 * we somehow want to say what the base name is. So the
2115 * syntax would be:
2116 *
2117 * pack-objects [options] base <refs...>
2118 *
2119 * in other words, we would treat the first non-option as the
2120 * base_name and send everything else to the internal revision
2121 * walker.
2122 */
2123
2124 if (!pack_to_stdout)
2125 base_name = argv[i++];
2126
2127 if (pack_to_stdout != !base_name)
2128 usage(pack_usage);
2129
01c12a23
DH
2130 if (pack_to_stdout && pack_size_limit)
2131 die("--max-pack-size cannot be used to build a pack for transfer.");
2132
8d1d8f83
JH
2133 if (!pack_to_stdout && thin)
2134 die("--thin cannot be used to build an indexable pack.");
b5d97e6b
JH
2135
2136 prepare_packed_git();
2137
13aaf148 2138 if (progress)
dc6a0757 2139 progress_state = start_progress("Counting objects", 0);
b5d97e6b
JH
2140 if (!use_internal_rev_list)
2141 read_object_list_from_stdin();
8d1d8f83
JH
2142 else {
2143 rp_av[rp_ac] = NULL;
2144 get_object_list(rp_ac, rp_av);
2145 }
4d4fcc54 2146 stop_progress(&progress_state);
96a02f8f 2147
f0b0af1b 2148 if (non_empty && !nr_result)
1c4a2912 2149 return 0;
f7ae6a93
NP
2150 if (nr_result)
2151 prepare_pack(window, depth);
d01fb92f 2152 write_pack_file();
ab7cd7bb 2153 if (progress)
7cadf491 2154 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
67c08ce1 2155 written, written_delta, reused, reused_delta);
c323ac7d
LT
2156 return 0;
2157}