Use GIT_OBJECT_DIR for temporary files of pack-objects
[git/git.git] / builtin-pack-objects.c
CommitLineData
5d4a6003 1#include "builtin.h"
c323ac7d
LT
2#include "cache.h"
3#include "object.h"
8e440259
PE
4#include "blob.h"
5#include "commit.h"
6#include "tag.h"
7#include "tree.h"
c323ac7d 8#include "delta.h"
a733cb60 9#include "pack.h"
c38138cd 10#include "csum-file.h"
1b0c7174 11#include "tree-walk.h"
b5d97e6b
JH
12#include "diff.h"
13#include "revision.h"
14#include "list-objects.h"
96a02f8f 15#include "progress.h"
c323ac7d 16
231f240b
NP
17static const char pack_usage[] = "\
18git-pack-objects [{ -q | --progress | --all-progress }] \n\
19 [--local] [--incremental] [--window=N] [--depth=N] \n\
20 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
63049292 21 [--revs [--unpacked | --all]*] [--reflog] [--stdout | base-name] \n\
231f240b 22 [<ref-list | <object-list]";
c323ac7d 23
c323ac7d
LT
24struct object_entry {
25 unsigned char sha1[20];
a3fbf4df
NP
26 uint32_t crc32; /* crc of raw pack data for this object */
27 off_t offset; /* offset into the final pack file */
3f9ac8d2 28 unsigned long size; /* uncompressed size */
3f9ac8d2 29 unsigned int hash; /* name hint hash */
a3fbf4df 30 unsigned int depth; /* delta depth */
3f9ac8d2 31 struct packed_git *in_pack; /* already in pack */
6777a59f 32 off_t in_pack_offset;
a3fbf4df 33 struct object_entry *delta; /* delta base object */
82e5a82f 34 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
35 struct object_entry *delta_sibling; /* other deltified objects who
36 * uses the same base as me
37 */
a3fbf4df
NP
38 unsigned long delta_size; /* delta data size (uncompressed) */
39 enum object_type type;
40 enum object_type in_pack_type; /* could be delta */
41 unsigned char in_pack_header_size;
42 unsigned char preferred_base; /* we do not pack this, but is available
43 * to be used as the base objectto delta
44 * objects against.
45 */
c323ac7d
LT
46};
47
3f9ac8d2 48/*
82e5a82f 49 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
50 * expanded). nr_objects & nr_alloc controls this array. They are stored
51 * in the order we see -- typically rev-list --objects order that gives us
52 * nice "minimum seek" order.
3f9ac8d2 53 */
e4d58311
NP
54static struct object_entry *objects;
55static uint32_t nr_objects, nr_alloc, nr_result;
3f9ac8d2 56
96f1e58f
DR
57static int non_empty;
58static int no_reuse_delta;
59static int local;
60static int incremental;
be6b1914 61static int allow_ofs_delta;
81a216a5
NP
62static const char *pack_tmp_name, *idx_tmp_name;
63static char tmpname[PATH_MAX];
e1808845 64static unsigned char pack_file_sha1[20];
024701f1 65static int progress = 1;
4812a93a 66static int window = 10;
df6d6101 67static int pack_to_stdout;
8d1d8f83 68static int num_preferred_base;
96a02f8f 69static struct progress progress_state;
c323ac7d 70
3f9ac8d2
JH
71/*
72 * The object names in objects array are hashed with this hashtable,
e4d58311 73 * to help looking up the entry by object name.
3f9ac8d2
JH
74 * This hashtable is built after all the objects are seen.
75 */
96f1e58f
DR
76static int *object_ix;
77static int object_ix_hashsz;
3f9ac8d2
JH
78
79/*
80 * Pack index for existing packs give us easy access to the offsets into
81 * corresponding pack file where each object's data starts, but the entries
82 * do not store the size of the compressed representation (uncompressed
780e6e73
NP
83 * size is easily available by examining the pack entry header). It is
84 * also rather expensive to find the sha1 for an object given its offset.
85 *
86 * We build a hashtable of existing packs (pack_revindex), and keep reverse
87 * index here -- pack index file is sorted by object name mapping to offset;
88 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
89 * ordered by offset, so if you know the offset of an object, next offset
90 * is where its packed representation ends and the index_nr can be used to
91 * get the object sha1 from the main index.
3f9ac8d2 92 */
780e6e73 93struct revindex_entry {
6777a59f 94 off_t offset;
780e6e73
NP
95 unsigned int nr;
96};
3f9ac8d2
JH
97struct pack_revindex {
98 struct packed_git *p;
780e6e73
NP
99 struct revindex_entry *revindex;
100};
101static struct pack_revindex *pack_revindex;
96f1e58f 102static int pack_revindex_hashsz;
3f9ac8d2
JH
103
104/*
105 * stats
106 */
7cadf491
SP
107static uint32_t written, written_delta;
108static uint32_t reused, reused_delta;
3f9ac8d2
JH
109
110static int pack_revindex_ix(struct packed_git *p)
111{
2b74cffa 112 unsigned long ui = (unsigned long)p;
3f9ac8d2
JH
113 int i;
114
115 ui = ui ^ (ui >> 16); /* defeat structure alignment */
116 i = (int)(ui % pack_revindex_hashsz);
117 while (pack_revindex[i].p) {
118 if (pack_revindex[i].p == p)
119 return i;
120 if (++i == pack_revindex_hashsz)
121 i = 0;
122 }
123 return -1 - i;
124}
125
126static void prepare_pack_ix(void)
127{
128 int num;
129 struct packed_git *p;
130 for (num = 0, p = packed_git; p; p = p->next)
131 num++;
132 if (!num)
133 return;
134 pack_revindex_hashsz = num * 11;
135 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
136 for (p = packed_git; p; p = p->next) {
137 num = pack_revindex_ix(p);
138 num = - 1 - num;
139 pack_revindex[num].p = p;
140 }
141 /* revindex elements are lazily initialized */
142}
143
144static int cmp_offset(const void *a_, const void *b_)
145{
780e6e73
NP
146 const struct revindex_entry *a = a_;
147 const struct revindex_entry *b = b_;
148 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
3f9ac8d2
JH
149}
150
151/*
152 * Ordered list of offsets of objects in the pack.
153 */
154static void prepare_pack_revindex(struct pack_revindex *rix)
155{
156 struct packed_git *p = rix->p;
57059091 157 int num_ent = p->num_objects;
3f9ac8d2 158 int i;
42873078 159 const char *index = p->index_data;
3f9ac8d2 160
780e6e73 161 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
c553ca25
NP
162 index += 4 * 256;
163
164 if (p->index_version > 1) {
165 const uint32_t *off_32 =
166 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
167 const uint32_t *off_64 = off_32 + p->num_objects;
168 for (i = 0; i < num_ent; i++) {
169 uint32_t off = ntohl(*off_32++);
170 if (!(off & 0x80000000)) {
171 rix->revindex[i].offset = off;
172 } else {
173 rix->revindex[i].offset =
174 ((uint64_t)ntohl(*off_64++)) << 32;
175 rix->revindex[i].offset |=
176 ntohl(*off_64++);
177 }
178 rix->revindex[i].nr = i;
179 }
180 } else {
181 for (i = 0; i < num_ent; i++) {
182 uint32_t hl = *((uint32_t *)(index + 24 * i));
183 rix->revindex[i].offset = ntohl(hl);
184 rix->revindex[i].nr = i;
185 }
3f9ac8d2 186 }
c553ca25 187
3f9ac8d2
JH
188 /* This knows the pack format -- the 20-byte trailer
189 * follows immediately after the last object data.
190 */
780e6e73
NP
191 rix->revindex[num_ent].offset = p->pack_size - 20;
192 rix->revindex[num_ent].nr = -1;
193 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
3f9ac8d2
JH
194}
195
780e6e73 196static struct revindex_entry * find_packed_object(struct packed_git *p,
6777a59f 197 off_t ofs)
3f9ac8d2
JH
198{
199 int num;
200 int lo, hi;
201 struct pack_revindex *rix;
780e6e73 202 struct revindex_entry *revindex;
3f9ac8d2
JH
203 num = pack_revindex_ix(p);
204 if (num < 0)
205 die("internal error: pack revindex uninitialized");
206 rix = &pack_revindex[num];
207 if (!rix->revindex)
208 prepare_pack_revindex(rix);
209 revindex = rix->revindex;
210 lo = 0;
57059091 211 hi = p->num_objects + 1;
3f9ac8d2
JH
212 do {
213 int mi = (lo + hi) / 2;
780e6e73
NP
214 if (revindex[mi].offset == ofs) {
215 return revindex + mi;
3f9ac8d2 216 }
780e6e73 217 else if (ofs < revindex[mi].offset)
3f9ac8d2
JH
218 hi = mi;
219 else
220 lo = mi + 1;
221 } while (lo < hi);
222 die("internal error: pack revindex corrupt");
223}
224
42873078
NP
225static const unsigned char *find_packed_object_name(struct packed_git *p,
226 off_t ofs)
780e6e73
NP
227{
228 struct revindex_entry *entry = find_packed_object(p, ofs);
d72308e0 229 return nth_packed_object_sha1(p, entry->nr);
780e6e73
NP
230}
231
c323ac7d
LT
232static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
233{
234 unsigned long othersize, delta_size;
21666f1a
NP
235 enum object_type type;
236 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
c323ac7d
LT
237 void *delta_buf;
238
239 if (!otherbuf)
240 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
8ee378a0 241 delta_buf = diff_delta(otherbuf, othersize,
dcde55bc 242 buf, size, &delta_size, 0);
75c42d8c 243 if (!delta_buf || delta_size != entry->delta_size)
c323ac7d
LT
244 die("delta size changed");
245 free(buf);
246 free(otherbuf);
247 return delta_buf;
248}
249
a733cb60
LT
250/*
251 * The per-object header is a pretty dense thing, which is
252 * - first byte: low four bits are "size", then three bits of "type",
253 * and the high bit is "size continues".
254 * - each byte afterwards: low seven bits are size continuation,
255 * with the high bit being "size continues"
256 */
257static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
258{
01247d87 259 int n = 1;
a733cb60
LT
260 unsigned char c;
261
eb32d236 262 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
263 die("bad type %d", type);
264
01247d87
LT
265 c = (type << 4) | (size & 15);
266 size >>= 4;
267 while (size) {
a733cb60 268 *hdr++ = c | 0x80;
01247d87
LT
269 c = size & 0x7f;
270 size >>= 7;
271 n++;
a733cb60
LT
272 }
273 *hdr = c;
274 return n;
275}
276
780e6e73
NP
277/*
278 * we are going to reuse the existing object data as is. make
279 * sure it is not corrupt.
280 */
079afb18
SP
281static int check_pack_inflate(struct packed_git *p,
282 struct pack_window **w_curs,
6777a59f
SP
283 off_t offset,
284 off_t len,
079afb18
SP
285 unsigned long expect)
286{
287 z_stream stream;
288 unsigned char fakebuf[4096], *in;
289 int st;
290
291 memset(&stream, 0, sizeof(stream));
292 inflateInit(&stream);
293 do {
294 in = use_pack(p, w_curs, offset, &stream.avail_in);
295 stream.next_in = in;
296 stream.next_out = fakebuf;
297 stream.avail_out = sizeof(fakebuf);
298 st = inflate(&stream, Z_FINISH);
299 offset += stream.next_in - in;
300 } while (st == Z_OK || st == Z_BUF_ERROR);
301 inflateEnd(&stream);
302 return (st == Z_STREAM_END &&
303 stream.total_out == expect &&
304 stream.total_in == len) ? 0 : -1;
305}
306
91ecbeca
NP
307static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
308 off_t offset, off_t len, unsigned int nr)
309{
310 const uint32_t *index_crc;
311 uint32_t data_crc = crc32(0, Z_NULL, 0);
312
313 do {
314 unsigned int avail;
315 void *data = use_pack(p, w_curs, offset, &avail);
316 if (avail > len)
317 avail = len;
318 data_crc = crc32(data_crc, data, avail);
319 offset += avail;
320 len -= avail;
321 } while (len);
322
323 index_crc = p->index_data;
324 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
325
326 return data_crc != ntohl(*index_crc);
327}
328
079afb18
SP
329static void copy_pack_data(struct sha1file *f,
330 struct packed_git *p,
331 struct pack_window **w_curs,
6777a59f
SP
332 off_t offset,
333 off_t len)
079afb18
SP
334{
335 unsigned char *in;
336 unsigned int avail;
337
338 while (len) {
339 in = use_pack(p, w_curs, offset, &avail);
340 if (avail > len)
6777a59f 341 avail = (unsigned int)len;
079afb18
SP
342 sha1write(f, in, avail);
343 offset += avail;
344 len -= avail;
345 }
346}
347
348static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
df6d6101 349{
72518e9c
JH
350 z_stream stream;
351 unsigned char fakebuf[4096];
352 int st;
353
354 memset(&stream, 0, sizeof(stream));
355 stream.next_in = data;
356 stream.avail_in = len;
357 stream.next_out = fakebuf;
358 stream.avail_out = sizeof(fakebuf);
359 inflateInit(&stream);
360
361 while (1) {
362 st = inflate(&stream, Z_FINISH);
363 if (st == Z_STREAM_END || st == Z_OK) {
364 st = (stream.total_out == expect &&
365 stream.total_in == len) ? 0 : -1;
366 break;
367 }
368 if (st != Z_BUF_ERROR) {
369 st = -1;
370 break;
371 }
372 stream.next_out = fakebuf;
373 stream.avail_out = sizeof(fakebuf);
374 }
375 inflateEnd(&stream);
376 return st;
df6d6101
JH
377}
378
df6d6101
JH
379static int revalidate_loose_object(struct object_entry *entry,
380 unsigned char *map,
381 unsigned long mapsize)
382{
383 /* we already know this is a loose object with new type header. */
72518e9c
JH
384 enum object_type type;
385 unsigned long size, used;
df6d6101
JH
386
387 if (pack_to_stdout)
388 return 0;
389
72518e9c
JH
390 used = unpack_object_header_gently(map, mapsize, &type, &size);
391 if (!used)
392 return -1;
393 map += used;
394 mapsize -= used;
079afb18 395 return check_loose_inflate(map, mapsize, size);
df6d6101
JH
396}
397
d7dd0223 398static unsigned long write_object(struct sha1file *f,
7a979d99 399 struct object_entry *entry)
c323ac7d
LT
400{
401 unsigned long size;
21666f1a 402 enum object_type type;
3f9ac8d2 403 void *buf;
a733cb60 404 unsigned char header[10];
6777a59f
SP
405 unsigned hdrlen;
406 off_t datalen;
a733cb60 407 enum object_type obj_type;
ab7cd7bb 408 int to_reuse = 0;
c323ac7d 409
78d1e84f
NP
410 if (!pack_to_stdout)
411 crc32_begin(f);
412
a733cb60 413 obj_type = entry->type;
ab7cd7bb
JH
414 if (! entry->in_pack)
415 to_reuse = 0; /* can't reuse what we don't have */
780e6e73 416 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
ab7cd7bb
JH
417 to_reuse = 1; /* check_object() decided it for us */
418 else if (obj_type != entry->in_pack_type)
419 to_reuse = 0; /* pack has delta which is unusable */
420 else if (entry->delta)
421 to_reuse = 0; /* we want to pack afresh */
422 else
423 to_reuse = 1; /* we have it in-pack undeltified,
424 * and we do not need to deltify it.
425 */
426
ceec1361
JH
427 if (!entry->in_pack && !entry->delta) {
428 unsigned char *map;
429 unsigned long mapsize;
430 map = map_sha1_file(entry->sha1, &mapsize);
431 if (map && !legacy_loose_object(map)) {
432 /* We can copy straight into the pack file */
df6d6101
JH
433 if (revalidate_loose_object(entry, map, mapsize))
434 die("corrupt loose object %s",
435 sha1_to_hex(entry->sha1));
ceec1361
JH
436 sha1write(f, map, mapsize);
437 munmap(map, mapsize);
438 written++;
439 reused++;
440 return mapsize;
441 }
442 if (map)
443 munmap(map, mapsize);
444 }
445
df6d6101 446 if (!to_reuse) {
21666f1a 447 buf = read_sha1_file(entry->sha1, &type, &size);
3f9ac8d2
JH
448 if (!buf)
449 die("unable to read %s", sha1_to_hex(entry->sha1));
450 if (size != entry->size)
451 die("object %s size inconsistency (%lu vs %lu)",
452 sha1_to_hex(entry->sha1), size, entry->size);
453 if (entry->delta) {
454 buf = delta_against(buf, size, entry);
455 size = entry->delta_size;
be6b1914
NP
456 obj_type = (allow_ofs_delta && entry->delta->offset) ?
457 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2
JH
458 }
459 /*
460 * The object header is a byte of 'type' followed by zero or
be6b1914 461 * more bytes of length.
3f9ac8d2
JH
462 */
463 hdrlen = encode_header(obj_type, size, header);
464 sha1write(f, header, hdrlen);
465
be6b1914
NP
466 if (obj_type == OBJ_OFS_DELTA) {
467 /*
468 * Deltas with relative base contain an additional
469 * encoding of the relative offset for the delta
470 * base from this object's position in the pack.
471 */
6777a59f 472 off_t ofs = entry->offset - entry->delta->offset;
be6b1914
NP
473 unsigned pos = sizeof(header) - 1;
474 header[pos] = ofs & 127;
475 while (ofs >>= 7)
476 header[--pos] = 128 | (--ofs & 127);
477 sha1write(f, header + pos, sizeof(header) - pos);
478 hdrlen += sizeof(header) - pos;
479 } else if (obj_type == OBJ_REF_DELTA) {
480 /*
481 * Deltas with a base reference contain
482 * an additional 20 bytes for the base sha1.
483 */
484 sha1write(f, entry->delta->sha1, 20);
3f9ac8d2
JH
485 hdrlen += 20;
486 }
487 datalen = sha1write_compressed(f, buf, size);
488 free(buf);
c323ac7d 489 }
3f9ac8d2
JH
490 else {
491 struct packed_git *p = entry->in_pack;
03e79c88 492 struct pack_window *w_curs = NULL;
91ecbeca 493 struct revindex_entry *revidx;
6777a59f 494 off_t offset;
3f9ac8d2 495
780e6e73
NP
496 if (entry->delta) {
497 obj_type = (allow_ofs_delta && entry->delta->offset) ?
498 OBJ_OFS_DELTA : OBJ_REF_DELTA;
499 reused_delta++;
500 }
501 hdrlen = encode_header(obj_type, entry->size, header);
502 sha1write(f, header, hdrlen);
503 if (obj_type == OBJ_OFS_DELTA) {
6777a59f 504 off_t ofs = entry->offset - entry->delta->offset;
780e6e73
NP
505 unsigned pos = sizeof(header) - 1;
506 header[pos] = ofs & 127;
507 while (ofs >>= 7)
508 header[--pos] = 128 | (--ofs & 127);
509 sha1write(f, header + pos, sizeof(header) - pos);
510 hdrlen += sizeof(header) - pos;
511 } else if (obj_type == OBJ_REF_DELTA) {
512 sha1write(f, entry->delta->sha1, 20);
513 hdrlen += 20;
514 }
df6d6101 515
91ecbeca
NP
516 offset = entry->in_pack_offset;
517 revidx = find_packed_object(p, offset);
518 datalen = revidx[1].offset - offset;
519 if (!pack_to_stdout && p->index_version > 1 &&
520 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
521 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
522 offset += entry->in_pack_header_size;
523 datalen -= entry->in_pack_header_size;
524 if (!pack_to_stdout && p->index_version == 1 &&
525 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
526 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
079afb18 527 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 528 unuse_pack(&w_curs);
3f9ac8d2 529 reused++;
a733cb60 530 }
be6b1914 531 if (entry->delta)
ab7cd7bb 532 written_delta++;
3f9ac8d2 533 written++;
78d1e84f
NP
534 if (!pack_to_stdout)
535 entry->crc32 = crc32_end(f);
c323ac7d
LT
536 return hdrlen + datalen;
537}
538
6777a59f 539static off_t write_one(struct sha1file *f,
9d5ab962 540 struct object_entry *e,
6777a59f 541 off_t offset)
9d5ab962 542{
d7dd0223
NP
543 unsigned long size;
544
545 /* offset is non zero if object is written already. */
be6b1914 546 if (e->offset || e->preferred_base)
9d5ab962 547 return offset;
d7dd0223
NP
548
549 /* if we are deltified, write out base object first. */
9d5ab962
JH
550 if (e->delta)
551 offset = write_one(f, e->delta, offset);
d7dd0223 552
be6b1914 553 e->offset = offset;
d7dd0223
NP
554 size = write_object(f, e);
555
556 /* make sure off_t is sufficiently large not to wrap */
557 if (offset > offset + size)
558 die("pack too large for current definition of off_t");
559 return offset + size;
9d5ab962
JH
560}
561
3082acfa
AR
562static int open_object_dir_tmp(const char *path)
563{
564 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
565 return mkstemp(tmpname);
566}
567
c553ca25 568static off_t write_pack_file(void)
c323ac7d 569{
7cadf491 570 uint32_t i;
d22b9290 571 struct sha1file *f;
c553ca25 572 off_t offset, last_obj_offset = 0;
a733cb60 573 struct pack_header hdr;
fa438a2e 574 int do_progress = progress;
c323ac7d 575
81a216a5 576 if (pack_to_stdout) {
d22b9290 577 f = sha1fd(1, "<stdout>");
fa438a2e 578 do_progress >>= 1;
81a216a5 579 } else {
3082acfa 580 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
81a216a5
NP
581 if (fd < 0)
582 die("unable to create %s: %s\n", tmpname, strerror(errno));
583 pack_tmp_name = xstrdup(tmpname);
584 f = sha1fd(fd, pack_tmp_name);
fa438a2e 585 }
81a216a5 586
13aaf148
NP
587 if (do_progress)
588 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
183bdb2c 589
a733cb60 590 hdr.hdr_signature = htonl(PACK_SIGNATURE);
01247d87 591 hdr.hdr_version = htonl(PACK_VERSION);
7a979d99 592 hdr.hdr_entries = htonl(nr_result);
a733cb60
LT
593 sha1write(f, &hdr, sizeof(hdr));
594 offset = sizeof(hdr);
f0b0af1b
JH
595 if (!nr_result)
596 goto done;
5e8dc750 597 for (i = 0; i < nr_objects; i++) {
c553ca25 598 last_obj_offset = offset;
9d5ab962 599 offset = write_one(f, objects + i, offset);
96a02f8f
NP
600 if (do_progress)
601 display_progress(&progress_state, written);
5e8dc750 602 }
183bdb2c 603 if (do_progress)
96a02f8f 604 stop_progress(&progress_state);
f0b0af1b 605 done:
67c08ce1 606 if (written != nr_result)
7cadf491 607 die("wrote %u objects while expecting %u", written, nr_result);
e1808845 608 sha1close(f, pack_file_sha1, 1);
c553ca25
NP
609
610 return last_obj_offset;
c323ac7d
LT
611}
612
81a216a5
NP
613static int sha1_sort(const void *_a, const void *_b)
614{
615 const struct object_entry *a = *(struct object_entry **)_a;
616 const struct object_entry *b = *(struct object_entry **)_b;
617 return hashcmp(a->sha1, b->sha1);
618}
619
4ba7d711
NP
620static uint32_t index_default_version = 1;
621static uint32_t index_off32_limit = 0x7fffffff;
622
81a216a5 623static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
c323ac7d 624{
81a216a5
NP
625 struct sha1file *f;
626 struct object_entry **sorted_by_sha, **list, **last;
b18b00a6 627 uint32_t array[256];
81a216a5
NP
628 uint32_t i, index_version;
629 SHA_CTX ctx;
81a216a5 630
3082acfa 631 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
81a216a5
NP
632 if (fd < 0)
633 die("unable to create %s: %s\n", tmpname, strerror(errno));
634 idx_tmp_name = xstrdup(tmpname);
635 f = sha1fd(fd, idx_tmp_name);
636
637 if (nr_result) {
638 uint32_t j = 0;
639 sorted_by_sha =
640 xcalloc(nr_result, sizeof(struct object_entry *));
641 for (i = 0; i < nr_objects; i++)
642 if (!objects[i].preferred_base)
643 sorted_by_sha[j++] = objects + i;
644 if (j != nr_result)
645 die("listed %u objects while expecting %u", j, nr_result);
646 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
647 list = sorted_by_sha;
648 last = sorted_by_sha + nr_result;
649 } else
650 sorted_by_sha = list = last = NULL;
c553ca25
NP
651
652 /* if last object's offset is >= 2^31 we should use index V2 */
4ba7d711 653 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
c553ca25
NP
654
655 /* index versions 2 and above need a header */
656 if (index_version >= 2) {
657 struct pack_idx_header hdr;
658 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
659 hdr.idx_version = htonl(index_version);
660 sha1write(f, &hdr, sizeof(hdr));
661 }
c323ac7d
LT
662
663 /*
664 * Write the first-level table (the list is sorted,
665 * but we use a 256-entry lookup to be able to avoid
e1808845 666 * having to do eight extra binary search iterations).
c323ac7d
LT
667 */
668 for (i = 0; i < 256; i++) {
669 struct object_entry **next = list;
670 while (next < last) {
671 struct object_entry *entry = *next;
672 if (entry->sha1[0] != i)
673 break;
674 next++;
675 }
676 array[i] = htonl(next - sorted_by_sha);
677 list = next;
678 }
b18b00a6 679 sha1write(f, array, 256 * 4);
c323ac7d 680
81a216a5
NP
681 /* Compute the SHA1 hash of sorted object names. */
682 SHA1_Init(&ctx);
683
684 /* Write the actual SHA1 entries. */
c323ac7d 685 list = sorted_by_sha;
7a979d99 686 for (i = 0; i < nr_result; i++) {
c323ac7d 687 struct object_entry *entry = *list++;
c553ca25
NP
688 if (index_version < 2) {
689 uint32_t offset = htonl(entry->offset);
690 sha1write(f, &offset, 4);
691 }
c38138cd 692 sha1write(f, entry->sha1, 20);
81a216a5 693 SHA1_Update(&ctx, entry->sha1, 20);
c323ac7d 694 }
c553ca25
NP
695
696 if (index_version >= 2) {
697 unsigned int nr_large_offset = 0;
698
699 /* write the crc32 table */
700 list = sorted_by_sha;
701 for (i = 0; i < nr_objects; i++) {
702 struct object_entry *entry = *list++;
703 uint32_t crc32_val = htonl(entry->crc32);
704 sha1write(f, &crc32_val, 4);
705 }
706
707 /* write the 32-bit offset table */
708 list = sorted_by_sha;
709 for (i = 0; i < nr_objects; i++) {
710 struct object_entry *entry = *list++;
4ba7d711 711 uint32_t offset = (entry->offset <= index_off32_limit) ?
c553ca25
NP
712 entry->offset : (0x80000000 | nr_large_offset++);
713 offset = htonl(offset);
714 sha1write(f, &offset, 4);
715 }
716
717 /* write the large offset table */
718 list = sorted_by_sha;
719 while (nr_large_offset) {
720 struct object_entry *entry = *list++;
721 uint64_t offset = entry->offset;
4ba7d711 722 if (offset > index_off32_limit) {
c553ca25
NP
723 uint32_t split[2];
724 split[0] = htonl(offset >> 32);
725 split[1] = htonl(offset & 0xffffffff);
726 sha1write(f, split, 8);
727 nr_large_offset--;
728 }
729 }
730 }
731
e1808845
LT
732 sha1write(f, pack_file_sha1, 20);
733 sha1close(f, NULL, 1);
81a216a5
NP
734 free(sorted_by_sha);
735 SHA1_Final(sha1, &ctx);
c323ac7d
LT
736}
737
7a979d99
JH
738static int locate_object_entry_hash(const unsigned char *sha1)
739{
740 int i;
741 unsigned int ui;
742 memcpy(&ui, sha1, sizeof(unsigned int));
743 i = ui % object_ix_hashsz;
744 while (0 < object_ix[i]) {
a89fccd2 745 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
7a979d99
JH
746 return i;
747 if (++i == object_ix_hashsz)
748 i = 0;
749 }
750 return -1 - i;
751}
752
753static struct object_entry *locate_object_entry(const unsigned char *sha1)
754{
755 int i;
756
757 if (!object_ix_hashsz)
758 return NULL;
759
760 i = locate_object_entry_hash(sha1);
761 if (0 <= i)
762 return &objects[object_ix[i]-1];
763 return NULL;
764}
765
766static void rehash_objects(void)
c323ac7d 767{
7cadf491 768 uint32_t i;
7a979d99
JH
769 struct object_entry *oe;
770
771 object_ix_hashsz = nr_objects * 3;
772 if (object_ix_hashsz < 1024)
773 object_ix_hashsz = 1024;
774 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 775 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99
JH
776 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
777 int ix = locate_object_entry_hash(oe->sha1);
778 if (0 <= ix)
779 continue;
780 ix = -1 - ix;
781 object_ix[ix] = i + 1;
782 }
783}
784
ce0bd642 785static unsigned name_hash(const char *name)
1d6b38cc 786{
ce0bd642
LT
787 unsigned char c;
788 unsigned hash = 0;
789
eeef7135 790 /*
ce0bd642
LT
791 * This effectively just creates a sortable number from the
792 * last sixteen non-whitespace characters. Last characters
793 * count "most", so things that end in ".c" sort together.
eeef7135 794 */
ce0bd642
LT
795 while ((c = *name++) != 0) {
796 if (isspace(c))
797 continue;
798 hash = (hash >> 2) + (c << 24);
799 }
1d6b38cc
JH
800 return hash;
801}
802
5c49c116
NP
803static int add_object_entry(const unsigned char *sha1, enum object_type type,
804 unsigned hash, int exclude)
c323ac7d 805{
c323ac7d 806 struct object_entry *entry;
29b734e4 807 struct packed_git *p, *found_pack = NULL;
6777a59f 808 off_t found_offset = 0;
29b734e4
NP
809 int ix;
810
811 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
812 if (ix >= 0) {
813 if (exclude) {
814 entry = objects + object_ix[ix] - 1;
81a216a5
NP
815 if (!entry->preferred_base)
816 nr_result--;
29b734e4
NP
817 entry->preferred_base = 1;
818 }
819 return 0;
820 }
c323ac7d 821
5c49c116
NP
822 for (p = packed_git; p; p = p->next) {
823 off_t offset = find_pack_entry_one(sha1, p);
824 if (offset) {
825 if (!found_pack) {
826 found_offset = offset;
827 found_pack = p;
64560374 828 }
5c49c116
NP
829 if (exclude)
830 break;
831 if (incremental)
832 return 0;
833 if (local && !p->pack_local)
834 return 0;
64560374
LT
835 }
836 }
eb019375 837
29b734e4
NP
838 if (nr_objects >= nr_alloc) {
839 nr_alloc = (nr_alloc + 1024) * 3 / 2;
7cadf491 840 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d 841 }
29b734e4
NP
842
843 entry = objects + nr_objects++;
c323ac7d 844 memset(entry, 0, sizeof(*entry));
e702496e 845 hashcpy(entry->sha1, sha1);
27225f2e 846 entry->hash = hash;
5c49c116
NP
847 if (type)
848 entry->type = type;
29b734e4
NP
849 if (exclude)
850 entry->preferred_base = 1;
81a216a5
NP
851 else
852 nr_result++;
29b734e4
NP
853 if (found_pack) {
854 entry->in_pack = found_pack;
855 entry->in_pack_offset = found_offset;
856 }
7a979d99
JH
857
858 if (object_ix_hashsz * 3 <= nr_objects * 4)
859 rehash_objects();
29b734e4
NP
860 else
861 object_ix[-1 - ix] = nr_objects;
7a979d99 862
96a02f8f
NP
863 if (progress)
864 display_progress(&progress_state, nr_objects);
29b734e4
NP
865
866 return 1;
c323ac7d
LT
867}
868
5379a5c5
JH
869struct pbase_tree_cache {
870 unsigned char sha1[20];
871 int ref;
872 int temporary;
873 void *tree_data;
874 unsigned long tree_size;
875};
876
877static struct pbase_tree_cache *(pbase_tree_cache[256]);
878static int pbase_tree_cache_ix(const unsigned char *sha1)
879{
880 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
881}
882static int pbase_tree_cache_ix_incr(int ix)
883{
884 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
885}
886
887static struct pbase_tree {
888 struct pbase_tree *next;
889 /* This is a phony "cache" entry; we are not
890 * going to evict it nor find it through _get()
891 * mechanism -- this is for the toplevel node that
892 * would almost always change with any commit.
893 */
894 struct pbase_tree_cache pcache;
895} *pbase_tree;
896
897static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
898{
899 struct pbase_tree_cache *ent, *nent;
900 void *data;
901 unsigned long size;
21666f1a 902 enum object_type type;
5379a5c5
JH
903 int neigh;
904 int my_ix = pbase_tree_cache_ix(sha1);
905 int available_ix = -1;
906
907 /* pbase-tree-cache acts as a limited hashtable.
908 * your object will be found at your index or within a few
909 * slots after that slot if it is cached.
910 */
911 for (neigh = 0; neigh < 8; neigh++) {
912 ent = pbase_tree_cache[my_ix];
a89fccd2 913 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
914 ent->ref++;
915 return ent;
916 }
917 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
918 ((0 <= available_ix) &&
919 (!ent && pbase_tree_cache[available_ix])))
920 available_ix = my_ix;
921 if (!ent)
922 break;
923 my_ix = pbase_tree_cache_ix_incr(my_ix);
924 }
925
926 /* Did not find one. Either we got a bogus request or
927 * we need to read and perhaps cache.
928 */
21666f1a 929 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
930 if (!data)
931 return NULL;
21666f1a 932 if (type != OBJ_TREE) {
5379a5c5
JH
933 free(data);
934 return NULL;
935 }
936
937 /* We need to either cache or return a throwaway copy */
938
939 if (available_ix < 0)
940 ent = NULL;
941 else {
942 ent = pbase_tree_cache[available_ix];
943 my_ix = available_ix;
944 }
945
946 if (!ent) {
947 nent = xmalloc(sizeof(*nent));
948 nent->temporary = (available_ix < 0);
949 }
950 else {
951 /* evict and reuse */
952 free(ent->tree_data);
953 nent = ent;
954 }
e702496e 955 hashcpy(nent->sha1, sha1);
5379a5c5
JH
956 nent->tree_data = data;
957 nent->tree_size = size;
958 nent->ref = 1;
959 if (!nent->temporary)
960 pbase_tree_cache[my_ix] = nent;
961 return nent;
962}
963
964static void pbase_tree_put(struct pbase_tree_cache *cache)
965{
966 if (!cache->temporary) {
967 cache->ref--;
968 return;
969 }
970 free(cache->tree_data);
971 free(cache);
972}
973
974static int name_cmp_len(const char *name)
975{
976 int i;
977 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
978 ;
979 return i;
980}
981
982static void add_pbase_object(struct tree_desc *tree,
5379a5c5 983 const char *name,
ce0bd642
LT
984 int cmplen,
985 const char *fullname)
3f9ac8d2 986{
4c068a98 987 struct name_entry entry;
8a5a8d6c 988 int cmp;
4c068a98
LT
989
990 while (tree_entry(tree,&entry)) {
8a5a8d6c
NP
991 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
992 memcmp(name, entry.path, cmplen);
993 if (cmp > 0)
7a979d99 994 continue;
8a5a8d6c
NP
995 if (cmp < 0)
996 return;
5379a5c5 997 if (name[cmplen] != '/') {
ce0bd642 998 unsigned hash = name_hash(fullname);
5c49c116
NP
999 add_object_entry(entry.sha1,
1000 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1001 hash, 1);
5379a5c5
JH
1002 return;
1003 }
8a5a8d6c 1004 if (S_ISDIR(entry.mode)) {
7a979d99 1005 struct tree_desc sub;
5379a5c5
JH
1006 struct pbase_tree_cache *tree;
1007 const char *down = name+cmplen+1;
1008 int downlen = name_cmp_len(down);
1009
4c068a98 1010 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
1011 if (!tree)
1012 return;
6fda5e51 1013 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 1014
ce0bd642 1015 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
1016 pbase_tree_put(tree);
1017 }
1018 }
1019}
1d6b38cc 1020
5379a5c5
JH
1021static unsigned *done_pbase_paths;
1022static int done_pbase_paths_num;
1023static int done_pbase_paths_alloc;
1024static int done_pbase_path_pos(unsigned hash)
1025{
1026 int lo = 0;
1027 int hi = done_pbase_paths_num;
1028 while (lo < hi) {
1029 int mi = (hi + lo) / 2;
1030 if (done_pbase_paths[mi] == hash)
1031 return mi;
1032 if (done_pbase_paths[mi] < hash)
1033 hi = mi;
1034 else
1035 lo = mi + 1;
1036 }
1037 return -lo-1;
1038}
1039
1040static int check_pbase_path(unsigned hash)
1041{
1042 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1043 if (0 <= pos)
1044 return 1;
1045 pos = -pos - 1;
1046 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1047 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1048 done_pbase_paths = xrealloc(done_pbase_paths,
1049 done_pbase_paths_alloc *
1050 sizeof(unsigned));
1051 }
1052 done_pbase_paths_num++;
1053 if (pos < done_pbase_paths_num)
1054 memmove(done_pbase_paths + pos + 1,
1055 done_pbase_paths + pos,
1056 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1057 done_pbase_paths[pos] = hash;
1058 return 0;
1059}
1060
8d1d8f83 1061static void add_preferred_base_object(const char *name, unsigned hash)
5379a5c5
JH
1062{
1063 struct pbase_tree *it;
8a5a8d6c 1064 int cmplen;
5379a5c5 1065
8a5a8d6c 1066 if (!num_preferred_base || check_pbase_path(hash))
5379a5c5
JH
1067 return;
1068
8a5a8d6c 1069 cmplen = name_cmp_len(name);
5379a5c5
JH
1070 for (it = pbase_tree; it; it = it->next) {
1071 if (cmplen == 0) {
5c49c116 1072 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
5379a5c5
JH
1073 }
1074 else {
1075 struct tree_desc tree;
6fda5e51 1076 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 1077 add_pbase_object(&tree, name, cmplen, name);
7a979d99 1078 }
3f9ac8d2 1079 }
3f9ac8d2
JH
1080}
1081
7a979d99 1082static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 1083{
5379a5c5
JH
1084 struct pbase_tree *it;
1085 void *data;
1086 unsigned long size;
1087 unsigned char tree_sha1[20];
1d6b38cc 1088
8d1d8f83
JH
1089 if (window <= num_preferred_base++)
1090 return;
1091
5379a5c5
JH
1092 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1093 if (!data)
7a979d99 1094 return;
5379a5c5
JH
1095
1096 for (it = pbase_tree; it; it = it->next) {
a89fccd2 1097 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
1098 free(data);
1099 return;
1100 }
1101 }
1102
1103 it = xcalloc(1, sizeof(*it));
1104 it->next = pbase_tree;
1105 pbase_tree = it;
1106
e702496e 1107 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
1108 it->pcache.tree_data = data;
1109 it->pcache.tree_size = size;
3f9ac8d2
JH
1110}
1111
c323ac7d
LT
1112static void check_object(struct object_entry *entry)
1113{
5c49c116 1114 if (entry->in_pack) {
780e6e73 1115 struct packed_git *p = entry->in_pack;
03e79c88 1116 struct pack_window *w_curs = NULL;
5c49c116
NP
1117 const unsigned char *base_ref = NULL;
1118 struct object_entry *base_entry;
1119 unsigned long used, used_0;
6777a59f 1120 unsigned int avail;
5c49c116
NP
1121 off_t ofs;
1122 unsigned char *buf, c;
780e6e73 1123
6777a59f 1124 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb 1125
5c49c116
NP
1126 /*
1127 * We want in_pack_type even if we do not reuse delta.
ab7cd7bb
JH
1128 * There is no point not reusing non-delta representations.
1129 */
6777a59f 1130 used = unpack_object_header_gently(buf, avail,
5c49c116
NP
1131 &entry->in_pack_type,
1132 &entry->size);
ab7cd7bb 1133
5c49c116
NP
1134 /*
1135 * Determine if this is a delta and if so whether we can
1136 * reuse it or not. Otherwise let's find out as cheaply as
1137 * possible what the actual type and size for this object is.
3f9ac8d2 1138 */
5c49c116
NP
1139 switch (entry->in_pack_type) {
1140 default:
1141 /* Not a delta hence we've already got all we need. */
1142 entry->type = entry->in_pack_type;
1143 entry->in_pack_header_size = used;
1144 unuse_pack(&w_curs);
1145 return;
1146 case OBJ_REF_DELTA:
1147 if (!no_reuse_delta && !entry->preferred_base)
1148 base_ref = use_pack(p, &w_curs,
1149 entry->in_pack_offset + used, NULL);
1150 entry->in_pack_header_size = used + 20;
1151 break;
1152 case OBJ_OFS_DELTA:
1153 buf = use_pack(p, &w_curs,
1154 entry->in_pack_offset + used, NULL);
1155 used_0 = 0;
1156 c = buf[used_0++];
1157 ofs = c & 127;
1158 while (c & 128) {
1159 ofs += 1;
1160 if (!ofs || MSB(ofs, 7))
1161 die("delta base offset overflow in pack for %s",
780e6e73 1162 sha1_to_hex(entry->sha1));
5c49c116
NP
1163 c = buf[used_0++];
1164 ofs = (ofs << 7) + (c & 127);
780e6e73 1165 }
5c49c116
NP
1166 if (ofs >= entry->in_pack_offset)
1167 die("delta base offset out of bound for %s",
1168 sha1_to_hex(entry->sha1));
1169 ofs = entry->in_pack_offset - ofs;
1170 if (!no_reuse_delta && !entry->preferred_base)
1171 base_ref = find_packed_object_name(p, ofs);
1172 entry->in_pack_header_size = used + used_0;
1173 break;
780e6e73 1174 }
780e6e73 1175
5c49c116
NP
1176 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1177 /*
1178 * If base_ref was set above that means we wish to
1179 * reuse delta data, and we even found that base
1180 * in the list of objects we want to pack. Goodie!
1181 *
1182 * Depth value does not matter - find_deltas() will
1183 * never consider reused delta as the base object to
1184 * deltify other objects against, in order to avoid
1185 * circular deltas.
3f9ac8d2 1186 */
780e6e73 1187 entry->type = entry->in_pack_type;
5c49c116 1188 entry->delta = base_entry;
15b4d577
JH
1189 entry->delta_sibling = base_entry->delta_child;
1190 base_entry->delta_child = entry;
5c49c116
NP
1191 unuse_pack(&w_curs);
1192 return;
1193 }
ab7cd7bb 1194
5c49c116
NP
1195 if (entry->type) {
1196 /*
1197 * This must be a delta and we already know what the
1198 * final object type is. Let's extract the actual
1199 * object size from the delta header.
1200 */
1201 entry->size = get_size_from_delta(p, &w_curs,
1202 entry->in_pack_offset + entry->in_pack_header_size);
1203 unuse_pack(&w_curs);
3f9ac8d2
JH
1204 return;
1205 }
5c49c116
NP
1206
1207 /*
1208 * No choice but to fall back to the recursive delta walk
1209 * with sha1_object_info() to find about the object type
1210 * at this point...
1211 */
1212 unuse_pack(&w_curs);
36e4d74a 1213 }
3f9ac8d2 1214
21666f1a
NP
1215 entry->type = sha1_object_info(entry->sha1, &entry->size);
1216 if (entry->type < 0)
36e4d74a
JH
1217 die("unable to get type of object %s",
1218 sha1_to_hex(entry->sha1));
3f9ac8d2
JH
1219}
1220
5c49c116
NP
1221static int pack_offset_sort(const void *_a, const void *_b)
1222{
1223 const struct object_entry *a = *(struct object_entry **)_a;
1224 const struct object_entry *b = *(struct object_entry **)_b;
1225
1226 /* avoid filesystem trashing with loose objects */
1227 if (!a->in_pack && !b->in_pack)
1228 return hashcmp(a->sha1, b->sha1);
1229
1230 if (a->in_pack < b->in_pack)
1231 return -1;
1232 if (a->in_pack > b->in_pack)
1233 return 1;
1234 return a->in_pack_offset < b->in_pack_offset ? -1 :
1235 (a->in_pack_offset > b->in_pack_offset);
1236}
1237
c323ac7d
LT
1238static void get_object_details(void)
1239{
7cadf491 1240 uint32_t i;
5c49c116
NP
1241 struct object_entry **sorted_by_offset;
1242
1243 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1244 for (i = 0; i < nr_objects; i++)
1245 sorted_by_offset[i] = objects + i;
1246 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
c323ac7d 1247
3f9ac8d2 1248 prepare_pack_ix();
5c49c116
NP
1249 for (i = 0; i < nr_objects; i++)
1250 check_object(sorted_by_offset[i]);
1251 free(sorted_by_offset);
c323ac7d
LT
1252}
1253
9668cf59 1254static int type_size_sort(const void *_a, const void *_b)
c323ac7d 1255{
9668cf59
NP
1256 const struct object_entry *a = *(struct object_entry **)_a;
1257 const struct object_entry *b = *(struct object_entry **)_b;
1258
c323ac7d
LT
1259 if (a->type < b->type)
1260 return -1;
1261 if (a->type > b->type)
1262 return 1;
27225f2e
LT
1263 if (a->hash < b->hash)
1264 return -1;
1265 if (a->hash > b->hash)
1266 return 1;
7a979d99
JH
1267 if (a->preferred_base < b->preferred_base)
1268 return -1;
1269 if (a->preferred_base > b->preferred_base)
1270 return 1;
c323ac7d
LT
1271 if (a->size < b->size)
1272 return -1;
1273 if (a->size > b->size)
1274 return 1;
adcc7095 1275 return a > b ? -1 : (a < b); /* newest last */
c323ac7d
LT
1276}
1277
1278struct unpacked {
1279 struct object_entry *entry;
1280 void *data;
f6c7081a 1281 struct delta_index *index;
c323ac7d
LT
1282};
1283
1284/*
521a4f4c
LT
1285 * We search for deltas _backwards_ in a list sorted by type and
1286 * by size, so that we see progressively smaller and smaller files.
1287 * That's because we prefer deltas to be from the bigger file
1288 * to the smaller - deletes are potentially cheaper, but perhaps
1289 * more importantly, the bigger file is likely the more recent
1290 * one.
c323ac7d 1291 */
f6c7081a 1292static int try_delta(struct unpacked *trg, struct unpacked *src,
560b25a8 1293 unsigned max_depth)
c323ac7d 1294{
f6c7081a
NP
1295 struct object_entry *trg_entry = trg->entry;
1296 struct object_entry *src_entry = src->entry;
560b25a8 1297 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
21666f1a 1298 enum object_type type;
c323ac7d
LT
1299 void *delta_buf;
1300
1301 /* Don't bother doing diffs between different types */
f6c7081a 1302 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1303 return -1;
1304
7a979d99
JH
1305 /* We do not compute delta to *create* objects we are not
1306 * going to pack.
1307 */
f6c7081a 1308 if (trg_entry->preferred_base)
75c42d8c 1309 return -1;
7a979d99 1310
51d1e83f
LT
1311 /*
1312 * We do not bother to try a delta that we discarded
8dbbd14e 1313 * on an earlier try, but only when reusing delta data.
51d1e83f 1314 */
8dbbd14e 1315 if (!no_reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1316 trg_entry->in_pack == src_entry->in_pack &&
1317 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1318 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1319 return 0;
1320
898b14ce 1321 /* Let's not bust the allowed depth. */
f6c7081a 1322 if (src_entry->depth >= max_depth)
d116a45a 1323 return 0;
c323ac7d 1324
c3b06a69 1325 /* Now some size filtering heuristics. */
560b25a8
NP
1326 trg_size = trg_entry->size;
1327 max_size = trg_size/2 - 20;
c3b06a69
NP
1328 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1329 if (max_size == 0)
1330 return 0;
4e8da195 1331 if (trg_entry->delta && trg_entry->delta_size <= max_size)
f6c7081a
NP
1332 max_size = trg_entry->delta_size-1;
1333 src_size = src_entry->size;
560b25a8 1334 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1335 if (sizediff >= max_size)
f527cb8c 1336 return 0;
f6c7081a 1337
560b25a8
NP
1338 /* Load data if not already done */
1339 if (!trg->data) {
21666f1a 1340 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
560b25a8
NP
1341 if (sz != trg_size)
1342 die("object %s inconsistent object length (%lu vs %lu)",
1343 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1344 }
1345 if (!src->data) {
21666f1a 1346 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
560b25a8
NP
1347 if (sz != src_size)
1348 die("object %s inconsistent object length (%lu vs %lu)",
1349 sha1_to_hex(src_entry->sha1), sz, src_size);
1350 }
1351 if (!src->index) {
1352 src->index = create_delta_index(src->data, src_size);
1353 if (!src->index)
1354 die("out of memory");
1355 }
1356
1357 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1358 if (!delta_buf)
75c42d8c 1359 return 0;
f6c7081a
NP
1360
1361 trg_entry->delta = src_entry;
1362 trg_entry->delta_size = delta_size;
1363 trg_entry->depth = src_entry->depth + 1;
c323ac7d 1364 free(delta_buf);
f6c7081a 1365 return 1;
c323ac7d
LT
1366}
1367
898b14ce 1368static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
b2504a0d 1369{
898b14ce
NP
1370 struct object_entry *child = me->delta_child;
1371 unsigned int m = n;
1372 while (child) {
1373 unsigned int c = check_delta_limit(child, n + 1);
1374 if (m < c)
1375 m = c;
1376 child = child->delta_sibling;
1377 }
1378 return m;
b2504a0d
NP
1379}
1380
d116a45a 1381static void find_deltas(struct object_entry **list, int window, int depth)
c323ac7d 1382{
7cadf491 1383 uint32_t i = nr_objects, idx = 0, processed = 0;
c323ac7d 1384 unsigned int array_size = window * sizeof(struct unpacked);
7cadf491 1385 struct unpacked *array;
898b14ce 1386 int max_depth;
c323ac7d 1387
7cadf491
SP
1388 if (!nr_objects)
1389 return;
1390 array = xmalloc(array_size);
c323ac7d 1391 memset(array, 0, array_size);
13aaf148
NP
1392 if (progress)
1393 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
21fcd1bd 1394
7cadf491
SP
1395 do {
1396 struct object_entry *entry = list[--i];
c323ac7d 1397 struct unpacked *n = array + idx;
c323ac7d
LT
1398 int j;
1399
f0b0af1b
JH
1400 if (!entry->preferred_base)
1401 processed++;
1402
96a02f8f
NP
1403 if (progress)
1404 display_progress(&progress_state, processed);
3f9ac8d2
JH
1405
1406 if (entry->delta)
1407 /* This happens if we decided to reuse existing
ab7cd7bb 1408 * delta from a pack. "!no_reuse_delta &&" is implied.
3f9ac8d2
JH
1409 */
1410 continue;
1411
9a8b6a0a
JH
1412 if (entry->size < 50)
1413 continue;
ff45715c
NP
1414 free_delta_index(n->index);
1415 n->index = NULL;
c323ac7d 1416 free(n->data);
560b25a8 1417 n->data = NULL;
c323ac7d 1418 n->entry = entry;
ab7cd7bb 1419
898b14ce
NP
1420 /*
1421 * If the current object is at pack edge, take the depth the
1422 * objects that depend on the current object into account
1423 * otherwise they would become too deep.
1424 */
1425 max_depth = depth;
1426 if (entry->delta_child) {
1427 max_depth -= check_delta_limit(entry, 0);
1428 if (max_depth <= 0)
1429 goto next;
1430 }
1431
78817c15
LT
1432 j = window;
1433 while (--j > 0) {
7cadf491 1434 uint32_t other_idx = idx + j;
c323ac7d 1435 struct unpacked *m;
78817c15
LT
1436 if (other_idx >= window)
1437 other_idx -= window;
c323ac7d
LT
1438 m = array + other_idx;
1439 if (!m->entry)
1440 break;
898b14ce 1441 if (try_delta(n, m, max_depth) < 0)
c323ac7d
LT
1442 break;
1443 }
898b14ce 1444
70ca1a3f
JH
1445 /* if we made n a delta, and if n is already at max
1446 * depth, leaving it in the window is pointless. we
1447 * should evict it first.
70ca1a3f
JH
1448 */
1449 if (entry->delta && depth <= entry->depth)
1450 continue;
ff45715c 1451
898b14ce 1452 next:
521a4f4c
LT
1453 idx++;
1454 if (idx >= window)
1455 idx = 0;
7cadf491 1456 } while (i > 0);
adee7bdf 1457
b2504a0d 1458 if (progress)
96a02f8f 1459 stop_progress(&progress_state);
b2504a0d 1460
f6c7081a 1461 for (i = 0; i < window; ++i) {
ff45715c 1462 free_delta_index(array[i].index);
adee7bdf 1463 free(array[i].data);
f6c7081a 1464 }
adee7bdf 1465 free(array);
c323ac7d
LT
1466}
1467
f3123c4a
JH
1468static void prepare_pack(int window, int depth)
1469{
9668cf59
NP
1470 struct object_entry **delta_list;
1471 uint32_t i;
1472
3f9ac8d2 1473 get_object_details();
9668cf59
NP
1474
1475 if (!window || !depth)
1476 return;
1477
1478 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1479 for (i = 0; i < nr_objects; i++)
1480 delta_list[i] = objects + i;
1481 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1482 find_deltas(delta_list, window+1, depth);
1483 free(delta_list);
f3123c4a
JH
1484}
1485
4812a93a
JK
1486static int git_pack_config(const char *k, const char *v)
1487{
1488 if(!strcmp(k, "pack.window")) {
1489 window = git_config_int(k, v);
1490 return 0;
1491 }
1492 return git_default_config(k, v);
1493}
1494
b5d97e6b 1495static void read_object_list_from_stdin(void)
c323ac7d 1496{
b5d97e6b
JH
1497 char line[40 + 1 + PATH_MAX + 2];
1498 unsigned char sha1[20];
1499 unsigned hash;
b2504a0d 1500
da93d12b 1501 for (;;) {
da93d12b
LT
1502 if (!fgets(line, sizeof(line), stdin)) {
1503 if (feof(stdin))
1504 break;
1505 if (!ferror(stdin))
1506 die("fgets returned NULL, not EOF, not error!");
687dd75c
JH
1507 if (errno != EINTR)
1508 die("fgets: %s", strerror(errno));
1509 clearerr(stdin);
1510 continue;
da93d12b 1511 }
7a979d99
JH
1512 if (line[0] == '-') {
1513 if (get_sha1_hex(line+1, sha1))
1514 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1515 line);
8d1d8f83 1516 add_preferred_base(sha1);
7a979d99 1517 continue;
21fcd1bd 1518 }
c323ac7d 1519 if (get_sha1_hex(line, sha1))
ef07618f 1520 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1521
ce0bd642 1522 hash = name_hash(line+41);
5379a5c5 1523 add_preferred_base_object(line+41, hash);
5c49c116 1524 add_object_entry(sha1, 0, hash, 0);
c323ac7d 1525 }
b5d97e6b
JH
1526}
1527
b5d97e6b
JH
1528static void show_commit(struct commit *commit)
1529{
5c49c116 1530 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
b5d97e6b
JH
1531}
1532
1533static void show_object(struct object_array_entry *p)
1534{
1535 unsigned hash = name_hash(p->name);
8d1d8f83 1536 add_preferred_base_object(p->name, hash);
5c49c116 1537 add_object_entry(p->item->sha1, p->item->type, hash, 0);
b5d97e6b
JH
1538}
1539
8d1d8f83
JH
1540static void show_edge(struct commit *commit)
1541{
1542 add_preferred_base(commit->object.sha1);
1543}
1544
1545static void get_object_list(int ac, const char **av)
b5d97e6b
JH
1546{
1547 struct rev_info revs;
1548 char line[1000];
b5d97e6b
JH
1549 int flags = 0;
1550
b5d97e6b
JH
1551 init_revisions(&revs, NULL);
1552 save_commit_buffer = 0;
1553 track_object_refs = 0;
1554 setup_revisions(ac, av, &revs, NULL);
1555
b5d97e6b
JH
1556 while (fgets(line, sizeof(line), stdin) != NULL) {
1557 int len = strlen(line);
1558 if (line[len - 1] == '\n')
1559 line[--len] = 0;
1560 if (!len)
1561 break;
1562 if (*line == '-') {
1563 if (!strcmp(line, "--not")) {
1564 flags ^= UNINTERESTING;
1565 continue;
1566 }
1567 die("not a rev '%s'", line);
1568 }
1569 if (handle_revision_arg(line, &revs, flags, 1))
1570 die("bad revision '%s'", line);
1571 }
1572
1573 prepare_revision_walk(&revs);
8d1d8f83 1574 mark_edges_uninteresting(revs.commits, &revs, show_edge);
b5d97e6b
JH
1575 traverse_commit_list(&revs, show_commit, show_object);
1576}
1577
b6b32ccb
JH
1578static int adjust_perm(const char *path, mode_t mode)
1579{
1580 if (chmod(path, mode))
1581 return -1;
1582 return adjust_shared_perm(path);
1583}
1584
b5d97e6b
JH
1585int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1586{
b5d97e6b 1587 int depth = 10;
b5d97e6b 1588 int use_internal_rev_list = 0;
8d1d8f83 1589 int thin = 0;
7cadf491 1590 uint32_t i;
81a216a5
NP
1591 off_t last_obj_offset;
1592 const char *base_name = NULL;
ffa84ffb
RD
1593 const char **rp_av;
1594 int rp_ac_alloc = 64;
8d1d8f83
JH
1595 int rp_ac;
1596
ffa84ffb
RD
1597 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1598
8d1d8f83
JH
1599 rp_av[0] = "pack-objects";
1600 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1601 rp_ac = 2;
b5d97e6b
JH
1602
1603 git_config(git_pack_config);
1604
1605 progress = isatty(2);
1606 for (i = 1; i < argc; i++) {
1607 const char *arg = argv[i];
1608
1609 if (*arg != '-')
1610 break;
1611
1612 if (!strcmp("--non-empty", arg)) {
1613 non_empty = 1;
1614 continue;
1615 }
1616 if (!strcmp("--local", arg)) {
1617 local = 1;
1618 continue;
1619 }
b5d97e6b
JH
1620 if (!strcmp("--incremental", arg)) {
1621 incremental = 1;
1622 continue;
1623 }
599065a3 1624 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
1625 char *end;
1626 window = strtoul(arg+9, &end, 0);
1627 if (!arg[9] || *end)
1628 usage(pack_usage);
1629 continue;
1630 }
599065a3 1631 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
1632 char *end;
1633 depth = strtoul(arg+8, &end, 0);
1634 if (!arg[8] || *end)
1635 usage(pack_usage);
1636 continue;
1637 }
1638 if (!strcmp("--progress", arg)) {
1639 progress = 1;
1640 continue;
1641 }
231f240b
NP
1642 if (!strcmp("--all-progress", arg)) {
1643 progress = 2;
1644 continue;
1645 }
b5d97e6b
JH
1646 if (!strcmp("-q", arg)) {
1647 progress = 0;
1648 continue;
1649 }
1650 if (!strcmp("--no-reuse-delta", arg)) {
1651 no_reuse_delta = 1;
1652 continue;
1653 }
be6b1914 1654 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 1655 allow_ofs_delta = 1;
be6b1914
NP
1656 continue;
1657 }
b5d97e6b
JH
1658 if (!strcmp("--stdout", arg)) {
1659 pack_to_stdout = 1;
1660 continue;
1661 }
1662 if (!strcmp("--revs", arg)) {
1663 use_internal_rev_list = 1;
1664 continue;
1665 }
8d1d8f83 1666 if (!strcmp("--unpacked", arg) ||
599065a3 1667 !prefixcmp(arg, "--unpacked=") ||
63049292 1668 !strcmp("--reflog", arg) ||
8d1d8f83
JH
1669 !strcmp("--all", arg)) {
1670 use_internal_rev_list = 1;
ffa84ffb
RD
1671 if (rp_ac >= rp_ac_alloc - 1) {
1672 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1673 rp_av = xrealloc(rp_av,
1674 rp_ac_alloc * sizeof(*rp_av));
1675 }
8d1d8f83 1676 rp_av[rp_ac++] = arg;
b5d97e6b
JH
1677 continue;
1678 }
8d1d8f83
JH
1679 if (!strcmp("--thin", arg)) {
1680 use_internal_rev_list = 1;
1681 thin = 1;
1682 rp_av[1] = "--objects-edge";
b5d97e6b 1683 continue;
4ba7d711
NP
1684 }
1685 if (!prefixcmp(arg, "--index-version=")) {
1686 char *c;
1687 index_default_version = strtoul(arg + 16, &c, 10);
1688 if (index_default_version > 2)
1689 die("bad %s", arg);
1690 if (*c == ',')
1691 index_off32_limit = strtoul(c+1, &c, 0);
1692 if (*c || index_off32_limit & 0x80000000)
1693 die("bad %s", arg);
1694 continue;
b5d97e6b
JH
1695 }
1696 usage(pack_usage);
1697 }
1698
1699 /* Traditionally "pack-objects [options] base extra" failed;
1700 * we would however want to take refs parameter that would
1701 * have been given to upstream rev-list ourselves, which means
1702 * we somehow want to say what the base name is. So the
1703 * syntax would be:
1704 *
1705 * pack-objects [options] base <refs...>
1706 *
1707 * in other words, we would treat the first non-option as the
1708 * base_name and send everything else to the internal revision
1709 * walker.
1710 */
1711
1712 if (!pack_to_stdout)
1713 base_name = argv[i++];
1714
1715 if (pack_to_stdout != !base_name)
1716 usage(pack_usage);
1717
8d1d8f83
JH
1718 if (!pack_to_stdout && thin)
1719 die("--thin cannot be used to build an indexable pack.");
b5d97e6b
JH
1720
1721 prepare_packed_git();
1722
13aaf148
NP
1723 if (progress)
1724 start_progress(&progress_state, "Generating pack...",
1725 "Counting objects: ", 0);
b5d97e6b
JH
1726 if (!use_internal_rev_list)
1727 read_object_list_from_stdin();
8d1d8f83
JH
1728 else {
1729 rp_av[rp_ac] = NULL;
1730 get_object_list(rp_ac, rp_av);
1731 }
96a02f8f
NP
1732 if (progress) {
1733 stop_progress(&progress_state);
7cadf491 1734 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
96a02f8f
NP
1735 }
1736
f0b0af1b 1737 if (non_empty && !nr_result)
1c4a2912 1738 return 0;
7a979d99 1739 if (progress && (nr_objects != nr_result))
7cadf491 1740 fprintf(stderr, "Result has %u objects.\n", nr_result);
f7ae6a93
NP
1741 if (nr_result)
1742 prepare_pack(window, depth);
f7ae6a93
NP
1743 last_obj_offset = write_pack_file();
1744 if (!pack_to_stdout) {
81a216a5 1745 unsigned char object_list_sha1[20];
b6b32ccb
JH
1746 mode_t mode = umask(0);
1747
1748 umask(mode);
d83c9af5 1749 mode = 0444 & ~mode;
b6b32ccb 1750
81a216a5
NP
1751 write_index_file(last_obj_offset, object_list_sha1);
1752 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
1753 base_name, sha1_to_hex(object_list_sha1));
b6b32ccb 1754 if (adjust_perm(pack_tmp_name, mode))
aef5aedd
JH
1755 die("unable to make temporary pack file readable: %s",
1756 strerror(errno));
81a216a5
NP
1757 if (rename(pack_tmp_name, tmpname))
1758 die("unable to rename temporary pack file: %s",
1759 strerror(errno));
1760 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
1761 base_name, sha1_to_hex(object_list_sha1));
b6b32ccb 1762 if (adjust_perm(idx_tmp_name, mode))
aef5aedd
JH
1763 die("unable to make temporary index file readable: %s",
1764 strerror(errno));
81a216a5
NP
1765 if (rename(idx_tmp_name, tmpname))
1766 die("unable to rename temporary index file: %s",
1767 strerror(errno));
f7ae6a93 1768 puts(sha1_to_hex(object_list_sha1));
5f3de58f 1769 }
ab7cd7bb 1770 if (progress)
7cadf491 1771 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
67c08ce1 1772 written, written_delta, reused, reused_delta);
c323ac7d
LT
1773 return 0;
1774}