Initialize tree descriptors with a helper function rather than by hand.
[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"
c323ac7d 15
231f240b
NP
16static const char pack_usage[] = "\
17git-pack-objects [{ -q | --progress | --all-progress }] \n\
18 [--local] [--incremental] [--window=N] [--depth=N] \n\
19 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
63049292 20 [--revs [--unpacked | --all]*] [--reflog] [--stdout | base-name] \n\
231f240b 21 [<ref-list | <object-list]";
c323ac7d 22
c323ac7d
LT
23struct object_entry {
24 unsigned char sha1[20];
3f9ac8d2 25 unsigned long size; /* uncompressed size */
6777a59f 26 off_t offset; /* offset into the final pack file;
15b4d577
JH
27 * nonzero if already written.
28 */
3f9ac8d2 29 unsigned int depth; /* delta depth */
15b4d577 30 unsigned int delta_limit; /* base adjustment for in-pack delta */
3f9ac8d2 31 unsigned int hash; /* name hint hash */
a733cb60 32 enum object_type type;
ab7cd7bb 33 enum object_type in_pack_type; /* could be delta */
3f9ac8d2 34 unsigned long delta_size; /* delta data size (uncompressed) */
780e6e73 35#define in_pack_header_size delta_size /* only when reusing pack data */
3f9ac8d2
JH
36 struct object_entry *delta; /* delta base object */
37 struct packed_git *in_pack; /* already in pack */
6777a59f 38 off_t in_pack_offset;
82e5a82f 39 struct object_entry *delta_child; /* deltified objects who bases me */
15b4d577
JH
40 struct object_entry *delta_sibling; /* other deltified objects who
41 * uses the same base as me
42 */
7a979d99
JH
43 int preferred_base; /* we do not pack this, but is encouraged to
44 * be used as the base objectto delta huge
45 * objects against.
46 */
c323ac7d
LT
47};
48
3f9ac8d2 49/*
82e5a82f 50 * Objects we are going to pack are collected in objects array (dynamically
3f9ac8d2
JH
51 * expanded). nr_objects & nr_alloc controls this array. They are stored
52 * in the order we see -- typically rev-list --objects order that gives us
53 * nice "minimum seek" order.
54 *
55 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
56 * elements in the objects array. The former is used to build the pack
57 * index (lists object names in the ascending order to help offset lookup),
58 * and the latter is used to group similar things together by try_delta()
59 * heuristics.
60 */
61
5f3de58f 62static unsigned char object_list_sha1[20];
96f1e58f
DR
63static int non_empty;
64static int no_reuse_delta;
65static int local;
66static int incremental;
be6b1914
NP
67static int allow_ofs_delta;
68
c323ac7d 69static struct object_entry **sorted_by_sha, **sorted_by_type;
96f1e58f 70static struct object_entry *objects;
7cadf491 71static uint32_t nr_objects, nr_alloc, nr_result;
c323ac7d 72static const char *base_name;
e1808845 73static unsigned char pack_file_sha1[20];
024701f1 74static int progress = 1;
96f1e58f 75static volatile sig_atomic_t progress_update;
4812a93a 76static int window = 10;
df6d6101 77static int pack_to_stdout;
8d1d8f83 78static int num_preferred_base;
c323ac7d 79
3f9ac8d2
JH
80/*
81 * The object names in objects array are hashed with this hashtable,
82 * to help looking up the entry by object name. Binary search from
83 * sorted_by_sha is also possible but this was easier to code and faster.
84 * This hashtable is built after all the objects are seen.
85 */
96f1e58f
DR
86static int *object_ix;
87static int object_ix_hashsz;
3f9ac8d2
JH
88
89/*
90 * Pack index for existing packs give us easy access to the offsets into
91 * corresponding pack file where each object's data starts, but the entries
92 * do not store the size of the compressed representation (uncompressed
780e6e73
NP
93 * size is easily available by examining the pack entry header). It is
94 * also rather expensive to find the sha1 for an object given its offset.
95 *
96 * We build a hashtable of existing packs (pack_revindex), and keep reverse
97 * index here -- pack index file is sorted by object name mapping to offset;
98 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
99 * ordered by offset, so if you know the offset of an object, next offset
100 * is where its packed representation ends and the index_nr can be used to
101 * get the object sha1 from the main index.
3f9ac8d2 102 */
780e6e73 103struct revindex_entry {
6777a59f 104 off_t offset;
780e6e73
NP
105 unsigned int nr;
106};
3f9ac8d2
JH
107struct pack_revindex {
108 struct packed_git *p;
780e6e73
NP
109 struct revindex_entry *revindex;
110};
111static struct pack_revindex *pack_revindex;
96f1e58f 112static int pack_revindex_hashsz;
3f9ac8d2
JH
113
114/*
115 * stats
116 */
7cadf491
SP
117static uint32_t written, written_delta;
118static uint32_t reused, reused_delta;
3f9ac8d2
JH
119
120static int pack_revindex_ix(struct packed_git *p)
121{
2b74cffa 122 unsigned long ui = (unsigned long)p;
3f9ac8d2
JH
123 int i;
124
125 ui = ui ^ (ui >> 16); /* defeat structure alignment */
126 i = (int)(ui % pack_revindex_hashsz);
127 while (pack_revindex[i].p) {
128 if (pack_revindex[i].p == p)
129 return i;
130 if (++i == pack_revindex_hashsz)
131 i = 0;
132 }
133 return -1 - i;
134}
135
136static void prepare_pack_ix(void)
137{
138 int num;
139 struct packed_git *p;
140 for (num = 0, p = packed_git; p; p = p->next)
141 num++;
142 if (!num)
143 return;
144 pack_revindex_hashsz = num * 11;
145 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
146 for (p = packed_git; p; p = p->next) {
147 num = pack_revindex_ix(p);
148 num = - 1 - num;
149 pack_revindex[num].p = p;
150 }
151 /* revindex elements are lazily initialized */
152}
153
154static int cmp_offset(const void *a_, const void *b_)
155{
780e6e73
NP
156 const struct revindex_entry *a = a_;
157 const struct revindex_entry *b = b_;
158 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
3f9ac8d2
JH
159}
160
161/*
162 * Ordered list of offsets of objects in the pack.
163 */
164static void prepare_pack_revindex(struct pack_revindex *rix)
165{
166 struct packed_git *p = rix->p;
167 int num_ent = num_packed_objects(p);
168 int i;
42873078 169 const char *index = p->index_data;
3f9ac8d2 170
42873078 171 index += 4 * 256;
780e6e73 172 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
3f9ac8d2 173 for (i = 0; i < num_ent; i++) {
42873078 174 uint32_t hl = *((uint32_t *)(index + 24 * i));
780e6e73
NP
175 rix->revindex[i].offset = ntohl(hl);
176 rix->revindex[i].nr = i;
3f9ac8d2
JH
177 }
178 /* This knows the pack format -- the 20-byte trailer
179 * follows immediately after the last object data.
180 */
780e6e73
NP
181 rix->revindex[num_ent].offset = p->pack_size - 20;
182 rix->revindex[num_ent].nr = -1;
183 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
3f9ac8d2
JH
184}
185
780e6e73 186static struct revindex_entry * find_packed_object(struct packed_git *p,
6777a59f 187 off_t ofs)
3f9ac8d2
JH
188{
189 int num;
190 int lo, hi;
191 struct pack_revindex *rix;
780e6e73 192 struct revindex_entry *revindex;
3f9ac8d2
JH
193 num = pack_revindex_ix(p);
194 if (num < 0)
195 die("internal error: pack revindex uninitialized");
196 rix = &pack_revindex[num];
197 if (!rix->revindex)
198 prepare_pack_revindex(rix);
199 revindex = rix->revindex;
200 lo = 0;
201 hi = num_packed_objects(p) + 1;
202 do {
203 int mi = (lo + hi) / 2;
780e6e73
NP
204 if (revindex[mi].offset == ofs) {
205 return revindex + mi;
3f9ac8d2 206 }
780e6e73 207 else if (ofs < revindex[mi].offset)
3f9ac8d2
JH
208 hi = mi;
209 else
210 lo = mi + 1;
211 } while (lo < hi);
212 die("internal error: pack revindex corrupt");
213}
214
6777a59f 215static off_t find_packed_object_size(struct packed_git *p, off_t ofs)
780e6e73
NP
216{
217 struct revindex_entry *entry = find_packed_object(p, ofs);
218 return entry[1].offset - ofs;
219}
220
42873078
NP
221static const unsigned char *find_packed_object_name(struct packed_git *p,
222 off_t ofs)
780e6e73
NP
223{
224 struct revindex_entry *entry = find_packed_object(p, ofs);
42873078 225 return ((unsigned char *)p->index_data) + 4 * 256 + 24 * entry->nr + 4;
780e6e73
NP
226}
227
c323ac7d
LT
228static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
229{
230 unsigned long othersize, delta_size;
21666f1a
NP
231 enum object_type type;
232 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
c323ac7d
LT
233 void *delta_buf;
234
235 if (!otherbuf)
236 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
8ee378a0 237 delta_buf = diff_delta(otherbuf, othersize,
dcde55bc 238 buf, size, &delta_size, 0);
75c42d8c 239 if (!delta_buf || delta_size != entry->delta_size)
c323ac7d
LT
240 die("delta size changed");
241 free(buf);
242 free(otherbuf);
243 return delta_buf;
244}
245
a733cb60
LT
246/*
247 * The per-object header is a pretty dense thing, which is
248 * - first byte: low four bits are "size", then three bits of "type",
249 * and the high bit is "size continues".
250 * - each byte afterwards: low seven bits are size continuation,
251 * with the high bit being "size continues"
252 */
253static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
254{
01247d87 255 int n = 1;
a733cb60
LT
256 unsigned char c;
257
eb32d236 258 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
a733cb60
LT
259 die("bad type %d", type);
260
01247d87
LT
261 c = (type << 4) | (size & 15);
262 size >>= 4;
263 while (size) {
a733cb60 264 *hdr++ = c | 0x80;
01247d87
LT
265 c = size & 0x7f;
266 size >>= 7;
267 n++;
a733cb60
LT
268 }
269 *hdr = c;
270 return n;
271}
272
780e6e73
NP
273/*
274 * we are going to reuse the existing object data as is. make
275 * sure it is not corrupt.
276 */
079afb18
SP
277static int check_pack_inflate(struct packed_git *p,
278 struct pack_window **w_curs,
6777a59f
SP
279 off_t offset,
280 off_t len,
079afb18
SP
281 unsigned long expect)
282{
283 z_stream stream;
284 unsigned char fakebuf[4096], *in;
285 int st;
286
287 memset(&stream, 0, sizeof(stream));
288 inflateInit(&stream);
289 do {
290 in = use_pack(p, w_curs, offset, &stream.avail_in);
291 stream.next_in = in;
292 stream.next_out = fakebuf;
293 stream.avail_out = sizeof(fakebuf);
294 st = inflate(&stream, Z_FINISH);
295 offset += stream.next_in - in;
296 } while (st == Z_OK || st == Z_BUF_ERROR);
297 inflateEnd(&stream);
298 return (st == Z_STREAM_END &&
299 stream.total_out == expect &&
300 stream.total_in == len) ? 0 : -1;
301}
302
303static void copy_pack_data(struct sha1file *f,
304 struct packed_git *p,
305 struct pack_window **w_curs,
6777a59f
SP
306 off_t offset,
307 off_t len)
079afb18
SP
308{
309 unsigned char *in;
310 unsigned int avail;
311
312 while (len) {
313 in = use_pack(p, w_curs, offset, &avail);
314 if (avail > len)
6777a59f 315 avail = (unsigned int)len;
079afb18
SP
316 sha1write(f, in, avail);
317 offset += avail;
318 len -= avail;
319 }
320}
321
322static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
df6d6101 323{
72518e9c
JH
324 z_stream stream;
325 unsigned char fakebuf[4096];
326 int st;
327
328 memset(&stream, 0, sizeof(stream));
329 stream.next_in = data;
330 stream.avail_in = len;
331 stream.next_out = fakebuf;
332 stream.avail_out = sizeof(fakebuf);
333 inflateInit(&stream);
334
335 while (1) {
336 st = inflate(&stream, Z_FINISH);
337 if (st == Z_STREAM_END || st == Z_OK) {
338 st = (stream.total_out == expect &&
339 stream.total_in == len) ? 0 : -1;
340 break;
341 }
342 if (st != Z_BUF_ERROR) {
343 st = -1;
344 break;
345 }
346 stream.next_out = fakebuf;
347 stream.avail_out = sizeof(fakebuf);
348 }
349 inflateEnd(&stream);
350 return st;
df6d6101
JH
351}
352
df6d6101
JH
353static int revalidate_loose_object(struct object_entry *entry,
354 unsigned char *map,
355 unsigned long mapsize)
356{
357 /* we already know this is a loose object with new type header. */
72518e9c
JH
358 enum object_type type;
359 unsigned long size, used;
df6d6101
JH
360
361 if (pack_to_stdout)
362 return 0;
363
72518e9c
JH
364 used = unpack_object_header_gently(map, mapsize, &type, &size);
365 if (!used)
366 return -1;
367 map += used;
368 mapsize -= used;
079afb18 369 return check_loose_inflate(map, mapsize, size);
df6d6101
JH
370}
371
6777a59f 372static off_t write_object(struct sha1file *f,
7a979d99 373 struct object_entry *entry)
c323ac7d
LT
374{
375 unsigned long size;
21666f1a 376 enum object_type type;
3f9ac8d2 377 void *buf;
a733cb60 378 unsigned char header[10];
6777a59f
SP
379 unsigned hdrlen;
380 off_t datalen;
a733cb60 381 enum object_type obj_type;
ab7cd7bb 382 int to_reuse = 0;
c323ac7d 383
a733cb60 384 obj_type = entry->type;
ab7cd7bb
JH
385 if (! entry->in_pack)
386 to_reuse = 0; /* can't reuse what we don't have */
780e6e73 387 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
ab7cd7bb
JH
388 to_reuse = 1; /* check_object() decided it for us */
389 else if (obj_type != entry->in_pack_type)
390 to_reuse = 0; /* pack has delta which is unusable */
391 else if (entry->delta)
392 to_reuse = 0; /* we want to pack afresh */
393 else
394 to_reuse = 1; /* we have it in-pack undeltified,
395 * and we do not need to deltify it.
396 */
397
ceec1361
JH
398 if (!entry->in_pack && !entry->delta) {
399 unsigned char *map;
400 unsigned long mapsize;
401 map = map_sha1_file(entry->sha1, &mapsize);
402 if (map && !legacy_loose_object(map)) {
403 /* We can copy straight into the pack file */
df6d6101
JH
404 if (revalidate_loose_object(entry, map, mapsize))
405 die("corrupt loose object %s",
406 sha1_to_hex(entry->sha1));
ceec1361
JH
407 sha1write(f, map, mapsize);
408 munmap(map, mapsize);
409 written++;
410 reused++;
411 return mapsize;
412 }
413 if (map)
414 munmap(map, mapsize);
415 }
416
df6d6101 417 if (!to_reuse) {
21666f1a 418 buf = read_sha1_file(entry->sha1, &type, &size);
3f9ac8d2
JH
419 if (!buf)
420 die("unable to read %s", sha1_to_hex(entry->sha1));
421 if (size != entry->size)
422 die("object %s size inconsistency (%lu vs %lu)",
423 sha1_to_hex(entry->sha1), size, entry->size);
424 if (entry->delta) {
425 buf = delta_against(buf, size, entry);
426 size = entry->delta_size;
be6b1914
NP
427 obj_type = (allow_ofs_delta && entry->delta->offset) ?
428 OBJ_OFS_DELTA : OBJ_REF_DELTA;
3f9ac8d2
JH
429 }
430 /*
431 * The object header is a byte of 'type' followed by zero or
be6b1914 432 * more bytes of length.
3f9ac8d2
JH
433 */
434 hdrlen = encode_header(obj_type, size, header);
435 sha1write(f, header, hdrlen);
436
be6b1914
NP
437 if (obj_type == OBJ_OFS_DELTA) {
438 /*
439 * Deltas with relative base contain an additional
440 * encoding of the relative offset for the delta
441 * base from this object's position in the pack.
442 */
6777a59f 443 off_t ofs = entry->offset - entry->delta->offset;
be6b1914
NP
444 unsigned pos = sizeof(header) - 1;
445 header[pos] = ofs & 127;
446 while (ofs >>= 7)
447 header[--pos] = 128 | (--ofs & 127);
448 sha1write(f, header + pos, sizeof(header) - pos);
449 hdrlen += sizeof(header) - pos;
450 } else if (obj_type == OBJ_REF_DELTA) {
451 /*
452 * Deltas with a base reference contain
453 * an additional 20 bytes for the base sha1.
454 */
455 sha1write(f, entry->delta->sha1, 20);
3f9ac8d2
JH
456 hdrlen += 20;
457 }
458 datalen = sha1write_compressed(f, buf, size);
459 free(buf);
c323ac7d 460 }
3f9ac8d2
JH
461 else {
462 struct packed_git *p = entry->in_pack;
03e79c88 463 struct pack_window *w_curs = NULL;
6777a59f 464 off_t offset;
3f9ac8d2 465
780e6e73
NP
466 if (entry->delta) {
467 obj_type = (allow_ofs_delta && entry->delta->offset) ?
468 OBJ_OFS_DELTA : OBJ_REF_DELTA;
469 reused_delta++;
470 }
471 hdrlen = encode_header(obj_type, entry->size, header);
472 sha1write(f, header, hdrlen);
473 if (obj_type == OBJ_OFS_DELTA) {
6777a59f 474 off_t ofs = entry->offset - entry->delta->offset;
780e6e73
NP
475 unsigned pos = sizeof(header) - 1;
476 header[pos] = ofs & 127;
477 while (ofs >>= 7)
478 header[--pos] = 128 | (--ofs & 127);
479 sha1write(f, header + pos, sizeof(header) - pos);
480 hdrlen += sizeof(header) - pos;
481 } else if (obj_type == OBJ_REF_DELTA) {
482 sha1write(f, entry->delta->sha1, 20);
483 hdrlen += 20;
484 }
df6d6101 485
079afb18 486 offset = entry->in_pack_offset + entry->in_pack_header_size;
780e6e73
NP
487 datalen = find_packed_object_size(p, entry->in_pack_offset)
488 - entry->in_pack_header_size;
079afb18
SP
489 if (!pack_to_stdout && check_pack_inflate(p, &w_curs,
490 offset, datalen, entry->size))
df6d6101 491 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
079afb18 492 copy_pack_data(f, p, &w_curs, offset, datalen);
03e79c88 493 unuse_pack(&w_curs);
3f9ac8d2 494 reused++;
a733cb60 495 }
be6b1914 496 if (entry->delta)
ab7cd7bb 497 written_delta++;
3f9ac8d2 498 written++;
c323ac7d
LT
499 return hdrlen + datalen;
500}
501
6777a59f 502static off_t write_one(struct sha1file *f,
9d5ab962 503 struct object_entry *e,
6777a59f 504 off_t offset)
9d5ab962 505{
be6b1914 506 if (e->offset || e->preferred_base)
9d5ab962
JH
507 /* offset starts from header size and cannot be zero
508 * if it is written already.
509 */
510 return offset;
be6b1914 511 /* if we are deltified, write out its base object first. */
9d5ab962
JH
512 if (e->delta)
513 offset = write_one(f, e->delta, offset);
be6b1914
NP
514 e->offset = offset;
515 return offset + write_object(f, e);
9d5ab962
JH
516}
517
c323ac7d
LT
518static void write_pack_file(void)
519{
7cadf491 520 uint32_t i;
d22b9290 521 struct sha1file *f;
6777a59f 522 off_t offset;
a733cb60 523 struct pack_header hdr;
183bdb2c 524 unsigned last_percent = 999;
fa438a2e 525 int do_progress = progress;
c323ac7d 526
fa438a2e 527 if (!base_name) {
d22b9290 528 f = sha1fd(1, "<stdout>");
fa438a2e
NP
529 do_progress >>= 1;
530 }
531 else
183bdb2c
JH
532 f = sha1create("%s-%s.%s", base_name,
533 sha1_to_hex(object_list_sha1), "pack");
183bdb2c 534 if (do_progress)
7cadf491 535 fprintf(stderr, "Writing %u objects.\n", nr_result);
183bdb2c 536
a733cb60 537 hdr.hdr_signature = htonl(PACK_SIGNATURE);
01247d87 538 hdr.hdr_version = htonl(PACK_VERSION);
7a979d99 539 hdr.hdr_entries = htonl(nr_result);
a733cb60
LT
540 sha1write(f, &hdr, sizeof(hdr));
541 offset = sizeof(hdr);
f0b0af1b
JH
542 if (!nr_result)
543 goto done;
5e8dc750 544 for (i = 0; i < nr_objects; i++) {
9d5ab962 545 offset = write_one(f, objects + i, offset);
183bdb2c 546 if (do_progress) {
f0b0af1b 547 unsigned percent = written * 100 / nr_result;
183bdb2c
JH
548 if (progress_update || percent != last_percent) {
549 fprintf(stderr, "%4u%% (%u/%u) done\r",
f0b0af1b 550 percent, written, nr_result);
183bdb2c
JH
551 progress_update = 0;
552 last_percent = percent;
553 }
5e8dc750
NP
554 }
555 }
183bdb2c
JH
556 if (do_progress)
557 fputc('\n', stderr);
f0b0af1b 558 done:
67c08ce1 559 if (written != nr_result)
7cadf491 560 die("wrote %u objects while expecting %u", written, nr_result);
e1808845 561 sha1close(f, pack_file_sha1, 1);
c323ac7d
LT
562}
563
564static void write_index_file(void)
565{
7cadf491 566 uint32_t i;
7a979d99
JH
567 struct sha1file *f = sha1create("%s-%s.%s", base_name,
568 sha1_to_hex(object_list_sha1), "idx");
c323ac7d 569 struct object_entry **list = sorted_by_sha;
7a979d99 570 struct object_entry **last = list + nr_result;
b18b00a6 571 uint32_t array[256];
c323ac7d
LT
572
573 /*
574 * Write the first-level table (the list is sorted,
575 * but we use a 256-entry lookup to be able to avoid
e1808845 576 * having to do eight extra binary search iterations).
c323ac7d
LT
577 */
578 for (i = 0; i < 256; i++) {
579 struct object_entry **next = list;
580 while (next < last) {
581 struct object_entry *entry = *next;
582 if (entry->sha1[0] != i)
583 break;
584 next++;
585 }
586 array[i] = htonl(next - sorted_by_sha);
587 list = next;
588 }
b18b00a6 589 sha1write(f, array, 256 * 4);
c323ac7d
LT
590
591 /*
592 * Write the actual SHA1 entries..
593 */
594 list = sorted_by_sha;
7a979d99 595 for (i = 0; i < nr_result; i++) {
c323ac7d 596 struct object_entry *entry = *list++;
b18b00a6 597 uint32_t offset = htonl(entry->offset);
c38138cd
LT
598 sha1write(f, &offset, 4);
599 sha1write(f, entry->sha1, 20);
c323ac7d 600 }
e1808845
LT
601 sha1write(f, pack_file_sha1, 20);
602 sha1close(f, NULL, 1);
c323ac7d
LT
603}
604
7a979d99
JH
605static int locate_object_entry_hash(const unsigned char *sha1)
606{
607 int i;
608 unsigned int ui;
609 memcpy(&ui, sha1, sizeof(unsigned int));
610 i = ui % object_ix_hashsz;
611 while (0 < object_ix[i]) {
a89fccd2 612 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
7a979d99
JH
613 return i;
614 if (++i == object_ix_hashsz)
615 i = 0;
616 }
617 return -1 - i;
618}
619
620static struct object_entry *locate_object_entry(const unsigned char *sha1)
621{
622 int i;
623
624 if (!object_ix_hashsz)
625 return NULL;
626
627 i = locate_object_entry_hash(sha1);
628 if (0 <= i)
629 return &objects[object_ix[i]-1];
630 return NULL;
631}
632
633static void rehash_objects(void)
c323ac7d 634{
7cadf491 635 uint32_t i;
7a979d99
JH
636 struct object_entry *oe;
637
638 object_ix_hashsz = nr_objects * 3;
639 if (object_ix_hashsz < 1024)
640 object_ix_hashsz = 1024;
641 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
5379a5c5 642 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
7a979d99
JH
643 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
644 int ix = locate_object_entry_hash(oe->sha1);
645 if (0 <= ix)
646 continue;
647 ix = -1 - ix;
648 object_ix[ix] = i + 1;
649 }
650}
651
ce0bd642 652static unsigned name_hash(const char *name)
1d6b38cc 653{
ce0bd642
LT
654 unsigned char c;
655 unsigned hash = 0;
656
eeef7135 657 /*
ce0bd642
LT
658 * This effectively just creates a sortable number from the
659 * last sixteen non-whitespace characters. Last characters
660 * count "most", so things that end in ".c" sort together.
eeef7135 661 */
ce0bd642
LT
662 while ((c = *name++) != 0) {
663 if (isspace(c))
664 continue;
665 hash = (hash >> 2) + (c << 24);
666 }
1d6b38cc
JH
667 return hash;
668}
669
670static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
c323ac7d 671{
7cadf491 672 uint32_t idx = nr_objects;
c323ac7d 673 struct object_entry *entry;
3f9ac8d2 674 struct packed_git *p;
6777a59f 675 off_t found_offset = 0;
ab7cd7bb 676 struct packed_git *found_pack = NULL;
b925410d 677 int ix, status = 0;
c323ac7d 678
7a979d99 679 if (!exclude) {
64560374 680 for (p = packed_git; p; p = p->next) {
6777a59f 681 off_t offset = find_pack_entry_one(sha1, p);
43057304 682 if (offset) {
64560374
LT
683 if (incremental)
684 return 0;
685 if (local && !p->pack_local)
686 return 0;
7a979d99 687 if (!found_pack) {
43057304
NP
688 found_offset = offset;
689 found_pack = p;
7a979d99 690 }
64560374
LT
691 }
692 }
693 }
7a979d99
JH
694 if ((entry = locate_object_entry(sha1)) != NULL)
695 goto already_added;
eb019375 696
c323ac7d 697 if (idx >= nr_alloc) {
7cadf491
SP
698 nr_alloc = (idx + 1024) * 3 / 2;
699 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
c323ac7d
LT
700 }
701 entry = objects + idx;
7a979d99 702 nr_objects = idx + 1;
c323ac7d 703 memset(entry, 0, sizeof(*entry));
e702496e 704 hashcpy(entry->sha1, sha1);
27225f2e 705 entry->hash = hash;
7a979d99
JH
706
707 if (object_ix_hashsz * 3 <= nr_objects * 4)
708 rehash_objects();
709 else {
710 ix = locate_object_entry_hash(entry->sha1);
711 if (0 <= ix)
712 die("internal error in object hashing.");
713 object_ix[-1 - ix] = idx + 1;
3f9ac8d2 714 }
b925410d 715 status = 1;
7a979d99
JH
716
717 already_added:
f0b0af1b 718 if (progress_update) {
7cadf491 719 fprintf(stderr, "Counting objects...%u\r", nr_objects);
f0b0af1b
JH
720 progress_update = 0;
721 }
7a979d99
JH
722 if (exclude)
723 entry->preferred_base = 1;
724 else {
725 if (found_pack) {
726 entry->in_pack = found_pack;
727 entry->in_pack_offset = found_offset;
728 }
a49dd05f 729 }
b925410d 730 return status;
c323ac7d
LT
731}
732
5379a5c5
JH
733struct pbase_tree_cache {
734 unsigned char sha1[20];
735 int ref;
736 int temporary;
737 void *tree_data;
738 unsigned long tree_size;
739};
740
741static struct pbase_tree_cache *(pbase_tree_cache[256]);
742static int pbase_tree_cache_ix(const unsigned char *sha1)
743{
744 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
745}
746static int pbase_tree_cache_ix_incr(int ix)
747{
748 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
749}
750
751static struct pbase_tree {
752 struct pbase_tree *next;
753 /* This is a phony "cache" entry; we are not
754 * going to evict it nor find it through _get()
755 * mechanism -- this is for the toplevel node that
756 * would almost always change with any commit.
757 */
758 struct pbase_tree_cache pcache;
759} *pbase_tree;
760
761static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
762{
763 struct pbase_tree_cache *ent, *nent;
764 void *data;
765 unsigned long size;
21666f1a 766 enum object_type type;
5379a5c5
JH
767 int neigh;
768 int my_ix = pbase_tree_cache_ix(sha1);
769 int available_ix = -1;
770
771 /* pbase-tree-cache acts as a limited hashtable.
772 * your object will be found at your index or within a few
773 * slots after that slot if it is cached.
774 */
775 for (neigh = 0; neigh < 8; neigh++) {
776 ent = pbase_tree_cache[my_ix];
a89fccd2 777 if (ent && !hashcmp(ent->sha1, sha1)) {
5379a5c5
JH
778 ent->ref++;
779 return ent;
780 }
781 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
782 ((0 <= available_ix) &&
783 (!ent && pbase_tree_cache[available_ix])))
784 available_ix = my_ix;
785 if (!ent)
786 break;
787 my_ix = pbase_tree_cache_ix_incr(my_ix);
788 }
789
790 /* Did not find one. Either we got a bogus request or
791 * we need to read and perhaps cache.
792 */
21666f1a 793 data = read_sha1_file(sha1, &type, &size);
5379a5c5
JH
794 if (!data)
795 return NULL;
21666f1a 796 if (type != OBJ_TREE) {
5379a5c5
JH
797 free(data);
798 return NULL;
799 }
800
801 /* We need to either cache or return a throwaway copy */
802
803 if (available_ix < 0)
804 ent = NULL;
805 else {
806 ent = pbase_tree_cache[available_ix];
807 my_ix = available_ix;
808 }
809
810 if (!ent) {
811 nent = xmalloc(sizeof(*nent));
812 nent->temporary = (available_ix < 0);
813 }
814 else {
815 /* evict and reuse */
816 free(ent->tree_data);
817 nent = ent;
818 }
e702496e 819 hashcpy(nent->sha1, sha1);
5379a5c5
JH
820 nent->tree_data = data;
821 nent->tree_size = size;
822 nent->ref = 1;
823 if (!nent->temporary)
824 pbase_tree_cache[my_ix] = nent;
825 return nent;
826}
827
828static void pbase_tree_put(struct pbase_tree_cache *cache)
829{
830 if (!cache->temporary) {
831 cache->ref--;
832 return;
833 }
834 free(cache->tree_data);
835 free(cache);
836}
837
838static int name_cmp_len(const char *name)
839{
840 int i;
841 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
842 ;
843 return i;
844}
845
846static void add_pbase_object(struct tree_desc *tree,
5379a5c5 847 const char *name,
ce0bd642
LT
848 int cmplen,
849 const char *fullname)
3f9ac8d2 850{
4c068a98
LT
851 struct name_entry entry;
852
853 while (tree_entry(tree,&entry)) {
7a979d99 854 unsigned long size;
21666f1a 855 enum object_type type;
7a979d99 856
a8c40471 857 if (tree_entry_len(entry.path, entry.sha1) != cmplen ||
4c068a98
LT
858 memcmp(entry.path, name, cmplen) ||
859 !has_sha1_file(entry.sha1) ||
21666f1a 860 (type = sha1_object_info(entry.sha1, &size)) < 0)
7a979d99 861 continue;
5379a5c5 862 if (name[cmplen] != '/') {
ce0bd642 863 unsigned hash = name_hash(fullname);
4c068a98 864 add_object_entry(entry.sha1, hash, 1);
5379a5c5
JH
865 return;
866 }
21666f1a 867 if (type == OBJ_TREE) {
7a979d99 868 struct tree_desc sub;
5379a5c5
JH
869 struct pbase_tree_cache *tree;
870 const char *down = name+cmplen+1;
871 int downlen = name_cmp_len(down);
872
4c068a98 873 tree = pbase_tree_get(entry.sha1);
5379a5c5
JH
874 if (!tree)
875 return;
6fda5e51 876 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
5379a5c5 877
ce0bd642 878 add_pbase_object(&sub, down, downlen, fullname);
5379a5c5
JH
879 pbase_tree_put(tree);
880 }
881 }
882}
1d6b38cc 883
5379a5c5
JH
884static unsigned *done_pbase_paths;
885static int done_pbase_paths_num;
886static int done_pbase_paths_alloc;
887static int done_pbase_path_pos(unsigned hash)
888{
889 int lo = 0;
890 int hi = done_pbase_paths_num;
891 while (lo < hi) {
892 int mi = (hi + lo) / 2;
893 if (done_pbase_paths[mi] == hash)
894 return mi;
895 if (done_pbase_paths[mi] < hash)
896 hi = mi;
897 else
898 lo = mi + 1;
899 }
900 return -lo-1;
901}
902
903static int check_pbase_path(unsigned hash)
904{
905 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
906 if (0 <= pos)
907 return 1;
908 pos = -pos - 1;
909 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
910 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
911 done_pbase_paths = xrealloc(done_pbase_paths,
912 done_pbase_paths_alloc *
913 sizeof(unsigned));
914 }
915 done_pbase_paths_num++;
916 if (pos < done_pbase_paths_num)
917 memmove(done_pbase_paths + pos + 1,
918 done_pbase_paths + pos,
919 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
920 done_pbase_paths[pos] = hash;
921 return 0;
922}
923
8d1d8f83 924static void add_preferred_base_object(const char *name, unsigned hash)
5379a5c5
JH
925{
926 struct pbase_tree *it;
927 int cmplen = name_cmp_len(name);
928
929 if (check_pbase_path(hash))
930 return;
931
932 for (it = pbase_tree; it; it = it->next) {
933 if (cmplen == 0) {
ce0bd642 934 hash = name_hash("");
5379a5c5
JH
935 add_object_entry(it->pcache.sha1, hash, 1);
936 }
937 else {
938 struct tree_desc tree;
6fda5e51 939 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
ce0bd642 940 add_pbase_object(&tree, name, cmplen, name);
7a979d99 941 }
3f9ac8d2 942 }
3f9ac8d2
JH
943}
944
7a979d99 945static void add_preferred_base(unsigned char *sha1)
3f9ac8d2 946{
5379a5c5
JH
947 struct pbase_tree *it;
948 void *data;
949 unsigned long size;
950 unsigned char tree_sha1[20];
1d6b38cc 951
8d1d8f83
JH
952 if (window <= num_preferred_base++)
953 return;
954
5379a5c5
JH
955 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
956 if (!data)
7a979d99 957 return;
5379a5c5
JH
958
959 for (it = pbase_tree; it; it = it->next) {
a89fccd2 960 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
5379a5c5
JH
961 free(data);
962 return;
963 }
964 }
965
966 it = xcalloc(1, sizeof(*it));
967 it->next = pbase_tree;
968 pbase_tree = it;
969
e702496e 970 hashcpy(it->pcache.sha1, tree_sha1);
5379a5c5
JH
971 it->pcache.tree_data = data;
972 it->pcache.tree_size = size;
3f9ac8d2
JH
973}
974
c323ac7d
LT
975static void check_object(struct object_entry *entry)
976{
7a979d99 977 if (entry->in_pack && !entry->preferred_base) {
780e6e73 978 struct packed_git *p = entry->in_pack;
03e79c88 979 struct pack_window *w_curs = NULL;
780e6e73 980 unsigned long size, used;
6777a59f 981 unsigned int avail;
780e6e73
NP
982 unsigned char *buf;
983 struct object_entry *base_entry = NULL;
984
6777a59f 985 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
ab7cd7bb
JH
986
987 /* We want in_pack_type even if we do not reuse delta.
988 * There is no point not reusing non-delta representations.
989 */
6777a59f 990 used = unpack_object_header_gently(buf, avail,
780e6e73 991 &entry->in_pack_type, &size);
ab7cd7bb 992
3f9ac8d2
JH
993 /* Check if it is delta, and the base is also an object
994 * we are going to pack. If so we will reuse the existing
995 * delta.
996 */
780e6e73 997 if (!no_reuse_delta) {
42873078
NP
998 unsigned char c;
999 const unsigned char *base_name;
6777a59f 1000 off_t ofs;
9c18df19 1001 unsigned long used_0;
780e6e73
NP
1002 /* there is at least 20 bytes left in the pack */
1003 switch (entry->in_pack_type) {
1004 case OBJ_REF_DELTA:
f5b1b5a0
SP
1005 base_name = use_pack(p, &w_curs,
1006 entry->in_pack_offset + used, NULL);
780e6e73
NP
1007 used += 20;
1008 break;
1009 case OBJ_OFS_DELTA:
f5b1b5a0
SP
1010 buf = use_pack(p, &w_curs,
1011 entry->in_pack_offset + used, NULL);
9c18df19
JH
1012 used_0 = 0;
1013 c = buf[used_0++];
780e6e73
NP
1014 ofs = c & 127;
1015 while (c & 128) {
1016 ofs += 1;
1017 if (!ofs || ofs & ~(~0UL >> 7))
1018 die("delta base offset overflow in pack for %s",
1019 sha1_to_hex(entry->sha1));
9c18df19 1020 c = buf[used_0++];
780e6e73
NP
1021 ofs = (ofs << 7) + (c & 127);
1022 }
1023 if (ofs >= entry->in_pack_offset)
1024 die("delta base offset out of bound for %s",
1025 sha1_to_hex(entry->sha1));
1026 ofs = entry->in_pack_offset - ofs;
1027 base_name = find_packed_object_name(p, ofs);
9c18df19 1028 used += used_0;
780e6e73
NP
1029 break;
1030 default:
1031 base_name = NULL;
1032 }
1033 if (base_name)
1034 base_entry = locate_object_entry(base_name);
1035 }
03e79c88 1036 unuse_pack(&w_curs);
780e6e73
NP
1037 entry->in_pack_header_size = used;
1038
a2700696 1039 if (base_entry) {
ab7cd7bb
JH
1040
1041 /* Depth value does not matter - find_deltas()
1042 * will never consider reused delta as the
1043 * base object to deltify other objects
1044 * against, in order to avoid circular deltas.
3f9ac8d2 1045 */
ab7cd7bb
JH
1046
1047 /* uncompressed size of the delta data */
780e6e73 1048 entry->size = size;
3f9ac8d2 1049 entry->delta = base_entry;
780e6e73 1050 entry->type = entry->in_pack_type;
ab7cd7bb 1051
15b4d577
JH
1052 entry->delta_sibling = base_entry->delta_child;
1053 base_entry->delta_child = entry;
ab7cd7bb 1054
3f9ac8d2
JH
1055 return;
1056 }
1057 /* Otherwise we would do the usual */
36e4d74a 1058 }
3f9ac8d2 1059
21666f1a
NP
1060 entry->type = sha1_object_info(entry->sha1, &entry->size);
1061 if (entry->type < 0)
36e4d74a
JH
1062 die("unable to get type of object %s",
1063 sha1_to_hex(entry->sha1));
3f9ac8d2
JH
1064}
1065
15b4d577
JH
1066static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1067{
1068 struct object_entry *child = me->delta_child;
1069 unsigned int m = n;
1070 while (child) {
1071 unsigned int c = check_delta_limit(child, n + 1);
1072 if (m < c)
1073 m = c;
1074 child = child->delta_sibling;
1075 }
1076 return m;
1077}
1078
c323ac7d
LT
1079static void get_object_details(void)
1080{
7cadf491 1081 uint32_t i;
15b4d577 1082 struct object_entry *entry;
c323ac7d 1083
3f9ac8d2 1084 prepare_pack_ix();
15b4d577
JH
1085 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1086 check_object(entry);
b76f6b62
JH
1087
1088 if (nr_objects == nr_result) {
1089 /*
1090 * Depth of objects that depend on the entry -- this
1091 * is subtracted from depth-max to break too deep
1092 * delta chain because of delta data reusing.
1093 * However, we loosen this restriction when we know we
1094 * are creating a thin pack -- it will have to be
1095 * expanded on the other end anyway, so do not
1096 * artificially cut the delta chain and let it go as
1097 * deep as it wants.
1098 */
1099 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1100 if (!entry->delta && entry->delta_child)
1101 entry->delta_limit =
1102 check_delta_limit(entry, 1);
1103 }
c323ac7d
LT
1104}
1105
1106typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1107
1108static entry_sort_t current_sort;
1109
1110static int sort_comparator(const void *_a, const void *_b)
1111{
1112 struct object_entry *a = *(struct object_entry **)_a;
1113 struct object_entry *b = *(struct object_entry **)_b;
1114 return current_sort(a,b);
1115}
1116
1117static struct object_entry **create_sorted_list(entry_sort_t sort)
1118{
1119 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
7cadf491 1120 uint32_t i;
c323ac7d
LT
1121
1122 for (i = 0; i < nr_objects; i++)
1123 list[i] = objects + i;
1124 current_sort = sort;
1125 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1126 return list;
1127}
1128
1129static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1130{
a89fccd2 1131 return hashcmp(a->sha1, b->sha1);
c323ac7d
LT
1132}
1133
962554c6 1134static struct object_entry **create_final_object_list(void)
7a979d99
JH
1135{
1136 struct object_entry **list;
7cadf491 1137 uint32_t i, j;
7a979d99
JH
1138
1139 for (i = nr_result = 0; i < nr_objects; i++)
1140 if (!objects[i].preferred_base)
1141 nr_result++;
1142 list = xmalloc(nr_result * sizeof(struct object_entry *));
1143 for (i = j = 0; i < nr_objects; i++) {
1144 if (!objects[i].preferred_base)
1145 list[j++] = objects + i;
1146 }
1147 current_sort = sha1_sort;
1148 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1149 return list;
1150}
1151
c323ac7d
LT
1152static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1153{
1154 if (a->type < b->type)
1155 return -1;
1156 if (a->type > b->type)
1157 return 1;
27225f2e
LT
1158 if (a->hash < b->hash)
1159 return -1;
1160 if (a->hash > b->hash)
1161 return 1;
7a979d99
JH
1162 if (a->preferred_base < b->preferred_base)
1163 return -1;
1164 if (a->preferred_base > b->preferred_base)
1165 return 1;
c323ac7d
LT
1166 if (a->size < b->size)
1167 return -1;
1168 if (a->size > b->size)
1169 return 1;
1170 return a < b ? -1 : (a > b);
1171}
1172
1173struct unpacked {
1174 struct object_entry *entry;
1175 void *data;
f6c7081a 1176 struct delta_index *index;
c323ac7d
LT
1177};
1178
1179/*
521a4f4c
LT
1180 * We search for deltas _backwards_ in a list sorted by type and
1181 * by size, so that we see progressively smaller and smaller files.
1182 * That's because we prefer deltas to be from the bigger file
1183 * to the smaller - deletes are potentially cheaper, but perhaps
1184 * more importantly, the bigger file is likely the more recent
1185 * one.
c323ac7d 1186 */
f6c7081a 1187static int try_delta(struct unpacked *trg, struct unpacked *src,
560b25a8 1188 unsigned max_depth)
c323ac7d 1189{
f6c7081a
NP
1190 struct object_entry *trg_entry = trg->entry;
1191 struct object_entry *src_entry = src->entry;
560b25a8 1192 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
21666f1a 1193 enum object_type type;
c323ac7d
LT
1194 void *delta_buf;
1195
1196 /* Don't bother doing diffs between different types */
f6c7081a 1197 if (trg_entry->type != src_entry->type)
c323ac7d
LT
1198 return -1;
1199
7a979d99
JH
1200 /* We do not compute delta to *create* objects we are not
1201 * going to pack.
1202 */
f6c7081a 1203 if (trg_entry->preferred_base)
75c42d8c 1204 return -1;
7a979d99 1205
51d1e83f
LT
1206 /*
1207 * We do not bother to try a delta that we discarded
8dbbd14e 1208 * on an earlier try, but only when reusing delta data.
51d1e83f 1209 */
8dbbd14e 1210 if (!no_reuse_delta && trg_entry->in_pack &&
e9195b58
JH
1211 trg_entry->in_pack == src_entry->in_pack &&
1212 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1213 trg_entry->in_pack_type != OBJ_OFS_DELTA)
51d1e83f
LT
1214 return 0;
1215
f6c7081a
NP
1216 /*
1217 * If the current object is at pack edge, take the depth the
7a979d99
JH
1218 * objects that depend on the current object into account --
1219 * otherwise they would become too deep.
ab7cd7bb 1220 */
f6c7081a
NP
1221 if (trg_entry->delta_child) {
1222 if (max_depth <= trg_entry->delta_limit)
15b4d577 1223 return 0;
f6c7081a 1224 max_depth -= trg_entry->delta_limit;
15b4d577 1225 }
f6c7081a 1226 if (src_entry->depth >= max_depth)
d116a45a 1227 return 0;
c323ac7d 1228
c3b06a69 1229 /* Now some size filtering heuristics. */
560b25a8
NP
1230 trg_size = trg_entry->size;
1231 max_size = trg_size/2 - 20;
c3b06a69
NP
1232 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1233 if (max_size == 0)
1234 return 0;
4e8da195 1235 if (trg_entry->delta && trg_entry->delta_size <= max_size)
f6c7081a
NP
1236 max_size = trg_entry->delta_size-1;
1237 src_size = src_entry->size;
560b25a8 1238 sizediff = src_size < trg_size ? trg_size - src_size : 0;
27225f2e 1239 if (sizediff >= max_size)
f527cb8c 1240 return 0;
f6c7081a 1241
560b25a8
NP
1242 /* Load data if not already done */
1243 if (!trg->data) {
21666f1a 1244 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
560b25a8
NP
1245 if (sz != trg_size)
1246 die("object %s inconsistent object length (%lu vs %lu)",
1247 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1248 }
1249 if (!src->data) {
21666f1a 1250 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
560b25a8
NP
1251 if (sz != src_size)
1252 die("object %s inconsistent object length (%lu vs %lu)",
1253 sha1_to_hex(src_entry->sha1), sz, src_size);
1254 }
1255 if (!src->index) {
1256 src->index = create_delta_index(src->data, src_size);
1257 if (!src->index)
1258 die("out of memory");
1259 }
1260
1261 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
c323ac7d 1262 if (!delta_buf)
75c42d8c 1263 return 0;
f6c7081a
NP
1264
1265 trg_entry->delta = src_entry;
1266 trg_entry->delta_size = delta_size;
1267 trg_entry->depth = src_entry->depth + 1;
c323ac7d 1268 free(delta_buf);
f6c7081a 1269 return 1;
c323ac7d
LT
1270}
1271
b2504a0d
NP
1272static void progress_interval(int signum)
1273{
b2504a0d
NP
1274 progress_update = 1;
1275}
1276
d116a45a 1277static void find_deltas(struct object_entry **list, int window, int depth)
c323ac7d 1278{
7cadf491 1279 uint32_t i = nr_objects, idx = 0, processed = 0;
c323ac7d 1280 unsigned int array_size = window * sizeof(struct unpacked);
7cadf491 1281 struct unpacked *array;
183bdb2c 1282 unsigned last_percent = 999;
c323ac7d 1283
7cadf491
SP
1284 if (!nr_objects)
1285 return;
1286 array = xmalloc(array_size);
c323ac7d 1287 memset(array, 0, array_size);
183bdb2c 1288 if (progress)
7cadf491 1289 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
21fcd1bd 1290
7cadf491
SP
1291 do {
1292 struct object_entry *entry = list[--i];
c323ac7d 1293 struct unpacked *n = array + idx;
c323ac7d
LT
1294 int j;
1295
f0b0af1b
JH
1296 if (!entry->preferred_base)
1297 processed++;
1298
183bdb2c 1299 if (progress) {
f0b0af1b 1300 unsigned percent = processed * 100 / nr_result;
183bdb2c
JH
1301 if (percent != last_percent || progress_update) {
1302 fprintf(stderr, "%4u%% (%u/%u) done\r",
f0b0af1b 1303 percent, processed, nr_result);
183bdb2c
JH
1304 progress_update = 0;
1305 last_percent = percent;
1306 }
21fcd1bd 1307 }
3f9ac8d2
JH
1308
1309 if (entry->delta)
1310 /* This happens if we decided to reuse existing
ab7cd7bb 1311 * delta from a pack. "!no_reuse_delta &&" is implied.
3f9ac8d2
JH
1312 */
1313 continue;
1314
9a8b6a0a
JH
1315 if (entry->size < 50)
1316 continue;
ff45715c
NP
1317 free_delta_index(n->index);
1318 n->index = NULL;
c323ac7d 1319 free(n->data);
560b25a8 1320 n->data = NULL;
c323ac7d 1321 n->entry = entry;
ab7cd7bb 1322
78817c15
LT
1323 j = window;
1324 while (--j > 0) {
7cadf491 1325 uint32_t other_idx = idx + j;
c323ac7d 1326 struct unpacked *m;
78817c15
LT
1327 if (other_idx >= window)
1328 other_idx -= window;
c323ac7d
LT
1329 m = array + other_idx;
1330 if (!m->entry)
1331 break;
560b25a8 1332 if (try_delta(n, m, depth) < 0)
c323ac7d
LT
1333 break;
1334 }
70ca1a3f
JH
1335 /* if we made n a delta, and if n is already at max
1336 * depth, leaving it in the window is pointless. we
1337 * should evict it first.
70ca1a3f
JH
1338 */
1339 if (entry->delta && depth <= entry->depth)
1340 continue;
ff45715c 1341
521a4f4c
LT
1342 idx++;
1343 if (idx >= window)
1344 idx = 0;
7cadf491 1345 } while (i > 0);
adee7bdf 1346
b2504a0d
NP
1347 if (progress)
1348 fputc('\n', stderr);
1349
f6c7081a 1350 for (i = 0; i < window; ++i) {
ff45715c 1351 free_delta_index(array[i].index);
adee7bdf 1352 free(array[i].data);
f6c7081a 1353 }
adee7bdf 1354 free(array);
c323ac7d
LT
1355}
1356
f3123c4a
JH
1357static void prepare_pack(int window, int depth)
1358{
3f9ac8d2 1359 get_object_details();
f3123c4a
JH
1360 sorted_by_type = create_sorted_list(type_size_sort);
1361 if (window && depth)
1362 find_deltas(sorted_by_type, window+1, depth);
f3123c4a
JH
1363}
1364
df6d6101 1365static int reuse_cached_pack(unsigned char *sha1)
f3123c4a
JH
1366{
1367 static const char cache[] = "pack-cache/pack-%s.%s";
1368 char *cached_pack, *cached_idx;
1369 int ifd, ofd, ifd_ix = -1;
1370
1371 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1372 ifd = open(cached_pack, O_RDONLY);
1373 if (ifd < 0)
1374 return 0;
1375
1376 if (!pack_to_stdout) {
1377 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1378 ifd_ix = open(cached_idx, O_RDONLY);
1379 if (ifd_ix < 0) {
1380 close(ifd);
1381 return 0;
1382 }
1383 }
1384
ab7cd7bb 1385 if (progress)
7cadf491 1386 fprintf(stderr, "Reusing %u objects pack %s\n", nr_objects,
ab7cd7bb 1387 sha1_to_hex(sha1));
f3123c4a
JH
1388
1389 if (pack_to_stdout) {
1390 if (copy_fd(ifd, 1))
1391 exit(1);
1392 close(ifd);
1393 }
1394 else {
1395 char name[PATH_MAX];
1396 snprintf(name, sizeof(name),
1397 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1398 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1399 if (ofd < 0)
1400 die("unable to open %s (%s)", name, strerror(errno));
1401 if (copy_fd(ifd, ofd))
1402 exit(1);
1403 close(ifd);
1404
1405 snprintf(name, sizeof(name),
1406 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1407 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1408 if (ofd < 0)
1409 die("unable to open %s (%s)", name, strerror(errno));
1410 if (copy_fd(ifd_ix, ofd))
1411 exit(1);
1412 close(ifd_ix);
1413 puts(sha1_to_hex(sha1));
1414 }
1415
1416 return 1;
1417}
1418
fb7a6531
LT
1419static void setup_progress_signal(void)
1420{
1421 struct sigaction sa;
1422 struct itimerval v;
1423
1424 memset(&sa, 0, sizeof(sa));
1425 sa.sa_handler = progress_interval;
1426 sigemptyset(&sa.sa_mask);
1427 sa.sa_flags = SA_RESTART;
1428 sigaction(SIGALRM, &sa, NULL);
1429
1430 v.it_interval.tv_sec = 1;
1431 v.it_interval.tv_usec = 0;
1432 v.it_value = v.it_interval;
1433 setitimer(ITIMER_REAL, &v, NULL);
1434}
1435
4812a93a
JK
1436static int git_pack_config(const char *k, const char *v)
1437{
1438 if(!strcmp(k, "pack.window")) {
1439 window = git_config_int(k, v);
1440 return 0;
1441 }
1442 return git_default_config(k, v);
1443}
1444
b5d97e6b 1445static void read_object_list_from_stdin(void)
c323ac7d 1446{
b5d97e6b
JH
1447 char line[40 + 1 + PATH_MAX + 2];
1448 unsigned char sha1[20];
1449 unsigned hash;
b2504a0d 1450
da93d12b 1451 for (;;) {
da93d12b
LT
1452 if (!fgets(line, sizeof(line), stdin)) {
1453 if (feof(stdin))
1454 break;
1455 if (!ferror(stdin))
1456 die("fgets returned NULL, not EOF, not error!");
687dd75c
JH
1457 if (errno != EINTR)
1458 die("fgets: %s", strerror(errno));
1459 clearerr(stdin);
1460 continue;
da93d12b 1461 }
7a979d99
JH
1462 if (line[0] == '-') {
1463 if (get_sha1_hex(line+1, sha1))
1464 die("expected edge sha1, got garbage:\n %s",
b5d97e6b 1465 line);
8d1d8f83 1466 add_preferred_base(sha1);
7a979d99 1467 continue;
21fcd1bd 1468 }
c323ac7d 1469 if (get_sha1_hex(line, sha1))
ef07618f 1470 die("expected sha1, got garbage:\n %s", line);
b5d97e6b 1471
ce0bd642 1472 hash = name_hash(line+41);
5379a5c5
JH
1473 add_preferred_base_object(line+41, hash);
1474 add_object_entry(sha1, hash, 0);
c323ac7d 1475 }
b5d97e6b
JH
1476}
1477
b5d97e6b
JH
1478static void show_commit(struct commit *commit)
1479{
1480 unsigned hash = name_hash("");
8d1d8f83 1481 add_preferred_base_object("", hash);
b5d97e6b
JH
1482 add_object_entry(commit->object.sha1, hash, 0);
1483}
1484
1485static void show_object(struct object_array_entry *p)
1486{
1487 unsigned hash = name_hash(p->name);
8d1d8f83 1488 add_preferred_base_object(p->name, hash);
b5d97e6b
JH
1489 add_object_entry(p->item->sha1, hash, 0);
1490}
1491
8d1d8f83
JH
1492static void show_edge(struct commit *commit)
1493{
1494 add_preferred_base(commit->object.sha1);
1495}
1496
1497static void get_object_list(int ac, const char **av)
b5d97e6b
JH
1498{
1499 struct rev_info revs;
1500 char line[1000];
b5d97e6b
JH
1501 int flags = 0;
1502
b5d97e6b
JH
1503 init_revisions(&revs, NULL);
1504 save_commit_buffer = 0;
1505 track_object_refs = 0;
1506 setup_revisions(ac, av, &revs, NULL);
1507
b5d97e6b
JH
1508 while (fgets(line, sizeof(line), stdin) != NULL) {
1509 int len = strlen(line);
1510 if (line[len - 1] == '\n')
1511 line[--len] = 0;
1512 if (!len)
1513 break;
1514 if (*line == '-') {
1515 if (!strcmp(line, "--not")) {
1516 flags ^= UNINTERESTING;
1517 continue;
1518 }
1519 die("not a rev '%s'", line);
1520 }
1521 if (handle_revision_arg(line, &revs, flags, 1))
1522 die("bad revision '%s'", line);
1523 }
1524
1525 prepare_revision_walk(&revs);
8d1d8f83 1526 mark_edges_uninteresting(revs.commits, &revs, show_edge);
b5d97e6b
JH
1527 traverse_commit_list(&revs, show_commit, show_object);
1528}
1529
1530int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1531{
1532 SHA_CTX ctx;
1533 int depth = 10;
1534 struct object_entry **list;
1535 int use_internal_rev_list = 0;
8d1d8f83 1536 int thin = 0;
7cadf491 1537 uint32_t i;
ffa84ffb
RD
1538 const char **rp_av;
1539 int rp_ac_alloc = 64;
8d1d8f83
JH
1540 int rp_ac;
1541
ffa84ffb
RD
1542 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1543
8d1d8f83
JH
1544 rp_av[0] = "pack-objects";
1545 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1546 rp_ac = 2;
b5d97e6b
JH
1547
1548 git_config(git_pack_config);
1549
1550 progress = isatty(2);
1551 for (i = 1; i < argc; i++) {
1552 const char *arg = argv[i];
1553
1554 if (*arg != '-')
1555 break;
1556
1557 if (!strcmp("--non-empty", arg)) {
1558 non_empty = 1;
1559 continue;
1560 }
1561 if (!strcmp("--local", arg)) {
1562 local = 1;
1563 continue;
1564 }
b5d97e6b
JH
1565 if (!strcmp("--incremental", arg)) {
1566 incremental = 1;
1567 continue;
1568 }
599065a3 1569 if (!prefixcmp(arg, "--window=")) {
b5d97e6b
JH
1570 char *end;
1571 window = strtoul(arg+9, &end, 0);
1572 if (!arg[9] || *end)
1573 usage(pack_usage);
1574 continue;
1575 }
599065a3 1576 if (!prefixcmp(arg, "--depth=")) {
b5d97e6b
JH
1577 char *end;
1578 depth = strtoul(arg+8, &end, 0);
1579 if (!arg[8] || *end)
1580 usage(pack_usage);
1581 continue;
1582 }
1583 if (!strcmp("--progress", arg)) {
1584 progress = 1;
1585 continue;
1586 }
231f240b
NP
1587 if (!strcmp("--all-progress", arg)) {
1588 progress = 2;
1589 continue;
1590 }
b5d97e6b
JH
1591 if (!strcmp("-q", arg)) {
1592 progress = 0;
1593 continue;
1594 }
1595 if (!strcmp("--no-reuse-delta", arg)) {
1596 no_reuse_delta = 1;
1597 continue;
1598 }
be6b1914 1599 if (!strcmp("--delta-base-offset", arg)) {
780e6e73 1600 allow_ofs_delta = 1;
be6b1914
NP
1601 continue;
1602 }
b5d97e6b
JH
1603 if (!strcmp("--stdout", arg)) {
1604 pack_to_stdout = 1;
1605 continue;
1606 }
1607 if (!strcmp("--revs", arg)) {
1608 use_internal_rev_list = 1;
1609 continue;
1610 }
8d1d8f83 1611 if (!strcmp("--unpacked", arg) ||
599065a3 1612 !prefixcmp(arg, "--unpacked=") ||
63049292 1613 !strcmp("--reflog", arg) ||
8d1d8f83
JH
1614 !strcmp("--all", arg)) {
1615 use_internal_rev_list = 1;
ffa84ffb
RD
1616 if (rp_ac >= rp_ac_alloc - 1) {
1617 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1618 rp_av = xrealloc(rp_av,
1619 rp_ac_alloc * sizeof(*rp_av));
1620 }
8d1d8f83 1621 rp_av[rp_ac++] = arg;
b5d97e6b
JH
1622 continue;
1623 }
8d1d8f83
JH
1624 if (!strcmp("--thin", arg)) {
1625 use_internal_rev_list = 1;
1626 thin = 1;
1627 rp_av[1] = "--objects-edge";
b5d97e6b
JH
1628 continue;
1629 }
1630 usage(pack_usage);
1631 }
1632
1633 /* Traditionally "pack-objects [options] base extra" failed;
1634 * we would however want to take refs parameter that would
1635 * have been given to upstream rev-list ourselves, which means
1636 * we somehow want to say what the base name is. So the
1637 * syntax would be:
1638 *
1639 * pack-objects [options] base <refs...>
1640 *
1641 * in other words, we would treat the first non-option as the
1642 * base_name and send everything else to the internal revision
1643 * walker.
1644 */
1645
1646 if (!pack_to_stdout)
1647 base_name = argv[i++];
1648
1649 if (pack_to_stdout != !base_name)
1650 usage(pack_usage);
1651
8d1d8f83
JH
1652 if (!pack_to_stdout && thin)
1653 die("--thin cannot be used to build an indexable pack.");
b5d97e6b
JH
1654
1655 prepare_packed_git();
1656
1657 if (progress) {
1658 fprintf(stderr, "Generating pack...\n");
1659 setup_progress_signal();
1660 }
1661
1662 if (!use_internal_rev_list)
1663 read_object_list_from_stdin();
8d1d8f83
JH
1664 else {
1665 rp_av[rp_ac] = NULL;
1666 get_object_list(rp_ac, rp_av);
1667 }
b5d97e6b 1668
21fcd1bd 1669 if (progress)
7cadf491 1670 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
f0b0af1b
JH
1671 sorted_by_sha = create_final_object_list();
1672 if (non_empty && !nr_result)
1c4a2912 1673 return 0;
c323ac7d 1674
84c8d8ae
JH
1675 SHA1_Init(&ctx);
1676 list = sorted_by_sha;
7a979d99 1677 for (i = 0; i < nr_result; i++) {
84c8d8ae
JH
1678 struct object_entry *entry = *list++;
1679 SHA1_Update(&ctx, entry->sha1, 20);
1680 }
1681 SHA1_Final(object_list_sha1, &ctx);
7a979d99 1682 if (progress && (nr_objects != nr_result))
7cadf491 1683 fprintf(stderr, "Result has %u objects.\n", nr_result);
84c8d8ae 1684
df6d6101 1685 if (reuse_cached_pack(object_list_sha1))
f3123c4a
JH
1686 ;
1687 else {
f0b0af1b
JH
1688 if (nr_result)
1689 prepare_pack(window, depth);
fa438a2e 1690 if (progress == 1 && pack_to_stdout) {
5e8dc750
NP
1691 /* the other end usually displays progress itself */
1692 struct itimerval v = {{0,},};
1693 setitimer(ITIMER_REAL, &v, NULL);
1694 signal(SIGALRM, SIG_IGN );
1695 progress_update = 0;
1696 }
1697 write_pack_file();
f3123c4a
JH
1698 if (!pack_to_stdout) {
1699 write_index_file();
1700 puts(sha1_to_hex(object_list_sha1));
1701 }
5f3de58f 1702 }
ab7cd7bb 1703 if (progress)
7cadf491 1704 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
67c08ce1 1705 written, written_delta, reused, reused_delta);
c323ac7d
LT
1706 return 0;
1707}