?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'maint'
[git/spearce.git] / builtin-pack-objects.c
blobe52332df99d4fc7c3d4643a682147a7397ebb568
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "csum-file.h"
12 #include "tree-walk.h"
13 #include "diff.h"
14 #include "revision.h"
15 #include "list-objects.h"
16 #include "progress.h"
17
18 static const char pack_usage[] = "\
19 git-pack-objects [{ -q | --progress | --all-progress }] [--max-pack-size=N] \n\
20 [--local] [--incremental] [--window=N] [--depth=N] \n\
21 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
22 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
23 [--stdout | base-name] [<ref-list | <object-list]";
24
25 struct object_entry {
26 unsigned char sha1[20];
27 uint32_t crc32; /* crc of raw pack data for this object */
28 off_t offset; /* offset into the final pack file */
29 unsigned long size; /* uncompressed size */
30 unsigned int hash; /* name hint hash */
31 unsigned int depth; /* delta depth */
32 struct packed_git *in_pack; /* already in pack */
33 off_t in_pack_offset;
34 struct object_entry *delta; /* delta base object */
35 struct object_entry *delta_child; /* deltified objects who bases me */
36 struct object_entry *delta_sibling; /* other deltified objects who
37 * uses the same base as me
38 */
39 unsigned long delta_size; /* delta data size (uncompressed) */
40 enum object_type type;
41 enum object_type in_pack_type; /* could be delta */
42 unsigned char in_pack_header_size;
43 unsigned char preferred_base; /* we do not pack this, but is available
44 * to be used as the base object to delta
45 * objects against.
46 */
47 unsigned char no_try_delta;
48 };
49
50 /*
51 * Objects we are going to pack are collected in objects array (dynamically
52 * expanded). nr_objects & nr_alloc controls this array. They are stored
53 * in the order we see -- typically rev-list --objects order that gives us
54 * nice "minimum seek" order.
55 */
56 static struct object_entry *objects;
57 static struct object_entry **written_list;
58 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
59
60 static int non_empty;
61 static int no_reuse_delta, no_reuse_object;
62 static int local;
63 static int incremental;
64 static int allow_ofs_delta;
65 static const char *pack_tmp_name, *idx_tmp_name;
66 static char tmpname[PATH_MAX];
67 static const char *base_name;
68 static unsigned char pack_file_sha1[20];
69 static int progress = 1;
70 static int window = 10;
71 static uint32_t pack_size_limit;
72 static int depth = 50;
73 static int pack_to_stdout;
74 static int num_preferred_base;
75 static struct progress progress_state;
76 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
77 static int pack_compression_seen;
78
79 /*
80 * The object names in objects array are hashed with this hashtable,
81 * to help looking up the entry by object name.
82 * This hashtable is built after all the objects are seen.
83 */
84 static int *object_ix;
85 static int object_ix_hashsz;
86
87 /*
88 * Pack index for existing packs give us easy access to the offsets into
89 * corresponding pack file where each object's data starts, but the entries
90 * do not store the size of the compressed representation (uncompressed
91 * size is easily available by examining the pack entry header). It is
92 * also rather expensive to find the sha1 for an object given its offset.
93 *
94 * We build a hashtable of existing packs (pack_revindex), and keep reverse
95 * index here -- pack index file is sorted by object name mapping to offset;
96 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
97 * ordered by offset, so if you know the offset of an object, next offset
98 * is where its packed representation ends and the index_nr can be used to
99 * get the object sha1 from the main index.
100 */
101 struct revindex_entry {
102 off_t offset;
103 unsigned int nr;
104 };
105 struct pack_revindex {
106 struct packed_git *p;
107 struct revindex_entry *revindex;
108 };
109 static struct pack_revindex *pack_revindex;
110 static int pack_revindex_hashsz;
111
112 /*
113 * stats
114 */
115 static uint32_t written, written_delta;
116 static uint32_t reused, reused_delta;
117
118 static int pack_revindex_ix(struct packed_git *p)
119 {
120 unsigned long ui = (unsigned long)p;
121 int i;
122
123 ui = ui ^ (ui >> 16); /* defeat structure alignment */
124 i = (int)(ui % pack_revindex_hashsz);
125 while (pack_revindex[i].p) {
126 if (pack_revindex[i].p == p)
127 return i;
128 if (++i == pack_revindex_hashsz)
129 i = 0;
130 }
131 return -1 - i;
132 }
133
134 static void prepare_pack_ix(void)
135 {
136 int num;
137 struct packed_git *p;
138 for (num = 0, p = packed_git; p; p = p->next)
139 num++;
140 if (!num)
141 return;
142 pack_revindex_hashsz = num * 11;
143 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
144 for (p = packed_git; p; p = p->next) {
145 num = pack_revindex_ix(p);
146 num = - 1 - num;
147 pack_revindex[num].p = p;
148 }
149 /* revindex elements are lazily initialized */
150 }
151
152 static int cmp_offset(const void *a_, const void *b_)
153 {
154 const struct revindex_entry *a = a_;
155 const struct revindex_entry *b = b_;
156 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
157 }
158
159 /*
160 * Ordered list of offsets of objects in the pack.
161 */
162 static void prepare_pack_revindex(struct pack_revindex *rix)
163 {
164 struct packed_git *p = rix->p;
165 int num_ent = p->num_objects;
166 int i;
167 const char *index = p->index_data;
168
169 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
170 index += 4 * 256;
171
172 if (p->index_version > 1) {
173 const uint32_t *off_32 =
174 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
175 const uint32_t *off_64 = off_32 + p->num_objects;
176 for (i = 0; i < num_ent; i++) {
177 uint32_t off = ntohl(*off_32++);
178 if (!(off & 0x80000000)) {
179 rix->revindex[i].offset = off;
180 } else {
181 rix->revindex[i].offset =
182 ((uint64_t)ntohl(*off_64++)) << 32;
183 rix->revindex[i].offset |=
184 ntohl(*off_64++);
185 }
186 rix->revindex[i].nr = i;
187 }
188 } else {
189 for (i = 0; i < num_ent; i++) {
190 uint32_t hl = *((uint32_t *)(index + 24 * i));
191 rix->revindex[i].offset = ntohl(hl);
192 rix->revindex[i].nr = i;
193 }
194 }
195
196 /* This knows the pack format -- the 20-byte trailer
197 * follows immediately after the last object data.
198 */
199 rix->revindex[num_ent].offset = p->pack_size - 20;
200 rix->revindex[num_ent].nr = -1;
201 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
202 }
203
204 static struct revindex_entry * find_packed_object(struct packed_git *p,
205 off_t ofs)
206 {
207 int num;
208 int lo, hi;
209 struct pack_revindex *rix;
210 struct revindex_entry *revindex;
211 num = pack_revindex_ix(p);
212 if (num < 0)
213 die("internal error: pack revindex uninitialized");
214 rix = &pack_revindex[num];
215 if (!rix->revindex)
216 prepare_pack_revindex(rix);
217 revindex = rix->revindex;
218 lo = 0;
219 hi = p->num_objects + 1;
220 do {
221 int mi = (lo + hi) / 2;
222 if (revindex[mi].offset == ofs) {
223 return revindex + mi;
224 }
225 else if (ofs < revindex[mi].offset)
226 hi = mi;
227 else
228 lo = mi + 1;
229 } while (lo < hi);
230 die("internal error: pack revindex corrupt");
231 }
232
233 static const unsigned char *find_packed_object_name(struct packed_git *p,
234 off_t ofs)
235 {
236 struct revindex_entry *entry = find_packed_object(p, ofs);
237 return nth_packed_object_sha1(p, entry->nr);
238 }
239
240 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
241 {
242 unsigned long othersize, delta_size;
243 enum object_type type;
244 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
245 void *delta_buf;
246
247 if (!otherbuf)
248 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
249 delta_buf = diff_delta(otherbuf, othersize,
250 buf, size, &delta_size, 0);
251 if (!delta_buf || delta_size != entry->delta_size)
252 die("delta size changed");
253 free(buf);
254 free(otherbuf);
255 return delta_buf;
256 }
257
258 /*
259 * The per-object header is a pretty dense thing, which is
260 * - first byte: low four bits are "size", then three bits of "type",
261 * and the high bit is "size continues".
262 * - each byte afterwards: low seven bits are size continuation,
263 * with the high bit being "size continues"
264 */
265 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
266 {
267 int n = 1;
268 unsigned char c;
269
270 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
271 die("bad type %d", type);
272
273 c = (type << 4) | (size & 15);
274 size >>= 4;
275 while (size) {
276 *hdr++ = c | 0x80;
277 c = size & 0x7f;
278 size >>= 7;
279 n++;
280 }
281 *hdr = c;
282 return n;
283 }
284
285 /*
286 * we are going to reuse the existing object data as is. make
287 * sure it is not corrupt.
288 */
289 static int check_pack_inflate(struct packed_git *p,
290 struct pack_window **w_curs,
291 off_t offset,
292 off_t len,
293 unsigned long expect)
294 {
295 z_stream stream;
296 unsigned char fakebuf[4096], *in;
297 int st;
298
299 memset(&stream, 0, sizeof(stream));
300 inflateInit(&stream);
301 do {
302 in = use_pack(p, w_curs, offset, &stream.avail_in);
303 stream.next_in = in;
304 stream.next_out = fakebuf;
305 stream.avail_out = sizeof(fakebuf);
306 st = inflate(&stream, Z_FINISH);
307 offset += stream.next_in - in;
308 } while (st == Z_OK || st == Z_BUF_ERROR);
309 inflateEnd(&stream);
310 return (st == Z_STREAM_END &&
311 stream.total_out == expect &&
312 stream.total_in == len) ? 0 : -1;
313 }
314
315 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
316 off_t offset, off_t len, unsigned int nr)
317 {
318 const uint32_t *index_crc;
319 uint32_t data_crc = crc32(0, Z_NULL, 0);
320
321 do {
322 unsigned int avail;
323 void *data = use_pack(p, w_curs, offset, &avail);
324 if (avail > len)
325 avail = len;
326 data_crc = crc32(data_crc, data, avail);
327 offset += avail;
328 len -= avail;
329 } while (len);
330
331 index_crc = p->index_data;
332 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
333
334 return data_crc != ntohl(*index_crc);
335 }
336
337 static void copy_pack_data(struct sha1file *f,
338 struct packed_git *p,
339 struct pack_window **w_curs,
340 off_t offset,
341 off_t len)
342 {
343 unsigned char *in;
344 unsigned int avail;
345
346 while (len) {
347 in = use_pack(p, w_curs, offset, &avail);
348 if (avail > len)
349 avail = (unsigned int)len;
350 sha1write(f, in, avail);
351 offset += avail;
352 len -= avail;
353 }
354 }
355
356 static unsigned long write_object(struct sha1file *f,
357 struct object_entry *entry,
358 off_t write_offset)
359 {
360 unsigned long size;
361 enum object_type type;
362 void *buf;
363 unsigned char header[10];
364 unsigned char dheader[10];
365 unsigned hdrlen;
366 off_t datalen;
367 enum object_type obj_type;
368 int to_reuse = 0;
369 /* write limit if limited packsize and not first object */
370 unsigned long limit = pack_size_limit && nr_written ?
371 pack_size_limit - write_offset : 0;
372 /* no if no delta */
373 int usable_delta = !entry->delta ? 0 :
374 /* yes if unlimited packfile */
375 !pack_size_limit ? 1 :
376 /* no if base written to previous pack */
377 entry->delta->offset == (off_t)-1 ? 0 :
378 /* otherwise double-check written to this
379 * pack, like we do below
380 */
381 entry->delta->offset ? 1 : 0;
382
383 if (!pack_to_stdout)
384 crc32_begin(f);
385
386 obj_type = entry->type;
387 if (no_reuse_object)
388 to_reuse = 0; /* explicit */
389 else if (!entry->in_pack)
390 to_reuse = 0; /* can't reuse what we don't have */
391 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
392 /* check_object() decided it for us ... */
393 to_reuse = usable_delta;
394 /* ... but pack split may override that */
395 else if (obj_type != entry->in_pack_type)
396 to_reuse = 0; /* pack has delta which is unusable */
397 else if (entry->delta)
398 to_reuse = 0; /* we want to pack afresh */
399 else
400 to_reuse = 1; /* we have it in-pack undeltified,
401 * and we do not need to deltify it.
402 */
403
404 if (!to_reuse) {
405 z_stream stream;
406 unsigned long maxsize;
407 void *out;
408 buf = read_sha1_file(entry->sha1, &type, &size);
409 if (!buf)
410 die("unable to read %s", sha1_to_hex(entry->sha1));
411 if (size != entry->size)
412 die("object %s size inconsistency (%lu vs %lu)",
413 sha1_to_hex(entry->sha1), size, entry->size);
414 if (usable_delta) {
415 buf = delta_against(buf, size, entry);
416 size = entry->delta_size;
417 obj_type = (allow_ofs_delta && entry->delta->offset) ?
418 OBJ_OFS_DELTA : OBJ_REF_DELTA;
419 } else {
420 /*
421 * recover real object type in case
422 * check_object() wanted to re-use a delta,
423 * but we couldn't since base was in previous split pack
424 */
425 obj_type = type;
426 }
427 /* compress the data to store and put compressed length in datalen */
428 memset(&stream, 0, sizeof(stream));
429 deflateInit(&stream, pack_compression_level);
430 maxsize = deflateBound(&stream, size);
431 out = xmalloc(maxsize);
432 /* Compress it */
433 stream.next_in = buf;
434 stream.avail_in = size;
435 stream.next_out = out;
436 stream.avail_out = maxsize;
437 while (deflate(&stream, Z_FINISH) == Z_OK)
438 /* nothing */;
439 deflateEnd(&stream);
440 datalen = stream.total_out;
441 deflateEnd(&stream);
442 /*
443 * The object header is a byte of 'type' followed by zero or
444 * more bytes of length.
445 */
446 hdrlen = encode_header(obj_type, size, header);
447
448 if (obj_type == OBJ_OFS_DELTA) {
449 /*
450 * Deltas with relative base contain an additional
451 * encoding of the relative offset for the delta
452 * base from this object's position in the pack.
453 */
454 off_t ofs = entry->offset - entry->delta->offset;
455 unsigned pos = sizeof(dheader) - 1;
456 dheader[pos] = ofs & 127;
457 while (ofs >>= 7)
458 dheader[--pos] = 128 | (--ofs & 127);
459 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
460 free(out);
461 free(buf);
462 return 0;
463 }
464 sha1write(f, header, hdrlen);
465 sha1write(f, dheader + pos, sizeof(dheader) - pos);
466 hdrlen += sizeof(dheader) - pos;
467 } else if (obj_type == OBJ_REF_DELTA) {
468 /*
469 * Deltas with a base reference contain
470 * an additional 20 bytes for the base sha1.
471 */
472 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
473 free(out);
474 free(buf);
475 return 0;
476 }
477 sha1write(f, header, hdrlen);
478 sha1write(f, entry->delta->sha1, 20);
479 hdrlen += 20;
480 } else {
481 if (limit && hdrlen + datalen + 20 >= limit) {
482 free(out);
483 free(buf);
484 return 0;
485 }
486 sha1write(f, header, hdrlen);
487 }
488 sha1write(f, out, datalen);
489 free(out);
490 free(buf);
491 }
492 else {
493 struct packed_git *p = entry->in_pack;
494 struct pack_window *w_curs = NULL;
495 struct revindex_entry *revidx;
496 off_t offset;
497
498 if (entry->delta) {
499 obj_type = (allow_ofs_delta && entry->delta->offset) ?
500 OBJ_OFS_DELTA : OBJ_REF_DELTA;
501 reused_delta++;
502 }
503 hdrlen = encode_header(obj_type, entry->size, header);
504 offset = entry->in_pack_offset;
505 revidx = find_packed_object(p, offset);
506 datalen = revidx[1].offset - offset;
507 if (!pack_to_stdout && p->index_version > 1 &&
508 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
509 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
510 offset += entry->in_pack_header_size;
511 datalen -= entry->in_pack_header_size;
512 if (obj_type == OBJ_OFS_DELTA) {
513 off_t ofs = entry->offset - entry->delta->offset;
514 unsigned pos = sizeof(dheader) - 1;
515 dheader[pos] = ofs & 127;
516 while (ofs >>= 7)
517 dheader[--pos] = 128 | (--ofs & 127);
518 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
519 return 0;
520 sha1write(f, header, hdrlen);
521 sha1write(f, dheader + pos, sizeof(dheader) - pos);
522 hdrlen += sizeof(dheader) - pos;
523 } else if (obj_type == OBJ_REF_DELTA) {
524 if (limit && hdrlen + 20 + datalen + 20 >= limit)
525 return 0;
526 sha1write(f, header, hdrlen);
527 sha1write(f, entry->delta->sha1, 20);
528 hdrlen += 20;
529 } else {
530 if (limit && hdrlen + datalen + 20 >= limit)
531 return 0;
532 sha1write(f, header, hdrlen);
533 }
534
535 if (!pack_to_stdout && p->index_version == 1 &&
536 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
537 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
538 copy_pack_data(f, p, &w_curs, offset, datalen);
539 unuse_pack(&w_curs);
540 reused++;
541 }
542 if (usable_delta)
543 written_delta++;
544 written++;
545 if (!pack_to_stdout)
546 entry->crc32 = crc32_end(f);
547 return hdrlen + datalen;
548 }
549
550 static off_t write_one(struct sha1file *f,
551 struct object_entry *e,
552 off_t offset)
553 {
554 unsigned long size;
555
556 /* offset is non zero if object is written already. */
557 if (e->offset || e->preferred_base)
558 return offset;
559
560 /* if we are deltified, write out base object first. */
561 if (e->delta) {
562 offset = write_one(f, e->delta, offset);
563 if (!offset)
564 return 0;
565 }
566
567 e->offset = offset;
568 size = write_object(f, e, offset);
569 if (!size) {
570 e->offset = 0;
571 return 0;
572 }
573 written_list[nr_written++] = e;
574
575 /* make sure off_t is sufficiently large not to wrap */
576 if (offset > offset + size)
577 die("pack too large for current definition of off_t");
578 return offset + size;
579 }
580
581 static int open_object_dir_tmp(const char *path)
582 {
583 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
584 return mkstemp(tmpname);
585 }
586
587 /* forward declarations for write_pack_file */
588 static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
589 static int adjust_perm(const char *path, mode_t mode);
590
591 static void write_pack_file(void)
592 {
593 uint32_t i = 0, j;
594 struct sha1file *f;
595 off_t offset, offset_one, last_obj_offset = 0;
596 struct pack_header hdr;
597 int do_progress = progress >> pack_to_stdout;
598 uint32_t nr_remaining = nr_result;
599
600 if (do_progress)
601 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
602 written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
603
604 do {
605 if (pack_to_stdout) {
606 f = sha1fd(1, "<stdout>");
607 } else {
608 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
609 if (fd < 0)
610 die("unable to create %s: %s\n", tmpname, strerror(errno));
611 pack_tmp_name = xstrdup(tmpname);
612 f = sha1fd(fd, pack_tmp_name);
613 }
614
615 hdr.hdr_signature = htonl(PACK_SIGNATURE);
616 hdr.hdr_version = htonl(PACK_VERSION);
617 hdr.hdr_entries = htonl(nr_remaining);
618 sha1write(f, &hdr, sizeof(hdr));
619 offset = sizeof(hdr);
620 nr_written = 0;
621 for (; i < nr_objects; i++) {
622 last_obj_offset = offset;
623 offset_one = write_one(f, objects + i, offset);
624 if (!offset_one)
625 break;
626 offset = offset_one;
627 if (do_progress)
628 display_progress(&progress_state, written);
629 }
630
631 /*
632 * Did we write the wrong # entries in the header?
633 * If so, rewrite it like in fast-import
634 */
635 if (pack_to_stdout || nr_written == nr_remaining) {
636 sha1close(f, pack_file_sha1, 1);
637 } else {
638 sha1close(f, pack_file_sha1, 0);
639 fixup_pack_header_footer(f->fd, pack_file_sha1, pack_tmp_name, nr_written);
640 close(f->fd);
641 }
642
643 if (!pack_to_stdout) {
644 unsigned char object_list_sha1[20];
645 mode_t mode = umask(0);
646
647 umask(mode);
648 mode = 0444 & ~mode;
649
650 write_index_file(last_obj_offset, object_list_sha1);
651 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
652 base_name, sha1_to_hex(object_list_sha1));
653 if (adjust_perm(pack_tmp_name, mode))
654 die("unable to make temporary pack file readable: %s",
655 strerror(errno));
656 if (rename(pack_tmp_name, tmpname))
657 die("unable to rename temporary pack file: %s",
658 strerror(errno));
659 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
660 base_name, sha1_to_hex(object_list_sha1));
661 if (adjust_perm(idx_tmp_name, mode))
662 die("unable to make temporary index file readable: %s",
663 strerror(errno));
664 if (rename(idx_tmp_name, tmpname))
665 die("unable to rename temporary index file: %s",
666 strerror(errno));
667 puts(sha1_to_hex(object_list_sha1));
668 }
669
670 /* mark written objects as written to previous pack */
671 for (j = 0; j < nr_written; j++) {
672 written_list[j]->offset = (off_t)-1;
673 }
674 nr_remaining -= nr_written;
675 } while (nr_remaining && i < nr_objects);
676
677 free(written_list);
678 if (do_progress)
679 stop_progress(&progress_state);
680 if (written != nr_result)
681 die("wrote %u objects while expecting %u", written, nr_result);
682 /*
683 * We have scanned through [0 ... i). Since we have written
684 * the correct number of objects, the remaining [i ... nr_objects)
685 * items must be either already written (due to out-of-order delta base)
686 * or a preferred base. Count those which are neither and complain if any.
687 */
688 for (j = 0; i < nr_objects; i++) {
689 struct object_entry *e = objects + i;
690 j += !e->offset && !e->preferred_base;
691 }
692 if (j)
693 die("wrote %u objects as expected but %u unwritten", written, j);
694 }
695
696 static int sha1_sort(const void *_a, const void *_b)
697 {
698 const struct object_entry *a = *(struct object_entry **)_a;
699 const struct object_entry *b = *(struct object_entry **)_b;
700 return hashcmp(a->sha1, b->sha1);
701 }
702
703 static uint32_t index_default_version = 1;
704 static uint32_t index_off32_limit = 0x7fffffff;
705
706 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
707 {
708 struct sha1file *f;
709 struct object_entry **sorted_by_sha, **list, **last;
710 uint32_t array[256];
711 uint32_t i, index_version;
712 SHA_CTX ctx;
713
714 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
715 if (fd < 0)
716 die("unable to create %s: %s\n", tmpname, strerror(errno));
717 idx_tmp_name = xstrdup(tmpname);
718 f = sha1fd(fd, idx_tmp_name);
719
720 if (nr_written) {
721 sorted_by_sha = written_list;
722 qsort(sorted_by_sha, nr_written, sizeof(*sorted_by_sha), sha1_sort);
723 list = sorted_by_sha;
724 last = sorted_by_sha + nr_written;
725 } else
726 sorted_by_sha = list = last = NULL;
727
728 /* if last object's offset is >= 2^31 we should use index V2 */
729 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
730
731 /* index versions 2 and above need a header */
732 if (index_version >= 2) {
733 struct pack_idx_header hdr;
734 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
735 hdr.idx_version = htonl(index_version);
736 sha1write(f, &hdr, sizeof(hdr));
737 }
738
739 /*
740 * Write the first-level table (the list is sorted,
741 * but we use a 256-entry lookup to be able to avoid
742 * having to do eight extra binary search iterations).
743 */
744 for (i = 0; i < 256; i++) {
745 struct object_entry **next = list;
746 while (next < last) {
747 struct object_entry *entry = *next;
748 if (entry->sha1[0] != i)
749 break;
750 next++;
751 }
752 array[i] = htonl(next - sorted_by_sha);
753 list = next;
754 }
755 sha1write(f, array, 256 * 4);
756
757 /* Compute the SHA1 hash of sorted object names. */
758 SHA1_Init(&ctx);
759
760 /* Write the actual SHA1 entries. */
761 list = sorted_by_sha;
762 for (i = 0; i < nr_written; i++) {
763 struct object_entry *entry = *list++;
764 if (index_version < 2) {
765 uint32_t offset = htonl(entry->offset);
766 sha1write(f, &offset, 4);
767 }
768 sha1write(f, entry->sha1, 20);
769 SHA1_Update(&ctx, entry->sha1, 20);
770 }
771
772 if (index_version >= 2) {
773 unsigned int nr_large_offset = 0;
774
775 /* write the crc32 table */
776 list = sorted_by_sha;
777 for (i = 0; i < nr_written; i++) {
778 struct object_entry *entry = *list++;
779 uint32_t crc32_val = htonl(entry->crc32);
780 sha1write(f, &crc32_val, 4);
781 }
782
783 /* write the 32-bit offset table */
784 list = sorted_by_sha;
785 for (i = 0; i < nr_written; i++) {
786 struct object_entry *entry = *list++;
787 uint32_t offset = (entry->offset <= index_off32_limit) ?
788 entry->offset : (0x80000000 | nr_large_offset++);
789 offset = htonl(offset);
790 sha1write(f, &offset, 4);
791 }
792
793 /* write the large offset table */
794 list = sorted_by_sha;
795 while (nr_large_offset) {
796 struct object_entry *entry = *list++;
797 uint64_t offset = entry->offset;
798 if (offset > index_off32_limit) {
799 uint32_t split[2];
800 split[0] = htonl(offset >> 32);
801 split[1] = htonl(offset & 0xffffffff);
802 sha1write(f, split, 8);
803 nr_large_offset--;
804 }
805 }
806 }
807
808 sha1write(f, pack_file_sha1, 20);
809 sha1close(f, NULL, 1);
810 SHA1_Final(sha1, &ctx);
811 }
812
813 static int locate_object_entry_hash(const unsigned char *sha1)
814 {
815 int i;
816 unsigned int ui;
817 memcpy(&ui, sha1, sizeof(unsigned int));
818 i = ui % object_ix_hashsz;
819 while (0 < object_ix[i]) {
820 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
821 return i;
822 if (++i == object_ix_hashsz)
823 i = 0;
824 }
825 return -1 - i;
826 }
827
828 static struct object_entry *locate_object_entry(const unsigned char *sha1)
829 {
830 int i;
831
832 if (!object_ix_hashsz)
833 return NULL;
834
835 i = locate_object_entry_hash(sha1);
836 if (0 <= i)
837 return &objects[object_ix[i]-1];
838 return NULL;
839 }
840
841 static void rehash_objects(void)
842 {
843 uint32_t i;
844 struct object_entry *oe;
845
846 object_ix_hashsz = nr_objects * 3;
847 if (object_ix_hashsz < 1024)
848 object_ix_hashsz = 1024;
849 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
850 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
851 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
852 int ix = locate_object_entry_hash(oe->sha1);
853 if (0 <= ix)
854 continue;
855 ix = -1 - ix;
856 object_ix[ix] = i + 1;
857 }
858 }
859
860 static unsigned name_hash(const char *name)
861 {
862 unsigned char c;
863 unsigned hash = 0;
864
865 if (!name)
866 return 0;
867
868 /*
869 * This effectively just creates a sortable number from the
870 * last sixteen non-whitespace characters. Last characters
871 * count "most", so things that end in ".c" sort together.
872 */
873 while ((c = *name++) != 0) {
874 if (isspace(c))
875 continue;
876 hash = (hash >> 2) + (c << 24);
877 }
878 return hash;
879 }
880
881 static void setup_delta_attr_check(struct git_attr_check *check)
882 {
883 static struct git_attr *attr_delta;
884
885 if (!attr_delta)
886 attr_delta = git_attr("delta", 5);
887
888 check[0].attr = attr_delta;
889 }
890
891 static int no_try_delta(const char *path)
892 {
893 struct git_attr_check check[1];
894
895 setup_delta_attr_check(check);
896 if (git_checkattr(path, ARRAY_SIZE(check), check))
897 return 0;
898 if (ATTR_FALSE(check->value))
899 return 1;
900 return 0;
901 }
902
903 static int add_object_entry(const unsigned char *sha1, enum object_type type,
904 const char *name, int exclude)
905 {
906 struct object_entry *entry;
907 struct packed_git *p, *found_pack = NULL;
908 off_t found_offset = 0;
909 int ix;
910 unsigned hash = name_hash(name);
911
912 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
913 if (ix >= 0) {
914 if (exclude) {
915 entry = objects + object_ix[ix] - 1;
916 if (!entry->preferred_base)
917 nr_result--;
918 entry->preferred_base = 1;
919 }
920 return 0;
921 }
922
923 for (p = packed_git; p; p = p->next) {
924 off_t offset = find_pack_entry_one(sha1, p);
925 if (offset) {
926 if (!found_pack) {
927 found_offset = offset;
928 found_pack = p;
929 }
930 if (exclude)
931 break;
932 if (incremental)
933 return 0;
934 if (local && !p->pack_local)
935 return 0;
936 }
937 }
938
939 if (nr_objects >= nr_alloc) {
940 nr_alloc = (nr_alloc + 1024) * 3 / 2;
941 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
942 }
943
944 entry = objects + nr_objects++;
945 memset(entry, 0, sizeof(*entry));
946 hashcpy(entry->sha1, sha1);
947 entry->hash = hash;
948 if (type)
949 entry->type = type;
950 if (exclude)
951 entry->preferred_base = 1;
952 else
953 nr_result++;
954 if (found_pack) {
955 entry->in_pack = found_pack;
956 entry->in_pack_offset = found_offset;
957 }
958
959 if (object_ix_hashsz * 3 <= nr_objects * 4)
960 rehash_objects();
961 else
962 object_ix[-1 - ix] = nr_objects;
963
964 if (progress)
965 display_progress(&progress_state, nr_objects);
966
967 if (name && no_try_delta(name))
968 entry->no_try_delta = 1;
969
970 return 1;
971 }
972
973 struct pbase_tree_cache {
974 unsigned char sha1[20];
975 int ref;
976 int temporary;
977 void *tree_data;
978 unsigned long tree_size;
979 };
980
981 static struct pbase_tree_cache *(pbase_tree_cache[256]);
982 static int pbase_tree_cache_ix(const unsigned char *sha1)
983 {
984 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
985 }
986 static int pbase_tree_cache_ix_incr(int ix)
987 {
988 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
989 }
990
991 static struct pbase_tree {
992 struct pbase_tree *next;
993 /* This is a phony "cache" entry; we are not
994 * going to evict it nor find it through _get()
995 * mechanism -- this is for the toplevel node that
996 * would almost always change with any commit.
997 */
998 struct pbase_tree_cache pcache;
999 } *pbase_tree;
1000
1001 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1002 {
1003 struct pbase_tree_cache *ent, *nent;
1004 void *data;
1005 unsigned long size;
1006 enum object_type type;
1007 int neigh;
1008 int my_ix = pbase_tree_cache_ix(sha1);
1009 int available_ix = -1;
1010
1011 /* pbase-tree-cache acts as a limited hashtable.
1012 * your object will be found at your index or within a few
1013 * slots after that slot if it is cached.
1014 */
1015 for (neigh = 0; neigh < 8; neigh++) {
1016 ent = pbase_tree_cache[my_ix];
1017 if (ent && !hashcmp(ent->sha1, sha1)) {
1018 ent->ref++;
1019 return ent;
1020 }
1021 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1022 ((0 <= available_ix) &&
1023 (!ent && pbase_tree_cache[available_ix])))
1024 available_ix = my_ix;
1025 if (!ent)
1026 break;
1027 my_ix = pbase_tree_cache_ix_incr(my_ix);
1028 }
1029
1030 /* Did not find one. Either we got a bogus request or
1031 * we need to read and perhaps cache.
1032 */
1033 data = read_sha1_file(sha1, &type, &size);
1034 if (!data)
1035 return NULL;
1036 if (type != OBJ_TREE) {
1037 free(data);
1038 return NULL;
1039 }
1040
1041 /* We need to either cache or return a throwaway copy */
1042
1043 if (available_ix < 0)
1044 ent = NULL;
1045 else {
1046 ent = pbase_tree_cache[available_ix];
1047 my_ix = available_ix;
1048 }
1049
1050 if (!ent) {
1051 nent = xmalloc(sizeof(*nent));
1052 nent->temporary = (available_ix < 0);
1053 }
1054 else {
1055 /* evict and reuse */
1056 free(ent->tree_data);
1057 nent = ent;
1058 }
1059 hashcpy(nent->sha1, sha1);
1060 nent->tree_data = data;
1061 nent->tree_size = size;
1062 nent->ref = 1;
1063 if (!nent->temporary)
1064 pbase_tree_cache[my_ix] = nent;
1065 return nent;
1066 }
1067
1068 static void pbase_tree_put(struct pbase_tree_cache *cache)
1069 {
1070 if (!cache->temporary) {
1071 cache->ref--;
1072 return;
1073 }
1074 free(cache->tree_data);
1075 free(cache);
1076 }
1077
1078 static int name_cmp_len(const char *name)
1079 {
1080 int i;
1081 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1082 ;
1083 return i;
1084 }
1085
1086 static void add_pbase_object(struct tree_desc *tree,
1087 const char *name,
1088 int cmplen,
1089 const char *fullname)
1090 {
1091 struct name_entry entry;
1092 int cmp;
1093
1094 while (tree_entry(tree,&entry)) {
1095 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1096 memcmp(name, entry.path, cmplen);
1097 if (cmp > 0)
1098 continue;
1099 if (cmp < 0)
1100 return;
1101 if (name[cmplen] != '/') {
1102 add_object_entry(entry.sha1,
1103 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1104 fullname, 1);
1105 return;
1106 }
1107 if (S_ISDIR(entry.mode)) {
1108 struct tree_desc sub;
1109 struct pbase_tree_cache *tree;
1110 const char *down = name+cmplen+1;
1111 int downlen = name_cmp_len(down);
1112
1113 tree = pbase_tree_get(entry.sha1);
1114 if (!tree)
1115 return;
1116 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1117
1118 add_pbase_object(&sub, down, downlen, fullname);
1119 pbase_tree_put(tree);
1120 }
1121 }
1122 }
1123
1124 static unsigned *done_pbase_paths;
1125 static int done_pbase_paths_num;
1126 static int done_pbase_paths_alloc;
1127 static int done_pbase_path_pos(unsigned hash)
1128 {
1129 int lo = 0;
1130 int hi = done_pbase_paths_num;
1131 while (lo < hi) {
1132 int mi = (hi + lo) / 2;
1133 if (done_pbase_paths[mi] == hash)
1134 return mi;
1135 if (done_pbase_paths[mi] < hash)
1136 hi = mi;
1137 else
1138 lo = mi + 1;
1139 }
1140 return -lo-1;
1141 }
1142
1143 static int check_pbase_path(unsigned hash)
1144 {
1145 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1146 if (0 <= pos)
1147 return 1;
1148 pos = -pos - 1;
1149 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1150 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1151 done_pbase_paths = xrealloc(done_pbase_paths,
1152 done_pbase_paths_alloc *
1153 sizeof(unsigned));
1154 }
1155 done_pbase_paths_num++;
1156 if (pos < done_pbase_paths_num)
1157 memmove(done_pbase_paths + pos + 1,
1158 done_pbase_paths + pos,
1159 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1160 done_pbase_paths[pos] = hash;
1161 return 0;
1162 }
1163
1164 static void add_preferred_base_object(const char *name)
1165 {
1166 struct pbase_tree *it;
1167 int cmplen;
1168 unsigned hash = name_hash(name);
1169
1170 if (!num_preferred_base || check_pbase_path(hash))
1171 return;
1172
1173 cmplen = name_cmp_len(name);
1174 for (it = pbase_tree; it; it = it->next) {
1175 if (cmplen == 0) {
1176 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1177 }
1178 else {
1179 struct tree_desc tree;
1180 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1181 add_pbase_object(&tree, name, cmplen, name);
1182 }
1183 }
1184 }
1185
1186 static void add_preferred_base(unsigned char *sha1)
1187 {
1188 struct pbase_tree *it;
1189 void *data;
1190 unsigned long size;
1191 unsigned char tree_sha1[20];
1192
1193 if (window <= num_preferred_base++)
1194 return;
1195
1196 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1197 if (!data)
1198 return;
1199
1200 for (it = pbase_tree; it; it = it->next) {
1201 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1202 free(data);
1203 return;
1204 }
1205 }
1206
1207 it = xcalloc(1, sizeof(*it));
1208 it->next = pbase_tree;
1209 pbase_tree = it;
1210
1211 hashcpy(it->pcache.sha1, tree_sha1);
1212 it->pcache.tree_data = data;
1213 it->pcache.tree_size = size;
1214 }
1215
1216 static void check_object(struct object_entry *entry)
1217 {
1218 if (entry->in_pack) {
1219 struct packed_git *p = entry->in_pack;
1220 struct pack_window *w_curs = NULL;
1221 const unsigned char *base_ref = NULL;
1222 struct object_entry *base_entry;
1223 unsigned long used, used_0;
1224 unsigned int avail;
1225 off_t ofs;
1226 unsigned char *buf, c;
1227
1228 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1229
1230 /*
1231 * We want in_pack_type even if we do not reuse delta
1232 * since non-delta representations could still be reused.
1233 */
1234 used = unpack_object_header_gently(buf, avail,
1235 &entry->in_pack_type,
1236 &entry->size);
1237
1238 /*
1239 * Determine if this is a delta and if so whether we can
1240 * reuse it or not. Otherwise let's find out as cheaply as
1241 * possible what the actual type and size for this object is.
1242 */
1243 switch (entry->in_pack_type) {
1244 default:
1245 /* Not a delta hence we've already got all we need. */
1246 entry->type = entry->in_pack_type;
1247 entry->in_pack_header_size = used;
1248 unuse_pack(&w_curs);
1249 return;
1250 case OBJ_REF_DELTA:
1251 if (!no_reuse_delta && !entry->preferred_base)
1252 base_ref = use_pack(p, &w_curs,
1253 entry->in_pack_offset + used, NULL);
1254 entry->in_pack_header_size = used + 20;
1255 break;
1256 case OBJ_OFS_DELTA:
1257 buf = use_pack(p, &w_curs,
1258 entry->in_pack_offset + used, NULL);
1259 used_0 = 0;
1260 c = buf[used_0++];
1261 ofs = c & 127;
1262 while (c & 128) {
1263 ofs += 1;
1264 if (!ofs || MSB(ofs, 7))
1265 die("delta base offset overflow in pack for %s",
1266 sha1_to_hex(entry->sha1));
1267 c = buf[used_0++];
1268 ofs = (ofs << 7) + (c & 127);
1269 }
1270 if (ofs >= entry->in_pack_offset)
1271 die("delta base offset out of bound for %s",
1272 sha1_to_hex(entry->sha1));
1273 ofs = entry->in_pack_offset - ofs;
1274 if (!no_reuse_delta && !entry->preferred_base)
1275 base_ref = find_packed_object_name(p, ofs);
1276 entry->in_pack_header_size = used + used_0;
1277 break;
1278 }
1279
1280 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1281 /*
1282 * If base_ref was set above that means we wish to
1283 * reuse delta data, and we even found that base
1284 * in the list of objects we want to pack. Goodie!
1285 *
1286 * Depth value does not matter - find_deltas() will
1287 * never consider reused delta as the base object to
1288 * deltify other objects against, in order to avoid
1289 * circular deltas.
1290 */
1291 entry->type = entry->in_pack_type;
1292 entry->delta = base_entry;
1293 entry->delta_sibling = base_entry->delta_child;
1294 base_entry->delta_child = entry;
1295 unuse_pack(&w_curs);
1296 return;
1297 }
1298
1299 if (entry->type) {
1300 /*
1301 * This must be a delta and we already know what the
1302 * final object type is. Let's extract the actual
1303 * object size from the delta header.
1304 */
1305 entry->size = get_size_from_delta(p, &w_curs,
1306 entry->in_pack_offset + entry->in_pack_header_size);
1307 unuse_pack(&w_curs);
1308 return;
1309 }
1310
1311 /*
1312 * No choice but to fall back to the recursive delta walk
1313 * with sha1_object_info() to find about the object type
1314 * at this point...
1315 */
1316 unuse_pack(&w_curs);
1317 }
1318
1319 entry->type = sha1_object_info(entry->sha1, &entry->size);
1320 if (entry->type < 0)
1321 die("unable to get type of object %s",
1322 sha1_to_hex(entry->sha1));
1323 }
1324
1325 static int pack_offset_sort(const void *_a, const void *_b)
1326 {
1327 const struct object_entry *a = *(struct object_entry **)_a;
1328 const struct object_entry *b = *(struct object_entry **)_b;
1329
1330 /* avoid filesystem trashing with loose objects */
1331 if (!a->in_pack && !b->in_pack)
1332 return hashcmp(a->sha1, b->sha1);
1333
1334 if (a->in_pack < b->in_pack)
1335 return -1;
1336 if (a->in_pack > b->in_pack)
1337 return 1;
1338 return a->in_pack_offset < b->in_pack_offset ? -1 :
1339 (a->in_pack_offset > b->in_pack_offset);
1340 }
1341
1342 static void get_object_details(void)
1343 {
1344 uint32_t i;
1345 struct object_entry **sorted_by_offset;
1346
1347 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1348 for (i = 0; i < nr_objects; i++)
1349 sorted_by_offset[i] = objects + i;
1350 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1351
1352 prepare_pack_ix();
1353 for (i = 0; i < nr_objects; i++)
1354 check_object(sorted_by_offset[i]);
1355 free(sorted_by_offset);
1356 }
1357
1358 static int type_size_sort(const void *_a, const void *_b)
1359 {
1360 const struct object_entry *a = *(struct object_entry **)_a;
1361 const struct object_entry *b = *(struct object_entry **)_b;
1362
1363 if (a->type < b->type)
1364 return -1;
1365 if (a->type > b->type)
1366 return 1;
1367 if (a->hash < b->hash)
1368 return -1;
1369 if (a->hash > b->hash)
1370 return 1;
1371 if (a->preferred_base < b->preferred_base)
1372 return -1;
1373 if (a->preferred_base > b->preferred_base)
1374 return 1;
1375 if (a->size < b->size)
1376 return -1;
1377 if (a->size > b->size)
1378 return 1;
1379 return a > b ? -1 : (a < b); /* newest last */
1380 }
1381
1382 struct unpacked {
1383 struct object_entry *entry;
1384 void *data;
1385 struct delta_index *index;
1386 };
1387
1388 /*
1389 * We search for deltas _backwards_ in a list sorted by type and
1390 * by size, so that we see progressively smaller and smaller files.
1391 * That's because we prefer deltas to be from the bigger file
1392 * to the smaller - deletes are potentially cheaper, but perhaps
1393 * more importantly, the bigger file is likely the more recent
1394 * one.
1395 */
1396 static int try_delta(struct unpacked *trg, struct unpacked *src,
1397 unsigned max_depth)
1398 {
1399 struct object_entry *trg_entry = trg->entry;
1400 struct object_entry *src_entry = src->entry;
1401 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1402 enum object_type type;
1403 void *delta_buf;
1404
1405 /* Don't bother doing diffs between different types */
1406 if (trg_entry->type != src_entry->type)
1407 return -1;
1408
1409 /* We do not compute delta to *create* objects we are not
1410 * going to pack.
1411 */
1412 if (trg_entry->preferred_base)
1413 return -1;
1414
1415 /*
1416 * We do not bother to try a delta that we discarded
1417 * on an earlier try, but only when reusing delta data.
1418 */
1419 if (!no_reuse_delta && trg_entry->in_pack &&
1420 trg_entry->in_pack == src_entry->in_pack &&
1421 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1422 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1423 return 0;
1424
1425 /* Let's not bust the allowed depth. */
1426 if (src_entry->depth >= max_depth)
1427 return 0;
1428
1429 /* Now some size filtering heuristics. */
1430 trg_size = trg_entry->size;
1431 max_size = trg_size/2 - 20;
1432 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1433 if (max_size == 0)
1434 return 0;
1435 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1436 max_size = trg_entry->delta_size-1;
1437 src_size = src_entry->size;
1438 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1439 if (sizediff >= max_size)
1440 return 0;
1441
1442 /* Load data if not already done */
1443 if (!trg->data) {
1444 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1445 if (sz != trg_size)
1446 die("object %s inconsistent object length (%lu vs %lu)",
1447 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1448 }
1449 if (!src->data) {
1450 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1451 if (sz != src_size)
1452 die("object %s inconsistent object length (%lu vs %lu)",
1453 sha1_to_hex(src_entry->sha1), sz, src_size);
1454 }
1455 if (!src->index) {
1456 src->index = create_delta_index(src->data, src_size);
1457 if (!src->index)
1458 die("out of memory");
1459 }
1460
1461 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1462 if (!delta_buf)
1463 return 0;
1464
1465 trg_entry->delta = src_entry;
1466 trg_entry->delta_size = delta_size;
1467 trg_entry->depth = src_entry->depth + 1;
1468 free(delta_buf);
1469 return 1;
1470 }
1471
1472 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1473 {
1474 struct object_entry *child = me->delta_child;
1475 unsigned int m = n;
1476 while (child) {
1477 unsigned int c = check_delta_limit(child, n + 1);
1478 if (m < c)
1479 m = c;
1480 child = child->delta_sibling;
1481 }
1482 return m;
1483 }
1484
1485 static void find_deltas(struct object_entry **list, int window, int depth)
1486 {
1487 uint32_t i = nr_objects, idx = 0, processed = 0;
1488 unsigned int array_size = window * sizeof(struct unpacked);
1489 struct unpacked *array;
1490 int max_depth;
1491
1492 if (!nr_objects)
1493 return;
1494 array = xmalloc(array_size);
1495 memset(array, 0, array_size);
1496 if (progress)
1497 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1498
1499 do {
1500 struct object_entry *entry = list[--i];
1501 struct unpacked *n = array + idx;
1502 int j;
1503
1504 if (!entry->preferred_base)
1505 processed++;
1506
1507 if (progress)
1508 display_progress(&progress_state, processed);
1509
1510 if (entry->delta)
1511 /* This happens if we decided to reuse existing
1512 * delta from a pack. "!no_reuse_delta &&" is implied.
1513 */
1514 continue;
1515
1516 if (entry->size < 50)
1517 continue;
1518
1519 if (entry->no_try_delta)
1520 continue;
1521
1522 free_delta_index(n->index);
1523 n->index = NULL;
1524 free(n->data);
1525 n->data = NULL;
1526 n->entry = entry;
1527
1528 /*
1529 * If the current object is at pack edge, take the depth the
1530 * objects that depend on the current object into account
1531 * otherwise they would become too deep.
1532 */
1533 max_depth = depth;
1534 if (entry->delta_child) {
1535 max_depth -= check_delta_limit(entry, 0);
1536 if (max_depth <= 0)
1537 goto next;
1538 }
1539
1540 j = window;
1541 while (--j > 0) {
1542 uint32_t other_idx = idx + j;
1543 struct unpacked *m;
1544 if (other_idx >= window)
1545 other_idx -= window;
1546 m = array + other_idx;
1547 if (!m->entry)
1548 break;
1549 if (try_delta(n, m, max_depth) < 0)
1550 break;
1551 }
1552
1553 /* if we made n a delta, and if n is already at max
1554 * depth, leaving it in the window is pointless. we
1555 * should evict it first.
1556 */
1557 if (entry->delta && depth <= entry->depth)
1558 continue;
1559
1560 next:
1561 idx++;
1562 if (idx >= window)
1563 idx = 0;
1564 } while (i > 0);
1565
1566 if (progress)
1567 stop_progress(&progress_state);
1568
1569 for (i = 0; i < window; ++i) {
1570 free_delta_index(array[i].index);
1571 free(array[i].data);
1572 }
1573 free(array);
1574 }
1575
1576 static void prepare_pack(int window, int depth)
1577 {
1578 struct object_entry **delta_list;
1579 uint32_t i;
1580
1581 get_object_details();
1582
1583 if (!window || !depth)
1584 return;
1585
1586 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1587 for (i = 0; i < nr_objects; i++)
1588 delta_list[i] = objects + i;
1589 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1590 find_deltas(delta_list, window+1, depth);
1591 free(delta_list);
1592 }
1593
1594 static int git_pack_config(const char *k, const char *v)
1595 {
1596 if(!strcmp(k, "pack.window")) {
1597 window = git_config_int(k, v);
1598 return 0;
1599 }
1600 if(!strcmp(k, "pack.depth")) {
1601 depth = git_config_int(k, v);
1602 return 0;
1603 }
1604 if (!strcmp(k, "pack.compression")) {
1605 int level = git_config_int(k, v);
1606 if (level == -1)
1607 level = Z_DEFAULT_COMPRESSION;
1608 else if (level < 0 || level > Z_BEST_COMPRESSION)
1609 die("bad pack compression level %d", level);
1610 pack_compression_level = level;
1611 pack_compression_seen = 1;
1612 return 0;
1613 }
1614 return git_default_config(k, v);
1615 }
1616
1617 static void read_object_list_from_stdin(void)
1618 {
1619 char line[40 + 1 + PATH_MAX + 2];
1620 unsigned char sha1[20];
1621
1622 for (;;) {
1623 if (!fgets(line, sizeof(line), stdin)) {
1624 if (feof(stdin))
1625 break;
1626 if (!ferror(stdin))
1627 die("fgets returned NULL, not EOF, not error!");
1628 if (errno != EINTR)
1629 die("fgets: %s", strerror(errno));
1630 clearerr(stdin);
1631 continue;
1632 }
1633 if (line[0] == '-') {
1634 if (get_sha1_hex(line+1, sha1))
1635 die("expected edge sha1, got garbage:\n %s",
1636 line);
1637 add_preferred_base(sha1);
1638 continue;
1639 }
1640 if (get_sha1_hex(line, sha1))
1641 die("expected sha1, got garbage:\n %s", line);
1642
1643 add_preferred_base_object(line+41);
1644 add_object_entry(sha1, 0, line+41, 0);
1645 }
1646 }
1647
1648 static void show_commit(struct commit *commit)
1649 {
1650 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1651 }
1652
1653 static void show_object(struct object_array_entry *p)
1654 {
1655 add_preferred_base_object(p->name);
1656 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1657 }
1658
1659 static void show_edge(struct commit *commit)
1660 {
1661 add_preferred_base(commit->object.sha1);
1662 }
1663
1664 static void get_object_list(int ac, const char **av)
1665 {
1666 struct rev_info revs;
1667 char line[1000];
1668 int flags = 0;
1669
1670 init_revisions(&revs, NULL);
1671 save_commit_buffer = 0;
1672 track_object_refs = 0;
1673 setup_revisions(ac, av, &revs, NULL);
1674
1675 while (fgets(line, sizeof(line), stdin) != NULL) {
1676 int len = strlen(line);
1677 if (line[len - 1] == '\n')
1678 line[--len] = 0;
1679 if (!len)
1680 break;
1681 if (*line == '-') {
1682 if (!strcmp(line, "--not")) {
1683 flags ^= UNINTERESTING;
1684 continue;
1685 }
1686 die("not a rev '%s'", line);
1687 }
1688 if (handle_revision_arg(line, &revs, flags, 1))
1689 die("bad revision '%s'", line);
1690 }
1691
1692 prepare_revision_walk(&revs);
1693 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1694 traverse_commit_list(&revs, show_commit, show_object);
1695 }
1696
1697 static int adjust_perm(const char *path, mode_t mode)
1698 {
1699 if (chmod(path, mode))
1700 return -1;
1701 return adjust_shared_perm(path);
1702 }
1703
1704 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1705 {
1706 int use_internal_rev_list = 0;
1707 int thin = 0;
1708 uint32_t i;
1709 const char **rp_av;
1710 int rp_ac_alloc = 64;
1711 int rp_ac;
1712
1713 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1714
1715 rp_av[0] = "pack-objects";
1716 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1717 rp_ac = 2;
1718
1719 git_config(git_pack_config);
1720 if (!pack_compression_seen && core_compression_seen)
1721 pack_compression_level = core_compression_level;
1722
1723 progress = isatty(2);
1724 for (i = 1; i < argc; i++) {
1725 const char *arg = argv[i];
1726
1727 if (*arg != '-')
1728 break;
1729
1730 if (!strcmp("--non-empty", arg)) {
1731 non_empty = 1;
1732 continue;
1733 }
1734 if (!strcmp("--local", arg)) {
1735 local = 1;
1736 continue;
1737 }
1738 if (!strcmp("--incremental", arg)) {
1739 incremental = 1;
1740 continue;
1741 }
1742 if (!prefixcmp(arg, "--compression=")) {
1743 char *end;
1744 int level = strtoul(arg+14, &end, 0);
1745 if (!arg[14] || *end)
1746 usage(pack_usage);
1747 if (level == -1)
1748 level = Z_DEFAULT_COMPRESSION;
1749 else if (level < 0 || level > Z_BEST_COMPRESSION)
1750 die("bad pack compression level %d", level);
1751 pack_compression_level = level;
1752 continue;
1753 }
1754 if (!prefixcmp(arg, "--max-pack-size=")) {
1755 char *end;
1756 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1757 if (!arg[16] || *end)
1758 usage(pack_usage);
1759 continue;
1760 }
1761 if (!prefixcmp(arg, "--window=")) {
1762 char *end;
1763 window = strtoul(arg+9, &end, 0);
1764 if (!arg[9] || *end)
1765 usage(pack_usage);
1766 continue;
1767 }
1768 if (!prefixcmp(arg, "--depth=")) {
1769 char *end;
1770 depth = strtoul(arg+8, &end, 0);
1771 if (!arg[8] || *end)
1772 usage(pack_usage);
1773 continue;
1774 }
1775 if (!strcmp("--progress", arg)) {
1776 progress = 1;
1777 continue;
1778 }
1779 if (!strcmp("--all-progress", arg)) {
1780 progress = 2;
1781 continue;
1782 }
1783 if (!strcmp("-q", arg)) {
1784 progress = 0;
1785 continue;
1786 }
1787 if (!strcmp("--no-reuse-delta", arg)) {
1788 no_reuse_delta = 1;
1789 continue;
1790 }
1791 if (!strcmp("--no-reuse-object", arg)) {
1792 no_reuse_object = no_reuse_delta = 1;
1793 continue;
1794 }
1795 if (!strcmp("--delta-base-offset", arg)) {
1796 allow_ofs_delta = 1;
1797 continue;
1798 }
1799 if (!strcmp("--stdout", arg)) {
1800 pack_to_stdout = 1;
1801 continue;
1802 }
1803 if (!strcmp("--revs", arg)) {
1804 use_internal_rev_list = 1;
1805 continue;
1806 }
1807 if (!strcmp("--unpacked", arg) ||
1808 !prefixcmp(arg, "--unpacked=") ||
1809 !strcmp("--reflog", arg) ||
1810 !strcmp("--all", arg)) {
1811 use_internal_rev_list = 1;
1812 if (rp_ac >= rp_ac_alloc - 1) {
1813 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1814 rp_av = xrealloc(rp_av,
1815 rp_ac_alloc * sizeof(*rp_av));
1816 }
1817 rp_av[rp_ac++] = arg;
1818 continue;
1819 }
1820 if (!strcmp("--thin", arg)) {
1821 use_internal_rev_list = 1;
1822 thin = 1;
1823 rp_av[1] = "--objects-edge";
1824 continue;
1825 }
1826 if (!prefixcmp(arg, "--index-version=")) {
1827 char *c;
1828 index_default_version = strtoul(arg + 16, &c, 10);
1829 if (index_default_version > 2)
1830 die("bad %s", arg);
1831 if (*c == ',')
1832 index_off32_limit = strtoul(c+1, &c, 0);
1833 if (*c || index_off32_limit & 0x80000000)
1834 die("bad %s", arg);
1835 continue;
1836 }
1837 usage(pack_usage);
1838 }
1839
1840 /* Traditionally "pack-objects [options] base extra" failed;
1841 * we would however want to take refs parameter that would
1842 * have been given to upstream rev-list ourselves, which means
1843 * we somehow want to say what the base name is. So the
1844 * syntax would be:
1845 *
1846 * pack-objects [options] base <refs...>
1847 *
1848 * in other words, we would treat the first non-option as the
1849 * base_name and send everything else to the internal revision
1850 * walker.
1851 */
1852
1853 if (!pack_to_stdout)
1854 base_name = argv[i++];
1855
1856 if (pack_to_stdout != !base_name)
1857 usage(pack_usage);
1858
1859 if (pack_to_stdout && pack_size_limit)
1860 die("--max-pack-size cannot be used to build a pack for transfer.");
1861
1862 if (!pack_to_stdout && thin)
1863 die("--thin cannot be used to build an indexable pack.");
1864
1865 prepare_packed_git();
1866
1867 if (progress)
1868 start_progress(&progress_state, "Generating pack...",
1869 "Counting objects: ", 0);
1870 if (!use_internal_rev_list)
1871 read_object_list_from_stdin();
1872 else {
1873 rp_av[rp_ac] = NULL;
1874 get_object_list(rp_ac, rp_av);
1875 }
1876 if (progress) {
1877 stop_progress(&progress_state);
1878 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1879 }
1880
1881 if (non_empty && !nr_result)
1882 return 0;
1883 if (progress && (nr_objects != nr_result))
1884 fprintf(stderr, "Result has %u objects.\n", nr_result);
1885 if (nr_result)
1886 prepare_pack(window, depth);
1887 write_pack_file();
1888 if (progress)
1889 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1890 written, written_delta, reused, reused_delta);
1891 return 0;
1892 }
Shawn's Git tree