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"
17static const char pack_usage[] = "\
19git-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]";
24struct 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 void *delta_data; /* cached delta (uncompressed) */
40 unsigned long delta_size; /* delta data size (uncompressed) */
41 enum object_type type;
42 enum object_type in_pack_type; /* could be delta */
43 unsigned char in_pack_header_size;
44 unsigned char preferred_base; /* we do not pack this, but is available
45 * to be used as the base object to delta
46 * objects against.
47 */
48 unsigned char no_try_delta;
49};
50/*
52 * Objects we are going to pack are collected in objects array (dynamically
53 * expanded). nr_objects & nr_alloc controls this array. They are stored
54 * in the order we see -- typically rev-list --objects order that gives us
55 * nice "minimum seek" order.
56 */
57static struct object_entry *objects;
58static struct object_entry **written_list;
59static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
60static int non_empty;
62static int no_reuse_delta, no_reuse_object;
63static int local;
64static int incremental;
65static int allow_ofs_delta;
66static const char *pack_tmp_name, *idx_tmp_name;
67static char tmpname[PATH_MAX];
68static const char *base_name;
69static unsigned char pack_file_sha1[20];
70static int progress = 1;
71static int window = 10;
72static uint32_t pack_size_limit;
73static int depth = 50;
74static int pack_to_stdout;
75static int num_preferred_base;
76static struct progress progress_state;
77static int pack_compression_level = Z_DEFAULT_COMPRESSION;
78static int pack_compression_seen;
79static unsigned long delta_cache_size = 0;
81static unsigned long max_delta_cache_size = 0;
82/*
84 * The object names in objects array are hashed with this hashtable,
85 * to help looking up the entry by object name.
86 * This hashtable is built after all the objects are seen.
87 */
88static int *object_ix;
89static int object_ix_hashsz;
90/*
92 * Pack index for existing packs give us easy access to the offsets into
93 * corresponding pack file where each object's data starts, but the entries
94 * do not store the size of the compressed representation (uncompressed
95 * size is easily available by examining the pack entry header). It is
96 * also rather expensive to find the sha1 for an object given its offset.
97 *
98 * We build a hashtable of existing packs (pack_revindex), and keep reverse
99 * index here -- pack index file is sorted by object name mapping to offset;
100 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
101 * ordered by offset, so if you know the offset of an object, next offset
102 * is where its packed representation ends and the index_nr can be used to
103 * get the object sha1 from the main index.
104 */
105struct revindex_entry {
106 off_t offset;
107 unsigned int nr;
108};
109struct pack_revindex {
110 struct packed_git *p;
111 struct revindex_entry *revindex;
112};
113static struct pack_revindex *pack_revindex;
114static int pack_revindex_hashsz;
115/*
117 * stats
118 */
119static uint32_t written, written_delta;
120static uint32_t reused, reused_delta;
121static int pack_revindex_ix(struct packed_git *p)
123{
124 unsigned long ui = (unsigned long)p;
125 int i;
126 ui = ui ^ (ui >> 16); /* defeat structure alignment */
128 i = (int)(ui % pack_revindex_hashsz);
129 while (pack_revindex[i].p) {
130 if (pack_revindex[i].p == p)
131 return i;
132 if (++i == pack_revindex_hashsz)
133 i = 0;
134 }
135 return -1 - i;
136}
137static void prepare_pack_ix(void)
139{
140 int num;
141 struct packed_git *p;
142 for (num = 0, p = packed_git; p; p = p->next)
143 num++;
144 if (!num)
145 return;
146 pack_revindex_hashsz = num * 11;
147 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
148 for (p = packed_git; p; p = p->next) {
149 num = pack_revindex_ix(p);
150 num = - 1 - num;
151 pack_revindex[num].p = p;
152 }
153 /* revindex elements are lazily initialized */
154}
155static int cmp_offset(const void *a_, const void *b_)
157{
158 const struct revindex_entry *a = a_;
159 const struct revindex_entry *b = b_;
160 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
161}
162/*
164 * Ordered list of offsets of objects in the pack.
165 */
166static void prepare_pack_revindex(struct pack_revindex *rix)
167{
168 struct packed_git *p = rix->p;
169 int num_ent = p->num_objects;
170 int i;
171 const char *index = p->index_data;
172 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
174 index += 4 * 256;
175 if (p->index_version > 1) {
177 const uint32_t *off_32 =
178 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
179 const uint32_t *off_64 = off_32 + p->num_objects;
180 for (i = 0; i < num_ent; i++) {
181 uint32_t off = ntohl(*off_32++);
182 if (!(off & 0x80000000)) {
183 rix->revindex[i].offset = off;
184 } else {
185 rix->revindex[i].offset =
186 ((uint64_t)ntohl(*off_64++)) << 32;
187 rix->revindex[i].offset |=
188 ntohl(*off_64++);
189 }
190 rix->revindex[i].nr = i;
191 }
192 } else {
193 for (i = 0; i < num_ent; i++) {
194 uint32_t hl = *((uint32_t *)(index + 24 * i));
195 rix->revindex[i].offset = ntohl(hl);
196 rix->revindex[i].nr = i;
197 }
198 }
199 /* This knows the pack format -- the 20-byte trailer
201 * follows immediately after the last object data.
202 */
203 rix->revindex[num_ent].offset = p->pack_size - 20;
204 rix->revindex[num_ent].nr = -1;
205 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
206}
207static struct revindex_entry * find_packed_object(struct packed_git *p,
209 off_t ofs)
210{
211 int num;
212 int lo, hi;
213 struct pack_revindex *rix;
214 struct revindex_entry *revindex;
215 num = pack_revindex_ix(p);
216 if (num < 0)
217 die("internal error: pack revindex uninitialized");
218 rix = &pack_revindex[num];
219 if (!rix->revindex)
220 prepare_pack_revindex(rix);
221 revindex = rix->revindex;
222 lo = 0;
223 hi = p->num_objects + 1;
224 do {
225 int mi = (lo + hi) / 2;
226 if (revindex[mi].offset == ofs) {
227 return revindex + mi;
228 }
229 else if (ofs < revindex[mi].offset)
230 hi = mi;
231 else
232 lo = mi + 1;
233 } while (lo < hi);
234 die("internal error: pack revindex corrupt");
235}
236static const unsigned char *find_packed_object_name(struct packed_git *p,
238 off_t ofs)
239{
240 struct revindex_entry *entry = find_packed_object(p, ofs);
241 return nth_packed_object_sha1(p, entry->nr);
242}
243static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
245{
246 unsigned long othersize, delta_size;
247 enum object_type type;
248 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
249 void *delta_buf;
250 if (!otherbuf)
252 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
253 delta_buf = diff_delta(otherbuf, othersize,
254 buf, size, &delta_size, 0);
255 if (!delta_buf || delta_size != entry->delta_size)
256 die("delta size changed");
257 free(buf);
258 free(otherbuf);
259 return delta_buf;
260}
261/*
263 * The per-object header is a pretty dense thing, which is
264 * - first byte: low four bits are "size", then three bits of "type",
265 * and the high bit is "size continues".
266 * - each byte afterwards: low seven bits are size continuation,
267 * with the high bit being "size continues"
268 */
269static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
270{
271 int n = 1;
272 unsigned char c;
273 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
275 die("bad type %d", type);
276 c = (type << 4) | (size & 15);
278 size >>= 4;
279 while (size) {
280 *hdr++ = c | 0x80;
281 c = size & 0x7f;
282 size >>= 7;
283 n++;
284 }
285 *hdr = c;
286 return n;
287}
288/*
290 * we are going to reuse the existing object data as is. make
291 * sure it is not corrupt.
292 */
293static int check_pack_inflate(struct packed_git *p,
294 struct pack_window **w_curs,
295 off_t offset,
296 off_t len,
297 unsigned long expect)
298{
299 z_stream stream;
300 unsigned char fakebuf[4096], *in;
301 int st;
302 memset(&stream, 0, sizeof(stream));
304 inflateInit(&stream);
305 do {
306 in = use_pack(p, w_curs, offset, &stream.avail_in);
307 stream.next_in = in;
308 stream.next_out = fakebuf;
309 stream.avail_out = sizeof(fakebuf);
310 st = inflate(&stream, Z_FINISH);
311 offset += stream.next_in - in;
312 } while (st == Z_OK || st == Z_BUF_ERROR);
313 inflateEnd(&stream);
314 return (st == Z_STREAM_END &&
315 stream.total_out == expect &&
316 stream.total_in == len) ? 0 : -1;
317}
318static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
320 off_t offset, off_t len, unsigned int nr)
321{
322 const uint32_t *index_crc;
323 uint32_t data_crc = crc32(0, Z_NULL, 0);
324 do {
326 unsigned int avail;
327 void *data = use_pack(p, w_curs, offset, &avail);
328 if (avail > len)
329 avail = len;
330 data_crc = crc32(data_crc, data, avail);
331 offset += avail;
332 len -= avail;
333 } while (len);
334 index_crc = p->index_data;
336 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
337 return data_crc != ntohl(*index_crc);
339}
340static void copy_pack_data(struct sha1file *f,
342 struct packed_git *p,
343 struct pack_window **w_curs,
344 off_t offset,
345 off_t len)
346{
347 unsigned char *in;
348 unsigned int avail;
349 while (len) {
351 in = use_pack(p, w_curs, offset, &avail);
352 if (avail > len)
353 avail = (unsigned int)len;
354 sha1write(f, in, avail);
355 offset += avail;
356 len -= avail;
357 }
358}
359static unsigned long write_object(struct sha1file *f,
361 struct object_entry *entry,
362 off_t write_offset)
363{
364 unsigned long size;
365 enum object_type type;
366 void *buf;
367 unsigned char header[10];
368 unsigned char dheader[10];
369 unsigned hdrlen;
370 off_t datalen;
371 enum object_type obj_type;
372 int to_reuse = 0;
373 /* write limit if limited packsize and not first object */
374 unsigned long limit = pack_size_limit && nr_written ?
375 pack_size_limit - write_offset : 0;
376 /* no if no delta */
377 int usable_delta = !entry->delta ? 0 :
378 /* yes if unlimited packfile */
379 !pack_size_limit ? 1 :
380 /* no if base written to previous pack */
381 entry->delta->offset == (off_t)-1 ? 0 :
382 /* otherwise double-check written to this
383 * pack, like we do below
384 */
385 entry->delta->offset ? 1 : 0;
386 if (!pack_to_stdout)
388 crc32_begin(f);
389 obj_type = entry->type;
391 if (no_reuse_object)
392 to_reuse = 0; /* explicit */
393 else if (!entry->in_pack)
394 to_reuse = 0; /* can't reuse what we don't have */
395 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
396 /* check_object() decided it for us ... */
397 to_reuse = usable_delta;
398 /* ... but pack split may override that */
399 else if (obj_type != entry->in_pack_type)
400 to_reuse = 0; /* pack has delta which is unusable */
401 else if (entry->delta)
402 to_reuse = 0; /* we want to pack afresh */
403 else
404 to_reuse = 1; /* we have it in-pack undeltified,
405 * and we do not need to deltify it.
406 */
407 if (!to_reuse) {
409 z_stream stream;
410 unsigned long maxsize;
411 void *out;
412 if (entry->delta_data && usable_delta) {
413 buf = entry->delta_data;
414 size = entry->delta_size;
415 obj_type = (allow_ofs_delta && entry->delta->offset) ?
416 OBJ_OFS_DELTA : OBJ_REF_DELTA;
417 } else {
418 buf = read_sha1_file(entry->sha1, &type, &size);
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 (usable_delta) {
425 buf = delta_against(buf, size, entry);
426 size = entry->delta_size;
427 obj_type = (allow_ofs_delta && entry->delta->offset) ?
428 OBJ_OFS_DELTA : OBJ_REF_DELTA;
429 } else {
430 /*
431 * recover real object type in case
432 * check_object() wanted to re-use a delta,
433 * but we couldn't since base was in previous split pack
434 */
435 obj_type = type;
436 }
437 }
438 /* compress the data to store and put compressed length in datalen */
439 memset(&stream, 0, sizeof(stream));
440 deflateInit(&stream, pack_compression_level);
441 maxsize = deflateBound(&stream, size);
442 out = xmalloc(maxsize);
443 /* Compress it */
444 stream.next_in = buf;
445 stream.avail_in = size;
446 stream.next_out = out;
447 stream.avail_out = maxsize;
448 while (deflate(&stream, Z_FINISH) == Z_OK)
449 /* nothing */;
450 deflateEnd(&stream);
451 datalen = stream.total_out;
452 deflateEnd(&stream);
453 /*
454 * The object header is a byte of 'type' followed by zero or
455 * more bytes of length.
456 */
457 hdrlen = encode_header(obj_type, size, header);
458 if (obj_type == OBJ_OFS_DELTA) {
460 /*
461 * Deltas with relative base contain an additional
462 * encoding of the relative offset for the delta
463 * base from this object's position in the pack.
464 */
465 off_t ofs = entry->offset - entry->delta->offset;
466 unsigned pos = sizeof(dheader) - 1;
467 dheader[pos] = ofs & 127;
468 while (ofs >>= 7)
469 dheader[--pos] = 128 | (--ofs & 127);
470 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
471 free(out);
472 free(buf);
473 return 0;
474 }
475 sha1write(f, header, hdrlen);
476 sha1write(f, dheader + pos, sizeof(dheader) - pos);
477 hdrlen += sizeof(dheader) - pos;
478 } else if (obj_type == OBJ_REF_DELTA) {
479 /*
480 * Deltas with a base reference contain
481 * an additional 20 bytes for the base sha1.
482 */
483 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
484 free(out);
485 free(buf);
486 return 0;
487 }
488 sha1write(f, header, hdrlen);
489 sha1write(f, entry->delta->sha1, 20);
490 hdrlen += 20;
491 } else {
492 if (limit && hdrlen + datalen + 20 >= limit) {
493 free(out);
494 free(buf);
495 return 0;
496 }
497 sha1write(f, header, hdrlen);
498 }
499 sha1write(f, out, datalen);
500 free(out);
501 free(buf);
502 }
503 else {
504 struct packed_git *p = entry->in_pack;
505 struct pack_window *w_curs = NULL;
506 struct revindex_entry *revidx;
507 off_t offset;
508 if (entry->delta) {
510 obj_type = (allow_ofs_delta && entry->delta->offset) ?
511 OBJ_OFS_DELTA : OBJ_REF_DELTA;
512 reused_delta++;
513 }
514 hdrlen = encode_header(obj_type, entry->size, header);
515 offset = entry->in_pack_offset;
516 revidx = find_packed_object(p, offset);
517 datalen = revidx[1].offset - offset;
518 if (!pack_to_stdout && p->index_version > 1 &&
519 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
520 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
521 offset += entry->in_pack_header_size;
522 datalen -= entry->in_pack_header_size;
523 if (obj_type == OBJ_OFS_DELTA) {
524 off_t ofs = entry->offset - entry->delta->offset;
525 unsigned pos = sizeof(dheader) - 1;
526 dheader[pos] = ofs & 127;
527 while (ofs >>= 7)
528 dheader[--pos] = 128 | (--ofs & 127);
529 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
530 return 0;
531 sha1write(f, header, hdrlen);
532 sha1write(f, dheader + pos, sizeof(dheader) - pos);
533 hdrlen += sizeof(dheader) - pos;
534 } else if (obj_type == OBJ_REF_DELTA) {
535 if (limit && hdrlen + 20 + datalen + 20 >= limit)
536 return 0;
537 sha1write(f, header, hdrlen);
538 sha1write(f, entry->delta->sha1, 20);
539 hdrlen += 20;
540 } else {
541 if (limit && hdrlen + datalen + 20 >= limit)
542 return 0;
543 sha1write(f, header, hdrlen);
544 }
545 if (!pack_to_stdout && p->index_version == 1 &&
547 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
548 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
549 copy_pack_data(f, p, &w_curs, offset, datalen);
550 unuse_pack(&w_curs);
551 reused++;
552 }
553 if (usable_delta)
554 written_delta++;
555 written++;
556 if (!pack_to_stdout)
557 entry->crc32 = crc32_end(f);
558 return hdrlen + datalen;
559}
560static off_t write_one(struct sha1file *f,
562 struct object_entry *e,
563 off_t offset)
564{
565 unsigned long size;
566 /* offset is non zero if object is written already. */
568 if (e->offset || e->preferred_base)
569 return offset;
570 /* if we are deltified, write out base object first. */
572 if (e->delta) {
573 offset = write_one(f, e->delta, offset);
574 if (!offset)
575 return 0;
576 }
577 e->offset = offset;
579 size = write_object(f, e, offset);
580 if (!size) {
581 e->offset = 0;
582 return 0;
583 }
584 written_list[nr_written++] = e;
585 /* make sure off_t is sufficiently large not to wrap */
587 if (offset > offset + size)
588 die("pack too large for current definition of off_t");
589 return offset + size;
590}
591static int open_object_dir_tmp(const char *path)
593{
594 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
595 return mkstemp(tmpname);
596}
597/* forward declarations for write_pack_file */
599static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
600static int adjust_perm(const char *path, mode_t mode);
601static void write_pack_file(void)
603{
604 uint32_t i = 0, j;
605 struct sha1file *f;
606 off_t offset, offset_one, last_obj_offset = 0;
607 struct pack_header hdr;
608 int do_progress = progress >> pack_to_stdout;
609 uint32_t nr_remaining = nr_result;
610 if (do_progress)
612 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
613 written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
614 do {
616 if (pack_to_stdout) {
617 f = sha1fd(1, "<stdout>");
618 } else {
619 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
620 if (fd < 0)
621 die("unable to create %s: %s\n", tmpname, strerror(errno));
622 pack_tmp_name = xstrdup(tmpname);
623 f = sha1fd(fd, pack_tmp_name);
624 }
625 hdr.hdr_signature = htonl(PACK_SIGNATURE);
627 hdr.hdr_version = htonl(PACK_VERSION);
628 hdr.hdr_entries = htonl(nr_remaining);
629 sha1write(f, &hdr, sizeof(hdr));
630 offset = sizeof(hdr);
631 nr_written = 0;
632 for (; i < nr_objects; i++) {
633 last_obj_offset = offset;
634 offset_one = write_one(f, objects + i, offset);
635 if (!offset_one)
636 break;
637 offset = offset_one;
638 if (do_progress)
639 display_progress(&progress_state, written);
640 }
641 /*
643 * Did we write the wrong # entries in the header?
644 * If so, rewrite it like in fast-import
645 */
646 if (pack_to_stdout || nr_written == nr_remaining) {
647 sha1close(f, pack_file_sha1, 1);
648 } else {
649 sha1close(f, pack_file_sha1, 0);
650 fixup_pack_header_footer(f->fd, pack_file_sha1, pack_tmp_name, nr_written);
651 close(f->fd);
652 }
653 if (!pack_to_stdout) {
655 unsigned char object_list_sha1[20];
656 mode_t mode = umask(0);
657 umask(mode);
659 mode = 0444 & ~mode;
660 write_index_file(last_obj_offset, object_list_sha1);
662 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
663 base_name, sha1_to_hex(object_list_sha1));
664 if (adjust_perm(pack_tmp_name, mode))
665 die("unable to make temporary pack file readable: %s",
666 strerror(errno));
667 if (rename(pack_tmp_name, tmpname))
668 die("unable to rename temporary pack file: %s",
669 strerror(errno));
670 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
671 base_name, sha1_to_hex(object_list_sha1));
672 if (adjust_perm(idx_tmp_name, mode))
673 die("unable to make temporary index file readable: %s",
674 strerror(errno));
675 if (rename(idx_tmp_name, tmpname))
676 die("unable to rename temporary index file: %s",
677 strerror(errno));
678 puts(sha1_to_hex(object_list_sha1));
679 }
680 /* mark written objects as written to previous pack */
682 for (j = 0; j < nr_written; j++) {
683 written_list[j]->offset = (off_t)-1;
684 }
685 nr_remaining -= nr_written;
686 } while (nr_remaining && i < nr_objects);
687 free(written_list);
689 if (do_progress)
690 stop_progress(&progress_state);
691 if (written != nr_result)
692 die("wrote %u objects while expecting %u", written, nr_result);
693 /*
694 * We have scanned through [0 ... i). Since we have written
695 * the correct number of objects, the remaining [i ... nr_objects)
696 * items must be either already written (due to out-of-order delta base)
697 * or a preferred base. Count those which are neither and complain if any.
698 */
699 for (j = 0; i < nr_objects; i++) {
700 struct object_entry *e = objects + i;
701 j += !e->offset && !e->preferred_base;
702 }
703 if (j)
704 die("wrote %u objects as expected but %u unwritten", written, j);
705}
706static int sha1_sort(const void *_a, const void *_b)
708{
709 const struct object_entry *a = *(struct object_entry **)_a;
710 const struct object_entry *b = *(struct object_entry **)_b;
711 return hashcmp(a->sha1, b->sha1);
712}
713static uint32_t index_default_version = 1;
715static uint32_t index_off32_limit = 0x7fffffff;
716static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
718{
719 struct sha1file *f;
720 struct object_entry **sorted_by_sha, **list, **last;
721 uint32_t array[256];
722 uint32_t i, index_version;
723 SHA_CTX ctx;
724 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
726 if (fd < 0)
727 die("unable to create %s: %s\n", tmpname, strerror(errno));
728 idx_tmp_name = xstrdup(tmpname);
729 f = sha1fd(fd, idx_tmp_name);
730 if (nr_written) {
732 sorted_by_sha = written_list;
733 qsort(sorted_by_sha, nr_written, sizeof(*sorted_by_sha), sha1_sort);
734 list = sorted_by_sha;
735 last = sorted_by_sha + nr_written;
736 } else
737 sorted_by_sha = list = last = NULL;
738 /* if last object's offset is >= 2^31 we should use index V2 */
740 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
741 /* index versions 2 and above need a header */
743 if (index_version >= 2) {
744 struct pack_idx_header hdr;
745 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
746 hdr.idx_version = htonl(index_version);
747 sha1write(f, &hdr, sizeof(hdr));
748 }
749 /*
751 * Write the first-level table (the list is sorted,
752 * but we use a 256-entry lookup to be able to avoid
753 * having to do eight extra binary search iterations).
754 */
755 for (i = 0; i < 256; i++) {
756 struct object_entry **next = list;
757 while (next < last) {
758 struct object_entry *entry = *next;
759 if (entry->sha1[0] != i)
760 break;
761 next++;
762 }
763 array[i] = htonl(next - sorted_by_sha);
764 list = next;
765 }
766 sha1write(f, array, 256 * 4);
767 /* Compute the SHA1 hash of sorted object names. */
769 SHA1_Init(&ctx);
770 /* Write the actual SHA1 entries. */
772 list = sorted_by_sha;
773 for (i = 0; i < nr_written; i++) {
774 struct object_entry *entry = *list++;
775 if (index_version < 2) {
776 uint32_t offset = htonl(entry->offset);
777 sha1write(f, &offset, 4);
778 }
779 sha1write(f, entry->sha1, 20);
780 SHA1_Update(&ctx, entry->sha1, 20);
781 }
782 if (index_version >= 2) {
784 unsigned int nr_large_offset = 0;
785 /* write the crc32 table */
787 list = sorted_by_sha;
788 for (i = 0; i < nr_written; i++) {
789 struct object_entry *entry = *list++;
790 uint32_t crc32_val = htonl(entry->crc32);
791 sha1write(f, &crc32_val, 4);
792 }
793 /* write the 32-bit offset table */
795 list = sorted_by_sha;
796 for (i = 0; i < nr_written; i++) {
797 struct object_entry *entry = *list++;
798 uint32_t offset = (entry->offset <= index_off32_limit) ?
799 entry->offset : (0x80000000 | nr_large_offset++);
800 offset = htonl(offset);
801 sha1write(f, &offset, 4);
802 }
803 /* write the large offset table */
805 list = sorted_by_sha;
806 while (nr_large_offset) {
807 struct object_entry *entry = *list++;
808 uint64_t offset = entry->offset;
809 if (offset > index_off32_limit) {
810 uint32_t split[2];
811 split[0] = htonl(offset >> 32);
812 split[1] = htonl(offset & 0xffffffff);
813 sha1write(f, split, 8);
814 nr_large_offset--;
815 }
816 }
817 }
818 sha1write(f, pack_file_sha1, 20);
820 sha1close(f, NULL, 1);
821 SHA1_Final(sha1, &ctx);
822}
823static int locate_object_entry_hash(const unsigned char *sha1)
825{
826 int i;
827 unsigned int ui;
828 memcpy(&ui, sha1, sizeof(unsigned int));
829 i = ui % object_ix_hashsz;
830 while (0 < object_ix[i]) {
831 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
832 return i;
833 if (++i == object_ix_hashsz)
834 i = 0;
835 }
836 return -1 - i;
837}
838static struct object_entry *locate_object_entry(const unsigned char *sha1)
840{
841 int i;
842 if (!object_ix_hashsz)
844 return NULL;
845 i = locate_object_entry_hash(sha1);
847 if (0 <= i)
848 return &objects[object_ix[i]-1];
849 return NULL;
850}
851static void rehash_objects(void)
853{
854 uint32_t i;
855 struct object_entry *oe;
856 object_ix_hashsz = nr_objects * 3;
858 if (object_ix_hashsz < 1024)
859 object_ix_hashsz = 1024;
860 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
861 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
862 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
863 int ix = locate_object_entry_hash(oe->sha1);
864 if (0 <= ix)
865 continue;
866 ix = -1 - ix;
867 object_ix[ix] = i + 1;
868 }
869}
870static unsigned name_hash(const char *name)
872{
873 unsigned char c;
874 unsigned hash = 0;
875 if (!name)
877 return 0;
878 /*
880 * This effectively just creates a sortable number from the
881 * last sixteen non-whitespace characters. Last characters
882 * count "most", so things that end in ".c" sort together.
883 */
884 while ((c = *name++) != 0) {
885 if (isspace(c))
886 continue;
887 hash = (hash >> 2) + (c << 24);
888 }
889 return hash;
890}
891static void setup_delta_attr_check(struct git_attr_check *check)
893{
894 static struct git_attr *attr_delta;
895 if (!attr_delta)
897 attr_delta = git_attr("delta", 5);
898 check[0].attr = attr_delta;
900}
901static int no_try_delta(const char *path)
903{
904 struct git_attr_check check[1];
905 setup_delta_attr_check(check);
907 if (git_checkattr(path, ARRAY_SIZE(check), check))
908 return 0;
909 if (ATTR_FALSE(check->value))
910 return 1;
911 return 0;
912}
913static int add_object_entry(const unsigned char *sha1, enum object_type type,
915 const char *name, int exclude)
916{
917 struct object_entry *entry;
918 struct packed_git *p, *found_pack = NULL;
919 off_t found_offset = 0;
920 int ix;
921 unsigned hash = name_hash(name);
922 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
924 if (ix >= 0) {
925 if (exclude) {
926 entry = objects + object_ix[ix] - 1;
927 if (!entry->preferred_base)
928 nr_result--;
929 entry->preferred_base = 1;
930 }
931 return 0;
932 }
933 for (p = packed_git; p; p = p->next) {
935 off_t offset = find_pack_entry_one(sha1, p);
936 if (offset) {
937 if (!found_pack) {
938 found_offset = offset;
939 found_pack = p;
940 }
941 if (exclude)
942 break;
943 if (incremental)
944 return 0;
945 if (local && !p->pack_local)
946 return 0;
947 }
948 }
949 if (nr_objects >= nr_alloc) {
951 nr_alloc = (nr_alloc + 1024) * 3 / 2;
952 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
953 }
954 entry = objects + nr_objects++;
956 memset(entry, 0, sizeof(*entry));
957 hashcpy(entry->sha1, sha1);
958 entry->hash = hash;
959 if (type)
960 entry->type = type;
961 if (exclude)
962 entry->preferred_base = 1;
963 else
964 nr_result++;
965 if (found_pack) {
966 entry->in_pack = found_pack;
967 entry->in_pack_offset = found_offset;
968 }
969 if (object_ix_hashsz * 3 <= nr_objects * 4)
971 rehash_objects();
972 else
973 object_ix[-1 - ix] = nr_objects;
974 if (progress)
976 display_progress(&progress_state, nr_objects);
977 if (name && no_try_delta(name))
979 entry->no_try_delta = 1;
980 return 1;
982}
983struct pbase_tree_cache {
985 unsigned char sha1[20];
986 int ref;
987 int temporary;
988 void *tree_data;
989 unsigned long tree_size;
990};
991static struct pbase_tree_cache *(pbase_tree_cache[256]);
993static int pbase_tree_cache_ix(const unsigned char *sha1)
994{
995 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
996}
997static int pbase_tree_cache_ix_incr(int ix)
998{
999 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1000}
1001static struct pbase_tree {
1003 struct pbase_tree *next;
1004 /* This is a phony "cache" entry; we are not
1005 * going to evict it nor find it through _get()
1006 * mechanism -- this is for the toplevel node that
1007 * would almost always change with any commit.
1008 */
1009 struct pbase_tree_cache pcache;
1010} *pbase_tree;
1011static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1013{
1014 struct pbase_tree_cache *ent, *nent;
1015 void *data;
1016 unsigned long size;
1017 enum object_type type;
1018 int neigh;
1019 int my_ix = pbase_tree_cache_ix(sha1);
1020 int available_ix = -1;
1021 /* pbase-tree-cache acts as a limited hashtable.
1023 * your object will be found at your index or within a few
1024 * slots after that slot if it is cached.
1025 */
1026 for (neigh = 0; neigh < 8; neigh++) {
1027 ent = pbase_tree_cache[my_ix];
1028 if (ent && !hashcmp(ent->sha1, sha1)) {
1029 ent->ref++;
1030 return ent;
1031 }
1032 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1033 ((0 <= available_ix) &&
1034 (!ent && pbase_tree_cache[available_ix])))
1035 available_ix = my_ix;
1036 if (!ent)
1037 break;
1038 my_ix = pbase_tree_cache_ix_incr(my_ix);
1039 }
1040 /* Did not find one. Either we got a bogus request or
1042 * we need to read and perhaps cache.
1043 */
1044 data = read_sha1_file(sha1, &type, &size);
1045 if (!data)
1046 return NULL;
1047 if (type != OBJ_TREE) {
1048 free(data);
1049 return NULL;
1050 }
1051 /* We need to either cache or return a throwaway copy */
1053 if (available_ix < 0)
1055 ent = NULL;
1056 else {
1057 ent = pbase_tree_cache[available_ix];
1058 my_ix = available_ix;
1059 }
1060 if (!ent) {
1062 nent = xmalloc(sizeof(*nent));
1063 nent->temporary = (available_ix < 0);
1064 }
1065 else {
1066 /* evict and reuse */
1067 free(ent->tree_data);
1068 nent = ent;
1069 }
1070 hashcpy(nent->sha1, sha1);
1071 nent->tree_data = data;
1072 nent->tree_size = size;
1073 nent->ref = 1;
1074 if (!nent->temporary)
1075 pbase_tree_cache[my_ix] = nent;
1076 return nent;
1077}
1078static void pbase_tree_put(struct pbase_tree_cache *cache)
1080{
1081 if (!cache->temporary) {
1082 cache->ref--;
1083 return;
1084 }
1085 free(cache->tree_data);
1086 free(cache);
1087}
1088static int name_cmp_len(const char *name)
1090{
1091 int i;
1092 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1093 ;
1094 return i;
1095}
1096static void add_pbase_object(struct tree_desc *tree,
1098 const char *name,
1099 int cmplen,
1100 const char *fullname)
1101{
1102 struct name_entry entry;
1103 int cmp;
1104 while (tree_entry(tree,&entry)) {
1106 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1107 memcmp(name, entry.path, cmplen);
1108 if (cmp > 0)
1109 continue;
1110 if (cmp < 0)
1111 return;
1112 if (name[cmplen] != '/') {
1113 add_object_entry(entry.sha1,
1114 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1115 fullname, 1);
1116 return;
1117 }
1118 if (S_ISDIR(entry.mode)) {
1119 struct tree_desc sub;
1120 struct pbase_tree_cache *tree;
1121 const char *down = name+cmplen+1;
1122 int downlen = name_cmp_len(down);
1123 tree = pbase_tree_get(entry.sha1);
1125 if (!tree)
1126 return;
1127 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1128 add_pbase_object(&sub, down, downlen, fullname);
1130 pbase_tree_put(tree);
1131 }
1132 }
1133}
1134static unsigned *done_pbase_paths;
1136static int done_pbase_paths_num;
1137static int done_pbase_paths_alloc;
1138static int done_pbase_path_pos(unsigned hash)
1139{
1140 int lo = 0;
1141 int hi = done_pbase_paths_num;
1142 while (lo < hi) {
1143 int mi = (hi + lo) / 2;
1144 if (done_pbase_paths[mi] == hash)
1145 return mi;
1146 if (done_pbase_paths[mi] < hash)
1147 hi = mi;
1148 else
1149 lo = mi + 1;
1150 }
1151 return -lo-1;
1152}
1153static int check_pbase_path(unsigned hash)
1155{
1156 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1157 if (0 <= pos)
1158 return 1;
1159 pos = -pos - 1;
1160 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1161 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1162 done_pbase_paths = xrealloc(done_pbase_paths,
1163 done_pbase_paths_alloc *
1164 sizeof(unsigned));
1165 }
1166 done_pbase_paths_num++;
1167 if (pos < done_pbase_paths_num)
1168 memmove(done_pbase_paths + pos + 1,
1169 done_pbase_paths + pos,
1170 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1171 done_pbase_paths[pos] = hash;
1172 return 0;
1173}
1174static void add_preferred_base_object(const char *name)
1176{
1177 struct pbase_tree *it;
1178 int cmplen;
1179 unsigned hash = name_hash(name);
1180 if (!num_preferred_base || check_pbase_path(hash))
1182 return;
1183 cmplen = name_cmp_len(name);
1185 for (it = pbase_tree; it; it = it->next) {
1186 if (cmplen == 0) {
1187 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1188 }
1189 else {
1190 struct tree_desc tree;
1191 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1192 add_pbase_object(&tree, name, cmplen, name);
1193 }
1194 }
1195}
1196static void add_preferred_base(unsigned char *sha1)
1198{
1199 struct pbase_tree *it;
1200 void *data;
1201 unsigned long size;
1202 unsigned char tree_sha1[20];
1203 if (window <= num_preferred_base++)
1205 return;
1206 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1208 if (!data)
1209 return;
1210 for (it = pbase_tree; it; it = it->next) {
1212 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1213 free(data);
1214 return;
1215 }
1216 }
1217 it = xcalloc(1, sizeof(*it));
1219 it->next = pbase_tree;
1220 pbase_tree = it;
1221 hashcpy(it->pcache.sha1, tree_sha1);
1223 it->pcache.tree_data = data;
1224 it->pcache.tree_size = size;
1225}
1226static void check_object(struct object_entry *entry)
1228{
1229 if (entry->in_pack) {
1230 struct packed_git *p = entry->in_pack;
1231 struct pack_window *w_curs = NULL;
1232 const unsigned char *base_ref = NULL;
1233 struct object_entry *base_entry;
1234 unsigned long used, used_0;
1235 unsigned int avail;
1236 off_t ofs;
1237 unsigned char *buf, c;
1238 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1240 /*
1242 * We want in_pack_type even if we do not reuse delta
1243 * since non-delta representations could still be reused.
1244 */
1245 used = unpack_object_header_gently(buf, avail,
1246 &entry->in_pack_type,
1247 &entry->size);
1248 /*
1250 * Determine if this is a delta and if so whether we can
1251 * reuse it or not. Otherwise let's find out as cheaply as
1252 * possible what the actual type and size for this object is.
1253 */
1254 switch (entry->in_pack_type) {
1255 default:
1256 /* Not a delta hence we've already got all we need. */
1257 entry->type = entry->in_pack_type;
1258 entry->in_pack_header_size = used;
1259 unuse_pack(&w_curs);
1260 return;
1261 case OBJ_REF_DELTA:
1262 if (!no_reuse_delta && !entry->preferred_base)
1263 base_ref = use_pack(p, &w_curs,
1264 entry->in_pack_offset + used, NULL);
1265 entry->in_pack_header_size = used + 20;
1266 break;
1267 case OBJ_OFS_DELTA:
1268 buf = use_pack(p, &w_curs,
1269 entry->in_pack_offset + used, NULL);
1270 used_0 = 0;
1271 c = buf[used_0++];
1272 ofs = c & 127;
1273 while (c & 128) {
1274 ofs += 1;
1275 if (!ofs || MSB(ofs, 7))
1276 die("delta base offset overflow in pack for %s",
1277 sha1_to_hex(entry->sha1));
1278 c = buf[used_0++];
1279 ofs = (ofs << 7) + (c & 127);
1280 }
1281 if (ofs >= entry->in_pack_offset)
1282 die("delta base offset out of bound for %s",
1283 sha1_to_hex(entry->sha1));
1284 ofs = entry->in_pack_offset - ofs;
1285 if (!no_reuse_delta && !entry->preferred_base)
1286 base_ref = find_packed_object_name(p, ofs);
1287 entry->in_pack_header_size = used + used_0;
1288 break;
1289 }
1290 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1292 /*
1293 * If base_ref was set above that means we wish to
1294 * reuse delta data, and we even found that base
1295 * in the list of objects we want to pack. Goodie!
1296 *
1297 * Depth value does not matter - find_deltas() will
1298 * never consider reused delta as the base object to
1299 * deltify other objects against, in order to avoid
1300 * circular deltas.
1301 */
1302 entry->type = entry->in_pack_type;
1303 entry->delta = base_entry;
1304 entry->delta_sibling = base_entry->delta_child;
1305 base_entry->delta_child = entry;
1306 unuse_pack(&w_curs);
1307 return;
1308 }
1309 if (entry->type) {
1311 /*
1312 * This must be a delta and we already know what the
1313 * final object type is. Let's extract the actual
1314 * object size from the delta header.
1315 */
1316 entry->size = get_size_from_delta(p, &w_curs,
1317 entry->in_pack_offset + entry->in_pack_header_size);
1318 unuse_pack(&w_curs);
1319 return;
1320 }
1321 /*
1323 * No choice but to fall back to the recursive delta walk
1324 * with sha1_object_info() to find about the object type
1325 * at this point...
1326 */
1327 unuse_pack(&w_curs);
1328 }
1329 entry->type = sha1_object_info(entry->sha1, &entry->size);
1331 if (entry->type < 0)
1332 die("unable to get type of object %s",
1333 sha1_to_hex(entry->sha1));
1334}
1335static int pack_offset_sort(const void *_a, const void *_b)
1337{
1338 const struct object_entry *a = *(struct object_entry **)_a;
1339 const struct object_entry *b = *(struct object_entry **)_b;
1340 /* avoid filesystem trashing with loose objects */
1342 if (!a->in_pack && !b->in_pack)
1343 return hashcmp(a->sha1, b->sha1);
1344 if (a->in_pack < b->in_pack)
1346 return -1;
1347 if (a->in_pack > b->in_pack)
1348 return 1;
1349 return a->in_pack_offset < b->in_pack_offset ? -1 :
1350 (a->in_pack_offset > b->in_pack_offset);
1351}
1352static void get_object_details(void)
1354{
1355 uint32_t i;
1356 struct object_entry **sorted_by_offset;
1357 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1359 for (i = 0; i < nr_objects; i++)
1360 sorted_by_offset[i] = objects + i;
1361 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1362 prepare_pack_ix();
1364 for (i = 0; i < nr_objects; i++)
1365 check_object(sorted_by_offset[i]);
1366 free(sorted_by_offset);
1367}
1368static int type_size_sort(const void *_a, const void *_b)
1370{
1371 const struct object_entry *a = *(struct object_entry **)_a;
1372 const struct object_entry *b = *(struct object_entry **)_b;
1373 if (a->type < b->type)
1375 return -1;
1376 if (a->type > b->type)
1377 return 1;
1378 if (a->hash < b->hash)
1379 return -1;
1380 if (a->hash > b->hash)
1381 return 1;
1382 if (a->preferred_base < b->preferred_base)
1383 return -1;
1384 if (a->preferred_base > b->preferred_base)
1385 return 1;
1386 if (a->size < b->size)
1387 return -1;
1388 if (a->size > b->size)
1389 return 1;
1390 return a > b ? -1 : (a < b); /* newest last */
1391}
1392struct unpacked {
1394 struct object_entry *entry;
1395 void *data;
1396 struct delta_index *index;
1397};
1398static int delta_cacheable(struct unpacked *trg, struct unpacked *src,
1400 unsigned long src_size, unsigned long trg_size,
1401 unsigned long delta_size)
1402{
1403 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1404 return 0;
1405 /* cache delta, if objects are large enough compared to delta size */
1407 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1408 return 1;
1409 return 0;
1411}
1412/*
1414 * We search for deltas _backwards_ in a list sorted by type and
1415 * by size, so that we see progressively smaller and smaller files.
1416 * That's because we prefer deltas to be from the bigger file
1417 * to the smaller - deletes are potentially cheaper, but perhaps
1418 * more importantly, the bigger file is likely the more recent
1419 * one.
1420 */
1421static int try_delta(struct unpacked *trg, struct unpacked *src,
1422 unsigned max_depth)
1423{
1424 struct object_entry *trg_entry = trg->entry;
1425 struct object_entry *src_entry = src->entry;
1426 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1427 enum object_type type;
1428 void *delta_buf;
1429 /* Don't bother doing diffs between different types */
1431 if (trg_entry->type != src_entry->type)
1432 return -1;
1433 /* We do not compute delta to *create* objects we are not
1435 * going to pack.
1436 */
1437 if (trg_entry->preferred_base)
1438 return -1;
1439 /*
1441 * We do not bother to try a delta that we discarded
1442 * on an earlier try, but only when reusing delta data.
1443 */
1444 if (!no_reuse_delta && trg_entry->in_pack &&
1445 trg_entry->in_pack == src_entry->in_pack &&
1446 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1447 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1448 return 0;
1449 /* Let's not bust the allowed depth. */
1451 if (src_entry->depth >= max_depth)
1452 return 0;
1453 /* Now some size filtering heuristics. */
1455 trg_size = trg_entry->size;
1456 max_size = trg_size/2 - 20;
1457 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1458 if (max_size == 0)
1459 return 0;
1460 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1461 max_size = trg_entry->delta_size-1;
1462 src_size = src_entry->size;
1463 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1464 if (sizediff >= max_size)
1465 return 0;
1466 /* Load data if not already done */
1468 if (!trg->data) {
1469 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1470 if (sz != trg_size)
1471 die("object %s inconsistent object length (%lu vs %lu)",
1472 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1473 }
1474 if (!src->data) {
1475 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1476 if (sz != src_size)
1477 die("object %s inconsistent object length (%lu vs %lu)",
1478 sha1_to_hex(src_entry->sha1), sz, src_size);
1479 }
1480 if (!src->index) {
1481 src->index = create_delta_index(src->data, src_size);
1482 if (!src->index) {
1483 static int warned = 0;
1484 if (!warned++)
1485 warning("suboptimal pack - out of memory");
1486 return 0;
1487 }
1488 }
1489 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1491 if (!delta_buf)
1492 return 0;
1493 if (trg_entry->delta_data) {
1495 delta_cache_size -= trg_entry->delta_size;
1496 free(trg_entry->delta_data);
1497 }
1498 trg_entry->delta_data = 0;
1499 trg_entry->delta = src_entry;
1500 trg_entry->delta_size = delta_size;
1501 trg_entry->depth = src_entry->depth + 1;
1502 if (delta_cacheable(src, trg, src_size, trg_size, delta_size)) {
1504 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1505 delta_cache_size += trg_entry->delta_size;
1506 } else
1507 free(delta_buf);
1508 return 1;
1509}
1510static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1512{
1513 struct object_entry *child = me->delta_child;
1514 unsigned int m = n;
1515 while (child) {
1516 unsigned int c = check_delta_limit(child, n + 1);
1517 if (m < c)
1518 m = c;
1519 child = child->delta_sibling;
1520 }
1521 return m;
1522}
1523static void find_deltas(struct object_entry **list, int window, int depth)
1525{
1526 uint32_t i = nr_objects, idx = 0, processed = 0;
1527 unsigned int array_size = window * sizeof(struct unpacked);
1528 struct unpacked *array;
1529 int max_depth;
1530 if (!nr_objects)
1532 return;
1533 array = xmalloc(array_size);
1534 memset(array, 0, array_size);
1535 if (progress)
1536 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1537 do {
1539 struct object_entry *entry = list[--i];
1540 struct unpacked *n = array + idx;
1541 int j;
1542 if (!entry->preferred_base)
1544 processed++;
1545 if (progress)
1547 display_progress(&progress_state, processed);
1548 if (entry->delta)
1550 /* This happens if we decided to reuse existing
1551 * delta from a pack. "!no_reuse_delta &&" is implied.
1552 */
1553 continue;
1554 if (entry->size < 50)
1556 continue;
1557 if (entry->no_try_delta)
1559 continue;
1560 free_delta_index(n->index);
1562 n->index = NULL;
1563 free(n->data);
1564 n->data = NULL;
1565 n->entry = entry;
1566 /*
1568 * If the current object is at pack edge, take the depth the
1569 * objects that depend on the current object into account
1570 * otherwise they would become too deep.
1571 */
1572 max_depth = depth;
1573 if (entry->delta_child) {
1574 max_depth -= check_delta_limit(entry, 0);
1575 if (max_depth <= 0)
1576 goto next;
1577 }
1578 j = window;
1580 while (--j > 0) {
1581 uint32_t other_idx = idx + j;
1582 struct unpacked *m;
1583 if (other_idx >= window)
1584 other_idx -= window;
1585 m = array + other_idx;
1586 if (!m->entry)
1587 break;
1588 if (try_delta(n, m, max_depth) < 0)
1589 break;
1590 }
1591 /* if we made n a delta, and if n is already at max
1593 * depth, leaving it in the window is pointless. we
1594 * should evict it first.
1595 */
1596 if (entry->delta && depth <= entry->depth)
1597 continue;
1598 next:
1600 idx++;
1601 if (idx >= window)
1602 idx = 0;
1603 } while (i > 0);
1604 if (progress)
1606 stop_progress(&progress_state);
1607 for (i = 0; i < window; ++i) {
1609 free_delta_index(array[i].index);
1610 free(array[i].data);
1611 }
1612 free(array);
1613}
1614static void prepare_pack(int window, int depth)
1616{
1617 struct object_entry **delta_list;
1618 uint32_t i;
1619 get_object_details();
1621 if (!window || !depth)
1623 return;
1624 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1626 for (i = 0; i < nr_objects; i++)
1627 delta_list[i] = objects + i;
1628 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1629 find_deltas(delta_list, window+1, depth);
1630 free(delta_list);
1631}
1632static int git_pack_config(const char *k, const char *v)
1634{
1635 if(!strcmp(k, "pack.window")) {
1636 window = git_config_int(k, v);
1637 return 0;
1638 }
1639 if(!strcmp(k, "pack.depth")) {
1640 depth = git_config_int(k, v);
1641 return 0;
1642 }
1643 if (!strcmp(k, "pack.compression")) {
1644 int level = git_config_int(k, v);
1645 if (level == -1)
1646 level = Z_DEFAULT_COMPRESSION;
1647 else if (level < 0 || level > Z_BEST_COMPRESSION)
1648 die("bad pack compression level %d", level);
1649 pack_compression_level = level;
1650 pack_compression_seen = 1;
1651 return 0;
1652 }
1653 if (!strcmp(k, "pack.deltacachesize")) {
1654 max_delta_cache_size = git_config_int(k, v);
1655 return 0;
1656 }
1657 return git_default_config(k, v);
1658}
1659static void read_object_list_from_stdin(void)
1661{
1662 char line[40 + 1 + PATH_MAX + 2];
1663 unsigned char sha1[20];
1664 for (;;) {
1666 if (!fgets(line, sizeof(line), stdin)) {
1667 if (feof(stdin))
1668 break;
1669 if (!ferror(stdin))
1670 die("fgets returned NULL, not EOF, not error!");
1671 if (errno != EINTR)
1672 die("fgets: %s", strerror(errno));
1673 clearerr(stdin);
1674 continue;
1675 }
1676 if (line[0] == '-') {
1677 if (get_sha1_hex(line+1, sha1))
1678 die("expected edge sha1, got garbage:\n %s",
1679 line);
1680 add_preferred_base(sha1);
1681 continue;
1682 }
1683 if (get_sha1_hex(line, sha1))
1684 die("expected sha1, got garbage:\n %s", line);
1685 add_preferred_base_object(line+41);
1687 add_object_entry(sha1, 0, line+41, 0);
1688 }
1689}
1690static void show_commit(struct commit *commit)
1692{
1693 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1694}
1695static void show_object(struct object_array_entry *p)
1697{
1698 add_preferred_base_object(p->name);
1699 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1700}
1701static void show_edge(struct commit *commit)
1703{
1704 add_preferred_base(commit->object.sha1);
1705}
1706static void get_object_list(int ac, const char **av)
1708{
1709 struct rev_info revs;
1710 char line[1000];
1711 int flags = 0;
1712 init_revisions(&revs, NULL);
1714 save_commit_buffer = 0;
1715 track_object_refs = 0;
1716 setup_revisions(ac, av, &revs, NULL);
1717 while (fgets(line, sizeof(line), stdin) != NULL) {
1719 int len = strlen(line);
1720 if (line[len - 1] == '\n')
1721 line[--len] = 0;
1722 if (!len)
1723 break;
1724 if (*line == '-') {
1725 if (!strcmp(line, "--not")) {
1726 flags ^= UNINTERESTING;
1727 continue;
1728 }
1729 die("not a rev '%s'", line);
1730 }
1731 if (handle_revision_arg(line, &revs, flags, 1))
1732 die("bad revision '%s'", line);
1733 }
1734 prepare_revision_walk(&revs);
1736 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1737 traverse_commit_list(&revs, show_commit, show_object);
1738}
1739static int adjust_perm(const char *path, mode_t mode)
1741{
1742 if (chmod(path, mode))
1743 return -1;
1744 return adjust_shared_perm(path);
1745}
1746int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1748{
1749 int use_internal_rev_list = 0;
1750 int thin = 0;
1751 uint32_t i;
1752 const char **rp_av;
1753 int rp_ac_alloc = 64;
1754 int rp_ac;
1755 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1757 rp_av[0] = "pack-objects";
1759 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1760 rp_ac = 2;
1761 git_config(git_pack_config);
1763 if (!pack_compression_seen && core_compression_seen)
1764 pack_compression_level = core_compression_level;
1765 progress = isatty(2);
1767 for (i = 1; i < argc; i++) {
1768 const char *arg = argv[i];
1769 if (*arg != '-')
1771 break;
1772 if (!strcmp("--non-empty", arg)) {
1774 non_empty = 1;
1775 continue;
1776 }
1777 if (!strcmp("--local", arg)) {
1778 local = 1;
1779 continue;
1780 }
1781 if (!strcmp("--incremental", arg)) {
1782 incremental = 1;
1783 continue;
1784 }
1785 if (!prefixcmp(arg, "--compression=")) {
1786 char *end;
1787 int level = strtoul(arg+14, &end, 0);
1788 if (!arg[14] || *end)
1789 usage(pack_usage);
1790 if (level == -1)
1791 level = Z_DEFAULT_COMPRESSION;
1792 else if (level < 0 || level > Z_BEST_COMPRESSION)
1793 die("bad pack compression level %d", level);
1794 pack_compression_level = level;
1795 continue;
1796 }
1797 if (!prefixcmp(arg, "--max-pack-size=")) {
1798 char *end;
1799 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1800 if (!arg[16] || *end)
1801 usage(pack_usage);
1802 continue;
1803 }
1804 if (!prefixcmp(arg, "--window=")) {
1805 char *end;
1806 window = strtoul(arg+9, &end, 0);
1807 if (!arg[9] || *end)
1808 usage(pack_usage);
1809 continue;
1810 }
1811 if (!prefixcmp(arg, "--depth=")) {
1812 char *end;
1813 depth = strtoul(arg+8, &end, 0);
1814 if (!arg[8] || *end)
1815 usage(pack_usage);
1816 continue;
1817 }
1818 if (!strcmp("--progress", arg)) {
1819 progress = 1;
1820 continue;
1821 }
1822 if (!strcmp("--all-progress", arg)) {
1823 progress = 2;
1824 continue;
1825 }
1826 if (!strcmp("-q", arg)) {
1827 progress = 0;
1828 continue;
1829 }
1830 if (!strcmp("--no-reuse-delta", arg)) {
1831 no_reuse_delta = 1;
1832 continue;
1833 }
1834 if (!strcmp("--no-reuse-object", arg)) {
1835 no_reuse_object = no_reuse_delta = 1;
1836 continue;
1837 }
1838 if (!strcmp("--delta-base-offset", arg)) {
1839 allow_ofs_delta = 1;
1840 continue;
1841 }
1842 if (!strcmp("--stdout", arg)) {
1843 pack_to_stdout = 1;
1844 continue;
1845 }
1846 if (!strcmp("--revs", arg)) {
1847 use_internal_rev_list = 1;
1848 continue;
1849 }
1850 if (!strcmp("--unpacked", arg) ||
1851 !prefixcmp(arg, "--unpacked=") ||
1852 !strcmp("--reflog", arg) ||
1853 !strcmp("--all", arg)) {
1854 use_internal_rev_list = 1;
1855 if (rp_ac >= rp_ac_alloc - 1) {
1856 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1857 rp_av = xrealloc(rp_av,
1858 rp_ac_alloc * sizeof(*rp_av));
1859 }
1860 rp_av[rp_ac++] = arg;
1861 continue;
1862 }
1863 if (!strcmp("--thin", arg)) {
1864 use_internal_rev_list = 1;
1865 thin = 1;
1866 rp_av[1] = "--objects-edge";
1867 continue;
1868 }
1869 if (!prefixcmp(arg, "--index-version=")) {
1870 char *c;
1871 index_default_version = strtoul(arg + 16, &c, 10);
1872 if (index_default_version > 2)
1873 die("bad %s", arg);
1874 if (*c == ',')
1875 index_off32_limit = strtoul(c+1, &c, 0);
1876 if (*c || index_off32_limit & 0x80000000)
1877 die("bad %s", arg);
1878 continue;
1879 }
1880 usage(pack_usage);
1881 }
1882 /* Traditionally "pack-objects [options] base extra" failed;
1884 * we would however want to take refs parameter that would
1885 * have been given to upstream rev-list ourselves, which means
1886 * we somehow want to say what the base name is. So the
1887 * syntax would be:
1888 *
1889 * pack-objects [options] base <refs...>
1890 *
1891 * in other words, we would treat the first non-option as the
1892 * base_name and send everything else to the internal revision
1893 * walker.
1894 */
1895 if (!pack_to_stdout)
1897 base_name = argv[i++];
1898 if (pack_to_stdout != !base_name)
1900 usage(pack_usage);
1901 if (pack_to_stdout && pack_size_limit)
1903 die("--max-pack-size cannot be used to build a pack for transfer.");
1904 if (!pack_to_stdout && thin)
1906 die("--thin cannot be used to build an indexable pack.");
1907 prepare_packed_git();
1909 if (progress)
1911 start_progress(&progress_state, "Generating pack...",
1912 "Counting objects: ", 0);
1913 if (!use_internal_rev_list)
1914 read_object_list_from_stdin();
1915 else {
1916 rp_av[rp_ac] = NULL;
1917 get_object_list(rp_ac, rp_av);
1918 }
1919 if (progress) {
1920 stop_progress(&progress_state);
1921 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1922 }
1923 if (non_empty && !nr_result)
1925 return 0;
1926 if (progress && (nr_objects != nr_result))
1927 fprintf(stderr, "Result has %u objects.\n", nr_result);
1928 if (nr_result)
1929 prepare_pack(window, depth);
1930 write_pack_file();
1931 if (progress)
1932 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1933 written, written_delta, reused, reused_delta);
1934 return 0;
1935}