930b57a34bf83f1570604af3c5a76e872faa7210
1#include "builtin.h"
2#include "cache.h"
3#include "object.h"
4#include "blob.h"
5#include "commit.h"
6#include "tag.h"
7#include "tree.h"
8#include "delta.h"
9#include "pack.h"
10#include "csum-file.h"
11#include "tree-walk.h"
12#include "diff.h"
13#include "revision.h"
14#include "list-objects.h"
15#include "progress.h"
16static const char pack_usage[] = "\
18git-pack-objects [{ -q | --progress | --all-progress }] [--max-pack-size=N] \n\
19 [--local] [--incremental] [--window=N] [--depth=N] \n\
20 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
21 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
22 [--stdout | base-name] [<ref-list | <object-list]";
23struct object_entry {
25 unsigned char sha1[20];
26 uint32_t crc32; /* crc of raw pack data for this object */
27 off_t offset; /* offset into the final pack file */
28 unsigned long size; /* uncompressed size */
29 unsigned int hash; /* name hint hash */
30 unsigned int depth; /* delta depth */
31 struct packed_git *in_pack; /* already in pack */
32 off_t in_pack_offset;
33 struct object_entry *delta; /* delta base object */
34 struct object_entry *delta_child; /* deltified objects who bases me */
35 struct object_entry *delta_sibling; /* other deltified objects who
36 * uses the same base as me
37 */
38 unsigned long delta_size; /* delta data size (uncompressed) */
39 enum object_type type;
40 enum object_type in_pack_type; /* could be delta */
41 unsigned char in_pack_header_size;
42 unsigned char preferred_base; /* we do not pack this, but is available
43 * to be used as the base objectto delta
44 * objects against.
45 */
46};
47/*
49 * Objects we are going to pack are collected in objects array (dynamically
50 * expanded). nr_objects & nr_alloc controls this array. They are stored
51 * in the order we see -- typically rev-list --objects order that gives us
52 * nice "minimum seek" order.
53 */
54static struct object_entry *objects;
55static struct object_entry **written_list;
56static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
57static int non_empty;
59static int no_reuse_delta, no_reuse_object;
60static int local;
61static int incremental;
62static int allow_ofs_delta;
63static const char *pack_tmp_name, *idx_tmp_name;
64static char tmpname[PATH_MAX];
65static const char *base_name;
66static unsigned char pack_file_sha1[20];
67static int progress = 1;
68static int window = 10;
69static uint32_t pack_size_limit;
70static int depth = 50;
71static int pack_to_stdout;
72static int num_preferred_base;
73static struct progress progress_state;
74static int pack_compression_level = Z_DEFAULT_COMPRESSION;
75static int pack_compression_seen;
76/*
78 * The object names in objects array are hashed with this hashtable,
79 * to help looking up the entry by object name.
80 * This hashtable is built after all the objects are seen.
81 */
82static int *object_ix;
83static int object_ix_hashsz;
84/*
86 * Pack index for existing packs give us easy access to the offsets into
87 * corresponding pack file where each object's data starts, but the entries
88 * do not store the size of the compressed representation (uncompressed
89 * size is easily available by examining the pack entry header). It is
90 * also rather expensive to find the sha1 for an object given its offset.
91 *
92 * We build a hashtable of existing packs (pack_revindex), and keep reverse
93 * index here -- pack index file is sorted by object name mapping to offset;
94 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
95 * ordered by offset, so if you know the offset of an object, next offset
96 * is where its packed representation ends and the index_nr can be used to
97 * get the object sha1 from the main index.
98 */
99struct revindex_entry {
100 off_t offset;
101 unsigned int nr;
102};
103struct pack_revindex {
104 struct packed_git *p;
105 struct revindex_entry *revindex;
106};
107static struct pack_revindex *pack_revindex;
108static int pack_revindex_hashsz;
109/*
111 * stats
112 */
113static uint32_t written, written_delta;
114static uint32_t reused, reused_delta;
115static int pack_revindex_ix(struct packed_git *p)
117{
118 unsigned long ui = (unsigned long)p;
119 int i;
120 ui = ui ^ (ui >> 16); /* defeat structure alignment */
122 i = (int)(ui % pack_revindex_hashsz);
123 while (pack_revindex[i].p) {
124 if (pack_revindex[i].p == p)
125 return i;
126 if (++i == pack_revindex_hashsz)
127 i = 0;
128 }
129 return -1 - i;
130}
131static void prepare_pack_ix(void)
133{
134 int num;
135 struct packed_git *p;
136 for (num = 0, p = packed_git; p; p = p->next)
137 num++;
138 if (!num)
139 return;
140 pack_revindex_hashsz = num * 11;
141 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
142 for (p = packed_git; p; p = p->next) {
143 num = pack_revindex_ix(p);
144 num = - 1 - num;
145 pack_revindex[num].p = p;
146 }
147 /* revindex elements are lazily initialized */
148}
149static int cmp_offset(const void *a_, const void *b_)
151{
152 const struct revindex_entry *a = a_;
153 const struct revindex_entry *b = b_;
154 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
155}
156/*
158 * Ordered list of offsets of objects in the pack.
159 */
160static void prepare_pack_revindex(struct pack_revindex *rix)
161{
162 struct packed_git *p = rix->p;
163 int num_ent = p->num_objects;
164 int i;
165 const char *index = p->index_data;
166 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
168 index += 4 * 256;
169 if (p->index_version > 1) {
171 const uint32_t *off_32 =
172 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
173 const uint32_t *off_64 = off_32 + p->num_objects;
174 for (i = 0; i < num_ent; i++) {
175 uint32_t off = ntohl(*off_32++);
176 if (!(off & 0x80000000)) {
177 rix->revindex[i].offset = off;
178 } else {
179 rix->revindex[i].offset =
180 ((uint64_t)ntohl(*off_64++)) << 32;
181 rix->revindex[i].offset |=
182 ntohl(*off_64++);
183 }
184 rix->revindex[i].nr = i;
185 }
186 } else {
187 for (i = 0; i < num_ent; i++) {
188 uint32_t hl = *((uint32_t *)(index + 24 * i));
189 rix->revindex[i].offset = ntohl(hl);
190 rix->revindex[i].nr = i;
191 }
192 }
193 /* This knows the pack format -- the 20-byte trailer
195 * follows immediately after the last object data.
196 */
197 rix->revindex[num_ent].offset = p->pack_size - 20;
198 rix->revindex[num_ent].nr = -1;
199 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
200}
201static struct revindex_entry * find_packed_object(struct packed_git *p,
203 off_t ofs)
204{
205 int num;
206 int lo, hi;
207 struct pack_revindex *rix;
208 struct revindex_entry *revindex;
209 num = pack_revindex_ix(p);
210 if (num < 0)
211 die("internal error: pack revindex uninitialized");
212 rix = &pack_revindex[num];
213 if (!rix->revindex)
214 prepare_pack_revindex(rix);
215 revindex = rix->revindex;
216 lo = 0;
217 hi = p->num_objects + 1;
218 do {
219 int mi = (lo + hi) / 2;
220 if (revindex[mi].offset == ofs) {
221 return revindex + mi;
222 }
223 else if (ofs < revindex[mi].offset)
224 hi = mi;
225 else
226 lo = mi + 1;
227 } while (lo < hi);
228 die("internal error: pack revindex corrupt");
229}
230static const unsigned char *find_packed_object_name(struct packed_git *p,
232 off_t ofs)
233{
234 struct revindex_entry *entry = find_packed_object(p, ofs);
235 return nth_packed_object_sha1(p, entry->nr);
236}
237static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
239{
240 unsigned long othersize, delta_size;
241 enum object_type type;
242 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
243 void *delta_buf;
244 if (!otherbuf)
246 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
247 delta_buf = diff_delta(otherbuf, othersize,
248 buf, size, &delta_size, 0);
249 if (!delta_buf || delta_size != entry->delta_size)
250 die("delta size changed");
251 free(buf);
252 free(otherbuf);
253 return delta_buf;
254}
255/*
257 * The per-object header is a pretty dense thing, which is
258 * - first byte: low four bits are "size", then three bits of "type",
259 * and the high bit is "size continues".
260 * - each byte afterwards: low seven bits are size continuation,
261 * with the high bit being "size continues"
262 */
263static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
264{
265 int n = 1;
266 unsigned char c;
267 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
269 die("bad type %d", type);
270 c = (type << 4) | (size & 15);
272 size >>= 4;
273 while (size) {
274 *hdr++ = c | 0x80;
275 c = size & 0x7f;
276 size >>= 7;
277 n++;
278 }
279 *hdr = c;
280 return n;
281}
282/*
284 * we are going to reuse the existing object data as is. make
285 * sure it is not corrupt.
286 */
287static int check_pack_inflate(struct packed_git *p,
288 struct pack_window **w_curs,
289 off_t offset,
290 off_t len,
291 unsigned long expect)
292{
293 z_stream stream;
294 unsigned char fakebuf[4096], *in;
295 int st;
296 memset(&stream, 0, sizeof(stream));
298 inflateInit(&stream);
299 do {
300 in = use_pack(p, w_curs, offset, &stream.avail_in);
301 stream.next_in = in;
302 stream.next_out = fakebuf;
303 stream.avail_out = sizeof(fakebuf);
304 st = inflate(&stream, Z_FINISH);
305 offset += stream.next_in - in;
306 } while (st == Z_OK || st == Z_BUF_ERROR);
307 inflateEnd(&stream);
308 return (st == Z_STREAM_END &&
309 stream.total_out == expect &&
310 stream.total_in == len) ? 0 : -1;
311}
312static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
314 off_t offset, off_t len, unsigned int nr)
315{
316 const uint32_t *index_crc;
317 uint32_t data_crc = crc32(0, Z_NULL, 0);
318 do {
320 unsigned int avail;
321 void *data = use_pack(p, w_curs, offset, &avail);
322 if (avail > len)
323 avail = len;
324 data_crc = crc32(data_crc, data, avail);
325 offset += avail;
326 len -= avail;
327 } while (len);
328 index_crc = p->index_data;
330 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
331 return data_crc != ntohl(*index_crc);
333}
334static void copy_pack_data(struct sha1file *f,
336 struct packed_git *p,
337 struct pack_window **w_curs,
338 off_t offset,
339 off_t len)
340{
341 unsigned char *in;
342 unsigned int avail;
343 while (len) {
345 in = use_pack(p, w_curs, offset, &avail);
346 if (avail > len)
347 avail = (unsigned int)len;
348 sha1write(f, in, avail);
349 offset += avail;
350 len -= avail;
351 }
352}
353static unsigned long write_object(struct sha1file *f,
355 struct object_entry *entry,
356 off_t write_offset)
357{
358 unsigned long size;
359 enum object_type type;
360 void *buf;
361 unsigned char header[10];
362 unsigned char dheader[10];
363 unsigned hdrlen;
364 off_t datalen;
365 enum object_type obj_type;
366 int to_reuse = 0;
367 /* write limit if limited packsize and not first object */
368 unsigned long limit = pack_size_limit && nr_written ?
369 pack_size_limit - write_offset : 0;
370 /* no if no delta */
371 int usable_delta = !entry->delta ? 0 :
372 /* yes if unlimited packfile */
373 !pack_size_limit ? 1 :
374 /* no if base written to previous pack */
375 entry->delta->offset == (off_t)-1 ? 0 :
376 /* otherwise double-check written to this
377 * pack, like we do below
378 */
379 entry->delta->offset ? 1 : 0;
380 if (!pack_to_stdout)
382 crc32_begin(f);
383 obj_type = entry->type;
385 if (no_reuse_object)
386 to_reuse = 0; /* explicit */
387 else if (!entry->in_pack)
388 to_reuse = 0; /* can't reuse what we don't have */
389 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
390 /* check_object() decided it for us ... */
391 to_reuse = usable_delta;
392 /* ... but pack split may override that */
393 else if (obj_type != entry->in_pack_type)
394 to_reuse = 0; /* pack has delta which is unusable */
395 else if (entry->delta)
396 to_reuse = 0; /* we want to pack afresh */
397 else
398 to_reuse = 1; /* we have it in-pack undeltified,
399 * and we do not need to deltify it.
400 */
401 if (!to_reuse) {
403 z_stream stream;
404 unsigned long maxsize;
405 void *out;
406 buf = read_sha1_file(entry->sha1, &type, &size);
407 if (!buf)
408 die("unable to read %s", sha1_to_hex(entry->sha1));
409 if (size != entry->size)
410 die("object %s size inconsistency (%lu vs %lu)",
411 sha1_to_hex(entry->sha1), size, entry->size);
412 if (usable_delta) {
413 buf = delta_against(buf, size, entry);
414 size = entry->delta_size;
415 obj_type = (allow_ofs_delta && entry->delta->offset) ?
416 OBJ_OFS_DELTA : OBJ_REF_DELTA;
417 } else {
418 /*
419 * recover real object type in case
420 * check_object() wanted to re-use a delta,
421 * but we couldn't since base was in previous split pack
422 */
423 obj_type = type;
424 }
425 /* compress the data to store and put compressed length in datalen */
426 memset(&stream, 0, sizeof(stream));
427 deflateInit(&stream, pack_compression_level);
428 maxsize = deflateBound(&stream, size);
429 out = xmalloc(maxsize);
430 /* Compress it */
431 stream.next_in = buf;
432 stream.avail_in = size;
433 stream.next_out = out;
434 stream.avail_out = maxsize;
435 while (deflate(&stream, Z_FINISH) == Z_OK)
436 /* nothing */;
437 deflateEnd(&stream);
438 datalen = stream.total_out;
439 deflateEnd(&stream);
440 /*
441 * The object header is a byte of 'type' followed by zero or
442 * more bytes of length.
443 */
444 hdrlen = encode_header(obj_type, size, header);
445 if (obj_type == OBJ_OFS_DELTA) {
447 /*
448 * Deltas with relative base contain an additional
449 * encoding of the relative offset for the delta
450 * base from this object's position in the pack.
451 */
452 off_t ofs = entry->offset - entry->delta->offset;
453 unsigned pos = sizeof(dheader) - 1;
454 dheader[pos] = ofs & 127;
455 while (ofs >>= 7)
456 dheader[--pos] = 128 | (--ofs & 127);
457 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
458 free(out);
459 free(buf);
460 return 0;
461 }
462 sha1write(f, header, hdrlen);
463 sha1write(f, dheader + pos, sizeof(dheader) - pos);
464 hdrlen += sizeof(dheader) - pos;
465 } else if (obj_type == OBJ_REF_DELTA) {
466 /*
467 * Deltas with a base reference contain
468 * an additional 20 bytes for the base sha1.
469 */
470 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
471 free(out);
472 free(buf);
473 return 0;
474 }
475 sha1write(f, header, hdrlen);
476 sha1write(f, entry->delta->sha1, 20);
477 hdrlen += 20;
478 } else {
479 if (limit && hdrlen + datalen + 20 >= limit) {
480 free(out);
481 free(buf);
482 return 0;
483 }
484 sha1write(f, header, hdrlen);
485 }
486 sha1write(f, out, datalen);
487 free(out);
488 free(buf);
489 }
490 else {
491 struct packed_git *p = entry->in_pack;
492 struct pack_window *w_curs = NULL;
493 struct revindex_entry *revidx;
494 off_t offset;
495 if (entry->delta) {
497 obj_type = (allow_ofs_delta && entry->delta->offset) ?
498 OBJ_OFS_DELTA : OBJ_REF_DELTA;
499 reused_delta++;
500 }
501 hdrlen = encode_header(obj_type, entry->size, header);
502 offset = entry->in_pack_offset;
503 revidx = find_packed_object(p, offset);
504 datalen = revidx[1].offset - offset;
505 if (!pack_to_stdout && p->index_version > 1 &&
506 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
507 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
508 offset += entry->in_pack_header_size;
509 datalen -= entry->in_pack_header_size;
510 if (obj_type == OBJ_OFS_DELTA) {
511 off_t ofs = entry->offset - entry->delta->offset;
512 unsigned pos = sizeof(dheader) - 1;
513 dheader[pos] = ofs & 127;
514 while (ofs >>= 7)
515 dheader[--pos] = 128 | (--ofs & 127);
516 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
517 return 0;
518 sha1write(f, header, hdrlen);
519 sha1write(f, dheader + pos, sizeof(dheader) - pos);
520 hdrlen += sizeof(dheader) - pos;
521 } else if (obj_type == OBJ_REF_DELTA) {
522 if (limit && hdrlen + 20 + datalen + 20 >= limit)
523 return 0;
524 sha1write(f, header, hdrlen);
525 sha1write(f, entry->delta->sha1, 20);
526 hdrlen += 20;
527 } else {
528 if (limit && hdrlen + datalen + 20 >= limit)
529 return 0;
530 sha1write(f, header, hdrlen);
531 }
532 if (!pack_to_stdout && p->index_version == 1 &&
534 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
535 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
536 copy_pack_data(f, p, &w_curs, offset, datalen);
537 unuse_pack(&w_curs);
538 reused++;
539 }
540 if (usable_delta)
541 written_delta++;
542 written++;
543 if (!pack_to_stdout)
544 entry->crc32 = crc32_end(f);
545 return hdrlen + datalen;
546}
547static off_t write_one(struct sha1file *f,
549 struct object_entry *e,
550 off_t offset)
551{
552 unsigned long size;
553 /* offset is non zero if object is written already. */
555 if (e->offset || e->preferred_base)
556 return offset;
557 /* if we are deltified, write out base object first. */
559 if (e->delta) {
560 offset = write_one(f, e->delta, offset);
561 if (!offset)
562 return 0;
563 }
564 e->offset = offset;
566 size = write_object(f, e, offset);
567 if (!size) {
568 e->offset = 0;
569 return 0;
570 }
571 written_list[nr_written++] = e;
572 /* make sure off_t is sufficiently large not to wrap */
574 if (offset > offset + size)
575 die("pack too large for current definition of off_t");
576 return offset + size;
577}
578static int open_object_dir_tmp(const char *path)
580{
581 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
582 return mkstemp(tmpname);
583}
584/* forward declarations for write_pack_file */
586static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
587static int adjust_perm(const char *path, mode_t mode);
588static void write_pack_file(void)
590{
591 uint32_t i = 0, j;
592 struct sha1file *f;
593 off_t offset, offset_one, last_obj_offset = 0;
594 struct pack_header hdr;
595 int do_progress = progress >> pack_to_stdout;
596 uint32_t nr_remaining = nr_result;
597 if (do_progress)
599 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
600 written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
601 do {
603 if (pack_to_stdout) {
604 f = sha1fd(1, "<stdout>");
605 } else {
606 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
607 if (fd < 0)
608 die("unable to create %s: %s\n", tmpname, strerror(errno));
609 pack_tmp_name = xstrdup(tmpname);
610 f = sha1fd(fd, pack_tmp_name);
611 }
612 hdr.hdr_signature = htonl(PACK_SIGNATURE);
614 hdr.hdr_version = htonl(PACK_VERSION);
615 hdr.hdr_entries = htonl(nr_remaining);
616 sha1write(f, &hdr, sizeof(hdr));
617 offset = sizeof(hdr);
618 nr_written = 0;
619 for (; i < nr_objects; i++) {
620 last_obj_offset = offset;
621 offset_one = write_one(f, objects + i, offset);
622 if (!offset_one)
623 break;
624 offset = offset_one;
625 if (do_progress)
626 display_progress(&progress_state, written);
627 }
628 /*
630 * Did we write the wrong # entries in the header?
631 * If so, rewrite it like in fast-import
632 */
633 if (pack_to_stdout || nr_written == nr_remaining) {
634 sha1close(f, pack_file_sha1, 1);
635 } else {
636 sha1close(f, pack_file_sha1, 0);
637 fixup_pack_header_footer(f->fd, pack_file_sha1, pack_tmp_name, nr_written);
638 close(f->fd);
639 }
640 if (!pack_to_stdout) {
642 unsigned char object_list_sha1[20];
643 mode_t mode = umask(0);
644 umask(mode);
646 mode = 0444 & ~mode;
647 write_index_file(last_obj_offset, object_list_sha1);
649 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
650 base_name, sha1_to_hex(object_list_sha1));
651 if (adjust_perm(pack_tmp_name, mode))
652 die("unable to make temporary pack file readable: %s",
653 strerror(errno));
654 if (rename(pack_tmp_name, tmpname))
655 die("unable to rename temporary pack file: %s",
656 strerror(errno));
657 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
658 base_name, sha1_to_hex(object_list_sha1));
659 if (adjust_perm(idx_tmp_name, mode))
660 die("unable to make temporary index file readable: %s",
661 strerror(errno));
662 if (rename(idx_tmp_name, tmpname))
663 die("unable to rename temporary index file: %s",
664 strerror(errno));
665 puts(sha1_to_hex(object_list_sha1));
666 }
667 /* mark written objects as written to previous pack */
669 for (j = 0; j < nr_written; j++) {
670 written_list[j]->offset = (off_t)-1;
671 }
672 nr_remaining -= nr_written;
673 } while (nr_remaining && i < nr_objects);
674 free(written_list);
676 if (do_progress)
677 stop_progress(&progress_state);
678 if (written != nr_result)
679 die("wrote %u objects while expecting %u", written, nr_result);
680 for (j = 0; i < nr_objects; i++) {
681 struct object_entry *e = objects + i;
682 j += !e->offset && !e->preferred_base;
683 }
684 if (j)
685 die("wrote %u objects as expected but %u unwritten", written, j);
686}
687static int sha1_sort(const void *_a, const void *_b)
689{
690 const struct object_entry *a = *(struct object_entry **)_a;
691 const struct object_entry *b = *(struct object_entry **)_b;
692 return hashcmp(a->sha1, b->sha1);
693}
694static uint32_t index_default_version = 1;
696static uint32_t index_off32_limit = 0x7fffffff;
697static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
699{
700 struct sha1file *f;
701 struct object_entry **sorted_by_sha, **list, **last;
702 uint32_t array[256];
703 uint32_t i, index_version;
704 SHA_CTX ctx;
705 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
707 if (fd < 0)
708 die("unable to create %s: %s\n", tmpname, strerror(errno));
709 idx_tmp_name = xstrdup(tmpname);
710 f = sha1fd(fd, idx_tmp_name);
711 if (nr_written) {
713 sorted_by_sha = written_list;
714 qsort(sorted_by_sha, nr_written, sizeof(*sorted_by_sha), sha1_sort);
715 list = sorted_by_sha;
716 last = sorted_by_sha + nr_written;
717 } else
718 sorted_by_sha = list = last = NULL;
719 /* if last object's offset is >= 2^31 we should use index V2 */
721 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
722 /* index versions 2 and above need a header */
724 if (index_version >= 2) {
725 struct pack_idx_header hdr;
726 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
727 hdr.idx_version = htonl(index_version);
728 sha1write(f, &hdr, sizeof(hdr));
729 }
730 /*
732 * Write the first-level table (the list is sorted,
733 * but we use a 256-entry lookup to be able to avoid
734 * having to do eight extra binary search iterations).
735 */
736 for (i = 0; i < 256; i++) {
737 struct object_entry **next = list;
738 while (next < last) {
739 struct object_entry *entry = *next;
740 if (entry->sha1[0] != i)
741 break;
742 next++;
743 }
744 array[i] = htonl(next - sorted_by_sha);
745 list = next;
746 }
747 sha1write(f, array, 256 * 4);
748 /* Compute the SHA1 hash of sorted object names. */
750 SHA1_Init(&ctx);
751 /* Write the actual SHA1 entries. */
753 list = sorted_by_sha;
754 for (i = 0; i < nr_written; i++) {
755 struct object_entry *entry = *list++;
756 if (index_version < 2) {
757 uint32_t offset = htonl(entry->offset);
758 sha1write(f, &offset, 4);
759 }
760 sha1write(f, entry->sha1, 20);
761 SHA1_Update(&ctx, entry->sha1, 20);
762 }
763 if (index_version >= 2) {
765 unsigned int nr_large_offset = 0;
766 /* write the crc32 table */
768 list = sorted_by_sha;
769 for (i = 0; i < nr_written; i++) {
770 struct object_entry *entry = *list++;
771 uint32_t crc32_val = htonl(entry->crc32);
772 sha1write(f, &crc32_val, 4);
773 }
774 /* write the 32-bit offset table */
776 list = sorted_by_sha;
777 for (i = 0; i < nr_written; i++) {
778 struct object_entry *entry = *list++;
779 uint32_t offset = (entry->offset <= index_off32_limit) ?
780 entry->offset : (0x80000000 | nr_large_offset++);
781 offset = htonl(offset);
782 sha1write(f, &offset, 4);
783 }
784 /* write the large offset table */
786 list = sorted_by_sha;
787 while (nr_large_offset) {
788 struct object_entry *entry = *list++;
789 uint64_t offset = entry->offset;
790 if (offset > index_off32_limit) {
791 uint32_t split[2];
792 split[0] = htonl(offset >> 32);
793 split[1] = htonl(offset & 0xffffffff);
794 sha1write(f, split, 8);
795 nr_large_offset--;
796 }
797 }
798 }
799 sha1write(f, pack_file_sha1, 20);
801 sha1close(f, NULL, 1);
802 SHA1_Final(sha1, &ctx);
803}
804static int locate_object_entry_hash(const unsigned char *sha1)
806{
807 int i;
808 unsigned int ui;
809 memcpy(&ui, sha1, sizeof(unsigned int));
810 i = ui % object_ix_hashsz;
811 while (0 < object_ix[i]) {
812 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
813 return i;
814 if (++i == object_ix_hashsz)
815 i = 0;
816 }
817 return -1 - i;
818}
819static struct object_entry *locate_object_entry(const unsigned char *sha1)
821{
822 int i;
823 if (!object_ix_hashsz)
825 return NULL;
826 i = locate_object_entry_hash(sha1);
828 if (0 <= i)
829 return &objects[object_ix[i]-1];
830 return NULL;
831}
832static void rehash_objects(void)
834{
835 uint32_t i;
836 struct object_entry *oe;
837 object_ix_hashsz = nr_objects * 3;
839 if (object_ix_hashsz < 1024)
840 object_ix_hashsz = 1024;
841 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
842 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
843 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
844 int ix = locate_object_entry_hash(oe->sha1);
845 if (0 <= ix)
846 continue;
847 ix = -1 - ix;
848 object_ix[ix] = i + 1;
849 }
850}
851static unsigned name_hash(const char *name)
853{
854 unsigned char c;
855 unsigned hash = 0;
856 /*
858 * This effectively just creates a sortable number from the
859 * last sixteen non-whitespace characters. Last characters
860 * count "most", so things that end in ".c" sort together.
861 */
862 while ((c = *name++) != 0) {
863 if (isspace(c))
864 continue;
865 hash = (hash >> 2) + (c << 24);
866 }
867 return hash;
868}
869static int add_object_entry(const unsigned char *sha1, enum object_type type,
871 unsigned hash, int exclude)
872{
873 struct object_entry *entry;
874 struct packed_git *p, *found_pack = NULL;
875 off_t found_offset = 0;
876 int ix;
877 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
879 if (ix >= 0) {
880 if (exclude) {
881 entry = objects + object_ix[ix] - 1;
882 if (!entry->preferred_base)
883 nr_result--;
884 entry->preferred_base = 1;
885 }
886 return 0;
887 }
888 for (p = packed_git; p; p = p->next) {
890 off_t offset = find_pack_entry_one(sha1, p);
891 if (offset) {
892 if (!found_pack) {
893 found_offset = offset;
894 found_pack = p;
895 }
896 if (exclude)
897 break;
898 if (incremental)
899 return 0;
900 if (local && !p->pack_local)
901 return 0;
902 }
903 }
904 if (nr_objects >= nr_alloc) {
906 nr_alloc = (nr_alloc + 1024) * 3 / 2;
907 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
908 }
909 entry = objects + nr_objects++;
911 memset(entry, 0, sizeof(*entry));
912 hashcpy(entry->sha1, sha1);
913 entry->hash = hash;
914 if (type)
915 entry->type = type;
916 if (exclude)
917 entry->preferred_base = 1;
918 else
919 nr_result++;
920 if (found_pack) {
921 entry->in_pack = found_pack;
922 entry->in_pack_offset = found_offset;
923 }
924 if (object_ix_hashsz * 3 <= nr_objects * 4)
926 rehash_objects();
927 else
928 object_ix[-1 - ix] = nr_objects;
929 if (progress)
931 display_progress(&progress_state, nr_objects);
932 return 1;
934}
935struct pbase_tree_cache {
937 unsigned char sha1[20];
938 int ref;
939 int temporary;
940 void *tree_data;
941 unsigned long tree_size;
942};
943static struct pbase_tree_cache *(pbase_tree_cache[256]);
945static int pbase_tree_cache_ix(const unsigned char *sha1)
946{
947 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
948}
949static int pbase_tree_cache_ix_incr(int ix)
950{
951 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
952}
953static struct pbase_tree {
955 struct pbase_tree *next;
956 /* This is a phony "cache" entry; we are not
957 * going to evict it nor find it through _get()
958 * mechanism -- this is for the toplevel node that
959 * would almost always change with any commit.
960 */
961 struct pbase_tree_cache pcache;
962} *pbase_tree;
963static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
965{
966 struct pbase_tree_cache *ent, *nent;
967 void *data;
968 unsigned long size;
969 enum object_type type;
970 int neigh;
971 int my_ix = pbase_tree_cache_ix(sha1);
972 int available_ix = -1;
973 /* pbase-tree-cache acts as a limited hashtable.
975 * your object will be found at your index or within a few
976 * slots after that slot if it is cached.
977 */
978 for (neigh = 0; neigh < 8; neigh++) {
979 ent = pbase_tree_cache[my_ix];
980 if (ent && !hashcmp(ent->sha1, sha1)) {
981 ent->ref++;
982 return ent;
983 }
984 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
985 ((0 <= available_ix) &&
986 (!ent && pbase_tree_cache[available_ix])))
987 available_ix = my_ix;
988 if (!ent)
989 break;
990 my_ix = pbase_tree_cache_ix_incr(my_ix);
991 }
992 /* Did not find one. Either we got a bogus request or
994 * we need to read and perhaps cache.
995 */
996 data = read_sha1_file(sha1, &type, &size);
997 if (!data)
998 return NULL;
999 if (type != OBJ_TREE) {
1000 free(data);
1001 return NULL;
1002 }
1003 /* We need to either cache or return a throwaway copy */
1005 if (available_ix < 0)
1007 ent = NULL;
1008 else {
1009 ent = pbase_tree_cache[available_ix];
1010 my_ix = available_ix;
1011 }
1012 if (!ent) {
1014 nent = xmalloc(sizeof(*nent));
1015 nent->temporary = (available_ix < 0);
1016 }
1017 else {
1018 /* evict and reuse */
1019 free(ent->tree_data);
1020 nent = ent;
1021 }
1022 hashcpy(nent->sha1, sha1);
1023 nent->tree_data = data;
1024 nent->tree_size = size;
1025 nent->ref = 1;
1026 if (!nent->temporary)
1027 pbase_tree_cache[my_ix] = nent;
1028 return nent;
1029}
1030static void pbase_tree_put(struct pbase_tree_cache *cache)
1032{
1033 if (!cache->temporary) {
1034 cache->ref--;
1035 return;
1036 }
1037 free(cache->tree_data);
1038 free(cache);
1039}
1040static int name_cmp_len(const char *name)
1042{
1043 int i;
1044 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1045 ;
1046 return i;
1047}
1048static void add_pbase_object(struct tree_desc *tree,
1050 const char *name,
1051 int cmplen,
1052 const char *fullname)
1053{
1054 struct name_entry entry;
1055 int cmp;
1056 while (tree_entry(tree,&entry)) {
1058 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1059 memcmp(name, entry.path, cmplen);
1060 if (cmp > 0)
1061 continue;
1062 if (cmp < 0)
1063 return;
1064 if (name[cmplen] != '/') {
1065 unsigned hash = name_hash(fullname);
1066 add_object_entry(entry.sha1,
1067 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1068 hash, 1);
1069 return;
1070 }
1071 if (S_ISDIR(entry.mode)) {
1072 struct tree_desc sub;
1073 struct pbase_tree_cache *tree;
1074 const char *down = name+cmplen+1;
1075 int downlen = name_cmp_len(down);
1076 tree = pbase_tree_get(entry.sha1);
1078 if (!tree)
1079 return;
1080 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1081 add_pbase_object(&sub, down, downlen, fullname);
1083 pbase_tree_put(tree);
1084 }
1085 }
1086}
1087static unsigned *done_pbase_paths;
1089static int done_pbase_paths_num;
1090static int done_pbase_paths_alloc;
1091static int done_pbase_path_pos(unsigned hash)
1092{
1093 int lo = 0;
1094 int hi = done_pbase_paths_num;
1095 while (lo < hi) {
1096 int mi = (hi + lo) / 2;
1097 if (done_pbase_paths[mi] == hash)
1098 return mi;
1099 if (done_pbase_paths[mi] < hash)
1100 hi = mi;
1101 else
1102 lo = mi + 1;
1103 }
1104 return -lo-1;
1105}
1106static int check_pbase_path(unsigned hash)
1108{
1109 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1110 if (0 <= pos)
1111 return 1;
1112 pos = -pos - 1;
1113 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1114 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1115 done_pbase_paths = xrealloc(done_pbase_paths,
1116 done_pbase_paths_alloc *
1117 sizeof(unsigned));
1118 }
1119 done_pbase_paths_num++;
1120 if (pos < done_pbase_paths_num)
1121 memmove(done_pbase_paths + pos + 1,
1122 done_pbase_paths + pos,
1123 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1124 done_pbase_paths[pos] = hash;
1125 return 0;
1126}
1127static void add_preferred_base_object(const char *name, unsigned hash)
1129{
1130 struct pbase_tree *it;
1131 int cmplen;
1132 if (!num_preferred_base || check_pbase_path(hash))
1134 return;
1135 cmplen = name_cmp_len(name);
1137 for (it = pbase_tree; it; it = it->next) {
1138 if (cmplen == 0) {
1139 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
1140 }
1141 else {
1142 struct tree_desc tree;
1143 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1144 add_pbase_object(&tree, name, cmplen, name);
1145 }
1146 }
1147}
1148static void add_preferred_base(unsigned char *sha1)
1150{
1151 struct pbase_tree *it;
1152 void *data;
1153 unsigned long size;
1154 unsigned char tree_sha1[20];
1155 if (window <= num_preferred_base++)
1157 return;
1158 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1160 if (!data)
1161 return;
1162 for (it = pbase_tree; it; it = it->next) {
1164 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1165 free(data);
1166 return;
1167 }
1168 }
1169 it = xcalloc(1, sizeof(*it));
1171 it->next = pbase_tree;
1172 pbase_tree = it;
1173 hashcpy(it->pcache.sha1, tree_sha1);
1175 it->pcache.tree_data = data;
1176 it->pcache.tree_size = size;
1177}
1178static void check_object(struct object_entry *entry)
1180{
1181 if (entry->in_pack) {
1182 struct packed_git *p = entry->in_pack;
1183 struct pack_window *w_curs = NULL;
1184 const unsigned char *base_ref = NULL;
1185 struct object_entry *base_entry;
1186 unsigned long used, used_0;
1187 unsigned int avail;
1188 off_t ofs;
1189 unsigned char *buf, c;
1190 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1192 /*
1194 * We want in_pack_type even if we do not reuse delta
1195 * since non-delta representations could still be reused.
1196 */
1197 used = unpack_object_header_gently(buf, avail,
1198 &entry->in_pack_type,
1199 &entry->size);
1200 /*
1202 * Determine if this is a delta and if so whether we can
1203 * reuse it or not. Otherwise let's find out as cheaply as
1204 * possible what the actual type and size for this object is.
1205 */
1206 switch (entry->in_pack_type) {
1207 default:
1208 /* Not a delta hence we've already got all we need. */
1209 entry->type = entry->in_pack_type;
1210 entry->in_pack_header_size = used;
1211 unuse_pack(&w_curs);
1212 return;
1213 case OBJ_REF_DELTA:
1214 if (!no_reuse_delta && !entry->preferred_base)
1215 base_ref = use_pack(p, &w_curs,
1216 entry->in_pack_offset + used, NULL);
1217 entry->in_pack_header_size = used + 20;
1218 break;
1219 case OBJ_OFS_DELTA:
1220 buf = use_pack(p, &w_curs,
1221 entry->in_pack_offset + used, NULL);
1222 used_0 = 0;
1223 c = buf[used_0++];
1224 ofs = c & 127;
1225 while (c & 128) {
1226 ofs += 1;
1227 if (!ofs || MSB(ofs, 7))
1228 die("delta base offset overflow in pack for %s",
1229 sha1_to_hex(entry->sha1));
1230 c = buf[used_0++];
1231 ofs = (ofs << 7) + (c & 127);
1232 }
1233 if (ofs >= entry->in_pack_offset)
1234 die("delta base offset out of bound for %s",
1235 sha1_to_hex(entry->sha1));
1236 ofs = entry->in_pack_offset - ofs;
1237 if (!no_reuse_delta && !entry->preferred_base)
1238 base_ref = find_packed_object_name(p, ofs);
1239 entry->in_pack_header_size = used + used_0;
1240 break;
1241 }
1242 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1244 /*
1245 * If base_ref was set above that means we wish to
1246 * reuse delta data, and we even found that base
1247 * in the list of objects we want to pack. Goodie!
1248 *
1249 * Depth value does not matter - find_deltas() will
1250 * never consider reused delta as the base object to
1251 * deltify other objects against, in order to avoid
1252 * circular deltas.
1253 */
1254 entry->type = entry->in_pack_type;
1255 entry->delta = base_entry;
1256 entry->delta_sibling = base_entry->delta_child;
1257 base_entry->delta_child = entry;
1258 unuse_pack(&w_curs);
1259 return;
1260 }
1261 if (entry->type) {
1263 /*
1264 * This must be a delta and we already know what the
1265 * final object type is. Let's extract the actual
1266 * object size from the delta header.
1267 */
1268 entry->size = get_size_from_delta(p, &w_curs,
1269 entry->in_pack_offset + entry->in_pack_header_size);
1270 unuse_pack(&w_curs);
1271 return;
1272 }
1273 /*
1275 * No choice but to fall back to the recursive delta walk
1276 * with sha1_object_info() to find about the object type
1277 * at this point...
1278 */
1279 unuse_pack(&w_curs);
1280 }
1281 entry->type = sha1_object_info(entry->sha1, &entry->size);
1283 if (entry->type < 0)
1284 die("unable to get type of object %s",
1285 sha1_to_hex(entry->sha1));
1286}
1287static int pack_offset_sort(const void *_a, const void *_b)
1289{
1290 const struct object_entry *a = *(struct object_entry **)_a;
1291 const struct object_entry *b = *(struct object_entry **)_b;
1292 /* avoid filesystem trashing with loose objects */
1294 if (!a->in_pack && !b->in_pack)
1295 return hashcmp(a->sha1, b->sha1);
1296 if (a->in_pack < b->in_pack)
1298 return -1;
1299 if (a->in_pack > b->in_pack)
1300 return 1;
1301 return a->in_pack_offset < b->in_pack_offset ? -1 :
1302 (a->in_pack_offset > b->in_pack_offset);
1303}
1304static void get_object_details(void)
1306{
1307 uint32_t i;
1308 struct object_entry **sorted_by_offset;
1309 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1311 for (i = 0; i < nr_objects; i++)
1312 sorted_by_offset[i] = objects + i;
1313 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1314 prepare_pack_ix();
1316 for (i = 0; i < nr_objects; i++)
1317 check_object(sorted_by_offset[i]);
1318 free(sorted_by_offset);
1319}
1320static int type_size_sort(const void *_a, const void *_b)
1322{
1323 const struct object_entry *a = *(struct object_entry **)_a;
1324 const struct object_entry *b = *(struct object_entry **)_b;
1325 if (a->type < b->type)
1327 return -1;
1328 if (a->type > b->type)
1329 return 1;
1330 if (a->hash < b->hash)
1331 return -1;
1332 if (a->hash > b->hash)
1333 return 1;
1334 if (a->preferred_base < b->preferred_base)
1335 return -1;
1336 if (a->preferred_base > b->preferred_base)
1337 return 1;
1338 if (a->size < b->size)
1339 return -1;
1340 if (a->size > b->size)
1341 return 1;
1342 return a > b ? -1 : (a < b); /* newest last */
1343}
1344struct unpacked {
1346 struct object_entry *entry;
1347 void *data;
1348 struct delta_index *index;
1349};
1350/*
1352 * We search for deltas _backwards_ in a list sorted by type and
1353 * by size, so that we see progressively smaller and smaller files.
1354 * That's because we prefer deltas to be from the bigger file
1355 * to the smaller - deletes are potentially cheaper, but perhaps
1356 * more importantly, the bigger file is likely the more recent
1357 * one.
1358 */
1359static int try_delta(struct unpacked *trg, struct unpacked *src,
1360 unsigned max_depth)
1361{
1362 struct object_entry *trg_entry = trg->entry;
1363 struct object_entry *src_entry = src->entry;
1364 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1365 enum object_type type;
1366 void *delta_buf;
1367 /* Don't bother doing diffs between different types */
1369 if (trg_entry->type != src_entry->type)
1370 return -1;
1371 /* We do not compute delta to *create* objects we are not
1373 * going to pack.
1374 */
1375 if (trg_entry->preferred_base)
1376 return -1;
1377 /*
1379 * We do not bother to try a delta that we discarded
1380 * on an earlier try, but only when reusing delta data.
1381 */
1382 if (!no_reuse_delta && trg_entry->in_pack &&
1383 trg_entry->in_pack == src_entry->in_pack &&
1384 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1385 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1386 return 0;
1387 /* Let's not bust the allowed depth. */
1389 if (src_entry->depth >= max_depth)
1390 return 0;
1391 /* Now some size filtering heuristics. */
1393 trg_size = trg_entry->size;
1394 max_size = trg_size/2 - 20;
1395 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1396 if (max_size == 0)
1397 return 0;
1398 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1399 max_size = trg_entry->delta_size-1;
1400 src_size = src_entry->size;
1401 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1402 if (sizediff >= max_size)
1403 return 0;
1404 /* Load data if not already done */
1406 if (!trg->data) {
1407 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1408 if (sz != trg_size)
1409 die("object %s inconsistent object length (%lu vs %lu)",
1410 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1411 }
1412 if (!src->data) {
1413 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1414 if (sz != src_size)
1415 die("object %s inconsistent object length (%lu vs %lu)",
1416 sha1_to_hex(src_entry->sha1), sz, src_size);
1417 }
1418 if (!src->index) {
1419 src->index = create_delta_index(src->data, src_size);
1420 if (!src->index)
1421 die("out of memory");
1422 }
1423 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1425 if (!delta_buf)
1426 return 0;
1427 trg_entry->delta = src_entry;
1429 trg_entry->delta_size = delta_size;
1430 trg_entry->depth = src_entry->depth + 1;
1431 free(delta_buf);
1432 return 1;
1433}
1434static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1436{
1437 struct object_entry *child = me->delta_child;
1438 unsigned int m = n;
1439 while (child) {
1440 unsigned int c = check_delta_limit(child, n + 1);
1441 if (m < c)
1442 m = c;
1443 child = child->delta_sibling;
1444 }
1445 return m;
1446}
1447static void find_deltas(struct object_entry **list, int window, int depth)
1449{
1450 uint32_t i = nr_objects, idx = 0, processed = 0;
1451 unsigned int array_size = window * sizeof(struct unpacked);
1452 struct unpacked *array;
1453 int max_depth;
1454 if (!nr_objects)
1456 return;
1457 array = xmalloc(array_size);
1458 memset(array, 0, array_size);
1459 if (progress)
1460 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1461 do {
1463 struct object_entry *entry = list[--i];
1464 struct unpacked *n = array + idx;
1465 int j;
1466 if (!entry->preferred_base)
1468 processed++;
1469 if (progress)
1471 display_progress(&progress_state, processed);
1472 if (entry->delta)
1474 /* This happens if we decided to reuse existing
1475 * delta from a pack. "!no_reuse_delta &&" is implied.
1476 */
1477 continue;
1478 if (entry->size < 50)
1480 continue;
1481 free_delta_index(n->index);
1482 n->index = NULL;
1483 free(n->data);
1484 n->data = NULL;
1485 n->entry = entry;
1486 /*
1488 * If the current object is at pack edge, take the depth the
1489 * objects that depend on the current object into account
1490 * otherwise they would become too deep.
1491 */
1492 max_depth = depth;
1493 if (entry->delta_child) {
1494 max_depth -= check_delta_limit(entry, 0);
1495 if (max_depth <= 0)
1496 goto next;
1497 }
1498 j = window;
1500 while (--j > 0) {
1501 uint32_t other_idx = idx + j;
1502 struct unpacked *m;
1503 if (other_idx >= window)
1504 other_idx -= window;
1505 m = array + other_idx;
1506 if (!m->entry)
1507 break;
1508 if (try_delta(n, m, max_depth) < 0)
1509 break;
1510 }
1511 /* if we made n a delta, and if n is already at max
1513 * depth, leaving it in the window is pointless. we
1514 * should evict it first.
1515 */
1516 if (entry->delta && depth <= entry->depth)
1517 continue;
1518 next:
1520 idx++;
1521 if (idx >= window)
1522 idx = 0;
1523 } while (i > 0);
1524 if (progress)
1526 stop_progress(&progress_state);
1527 for (i = 0; i < window; ++i) {
1529 free_delta_index(array[i].index);
1530 free(array[i].data);
1531 }
1532 free(array);
1533}
1534static void prepare_pack(int window, int depth)
1536{
1537 struct object_entry **delta_list;
1538 uint32_t i;
1539 get_object_details();
1541 if (!window || !depth)
1543 return;
1544 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1546 for (i = 0; i < nr_objects; i++)
1547 delta_list[i] = objects + i;
1548 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1549 find_deltas(delta_list, window+1, depth);
1550 free(delta_list);
1551}
1552static int git_pack_config(const char *k, const char *v)
1554{
1555 if(!strcmp(k, "pack.window")) {
1556 window = git_config_int(k, v);
1557 return 0;
1558 }
1559 if(!strcmp(k, "pack.depth")) {
1560 depth = git_config_int(k, v);
1561 return 0;
1562 }
1563 if (!strcmp(k, "pack.compression")) {
1564 int level = git_config_int(k, v);
1565 if (level == -1)
1566 level = Z_DEFAULT_COMPRESSION;
1567 else if (level < 0 || level > Z_BEST_COMPRESSION)
1568 die("bad pack compression level %d", level);
1569 pack_compression_level = level;
1570 pack_compression_seen = 1;
1571 return 0;
1572 }
1573 return git_default_config(k, v);
1574}
1575static void read_object_list_from_stdin(void)
1577{
1578 char line[40 + 1 + PATH_MAX + 2];
1579 unsigned char sha1[20];
1580 unsigned hash;
1581 for (;;) {
1583 if (!fgets(line, sizeof(line), stdin)) {
1584 if (feof(stdin))
1585 break;
1586 if (!ferror(stdin))
1587 die("fgets returned NULL, not EOF, not error!");
1588 if (errno != EINTR)
1589 die("fgets: %s", strerror(errno));
1590 clearerr(stdin);
1591 continue;
1592 }
1593 if (line[0] == '-') {
1594 if (get_sha1_hex(line+1, sha1))
1595 die("expected edge sha1, got garbage:\n %s",
1596 line);
1597 add_preferred_base(sha1);
1598 continue;
1599 }
1600 if (get_sha1_hex(line, sha1))
1601 die("expected sha1, got garbage:\n %s", line);
1602 hash = name_hash(line+41);
1604 add_preferred_base_object(line+41, hash);
1605 add_object_entry(sha1, 0, hash, 0);
1606 }
1607}
1608static void show_commit(struct commit *commit)
1610{
1611 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
1612}
1613static void show_object(struct object_array_entry *p)
1615{
1616 unsigned hash = name_hash(p->name);
1617 add_preferred_base_object(p->name, hash);
1618 add_object_entry(p->item->sha1, p->item->type, hash, 0);
1619}
1620static void show_edge(struct commit *commit)
1622{
1623 add_preferred_base(commit->object.sha1);
1624}
1625static void get_object_list(int ac, const char **av)
1627{
1628 struct rev_info revs;
1629 char line[1000];
1630 int flags = 0;
1631 init_revisions(&revs, NULL);
1633 save_commit_buffer = 0;
1634 track_object_refs = 0;
1635 setup_revisions(ac, av, &revs, NULL);
1636 while (fgets(line, sizeof(line), stdin) != NULL) {
1638 int len = strlen(line);
1639 if (line[len - 1] == '\n')
1640 line[--len] = 0;
1641 if (!len)
1642 break;
1643 if (*line == '-') {
1644 if (!strcmp(line, "--not")) {
1645 flags ^= UNINTERESTING;
1646 continue;
1647 }
1648 die("not a rev '%s'", line);
1649 }
1650 if (handle_revision_arg(line, &revs, flags, 1))
1651 die("bad revision '%s'", line);
1652 }
1653 prepare_revision_walk(&revs);
1655 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1656 traverse_commit_list(&revs, show_commit, show_object);
1657}
1658static int adjust_perm(const char *path, mode_t mode)
1660{
1661 if (chmod(path, mode))
1662 return -1;
1663 return adjust_shared_perm(path);
1664}
1665int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1667{
1668 int use_internal_rev_list = 0;
1669 int thin = 0;
1670 uint32_t i;
1671 const char **rp_av;
1672 int rp_ac_alloc = 64;
1673 int rp_ac;
1674 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1676 rp_av[0] = "pack-objects";
1678 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1679 rp_ac = 2;
1680 git_config(git_pack_config);
1682 if (!pack_compression_seen && core_compression_seen)
1683 pack_compression_level = core_compression_level;
1684 progress = isatty(2);
1686 for (i = 1; i < argc; i++) {
1687 const char *arg = argv[i];
1688 if (*arg != '-')
1690 break;
1691 if (!strcmp("--non-empty", arg)) {
1693 non_empty = 1;
1694 continue;
1695 }
1696 if (!strcmp("--local", arg)) {
1697 local = 1;
1698 continue;
1699 }
1700 if (!strcmp("--incremental", arg)) {
1701 incremental = 1;
1702 continue;
1703 }
1704 if (!prefixcmp(arg, "--compression=")) {
1705 char *end;
1706 int level = strtoul(arg+14, &end, 0);
1707 if (!arg[14] || *end)
1708 usage(pack_usage);
1709 if (level == -1)
1710 level = Z_DEFAULT_COMPRESSION;
1711 else if (level < 0 || level > Z_BEST_COMPRESSION)
1712 die("bad pack compression level %d", level);
1713 pack_compression_level = level;
1714 continue;
1715 }
1716 if (!prefixcmp(arg, "--max-pack-size=")) {
1717 char *end;
1718 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1719 if (!arg[16] || *end)
1720 usage(pack_usage);
1721 continue;
1722 }
1723 if (!prefixcmp(arg, "--window=")) {
1724 char *end;
1725 window = strtoul(arg+9, &end, 0);
1726 if (!arg[9] || *end)
1727 usage(pack_usage);
1728 continue;
1729 }
1730 if (!prefixcmp(arg, "--depth=")) {
1731 char *end;
1732 depth = strtoul(arg+8, &end, 0);
1733 if (!arg[8] || *end)
1734 usage(pack_usage);
1735 continue;
1736 }
1737 if (!strcmp("--progress", arg)) {
1738 progress = 1;
1739 continue;
1740 }
1741 if (!strcmp("--all-progress", arg)) {
1742 progress = 2;
1743 continue;
1744 }
1745 if (!strcmp("-q", arg)) {
1746 progress = 0;
1747 continue;
1748 }
1749 if (!strcmp("--no-reuse-delta", arg)) {
1750 no_reuse_delta = 1;
1751 continue;
1752 }
1753 if (!strcmp("--no-reuse-object", arg)) {
1754 no_reuse_object = no_reuse_delta = 1;
1755 continue;
1756 }
1757 if (!strcmp("--delta-base-offset", arg)) {
1758 allow_ofs_delta = 1;
1759 continue;
1760 }
1761 if (!strcmp("--stdout", arg)) {
1762 pack_to_stdout = 1;
1763 continue;
1764 }
1765 if (!strcmp("--revs", arg)) {
1766 use_internal_rev_list = 1;
1767 continue;
1768 }
1769 if (!strcmp("--unpacked", arg) ||
1770 !prefixcmp(arg, "--unpacked=") ||
1771 !strcmp("--reflog", arg) ||
1772 !strcmp("--all", arg)) {
1773 use_internal_rev_list = 1;
1774 if (rp_ac >= rp_ac_alloc - 1) {
1775 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1776 rp_av = xrealloc(rp_av,
1777 rp_ac_alloc * sizeof(*rp_av));
1778 }
1779 rp_av[rp_ac++] = arg;
1780 continue;
1781 }
1782 if (!strcmp("--thin", arg)) {
1783 use_internal_rev_list = 1;
1784 thin = 1;
1785 rp_av[1] = "--objects-edge";
1786 continue;
1787 }
1788 if (!prefixcmp(arg, "--index-version=")) {
1789 char *c;
1790 index_default_version = strtoul(arg + 16, &c, 10);
1791 if (index_default_version > 2)
1792 die("bad %s", arg);
1793 if (*c == ',')
1794 index_off32_limit = strtoul(c+1, &c, 0);
1795 if (*c || index_off32_limit & 0x80000000)
1796 die("bad %s", arg);
1797 continue;
1798 }
1799 usage(pack_usage);
1800 }
1801 /* Traditionally "pack-objects [options] base extra" failed;
1803 * we would however want to take refs parameter that would
1804 * have been given to upstream rev-list ourselves, which means
1805 * we somehow want to say what the base name is. So the
1806 * syntax would be:
1807 *
1808 * pack-objects [options] base <refs...>
1809 *
1810 * in other words, we would treat the first non-option as the
1811 * base_name and send everything else to the internal revision
1812 * walker.
1813 */
1814 if (!pack_to_stdout)
1816 base_name = argv[i++];
1817 if (pack_to_stdout != !base_name)
1819 usage(pack_usage);
1820 if (!pack_to_stdout && thin)
1822 die("--thin cannot be used to build an indexable pack.");
1823 prepare_packed_git();
1825 if (progress)
1827 start_progress(&progress_state, "Generating pack...",
1828 "Counting objects: ", 0);
1829 if (!use_internal_rev_list)
1830 read_object_list_from_stdin();
1831 else {
1832 rp_av[rp_ac] = NULL;
1833 get_object_list(rp_ac, rp_av);
1834 }
1835 if (progress) {
1836 stop_progress(&progress_state);
1837 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1838 }
1839 if (non_empty && !nr_result)
1841 return 0;
1842 if (progress && (nr_objects != nr_result))
1843 fprintf(stderr, "Result has %u objects.\n", nr_result);
1844 if (nr_result)
1845 prepare_pack(window, depth);
1846 write_pack_file();
1847 if (progress)
1848 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1849 written, written_delta, reused, reused_delta);
1850 return 0;
1851}