11 #include "csum-file.h"
12 #include "tree-walk.h"
15 #include "list-objects.h"
18 static const char pack_usage[] = "\
19 git-pack-objects [{ -q | --progress | --all-progress }] [--max-pack-size=N] \n\
20 [--local] [--incremental] [--window=N] [--depth=N] \n\
21 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
22 [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
23 [--stdout | base-name] [<ref-list | <object-list]";
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 */
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
39 unsigned long delta_size; /* delta data size (uncompressed) */
40 enum object_type type;
41 enum object_type in_pack_type; /* could be delta */
42 unsigned char in_pack_header_size;
43 unsigned char preferred_base; /* we do not pack this, but is available
44 * to be used as the base object to delta
47 unsigned char no_try_delta;
51 * Objects we are going to pack are collected in objects array (dynamically
52 * expanded). nr_objects & nr_alloc controls this array. They are stored
53 * in the order we see -- typically rev-list --objects order that gives us
54 * nice "minimum seek" order.
56 static struct object_entry *objects;
57 static struct object_entry **written_list;
58 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
61 static int no_reuse_delta, no_reuse_object;
63 static int incremental;
64 static int allow_ofs_delta;
65 static const char *pack_tmp_name, *idx_tmp_name;
66 static char tmpname[PATH_MAX];
67 static const char *base_name;
68 static unsigned char pack_file_sha1[20];
69 static int progress = 1;
70 static int window = 10;
71 static uint32_t pack_size_limit;
72 static int depth = 50;
73 static int pack_to_stdout;
74 static int num_preferred_base;
75 static struct progress progress_state;
76 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
77 static int pack_compression_seen;
80 * The object names in objects array are hashed with this hashtable,
81 * to help looking up the entry by object name.
82 * This hashtable is built after all the objects are seen.
84 static int *object_ix;
85 static int object_ix_hashsz;
88 * Pack index for existing packs give us easy access to the offsets into
89 * corresponding pack file where each object's data starts, but the entries
90 * do not store the size of the compressed representation (uncompressed
91 * size is easily available by examining the pack entry header). It is
92 * also rather expensive to find the sha1 for an object given its offset.
94 * We build a hashtable of existing packs (pack_revindex), and keep reverse
95 * index here -- pack index file is sorted by object name mapping to offset;
96 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
97 * ordered by offset, so if you know the offset of an object, next offset
98 * is where its packed representation ends and the index_nr can be used to
99 * get the object sha1 from the main index.
101 struct revindex_entry {
105 struct pack_revindex {
106 struct packed_git *p;
107 struct revindex_entry *revindex;
109 static struct pack_revindex *pack_revindex;
110 static int pack_revindex_hashsz;
115 static uint32_t written, written_delta;
116 static uint32_t reused, reused_delta;
118 static int pack_revindex_ix(struct packed_git *p)
120 unsigned long ui = (unsigned long)p;
123 ui = ui ^ (ui >> 16); /* defeat structure alignment */
124 i = (int)(ui % pack_revindex_hashsz);
125 while (pack_revindex[i].p) {
126 if (pack_revindex[i].p == p)
128 if (++i == pack_revindex_hashsz)
134 static void prepare_pack_ix(void)
137 struct packed_git *p;
138 for (num = 0, p = packed_git; p; p = p->next)
142 pack_revindex_hashsz = num * 11;
143 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
144 for (p = packed_git; p; p = p->next) {
145 num = pack_revindex_ix(p);
147 pack_revindex[num].p = p;
149 /* revindex elements are lazily initialized */
152 static int cmp_offset(const void *a_, const void *b_)
154 const struct revindex_entry *a = a_;
155 const struct revindex_entry *b = b_;
156 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
160 * Ordered list of offsets of objects in the pack.
162 static void prepare_pack_revindex(struct pack_revindex *rix)
164 struct packed_git *p = rix->p;
165 int num_ent = p->num_objects;
167 const char *index = p->index_data;
169 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
172 if (p->index_version > 1) {
173 const uint32_t *off_32 =
174 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
175 const uint32_t *off_64 = off_32 + p->num_objects;
176 for (i = 0; i < num_ent; i++) {
177 uint32_t off = ntohl(*off_32++);
178 if (!(off & 0x80000000)) {
179 rix->revindex[i].offset = off;
181 rix->revindex[i].offset =
182 ((uint64_t)ntohl(*off_64++)) << 32;
183 rix->revindex[i].offset |=
186 rix->revindex[i].nr = i;
189 for (i = 0; i < num_ent; i++) {
190 uint32_t hl = *((uint32_t *)(index + 24 * i));
191 rix->revindex[i].offset = ntohl(hl);
192 rix->revindex[i].nr = i;
196 /* This knows the pack format -- the 20-byte trailer
197 * follows immediately after the last object data.
199 rix->revindex[num_ent].offset = p->pack_size - 20;
200 rix->revindex[num_ent].nr = -1;
201 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
204 static struct revindex_entry * find_packed_object(struct packed_git *p,
209 struct pack_revindex *rix;
210 struct revindex_entry *revindex;
211 num = pack_revindex_ix(p);
213 die("internal error: pack revindex uninitialized");
214 rix = &pack_revindex[num];
216 prepare_pack_revindex(rix);
217 revindex = rix->revindex;
219 hi = p->num_objects + 1;
221 int mi = (lo + hi) / 2;
222 if (revindex[mi].offset == ofs) {
223 return revindex + mi;
225 else if (ofs < revindex[mi].offset)
230 die("internal error: pack revindex corrupt");
233 static const unsigned char *find_packed_object_name(struct packed_git *p,
236 struct revindex_entry *entry = find_packed_object(p, ofs);
237 return nth_packed_object_sha1(p, entry->nr);
240 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
242 unsigned long othersize, delta_size;
243 enum object_type type;
244 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
248 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
249 delta_buf = diff_delta(otherbuf, othersize,
250 buf, size, &delta_size, 0);
251 if (!delta_buf || delta_size != entry->delta_size)
252 die("delta size changed");
259 * The per-object header is a pretty dense thing, which is
260 * - first byte: low four bits are "size", then three bits of "type",
261 * and the high bit is "size continues".
262 * - each byte afterwards: low seven bits are size continuation,
263 * with the high bit being "size continues"
265 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
270 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
271 die("bad type %d", type);
273 c = (type << 4) | (size & 15);
286 * we are going to reuse the existing object data as is. make
287 * sure it is not corrupt.
289 static int check_pack_inflate(struct packed_git *p,
290 struct pack_window **w_curs,
293 unsigned long expect)
296 unsigned char fakebuf[4096], *in;
299 memset(&stream, 0, sizeof(stream));
300 inflateInit(&stream);
302 in = use_pack(p, w_curs, offset, &stream.avail_in);
304 stream.next_out = fakebuf;
305 stream.avail_out = sizeof(fakebuf);
306 st = inflate(&stream, Z_FINISH);
307 offset += stream.next_in - in;
308 } while (st == Z_OK || st == Z_BUF_ERROR);
310 return (st == Z_STREAM_END &&
311 stream.total_out == expect &&
312 stream.total_in == len) ? 0 : -1;
315 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
316 off_t offset, off_t len, unsigned int nr)
318 const uint32_t *index_crc;
319 uint32_t data_crc = crc32(0, Z_NULL, 0);
323 void *data = use_pack(p, w_curs, offset, &avail);
326 data_crc = crc32(data_crc, data, avail);
331 index_crc = p->index_data;
332 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
334 return data_crc != ntohl(*index_crc);
337 static void copy_pack_data(struct sha1file *f,
338 struct packed_git *p,
339 struct pack_window **w_curs,
347 in = use_pack(p, w_curs, offset, &avail);
349 avail = (unsigned int)len;
350 sha1write(f, in, avail);
356 static unsigned long write_object(struct sha1file *f,
357 struct object_entry *entry,
361 enum object_type type;
363 unsigned char header[10];
364 unsigned char dheader[10];
367 enum object_type obj_type;
369 /* write limit if limited packsize and not first object */
370 unsigned long limit = pack_size_limit && nr_written ?
371 pack_size_limit - write_offset : 0;
373 int usable_delta = !entry->delta ? 0 :
374 /* yes if unlimited packfile */
375 !pack_size_limit ? 1 :
376 /* no if base written to previous pack */
377 entry->delta->offset == (off_t)-1 ? 0 :
378 /* otherwise double-check written to this
379 * pack, like we do below
381 entry->delta->offset ? 1 : 0;
386 obj_type = entry->type;
388 to_reuse = 0; /* explicit */
389 else if (!entry->in_pack)
390 to_reuse = 0; /* can't reuse what we don't have */
391 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
392 /* check_object() decided it for us ... */
393 to_reuse = usable_delta;
394 /* ... but pack split may override that */
395 else if (obj_type != entry->in_pack_type)
396 to_reuse = 0; /* pack has delta which is unusable */
397 else if (entry->delta)
398 to_reuse = 0; /* we want to pack afresh */
400 to_reuse = 1; /* we have it in-pack undeltified,
401 * and we do not need to deltify it.
406 unsigned long maxsize;
408 buf = read_sha1_file(entry->sha1, &type, &size);
410 die("unable to read %s", sha1_to_hex(entry->sha1));
411 if (size != entry->size)
412 die("object %s size inconsistency (%lu vs %lu)",
413 sha1_to_hex(entry->sha1), size, entry->size);
415 buf = delta_against(buf, size, entry);
416 size = entry->delta_size;
417 obj_type = (allow_ofs_delta && entry->delta->offset) ?
418 OBJ_OFS_DELTA : OBJ_REF_DELTA;
421 * recover real object type in case
422 * check_object() wanted to re-use a delta,
423 * but we couldn't since base was in previous split pack
427 /* compress the data to store and put compressed length in datalen */
428 memset(&stream, 0, sizeof(stream));
429 deflateInit(&stream, pack_compression_level);
430 maxsize = deflateBound(&stream, size);
431 out = xmalloc(maxsize);
433 stream.next_in = buf;
434 stream.avail_in = size;
435 stream.next_out = out;
436 stream.avail_out = maxsize;
437 while (deflate(&stream, Z_FINISH) == Z_OK)
440 datalen = stream.total_out;
443 * The object header is a byte of 'type' followed by zero or
444 * more bytes of length.
446 hdrlen = encode_header(obj_type, size, header);
448 if (obj_type == OBJ_OFS_DELTA) {
450 * Deltas with relative base contain an additional
451 * encoding of the relative offset for the delta
452 * base from this object's position in the pack.
454 off_t ofs = entry->offset - entry->delta->offset;
455 unsigned pos = sizeof(dheader) - 1;
456 dheader[pos] = ofs & 127;
458 dheader[--pos] = 128 | (--ofs & 127);
459 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
464 sha1write(f, header, hdrlen);
465 sha1write(f, dheader + pos, sizeof(dheader) - pos);
466 hdrlen += sizeof(dheader) - pos;
467 } else if (obj_type == OBJ_REF_DELTA) {
469 * Deltas with a base reference contain
470 * an additional 20 bytes for the base sha1.
472 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
477 sha1write(f, header, hdrlen);
478 sha1write(f, entry->delta->sha1, 20);
481 if (limit && hdrlen + datalen + 20 >= limit) {
486 sha1write(f, header, hdrlen);
488 sha1write(f, out, datalen);
493 struct packed_git *p = entry->in_pack;
494 struct pack_window *w_curs = NULL;
495 struct revindex_entry *revidx;
499 obj_type = (allow_ofs_delta && entry->delta->offset) ?
500 OBJ_OFS_DELTA : OBJ_REF_DELTA;
503 hdrlen = encode_header(obj_type, entry->size, header);
504 offset = entry->in_pack_offset;
505 revidx = find_packed_object(p, offset);
506 datalen = revidx[1].offset - offset;
507 if (!pack_to_stdout && p->index_version > 1 &&
508 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
509 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
510 offset += entry->in_pack_header_size;
511 datalen -= entry->in_pack_header_size;
512 if (obj_type == OBJ_OFS_DELTA) {
513 off_t ofs = entry->offset - entry->delta->offset;
514 unsigned pos = sizeof(dheader) - 1;
515 dheader[pos] = ofs & 127;
517 dheader[--pos] = 128 | (--ofs & 127);
518 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
520 sha1write(f, header, hdrlen);
521 sha1write(f, dheader + pos, sizeof(dheader) - pos);
522 hdrlen += sizeof(dheader) - pos;
523 } else if (obj_type == OBJ_REF_DELTA) {
524 if (limit && hdrlen + 20 + datalen + 20 >= limit)
526 sha1write(f, header, hdrlen);
527 sha1write(f, entry->delta->sha1, 20);
530 if (limit && hdrlen + datalen + 20 >= limit)
532 sha1write(f, header, hdrlen);
535 if (!pack_to_stdout && p->index_version == 1 &&
536 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
537 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
538 copy_pack_data(f, p, &w_curs, offset, datalen);
546 entry->crc32 = crc32_end(f);
547 return hdrlen + datalen;
550 static off_t write_one(struct sha1file *f,
551 struct object_entry *e,
556 /* offset is non zero if object is written already. */
557 if (e->offset || e->preferred_base)
560 /* if we are deltified, write out base object first. */
562 offset = write_one(f, e->delta, offset);
568 size = write_object(f, e, offset);
573 written_list[nr_written++] = e;
575 /* make sure off_t is sufficiently large not to wrap */
576 if (offset > offset + size)
577 die("pack too large for current definition of off_t");
578 return offset + size;
581 static int open_object_dir_tmp(const char *path)
583 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
584 return mkstemp(tmpname);
587 /* forward declarations for write_pack_file */
588 static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
589 static int adjust_perm(const char *path, mode_t mode);
591 static void write_pack_file(void)
595 off_t offset, offset_one, last_obj_offset = 0;
596 struct pack_header hdr;
597 int do_progress = progress >> pack_to_stdout;
598 uint32_t nr_remaining = nr_result;
601 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
602 written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
605 if (pack_to_stdout) {
606 f = sha1fd(1, "<stdout>");
608 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
610 die("unable to create %s: %s\n", tmpname, strerror(errno));
611 pack_tmp_name = xstrdup(tmpname);
612 f = sha1fd(fd, pack_tmp_name);
615 hdr.hdr_signature = htonl(PACK_SIGNATURE);
616 hdr.hdr_version = htonl(PACK_VERSION);
617 hdr.hdr_entries = htonl(nr_remaining);
618 sha1write(f, &hdr, sizeof(hdr));
619 offset = sizeof(hdr);
621 for (; i < nr_objects; i++) {
622 last_obj_offset = offset;
623 offset_one = write_one(f, objects + i, offset);
628 display_progress(&progress_state, written);
632 * Did we write the wrong # entries in the header?
633 * If so, rewrite it like in fast-import
635 if (pack_to_stdout || nr_written == nr_remaining) {
636 sha1close(f, pack_file_sha1, 1);
638 sha1close(f, pack_file_sha1, 0);
639 fixup_pack_header_footer(f->fd, pack_file_sha1, pack_tmp_name, nr_written);
643 if (!pack_to_stdout) {
644 unsigned char object_list_sha1[20];
645 mode_t mode = umask(0);
650 write_index_file(last_obj_offset, object_list_sha1);
651 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
652 base_name, sha1_to_hex(object_list_sha1));
653 if (adjust_perm(pack_tmp_name, mode))
654 die("unable to make temporary pack file readable: %s",
656 if (rename(pack_tmp_name, tmpname))
657 die("unable to rename temporary pack file: %s",
659 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
660 base_name, sha1_to_hex(object_list_sha1));
661 if (adjust_perm(idx_tmp_name, mode))
662 die("unable to make temporary index file readable: %s",
664 if (rename(idx_tmp_name, tmpname))
665 die("unable to rename temporary index file: %s",
667 puts(sha1_to_hex(object_list_sha1));
670 /* mark written objects as written to previous pack */
671 for (j = 0; j < nr_written; j++) {
672 written_list[j]->offset = (off_t)-1;
674 nr_remaining -= nr_written;
675 } while (nr_remaining && i < nr_objects);
679 stop_progress(&progress_state);
680 if (written != nr_result)
681 die("wrote %u objects while expecting %u", written, nr_result);
683 * We have scanned through [0 ... i). Since we have written
684 * the correct number of objects, the remaining [i ... nr_objects)
685 * items must be either already written (due to out-of-order delta base)
686 * or a preferred base. Count those which are neither and complain if any.
688 for (j = 0; i < nr_objects; i++) {
689 struct object_entry *e = objects + i;
690 j += !e->offset && !e->preferred_base;
693 die("wrote %u objects as expected but %u unwritten", written, j);
696 static int sha1_sort(const void *_a, const void *_b)
698 const struct object_entry *a = *(struct object_entry **)_a;
699 const struct object_entry *b = *(struct object_entry **)_b;
700 return hashcmp(a->sha1, b->sha1);
703 static uint32_t index_default_version = 1;
704 static uint32_t index_off32_limit = 0x7fffffff;
706 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
709 struct object_entry **sorted_by_sha, **list, **last;
711 uint32_t i, index_version;
714 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
716 die("unable to create %s: %s\n", tmpname, strerror(errno));
717 idx_tmp_name = xstrdup(tmpname);
718 f = sha1fd(fd, idx_tmp_name);
721 sorted_by_sha = written_list;
722 qsort(sorted_by_sha, nr_written, sizeof(*sorted_by_sha), sha1_sort);
723 list = sorted_by_sha;
724 last = sorted_by_sha + nr_written;
726 sorted_by_sha = list = last = NULL;
728 /* if last object's offset is >= 2^31 we should use index V2 */
729 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
731 /* index versions 2 and above need a header */
732 if (index_version >= 2) {
733 struct pack_idx_header hdr;
734 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
735 hdr.idx_version = htonl(index_version);
736 sha1write(f, &hdr, sizeof(hdr));
740 * Write the first-level table (the list is sorted,
741 * but we use a 256-entry lookup to be able to avoid
742 * having to do eight extra binary search iterations).
744 for (i = 0; i < 256; i++) {
745 struct object_entry **next = list;
746 while (next < last) {
747 struct object_entry *entry = *next;
748 if (entry->sha1[0] != i)
752 array[i] = htonl(next - sorted_by_sha);
755 sha1write(f, array, 256 * 4);
757 /* Compute the SHA1 hash of sorted object names. */
760 /* Write the actual SHA1 entries. */
761 list = sorted_by_sha;
762 for (i = 0; i < nr_written; i++) {
763 struct object_entry *entry = *list++;
764 if (index_version < 2) {
765 uint32_t offset = htonl(entry->offset);
766 sha1write(f, &offset, 4);
768 sha1write(f, entry->sha1, 20);
769 SHA1_Update(&ctx, entry->sha1, 20);
772 if (index_version >= 2) {
773 unsigned int nr_large_offset = 0;
775 /* write the crc32 table */
776 list = sorted_by_sha;
777 for (i = 0; i < nr_written; i++) {
778 struct object_entry *entry = *list++;
779 uint32_t crc32_val = htonl(entry->crc32);
780 sha1write(f, &crc32_val, 4);
783 /* write the 32-bit offset table */
784 list = sorted_by_sha;
785 for (i = 0; i < nr_written; i++) {
786 struct object_entry *entry = *list++;
787 uint32_t offset = (entry->offset <= index_off32_limit) ?
788 entry->offset : (0x80000000 | nr_large_offset++);
789 offset = htonl(offset);
790 sha1write(f, &offset, 4);
793 /* write the large offset table */
794 list = sorted_by_sha;
795 while (nr_large_offset) {
796 struct object_entry *entry = *list++;
797 uint64_t offset = entry->offset;
798 if (offset > index_off32_limit) {
800 split[0] = htonl(offset >> 32);
801 split[1] = htonl(offset & 0xffffffff);
802 sha1write(f, split, 8);
808 sha1write(f, pack_file_sha1, 20);
809 sha1close(f, NULL, 1);
810 SHA1_Final(sha1, &ctx);
813 static int locate_object_entry_hash(const unsigned char *sha1)
817 memcpy(&ui, sha1, sizeof(unsigned int));
818 i = ui % object_ix_hashsz;
819 while (0 < object_ix[i]) {
820 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
822 if (++i == object_ix_hashsz)
828 static struct object_entry *locate_object_entry(const unsigned char *sha1)
832 if (!object_ix_hashsz)
835 i = locate_object_entry_hash(sha1);
837 return &objects[object_ix[i]-1];
841 static void rehash_objects(void)
844 struct object_entry *oe;
846 object_ix_hashsz = nr_objects * 3;
847 if (object_ix_hashsz < 1024)
848 object_ix_hashsz = 1024;
849 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
850 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
851 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
852 int ix = locate_object_entry_hash(oe->sha1);
856 object_ix[ix] = i + 1;
860 static unsigned name_hash(const char *name)
869 * This effectively just creates a sortable number from the
870 * last sixteen non-whitespace characters. Last characters
871 * count "most", so things that end in ".c" sort together.
873 while ((c = *name++) != 0) {
876 hash = (hash >> 2) + (c << 24);
881 static void setup_delta_attr_check(struct git_attr_check *check)
883 static struct git_attr *attr_delta;
886 attr_delta = git_attr("delta", 5);
888 check[0].attr = attr_delta;
891 static int no_try_delta(const char *path)
893 struct git_attr_check check[1];
895 setup_delta_attr_check(check);
896 if (git_checkattr(path, ARRAY_SIZE(check), check))
898 if (ATTR_FALSE(check->value))
903 static int add_object_entry(const unsigned char *sha1, enum object_type type,
904 const char *name, int exclude)
906 struct object_entry *entry;
907 struct packed_git *p, *found_pack = NULL;
908 off_t found_offset = 0;
910 unsigned hash = name_hash(name);
912 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
915 entry = objects + object_ix[ix] - 1;
916 if (!entry->preferred_base)
918 entry->preferred_base = 1;
923 for (p = packed_git; p; p = p->next) {
924 off_t offset = find_pack_entry_one(sha1, p);
927 found_offset = offset;
934 if (local && !p->pack_local)
939 if (nr_objects >= nr_alloc) {
940 nr_alloc = (nr_alloc + 1024) * 3 / 2;
941 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
944 entry = objects + nr_objects++;
945 memset(entry, 0, sizeof(*entry));
946 hashcpy(entry->sha1, sha1);
951 entry->preferred_base = 1;
955 entry->in_pack = found_pack;
956 entry->in_pack_offset = found_offset;
959 if (object_ix_hashsz * 3 <= nr_objects * 4)
962 object_ix[-1 - ix] = nr_objects;
965 display_progress(&progress_state, nr_objects);
967 if (name && no_try_delta(name))
968 entry->no_try_delta = 1;
973 struct pbase_tree_cache {
974 unsigned char sha1[20];
978 unsigned long tree_size;
981 static struct pbase_tree_cache *(pbase_tree_cache[256]);
982 static int pbase_tree_cache_ix(const unsigned char *sha1)
984 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
986 static int pbase_tree_cache_ix_incr(int ix)
988 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
991 static struct pbase_tree {
992 struct pbase_tree *next;
993 /* This is a phony "cache" entry; we are not
994 * going to evict it nor find it through _get()
995 * mechanism -- this is for the toplevel node that
996 * would almost always change with any commit.
998 struct pbase_tree_cache pcache;
1001 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1003 struct pbase_tree_cache *ent, *nent;
1006 enum object_type type;
1008 int my_ix = pbase_tree_cache_ix(sha1);
1009 int available_ix = -1;
1011 /* pbase-tree-cache acts as a limited hashtable.
1012 * your object will be found at your index or within a few
1013 * slots after that slot if it is cached.
1015 for (neigh = 0; neigh < 8; neigh++) {
1016 ent = pbase_tree_cache[my_ix];
1017 if (ent && !hashcmp(ent->sha1, sha1)) {
1021 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1022 ((0 <= available_ix) &&
1023 (!ent && pbase_tree_cache[available_ix])))
1024 available_ix = my_ix;
1027 my_ix = pbase_tree_cache_ix_incr(my_ix);
1030 /* Did not find one. Either we got a bogus request or
1031 * we need to read and perhaps cache.
1033 data = read_sha1_file(sha1, &type, &size);
1036 if (type != OBJ_TREE) {
1041 /* We need to either cache or return a throwaway copy */
1043 if (available_ix < 0)
1046 ent = pbase_tree_cache[available_ix];
1047 my_ix = available_ix;
1051 nent = xmalloc(sizeof(*nent));
1052 nent->temporary = (available_ix < 0);
1055 /* evict and reuse */
1056 free(ent->tree_data);
1059 hashcpy(nent->sha1, sha1);
1060 nent->tree_data = data;
1061 nent->tree_size = size;
1063 if (!nent->temporary)
1064 pbase_tree_cache[my_ix] = nent;
1068 static void pbase_tree_put(struct pbase_tree_cache *cache)
1070 if (!cache->temporary) {
1074 free(cache->tree_data);
1078 static int name_cmp_len(const char *name)
1081 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1086 static void add_pbase_object(struct tree_desc *tree,
1089 const char *fullname)
1091 struct name_entry entry;
1094 while (tree_entry(tree,&entry)) {
1095 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1096 memcmp(name, entry.path, cmplen);
1101 if (name[cmplen] != '/') {
1102 add_object_entry(entry.sha1,
1103 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1107 if (S_ISDIR(entry.mode)) {
1108 struct tree_desc sub;
1109 struct pbase_tree_cache *tree;
1110 const char *down = name+cmplen+1;
1111 int downlen = name_cmp_len(down);
1113 tree = pbase_tree_get(entry.sha1);
1116 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1118 add_pbase_object(&sub, down, downlen, fullname);
1119 pbase_tree_put(tree);
1124 static unsigned *done_pbase_paths;
1125 static int done_pbase_paths_num;
1126 static int done_pbase_paths_alloc;
1127 static int done_pbase_path_pos(unsigned hash)
1130 int hi = done_pbase_paths_num;
1132 int mi = (hi + lo) / 2;
1133 if (done_pbase_paths[mi] == hash)
1135 if (done_pbase_paths[mi] < hash)
1143 static int check_pbase_path(unsigned hash)
1145 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1149 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1150 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1151 done_pbase_paths = xrealloc(done_pbase_paths,
1152 done_pbase_paths_alloc *
1155 done_pbase_paths_num++;
1156 if (pos < done_pbase_paths_num)
1157 memmove(done_pbase_paths + pos + 1,
1158 done_pbase_paths + pos,
1159 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1160 done_pbase_paths[pos] = hash;
1164 static void add_preferred_base_object(const char *name)
1166 struct pbase_tree *it;
1168 unsigned hash = name_hash(name);
1170 if (!num_preferred_base || check_pbase_path(hash))
1173 cmplen = name_cmp_len(name);
1174 for (it = pbase_tree; it; it = it->next) {
1176 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1179 struct tree_desc tree;
1180 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1181 add_pbase_object(&tree, name, cmplen, name);
1186 static void add_preferred_base(unsigned char *sha1)
1188 struct pbase_tree *it;
1191 unsigned char tree_sha1[20];
1193 if (window <= num_preferred_base++)
1196 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1200 for (it = pbase_tree; it; it = it->next) {
1201 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1207 it = xcalloc(1, sizeof(*it));
1208 it->next = pbase_tree;
1211 hashcpy(it->pcache.sha1, tree_sha1);
1212 it->pcache.tree_data = data;
1213 it->pcache.tree_size = size;
1216 static void check_object(struct object_entry *entry)
1218 if (entry->in_pack) {
1219 struct packed_git *p = entry->in_pack;
1220 struct pack_window *w_curs = NULL;
1221 const unsigned char *base_ref = NULL;
1222 struct object_entry *base_entry;
1223 unsigned long used, used_0;
1226 unsigned char *buf, c;
1228 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1231 * We want in_pack_type even if we do not reuse delta
1232 * since non-delta representations could still be reused.
1234 used = unpack_object_header_gently(buf, avail,
1235 &entry->in_pack_type,
1239 * Determine if this is a delta and if so whether we can
1240 * reuse it or not. Otherwise let's find out as cheaply as
1241 * possible what the actual type and size for this object is.
1243 switch (entry->in_pack_type) {
1245 /* Not a delta hence we've already got all we need. */
1246 entry->type = entry->in_pack_type;
1247 entry->in_pack_header_size = used;
1248 unuse_pack(&w_curs);
1251 if (!no_reuse_delta && !entry->preferred_base)
1252 base_ref = use_pack(p, &w_curs,
1253 entry->in_pack_offset + used, NULL);
1254 entry->in_pack_header_size = used + 20;
1257 buf = use_pack(p, &w_curs,
1258 entry->in_pack_offset + used, NULL);
1264 if (!ofs || MSB(ofs, 7))
1265 die("delta base offset overflow in pack for %s",
1266 sha1_to_hex(entry->sha1));
1268 ofs = (ofs << 7) + (c & 127);
1270 if (ofs >= entry->in_pack_offset)
1271 die("delta base offset out of bound for %s",
1272 sha1_to_hex(entry->sha1));
1273 ofs = entry->in_pack_offset - ofs;
1274 if (!no_reuse_delta && !entry->preferred_base)
1275 base_ref = find_packed_object_name(p, ofs);
1276 entry->in_pack_header_size = used + used_0;
1280 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1282 * If base_ref was set above that means we wish to
1283 * reuse delta data, and we even found that base
1284 * in the list of objects we want to pack. Goodie!
1286 * Depth value does not matter - find_deltas() will
1287 * never consider reused delta as the base object to
1288 * deltify other objects against, in order to avoid
1291 entry->type = entry->in_pack_type;
1292 entry->delta = base_entry;
1293 entry->delta_sibling = base_entry->delta_child;
1294 base_entry->delta_child = entry;
1295 unuse_pack(&w_curs);
1301 * This must be a delta and we already know what the
1302 * final object type is. Let's extract the actual
1303 * object size from the delta header.
1305 entry->size = get_size_from_delta(p, &w_curs,
1306 entry->in_pack_offset + entry->in_pack_header_size);
1307 unuse_pack(&w_curs);
1312 * No choice but to fall back to the recursive delta walk
1313 * with sha1_object_info() to find about the object type
1316 unuse_pack(&w_curs);
1319 entry->type = sha1_object_info(entry->sha1, &entry->size);
1320 if (entry->type < 0)
1321 die("unable to get type of object %s",
1322 sha1_to_hex(entry->sha1));
1325 static int pack_offset_sort(const void *_a, const void *_b)
1327 const struct object_entry *a = *(struct object_entry **)_a;
1328 const struct object_entry *b = *(struct object_entry **)_b;
1330 /* avoid filesystem trashing with loose objects */
1331 if (!a->in_pack && !b->in_pack)
1332 return hashcmp(a->sha1, b->sha1);
1334 if (a->in_pack < b->in_pack)
1336 if (a->in_pack > b->in_pack)
1338 return a->in_pack_offset < b->in_pack_offset ? -1 :
1339 (a->in_pack_offset > b->in_pack_offset);
1342 static void get_object_details(void)
1345 struct object_entry **sorted_by_offset;
1347 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1348 for (i = 0; i < nr_objects; i++)
1349 sorted_by_offset[i] = objects + i;
1350 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1353 for (i = 0; i < nr_objects; i++)
1354 check_object(sorted_by_offset[i]);
1355 free(sorted_by_offset);
1358 static int type_size_sort(const void *_a, const void *_b)
1360 const struct object_entry *a = *(struct object_entry **)_a;
1361 const struct object_entry *b = *(struct object_entry **)_b;
1363 if (a->type < b->type)
1365 if (a->type > b->type)
1367 if (a->hash < b->hash)
1369 if (a->hash > b->hash)
1371 if (a->preferred_base < b->preferred_base)
1373 if (a->preferred_base > b->preferred_base)
1375 if (a->size < b->size)
1377 if (a->size > b->size)
1379 return a > b ? -1 : (a < b); /* newest last */
1383 struct object_entry *entry;
1385 struct delta_index *index;
1389 * We search for deltas _backwards_ in a list sorted by type and
1390 * by size, so that we see progressively smaller and smaller files.
1391 * That's because we prefer deltas to be from the bigger file
1392 * to the smaller - deletes are potentially cheaper, but perhaps
1393 * more importantly, the bigger file is likely the more recent
1396 static int try_delta(struct unpacked *trg, struct unpacked *src,
1399 struct object_entry *trg_entry = trg->entry;
1400 struct object_entry *src_entry = src->entry;
1401 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1402 enum object_type type;
1405 /* Don't bother doing diffs between different types */
1406 if (trg_entry->type != src_entry->type)
1409 /* We do not compute delta to *create* objects we are not
1412 if (trg_entry->preferred_base)
1416 * We do not bother to try a delta that we discarded
1417 * on an earlier try, but only when reusing delta data.
1419 if (!no_reuse_delta && trg_entry->in_pack &&
1420 trg_entry->in_pack == src_entry->in_pack &&
1421 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1422 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1425 /* Let's not bust the allowed depth. */
1426 if (src_entry->depth >= max_depth)
1429 /* Now some size filtering heuristics. */
1430 trg_size = trg_entry->size;
1431 max_size = trg_size/2 - 20;
1432 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1435 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1436 max_size = trg_entry->delta_size-1;
1437 src_size = src_entry->size;
1438 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1439 if (sizediff >= max_size)
1442 /* Load data if not already done */
1444 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1446 die("object %s inconsistent object length (%lu vs %lu)",
1447 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1450 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1452 die("object %s inconsistent object length (%lu vs %lu)",
1453 sha1_to_hex(src_entry->sha1), sz, src_size);
1456 src->index = create_delta_index(src->data, src_size);
1458 die("out of memory");
1461 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1465 trg_entry->delta = src_entry;
1466 trg_entry->delta_size = delta_size;
1467 trg_entry->depth = src_entry->depth + 1;
1472 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1474 struct object_entry *child = me->delta_child;
1477 unsigned int c = check_delta_limit(child, n + 1);
1480 child = child->delta_sibling;
1485 static void find_deltas(struct object_entry **list, int window, int depth)
1487 uint32_t i = nr_objects, idx = 0, processed = 0;
1488 unsigned int array_size = window * sizeof(struct unpacked);
1489 struct unpacked *array;
1494 array = xmalloc(array_size);
1495 memset(array, 0, array_size);
1497 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1500 struct object_entry *entry = list[--i];
1501 struct unpacked *n = array + idx;
1504 if (!entry->preferred_base)
1508 display_progress(&progress_state, processed);
1511 /* This happens if we decided to reuse existing
1512 * delta from a pack. "!no_reuse_delta &&" is implied.
1516 if (entry->size < 50)
1519 if (entry->no_try_delta)
1522 free_delta_index(n->index);
1529 * If the current object is at pack edge, take the depth the
1530 * objects that depend on the current object into account
1531 * otherwise they would become too deep.
1534 if (entry->delta_child) {
1535 max_depth -= check_delta_limit(entry, 0);
1542 uint32_t other_idx = idx + j;
1544 if (other_idx >= window)
1545 other_idx -= window;
1546 m = array + other_idx;
1549 if (try_delta(n, m, max_depth) < 0)
1553 /* if we made n a delta, and if n is already at max
1554 * depth, leaving it in the window is pointless. we
1555 * should evict it first.
1557 if (entry->delta && depth <= entry->depth)
1567 stop_progress(&progress_state);
1569 for (i = 0; i < window; ++i) {
1570 free_delta_index(array[i].index);
1571 free(array[i].data);
1576 static void prepare_pack(int window, int depth)
1578 struct object_entry **delta_list;
1581 get_object_details();
1583 if (!window || !depth)
1586 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1587 for (i = 0; i < nr_objects; i++)
1588 delta_list[i] = objects + i;
1589 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1590 find_deltas(delta_list, window+1, depth);
1594 static int git_pack_config(const char *k, const char *v)
1596 if(!strcmp(k, "pack.window")) {
1597 window = git_config_int(k, v);
1600 if(!strcmp(k, "pack.depth")) {
1601 depth = git_config_int(k, v);
1604 if (!strcmp(k, "pack.compression")) {
1605 int level = git_config_int(k, v);
1607 level = Z_DEFAULT_COMPRESSION;
1608 else if (level < 0 || level > Z_BEST_COMPRESSION)
1609 die("bad pack compression level %d", level);
1610 pack_compression_level = level;
1611 pack_compression_seen = 1;
1614 return git_default_config(k, v);
1617 static void read_object_list_from_stdin(void)
1619 char line[40 + 1 + PATH_MAX + 2];
1620 unsigned char sha1[20];
1623 if (!fgets(line, sizeof(line), stdin)) {
1627 die("fgets returned NULL, not EOF, not error!");
1629 die("fgets: %s", strerror(errno));
1633 if (line[0] == '-') {
1634 if (get_sha1_hex(line+1, sha1))
1635 die("expected edge sha1, got garbage:\n %s",
1637 add_preferred_base(sha1);
1640 if (get_sha1_hex(line, sha1))
1641 die("expected sha1, got garbage:\n %s", line);
1643 add_preferred_base_object(line+41);
1644 add_object_entry(sha1, 0, line+41, 0);
1648 static void show_commit(struct commit *commit)
1650 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1653 static void show_object(struct object_array_entry *p)
1655 add_preferred_base_object(p->name);
1656 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1659 static void show_edge(struct commit *commit)
1661 add_preferred_base(commit->object.sha1);
1664 static void get_object_list(int ac, const char **av)
1666 struct rev_info revs;
1670 init_revisions(&revs, NULL);
1671 save_commit_buffer = 0;
1672 track_object_refs = 0;
1673 setup_revisions(ac, av, &revs, NULL);
1675 while (fgets(line, sizeof(line), stdin) != NULL) {
1676 int len = strlen(line);
1677 if (line[len - 1] == '\n')
1682 if (!strcmp(line, "--not")) {
1683 flags ^= UNINTERESTING;
1686 die("not a rev '%s'", line);
1688 if (handle_revision_arg(line, &revs, flags, 1))
1689 die("bad revision '%s'", line);
1692 prepare_revision_walk(&revs);
1693 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1694 traverse_commit_list(&revs, show_commit, show_object);
1697 static int adjust_perm(const char *path, mode_t mode)
1699 if (chmod(path, mode))
1701 return adjust_shared_perm(path);
1704 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1706 int use_internal_rev_list = 0;
1710 int rp_ac_alloc = 64;
1713 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1715 rp_av[0] = "pack-objects";
1716 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1719 git_config(git_pack_config);
1720 if (!pack_compression_seen && core_compression_seen)
1721 pack_compression_level = core_compression_level;
1723 progress = isatty(2);
1724 for (i = 1; i < argc; i++) {
1725 const char *arg = argv[i];
1730 if (!strcmp("--non-empty", arg)) {
1734 if (!strcmp("--local", arg)) {
1738 if (!strcmp("--incremental", arg)) {
1742 if (!prefixcmp(arg, "--compression=")) {
1744 int level = strtoul(arg+14, &end, 0);
1745 if (!arg[14] || *end)
1748 level = Z_DEFAULT_COMPRESSION;
1749 else if (level < 0 || level > Z_BEST_COMPRESSION)
1750 die("bad pack compression level %d", level);
1751 pack_compression_level = level;
1754 if (!prefixcmp(arg, "--max-pack-size=")) {
1756 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1757 if (!arg[16] || *end)
1761 if (!prefixcmp(arg, "--window=")) {
1763 window = strtoul(arg+9, &end, 0);
1764 if (!arg[9] || *end)
1768 if (!prefixcmp(arg, "--depth=")) {
1770 depth = strtoul(arg+8, &end, 0);
1771 if (!arg[8] || *end)
1775 if (!strcmp("--progress", arg)) {
1779 if (!strcmp("--all-progress", arg)) {
1783 if (!strcmp("-q", arg)) {
1787 if (!strcmp("--no-reuse-delta", arg)) {
1791 if (!strcmp("--no-reuse-object", arg)) {
1792 no_reuse_object = no_reuse_delta = 1;
1795 if (!strcmp("--delta-base-offset", arg)) {
1796 allow_ofs_delta = 1;
1799 if (!strcmp("--stdout", arg)) {
1803 if (!strcmp("--revs", arg)) {
1804 use_internal_rev_list = 1;
1807 if (!strcmp("--unpacked", arg) ||
1808 !prefixcmp(arg, "--unpacked=") ||
1809 !strcmp("--reflog", arg) ||
1810 !strcmp("--all", arg)) {
1811 use_internal_rev_list = 1;
1812 if (rp_ac >= rp_ac_alloc - 1) {
1813 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1814 rp_av = xrealloc(rp_av,
1815 rp_ac_alloc * sizeof(*rp_av));
1817 rp_av[rp_ac++] = arg;
1820 if (!strcmp("--thin", arg)) {
1821 use_internal_rev_list = 1;
1823 rp_av[1] = "--objects-edge";
1826 if (!prefixcmp(arg, "--index-version=")) {
1828 index_default_version = strtoul(arg + 16, &c, 10);
1829 if (index_default_version > 2)
1832 index_off32_limit = strtoul(c+1, &c, 0);
1833 if (*c || index_off32_limit & 0x80000000)
1840 /* Traditionally "pack-objects [options] base extra" failed;
1841 * we would however want to take refs parameter that would
1842 * have been given to upstream rev-list ourselves, which means
1843 * we somehow want to say what the base name is. So the
1846 * pack-objects [options] base <refs...>
1848 * in other words, we would treat the first non-option as the
1849 * base_name and send everything else to the internal revision
1853 if (!pack_to_stdout)
1854 base_name = argv[i++];
1856 if (pack_to_stdout != !base_name)
1859 if (pack_to_stdout && pack_size_limit)
1860 die("--max-pack-size cannot be used to build a pack for transfer.");
1862 if (!pack_to_stdout && thin)
1863 die("--thin cannot be used to build an indexable pack.");
1865 prepare_packed_git();
1868 start_progress(&progress_state, "Generating pack...",
1869 "Counting objects: ", 0);
1870 if (!use_internal_rev_list)
1871 read_object_list_from_stdin();
1873 rp_av[rp_ac] = NULL;
1874 get_object_list(rp_ac, rp_av);
1877 stop_progress(&progress_state);
1878 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1881 if (non_empty && !nr_result)
1883 if (progress && (nr_objects != nr_result))
1884 fprintf(stderr, "Result has %u objects.\n", nr_result);
1886 prepare_pack(window, depth);
1889 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1890 written, written_delta, reused, reused_delta);