10 #include "csum-file.h"
11 #include "tree-walk.h"
14 #include "list-objects.h"
17 static const char pack_usage[] = "\
18 git-pack-objects [{ -q | --progress | --all-progress }] \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]";
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 */
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
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
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.
54 static struct object_entry *objects;
55 static uint32_t nr_objects, nr_alloc, nr_result;
58 static int no_reuse_delta, no_reuse_object;
60 static int incremental;
61 static int allow_ofs_delta;
62 static const char *pack_tmp_name, *idx_tmp_name;
63 static char tmpname[PATH_MAX];
64 static unsigned char pack_file_sha1[20];
65 static int progress = 1;
66 static int window = 10;
67 static int depth = 50;
68 static int pack_to_stdout;
69 static int num_preferred_base;
70 static struct progress progress_state;
73 * The object names in objects array are hashed with this hashtable,
74 * to help looking up the entry by object name.
75 * This hashtable is built after all the objects are seen.
77 static int *object_ix;
78 static int object_ix_hashsz;
81 * Pack index for existing packs give us easy access to the offsets into
82 * corresponding pack file where each object's data starts, but the entries
83 * do not store the size of the compressed representation (uncompressed
84 * size is easily available by examining the pack entry header). It is
85 * also rather expensive to find the sha1 for an object given its offset.
87 * We build a hashtable of existing packs (pack_revindex), and keep reverse
88 * index here -- pack index file is sorted by object name mapping to offset;
89 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
90 * ordered by offset, so if you know the offset of an object, next offset
91 * is where its packed representation ends and the index_nr can be used to
92 * get the object sha1 from the main index.
94 struct revindex_entry {
98 struct pack_revindex {
100 struct revindex_entry *revindex;
102 static struct pack_revindex *pack_revindex;
103 static int pack_revindex_hashsz;
108 static uint32_t written, written_delta;
109 static uint32_t reused, reused_delta;
111 static int pack_revindex_ix(struct packed_git *p)
113 unsigned long ui = (unsigned long)p;
116 ui = ui ^ (ui >> 16); /* defeat structure alignment */
117 i = (int)(ui % pack_revindex_hashsz);
118 while (pack_revindex[i].p) {
119 if (pack_revindex[i].p == p)
121 if (++i == pack_revindex_hashsz)
127 static void prepare_pack_ix(void)
130 struct packed_git *p;
131 for (num = 0, p = packed_git; p; p = p->next)
135 pack_revindex_hashsz = num * 11;
136 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
137 for (p = packed_git; p; p = p->next) {
138 num = pack_revindex_ix(p);
140 pack_revindex[num].p = p;
142 /* revindex elements are lazily initialized */
145 static int cmp_offset(const void *a_, const void *b_)
147 const struct revindex_entry *a = a_;
148 const struct revindex_entry *b = b_;
149 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
153 * Ordered list of offsets of objects in the pack.
155 static void prepare_pack_revindex(struct pack_revindex *rix)
157 struct packed_git *p = rix->p;
158 int num_ent = p->num_objects;
160 const char *index = p->index_data;
162 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
165 if (p->index_version > 1) {
166 const uint32_t *off_32 =
167 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
168 const uint32_t *off_64 = off_32 + p->num_objects;
169 for (i = 0; i < num_ent; i++) {
170 uint32_t off = ntohl(*off_32++);
171 if (!(off & 0x80000000)) {
172 rix->revindex[i].offset = off;
174 rix->revindex[i].offset =
175 ((uint64_t)ntohl(*off_64++)) << 32;
176 rix->revindex[i].offset |=
179 rix->revindex[i].nr = i;
182 for (i = 0; i < num_ent; i++) {
183 uint32_t hl = *((uint32_t *)(index + 24 * i));
184 rix->revindex[i].offset = ntohl(hl);
185 rix->revindex[i].nr = i;
189 /* This knows the pack format -- the 20-byte trailer
190 * follows immediately after the last object data.
192 rix->revindex[num_ent].offset = p->pack_size - 20;
193 rix->revindex[num_ent].nr = -1;
194 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
197 static struct revindex_entry * find_packed_object(struct packed_git *p,
202 struct pack_revindex *rix;
203 struct revindex_entry *revindex;
204 num = pack_revindex_ix(p);
206 die("internal error: pack revindex uninitialized");
207 rix = &pack_revindex[num];
209 prepare_pack_revindex(rix);
210 revindex = rix->revindex;
212 hi = p->num_objects + 1;
214 int mi = (lo + hi) / 2;
215 if (revindex[mi].offset == ofs) {
216 return revindex + mi;
218 else if (ofs < revindex[mi].offset)
223 die("internal error: pack revindex corrupt");
226 static const unsigned char *find_packed_object_name(struct packed_git *p,
229 struct revindex_entry *entry = find_packed_object(p, ofs);
230 return nth_packed_object_sha1(p, entry->nr);
233 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
235 unsigned long othersize, delta_size;
236 enum object_type type;
237 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
241 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
242 delta_buf = diff_delta(otherbuf, othersize,
243 buf, size, &delta_size, 0);
244 if (!delta_buf || delta_size != entry->delta_size)
245 die("delta size changed");
252 * The per-object header is a pretty dense thing, which is
253 * - first byte: low four bits are "size", then three bits of "type",
254 * and the high bit is "size continues".
255 * - each byte afterwards: low seven bits are size continuation,
256 * with the high bit being "size continues"
258 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
263 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
264 die("bad type %d", type);
266 c = (type << 4) | (size & 15);
279 * we are going to reuse the existing object data as is. make
280 * sure it is not corrupt.
282 static int check_pack_inflate(struct packed_git *p,
283 struct pack_window **w_curs,
286 unsigned long expect)
289 unsigned char fakebuf[4096], *in;
292 memset(&stream, 0, sizeof(stream));
293 inflateInit(&stream);
295 in = use_pack(p, w_curs, offset, &stream.avail_in);
297 stream.next_out = fakebuf;
298 stream.avail_out = sizeof(fakebuf);
299 st = inflate(&stream, Z_FINISH);
300 offset += stream.next_in - in;
301 } while (st == Z_OK || st == Z_BUF_ERROR);
303 return (st == Z_STREAM_END &&
304 stream.total_out == expect &&
305 stream.total_in == len) ? 0 : -1;
308 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
309 off_t offset, off_t len, unsigned int nr)
311 const uint32_t *index_crc;
312 uint32_t data_crc = crc32(0, Z_NULL, 0);
316 void *data = use_pack(p, w_curs, offset, &avail);
319 data_crc = crc32(data_crc, data, avail);
324 index_crc = p->index_data;
325 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
327 return data_crc != ntohl(*index_crc);
330 static void copy_pack_data(struct sha1file *f,
331 struct packed_git *p,
332 struct pack_window **w_curs,
340 in = use_pack(p, w_curs, offset, &avail);
342 avail = (unsigned int)len;
343 sha1write(f, in, avail);
349 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
352 unsigned char fakebuf[4096];
355 memset(&stream, 0, sizeof(stream));
356 stream.next_in = data;
357 stream.avail_in = len;
358 stream.next_out = fakebuf;
359 stream.avail_out = sizeof(fakebuf);
360 inflateInit(&stream);
363 st = inflate(&stream, Z_FINISH);
364 if (st == Z_STREAM_END || st == Z_OK) {
365 st = (stream.total_out == expect &&
366 stream.total_in == len) ? 0 : -1;
369 if (st != Z_BUF_ERROR) {
373 stream.next_out = fakebuf;
374 stream.avail_out = sizeof(fakebuf);
380 static int revalidate_loose_object(struct object_entry *entry,
382 unsigned long mapsize)
384 /* we already know this is a loose object with new type header. */
385 enum object_type type;
386 unsigned long size, used;
391 used = unpack_object_header_gently(map, mapsize, &type, &size);
396 return check_loose_inflate(map, mapsize, size);
399 static unsigned long write_object(struct sha1file *f,
400 struct object_entry *entry)
403 enum object_type type;
405 unsigned char header[10];
408 enum object_type obj_type;
414 obj_type = entry->type;
416 to_reuse = 0; /* explicit */
417 else if (!entry->in_pack)
418 to_reuse = 0; /* can't reuse what we don't have */
419 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
420 to_reuse = 1; /* check_object() decided it for us */
421 else if (obj_type != entry->in_pack_type)
422 to_reuse = 0; /* pack has delta which is unusable */
423 else if (entry->delta)
424 to_reuse = 0; /* we want to pack afresh */
426 to_reuse = 1; /* we have it in-pack undeltified,
427 * and we do not need to deltify it.
430 if (!no_reuse_object && !entry->in_pack && !entry->delta) {
432 unsigned long mapsize;
433 map = map_sha1_file(entry->sha1, &mapsize);
434 if (map && !legacy_loose_object(map)) {
435 /* We can copy straight into the pack file */
436 if (revalidate_loose_object(entry, map, mapsize))
437 die("corrupt loose object %s",
438 sha1_to_hex(entry->sha1));
439 sha1write(f, map, mapsize);
440 munmap(map, mapsize);
446 munmap(map, mapsize);
450 buf = read_sha1_file(entry->sha1, &type, &size);
452 die("unable to read %s", sha1_to_hex(entry->sha1));
453 if (size != entry->size)
454 die("object %s size inconsistency (%lu vs %lu)",
455 sha1_to_hex(entry->sha1), size, entry->size);
457 buf = delta_against(buf, size, entry);
458 size = entry->delta_size;
459 obj_type = (allow_ofs_delta && entry->delta->offset) ?
460 OBJ_OFS_DELTA : OBJ_REF_DELTA;
463 * The object header is a byte of 'type' followed by zero or
464 * more bytes of length.
466 hdrlen = encode_header(obj_type, size, header);
467 sha1write(f, header, hdrlen);
469 if (obj_type == OBJ_OFS_DELTA) {
471 * Deltas with relative base contain an additional
472 * encoding of the relative offset for the delta
473 * base from this object's position in the pack.
475 off_t ofs = entry->offset - entry->delta->offset;
476 unsigned pos = sizeof(header) - 1;
477 header[pos] = ofs & 127;
479 header[--pos] = 128 | (--ofs & 127);
480 sha1write(f, header + pos, sizeof(header) - pos);
481 hdrlen += sizeof(header) - pos;
482 } else if (obj_type == OBJ_REF_DELTA) {
484 * Deltas with a base reference contain
485 * an additional 20 bytes for the base sha1.
487 sha1write(f, entry->delta->sha1, 20);
490 datalen = sha1write_compressed(f, buf, size);
494 struct packed_git *p = entry->in_pack;
495 struct pack_window *w_curs = NULL;
496 struct revindex_entry *revidx;
500 obj_type = (allow_ofs_delta && entry->delta->offset) ?
501 OBJ_OFS_DELTA : OBJ_REF_DELTA;
504 hdrlen = encode_header(obj_type, entry->size, header);
505 sha1write(f, header, hdrlen);
506 if (obj_type == OBJ_OFS_DELTA) {
507 off_t ofs = entry->offset - entry->delta->offset;
508 unsigned pos = sizeof(header) - 1;
509 header[pos] = ofs & 127;
511 header[--pos] = 128 | (--ofs & 127);
512 sha1write(f, header + pos, sizeof(header) - pos);
513 hdrlen += sizeof(header) - pos;
514 } else if (obj_type == OBJ_REF_DELTA) {
515 sha1write(f, entry->delta->sha1, 20);
519 offset = entry->in_pack_offset;
520 revidx = find_packed_object(p, offset);
521 datalen = revidx[1].offset - offset;
522 if (!pack_to_stdout && p->index_version > 1 &&
523 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
524 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
525 offset += entry->in_pack_header_size;
526 datalen -= entry->in_pack_header_size;
527 if (!pack_to_stdout && p->index_version == 1 &&
528 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
529 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
530 copy_pack_data(f, p, &w_curs, offset, datalen);
538 entry->crc32 = crc32_end(f);
539 return hdrlen + datalen;
542 static off_t write_one(struct sha1file *f,
543 struct object_entry *e,
548 /* offset is non zero if object is written already. */
549 if (e->offset || e->preferred_base)
552 /* if we are deltified, write out base object first. */
554 offset = write_one(f, e->delta, offset);
557 size = write_object(f, e);
559 /* make sure off_t is sufficiently large not to wrap */
560 if (offset > offset + size)
561 die("pack too large for current definition of off_t");
562 return offset + size;
565 static int open_object_dir_tmp(const char *path)
567 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
568 return mkstemp(tmpname);
571 static off_t write_pack_file(void)
575 off_t offset, last_obj_offset = 0;
576 struct pack_header hdr;
577 int do_progress = progress;
579 if (pack_to_stdout) {
580 f = sha1fd(1, "<stdout>");
583 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
585 die("unable to create %s: %s\n", tmpname, strerror(errno));
586 pack_tmp_name = xstrdup(tmpname);
587 f = sha1fd(fd, pack_tmp_name);
591 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
593 hdr.hdr_signature = htonl(PACK_SIGNATURE);
594 hdr.hdr_version = htonl(PACK_VERSION);
595 hdr.hdr_entries = htonl(nr_result);
596 sha1write(f, &hdr, sizeof(hdr));
597 offset = sizeof(hdr);
600 for (i = 0; i < nr_objects; i++) {
601 last_obj_offset = offset;
602 offset = write_one(f, objects + i, offset);
604 display_progress(&progress_state, written);
607 stop_progress(&progress_state);
609 if (written != nr_result)
610 die("wrote %u objects while expecting %u", written, nr_result);
611 sha1close(f, pack_file_sha1, 1);
613 return last_obj_offset;
616 static int sha1_sort(const void *_a, const void *_b)
618 const struct object_entry *a = *(struct object_entry **)_a;
619 const struct object_entry *b = *(struct object_entry **)_b;
620 return hashcmp(a->sha1, b->sha1);
623 static uint32_t index_default_version = 1;
624 static uint32_t index_off32_limit = 0x7fffffff;
626 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
629 struct object_entry **sorted_by_sha, **list, **last;
631 uint32_t i, index_version;
634 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
636 die("unable to create %s: %s\n", tmpname, strerror(errno));
637 idx_tmp_name = xstrdup(tmpname);
638 f = sha1fd(fd, idx_tmp_name);
643 xcalloc(nr_result, sizeof(struct object_entry *));
644 for (i = 0; i < nr_objects; i++)
645 if (!objects[i].preferred_base)
646 sorted_by_sha[j++] = objects + i;
648 die("listed %u objects while expecting %u", j, nr_result);
649 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
650 list = sorted_by_sha;
651 last = sorted_by_sha + nr_result;
653 sorted_by_sha = list = last = NULL;
655 /* if last object's offset is >= 2^31 we should use index V2 */
656 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
658 /* index versions 2 and above need a header */
659 if (index_version >= 2) {
660 struct pack_idx_header hdr;
661 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
662 hdr.idx_version = htonl(index_version);
663 sha1write(f, &hdr, sizeof(hdr));
667 * Write the first-level table (the list is sorted,
668 * but we use a 256-entry lookup to be able to avoid
669 * having to do eight extra binary search iterations).
671 for (i = 0; i < 256; i++) {
672 struct object_entry **next = list;
673 while (next < last) {
674 struct object_entry *entry = *next;
675 if (entry->sha1[0] != i)
679 array[i] = htonl(next - sorted_by_sha);
682 sha1write(f, array, 256 * 4);
684 /* Compute the SHA1 hash of sorted object names. */
687 /* Write the actual SHA1 entries. */
688 list = sorted_by_sha;
689 for (i = 0; i < nr_result; i++) {
690 struct object_entry *entry = *list++;
691 if (index_version < 2) {
692 uint32_t offset = htonl(entry->offset);
693 sha1write(f, &offset, 4);
695 sha1write(f, entry->sha1, 20);
696 SHA1_Update(&ctx, entry->sha1, 20);
699 if (index_version >= 2) {
700 unsigned int nr_large_offset = 0;
702 /* write the crc32 table */
703 list = sorted_by_sha;
704 for (i = 0; i < nr_objects; i++) {
705 struct object_entry *entry = *list++;
706 uint32_t crc32_val = htonl(entry->crc32);
707 sha1write(f, &crc32_val, 4);
710 /* write the 32-bit offset table */
711 list = sorted_by_sha;
712 for (i = 0; i < nr_objects; i++) {
713 struct object_entry *entry = *list++;
714 uint32_t offset = (entry->offset <= index_off32_limit) ?
715 entry->offset : (0x80000000 | nr_large_offset++);
716 offset = htonl(offset);
717 sha1write(f, &offset, 4);
720 /* write the large offset table */
721 list = sorted_by_sha;
722 while (nr_large_offset) {
723 struct object_entry *entry = *list++;
724 uint64_t offset = entry->offset;
725 if (offset > index_off32_limit) {
727 split[0] = htonl(offset >> 32);
728 split[1] = htonl(offset & 0xffffffff);
729 sha1write(f, split, 8);
735 sha1write(f, pack_file_sha1, 20);
736 sha1close(f, NULL, 1);
738 SHA1_Final(sha1, &ctx);
741 static int locate_object_entry_hash(const unsigned char *sha1)
745 memcpy(&ui, sha1, sizeof(unsigned int));
746 i = ui % object_ix_hashsz;
747 while (0 < object_ix[i]) {
748 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
750 if (++i == object_ix_hashsz)
756 static struct object_entry *locate_object_entry(const unsigned char *sha1)
760 if (!object_ix_hashsz)
763 i = locate_object_entry_hash(sha1);
765 return &objects[object_ix[i]-1];
769 static void rehash_objects(void)
772 struct object_entry *oe;
774 object_ix_hashsz = nr_objects * 3;
775 if (object_ix_hashsz < 1024)
776 object_ix_hashsz = 1024;
777 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
778 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
779 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
780 int ix = locate_object_entry_hash(oe->sha1);
784 object_ix[ix] = i + 1;
788 static unsigned name_hash(const char *name)
794 * This effectively just creates a sortable number from the
795 * last sixteen non-whitespace characters. Last characters
796 * count "most", so things that end in ".c" sort together.
798 while ((c = *name++) != 0) {
801 hash = (hash >> 2) + (c << 24);
806 static int add_object_entry(const unsigned char *sha1, enum object_type type,
807 unsigned hash, int exclude)
809 struct object_entry *entry;
810 struct packed_git *p, *found_pack = NULL;
811 off_t found_offset = 0;
814 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
817 entry = objects + object_ix[ix] - 1;
818 if (!entry->preferred_base)
820 entry->preferred_base = 1;
825 for (p = packed_git; p; p = p->next) {
826 off_t offset = find_pack_entry_one(sha1, p);
829 found_offset = offset;
836 if (local && !p->pack_local)
841 if (nr_objects >= nr_alloc) {
842 nr_alloc = (nr_alloc + 1024) * 3 / 2;
843 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
846 entry = objects + nr_objects++;
847 memset(entry, 0, sizeof(*entry));
848 hashcpy(entry->sha1, sha1);
853 entry->preferred_base = 1;
857 entry->in_pack = found_pack;
858 entry->in_pack_offset = found_offset;
861 if (object_ix_hashsz * 3 <= nr_objects * 4)
864 object_ix[-1 - ix] = nr_objects;
867 display_progress(&progress_state, nr_objects);
872 struct pbase_tree_cache {
873 unsigned char sha1[20];
877 unsigned long tree_size;
880 static struct pbase_tree_cache *(pbase_tree_cache[256]);
881 static int pbase_tree_cache_ix(const unsigned char *sha1)
883 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
885 static int pbase_tree_cache_ix_incr(int ix)
887 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
890 static struct pbase_tree {
891 struct pbase_tree *next;
892 /* This is a phony "cache" entry; we are not
893 * going to evict it nor find it through _get()
894 * mechanism -- this is for the toplevel node that
895 * would almost always change with any commit.
897 struct pbase_tree_cache pcache;
900 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
902 struct pbase_tree_cache *ent, *nent;
905 enum object_type type;
907 int my_ix = pbase_tree_cache_ix(sha1);
908 int available_ix = -1;
910 /* pbase-tree-cache acts as a limited hashtable.
911 * your object will be found at your index or within a few
912 * slots after that slot if it is cached.
914 for (neigh = 0; neigh < 8; neigh++) {
915 ent = pbase_tree_cache[my_ix];
916 if (ent && !hashcmp(ent->sha1, sha1)) {
920 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
921 ((0 <= available_ix) &&
922 (!ent && pbase_tree_cache[available_ix])))
923 available_ix = my_ix;
926 my_ix = pbase_tree_cache_ix_incr(my_ix);
929 /* Did not find one. Either we got a bogus request or
930 * we need to read and perhaps cache.
932 data = read_sha1_file(sha1, &type, &size);
935 if (type != OBJ_TREE) {
940 /* We need to either cache or return a throwaway copy */
942 if (available_ix < 0)
945 ent = pbase_tree_cache[available_ix];
946 my_ix = available_ix;
950 nent = xmalloc(sizeof(*nent));
951 nent->temporary = (available_ix < 0);
954 /* evict and reuse */
955 free(ent->tree_data);
958 hashcpy(nent->sha1, sha1);
959 nent->tree_data = data;
960 nent->tree_size = size;
962 if (!nent->temporary)
963 pbase_tree_cache[my_ix] = nent;
967 static void pbase_tree_put(struct pbase_tree_cache *cache)
969 if (!cache->temporary) {
973 free(cache->tree_data);
977 static int name_cmp_len(const char *name)
980 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
985 static void add_pbase_object(struct tree_desc *tree,
988 const char *fullname)
990 struct name_entry entry;
993 while (tree_entry(tree,&entry)) {
994 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
995 memcmp(name, entry.path, cmplen);
1000 if (name[cmplen] != '/') {
1001 unsigned hash = name_hash(fullname);
1002 add_object_entry(entry.sha1,
1003 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1007 if (S_ISDIR(entry.mode)) {
1008 struct tree_desc sub;
1009 struct pbase_tree_cache *tree;
1010 const char *down = name+cmplen+1;
1011 int downlen = name_cmp_len(down);
1013 tree = pbase_tree_get(entry.sha1);
1016 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1018 add_pbase_object(&sub, down, downlen, fullname);
1019 pbase_tree_put(tree);
1024 static unsigned *done_pbase_paths;
1025 static int done_pbase_paths_num;
1026 static int done_pbase_paths_alloc;
1027 static int done_pbase_path_pos(unsigned hash)
1030 int hi = done_pbase_paths_num;
1032 int mi = (hi + lo) / 2;
1033 if (done_pbase_paths[mi] == hash)
1035 if (done_pbase_paths[mi] < hash)
1043 static int check_pbase_path(unsigned hash)
1045 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1049 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1050 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1051 done_pbase_paths = xrealloc(done_pbase_paths,
1052 done_pbase_paths_alloc *
1055 done_pbase_paths_num++;
1056 if (pos < done_pbase_paths_num)
1057 memmove(done_pbase_paths + pos + 1,
1058 done_pbase_paths + pos,
1059 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1060 done_pbase_paths[pos] = hash;
1064 static void add_preferred_base_object(const char *name, unsigned hash)
1066 struct pbase_tree *it;
1069 if (!num_preferred_base || check_pbase_path(hash))
1072 cmplen = name_cmp_len(name);
1073 for (it = pbase_tree; it; it = it->next) {
1075 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
1078 struct tree_desc tree;
1079 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1080 add_pbase_object(&tree, name, cmplen, name);
1085 static void add_preferred_base(unsigned char *sha1)
1087 struct pbase_tree *it;
1090 unsigned char tree_sha1[20];
1092 if (window <= num_preferred_base++)
1095 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1099 for (it = pbase_tree; it; it = it->next) {
1100 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1106 it = xcalloc(1, sizeof(*it));
1107 it->next = pbase_tree;
1110 hashcpy(it->pcache.sha1, tree_sha1);
1111 it->pcache.tree_data = data;
1112 it->pcache.tree_size = size;
1115 static void check_object(struct object_entry *entry)
1117 if (entry->in_pack) {
1118 struct packed_git *p = entry->in_pack;
1119 struct pack_window *w_curs = NULL;
1120 const unsigned char *base_ref = NULL;
1121 struct object_entry *base_entry;
1122 unsigned long used, used_0;
1125 unsigned char *buf, c;
1127 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1130 * We want in_pack_type even if we do not reuse delta
1131 * since non-delta representations could still be reused.
1133 used = unpack_object_header_gently(buf, avail,
1134 &entry->in_pack_type,
1138 * Determine if this is a delta and if so whether we can
1139 * reuse it or not. Otherwise let's find out as cheaply as
1140 * possible what the actual type and size for this object is.
1142 switch (entry->in_pack_type) {
1144 /* Not a delta hence we've already got all we need. */
1145 entry->type = entry->in_pack_type;
1146 entry->in_pack_header_size = used;
1147 unuse_pack(&w_curs);
1150 if (!no_reuse_delta && !entry->preferred_base)
1151 base_ref = use_pack(p, &w_curs,
1152 entry->in_pack_offset + used, NULL);
1153 entry->in_pack_header_size = used + 20;
1156 buf = use_pack(p, &w_curs,
1157 entry->in_pack_offset + used, NULL);
1163 if (!ofs || MSB(ofs, 7))
1164 die("delta base offset overflow in pack for %s",
1165 sha1_to_hex(entry->sha1));
1167 ofs = (ofs << 7) + (c & 127);
1169 if (ofs >= entry->in_pack_offset)
1170 die("delta base offset out of bound for %s",
1171 sha1_to_hex(entry->sha1));
1172 ofs = entry->in_pack_offset - ofs;
1173 if (!no_reuse_delta && !entry->preferred_base)
1174 base_ref = find_packed_object_name(p, ofs);
1175 entry->in_pack_header_size = used + used_0;
1179 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1181 * If base_ref was set above that means we wish to
1182 * reuse delta data, and we even found that base
1183 * in the list of objects we want to pack. Goodie!
1185 * Depth value does not matter - find_deltas() will
1186 * never consider reused delta as the base object to
1187 * deltify other objects against, in order to avoid
1190 entry->type = entry->in_pack_type;
1191 entry->delta = base_entry;
1192 entry->delta_sibling = base_entry->delta_child;
1193 base_entry->delta_child = entry;
1194 unuse_pack(&w_curs);
1200 * This must be a delta and we already know what the
1201 * final object type is. Let's extract the actual
1202 * object size from the delta header.
1204 entry->size = get_size_from_delta(p, &w_curs,
1205 entry->in_pack_offset + entry->in_pack_header_size);
1206 unuse_pack(&w_curs);
1211 * No choice but to fall back to the recursive delta walk
1212 * with sha1_object_info() to find about the object type
1215 unuse_pack(&w_curs);
1218 entry->type = sha1_object_info(entry->sha1, &entry->size);
1219 if (entry->type < 0)
1220 die("unable to get type of object %s",
1221 sha1_to_hex(entry->sha1));
1224 static int pack_offset_sort(const void *_a, const void *_b)
1226 const struct object_entry *a = *(struct object_entry **)_a;
1227 const struct object_entry *b = *(struct object_entry **)_b;
1229 /* avoid filesystem trashing with loose objects */
1230 if (!a->in_pack && !b->in_pack)
1231 return hashcmp(a->sha1, b->sha1);
1233 if (a->in_pack < b->in_pack)
1235 if (a->in_pack > b->in_pack)
1237 return a->in_pack_offset < b->in_pack_offset ? -1 :
1238 (a->in_pack_offset > b->in_pack_offset);
1241 static void get_object_details(void)
1244 struct object_entry **sorted_by_offset;
1246 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1247 for (i = 0; i < nr_objects; i++)
1248 sorted_by_offset[i] = objects + i;
1249 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1252 for (i = 0; i < nr_objects; i++)
1253 check_object(sorted_by_offset[i]);
1254 free(sorted_by_offset);
1257 static int type_size_sort(const void *_a, const void *_b)
1259 const struct object_entry *a = *(struct object_entry **)_a;
1260 const struct object_entry *b = *(struct object_entry **)_b;
1262 if (a->type < b->type)
1264 if (a->type > b->type)
1266 if (a->hash < b->hash)
1268 if (a->hash > b->hash)
1270 if (a->preferred_base < b->preferred_base)
1272 if (a->preferred_base > b->preferred_base)
1274 if (a->size < b->size)
1276 if (a->size > b->size)
1278 return a > b ? -1 : (a < b); /* newest last */
1282 struct object_entry *entry;
1284 struct delta_index *index;
1288 * We search for deltas _backwards_ in a list sorted by type and
1289 * by size, so that we see progressively smaller and smaller files.
1290 * That's because we prefer deltas to be from the bigger file
1291 * to the smaller - deletes are potentially cheaper, but perhaps
1292 * more importantly, the bigger file is likely the more recent
1295 static int try_delta(struct unpacked *trg, struct unpacked *src,
1298 struct object_entry *trg_entry = trg->entry;
1299 struct object_entry *src_entry = src->entry;
1300 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1301 enum object_type type;
1304 /* Don't bother doing diffs between different types */
1305 if (trg_entry->type != src_entry->type)
1308 /* We do not compute delta to *create* objects we are not
1311 if (trg_entry->preferred_base)
1315 * We do not bother to try a delta that we discarded
1316 * on an earlier try, but only when reusing delta data.
1318 if (!no_reuse_delta && trg_entry->in_pack &&
1319 trg_entry->in_pack == src_entry->in_pack &&
1320 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1321 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1324 /* Let's not bust the allowed depth. */
1325 if (src_entry->depth >= max_depth)
1328 /* Now some size filtering heuristics. */
1329 trg_size = trg_entry->size;
1330 max_size = trg_size/2 - 20;
1331 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1334 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1335 max_size = trg_entry->delta_size-1;
1336 src_size = src_entry->size;
1337 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1338 if (sizediff >= max_size)
1341 /* Load data if not already done */
1343 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1345 die("object %s inconsistent object length (%lu vs %lu)",
1346 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1349 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1351 die("object %s inconsistent object length (%lu vs %lu)",
1352 sha1_to_hex(src_entry->sha1), sz, src_size);
1355 src->index = create_delta_index(src->data, src_size);
1357 die("out of memory");
1360 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1364 trg_entry->delta = src_entry;
1365 trg_entry->delta_size = delta_size;
1366 trg_entry->depth = src_entry->depth + 1;
1371 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1373 struct object_entry *child = me->delta_child;
1376 unsigned int c = check_delta_limit(child, n + 1);
1379 child = child->delta_sibling;
1384 static void find_deltas(struct object_entry **list, int window, int depth)
1386 uint32_t i = nr_objects, idx = 0, processed = 0;
1387 unsigned int array_size = window * sizeof(struct unpacked);
1388 struct unpacked *array;
1393 array = xmalloc(array_size);
1394 memset(array, 0, array_size);
1396 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1399 struct object_entry *entry = list[--i];
1400 struct unpacked *n = array + idx;
1403 if (!entry->preferred_base)
1407 display_progress(&progress_state, processed);
1410 /* This happens if we decided to reuse existing
1411 * delta from a pack. "!no_reuse_delta &&" is implied.
1415 if (entry->size < 50)
1417 free_delta_index(n->index);
1424 * If the current object is at pack edge, take the depth the
1425 * objects that depend on the current object into account
1426 * otherwise they would become too deep.
1429 if (entry->delta_child) {
1430 max_depth -= check_delta_limit(entry, 0);
1437 uint32_t other_idx = idx + j;
1439 if (other_idx >= window)
1440 other_idx -= window;
1441 m = array + other_idx;
1444 if (try_delta(n, m, max_depth) < 0)
1448 /* if we made n a delta, and if n is already at max
1449 * depth, leaving it in the window is pointless. we
1450 * should evict it first.
1452 if (entry->delta && depth <= entry->depth)
1462 stop_progress(&progress_state);
1464 for (i = 0; i < window; ++i) {
1465 free_delta_index(array[i].index);
1466 free(array[i].data);
1471 static void prepare_pack(int window, int depth)
1473 struct object_entry **delta_list;
1476 get_object_details();
1478 if (!window || !depth)
1481 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1482 for (i = 0; i < nr_objects; i++)
1483 delta_list[i] = objects + i;
1484 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1485 find_deltas(delta_list, window+1, depth);
1489 static int git_pack_config(const char *k, const char *v)
1491 if(!strcmp(k, "pack.window")) {
1492 window = git_config_int(k, v);
1495 if(!strcmp(k, "pack.depth")) {
1496 depth = git_config_int(k, v);
1499 return git_default_config(k, v);
1502 static void read_object_list_from_stdin(void)
1504 char line[40 + 1 + PATH_MAX + 2];
1505 unsigned char sha1[20];
1509 if (!fgets(line, sizeof(line), stdin)) {
1513 die("fgets returned NULL, not EOF, not error!");
1515 die("fgets: %s", strerror(errno));
1519 if (line[0] == '-') {
1520 if (get_sha1_hex(line+1, sha1))
1521 die("expected edge sha1, got garbage:\n %s",
1523 add_preferred_base(sha1);
1526 if (get_sha1_hex(line, sha1))
1527 die("expected sha1, got garbage:\n %s", line);
1529 hash = name_hash(line+41);
1530 add_preferred_base_object(line+41, hash);
1531 add_object_entry(sha1, 0, hash, 0);
1535 static void show_commit(struct commit *commit)
1537 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
1540 static void show_object(struct object_array_entry *p)
1542 unsigned hash = name_hash(p->name);
1543 add_preferred_base_object(p->name, hash);
1544 add_object_entry(p->item->sha1, p->item->type, hash, 0);
1547 static void show_edge(struct commit *commit)
1549 add_preferred_base(commit->object.sha1);
1552 static void get_object_list(int ac, const char **av)
1554 struct rev_info revs;
1558 init_revisions(&revs, NULL);
1559 save_commit_buffer = 0;
1560 track_object_refs = 0;
1561 setup_revisions(ac, av, &revs, NULL);
1563 while (fgets(line, sizeof(line), stdin) != NULL) {
1564 int len = strlen(line);
1565 if (line[len - 1] == '\n')
1570 if (!strcmp(line, "--not")) {
1571 flags ^= UNINTERESTING;
1574 die("not a rev '%s'", line);
1576 if (handle_revision_arg(line, &revs, flags, 1))
1577 die("bad revision '%s'", line);
1580 prepare_revision_walk(&revs);
1581 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1582 traverse_commit_list(&revs, show_commit, show_object);
1585 static int adjust_perm(const char *path, mode_t mode)
1587 if (chmod(path, mode))
1589 return adjust_shared_perm(path);
1592 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1594 int use_internal_rev_list = 0;
1597 off_t last_obj_offset;
1598 const char *base_name = NULL;
1600 int rp_ac_alloc = 64;
1603 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1605 rp_av[0] = "pack-objects";
1606 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1609 git_config(git_pack_config);
1611 progress = isatty(2);
1612 for (i = 1; i < argc; i++) {
1613 const char *arg = argv[i];
1618 if (!strcmp("--non-empty", arg)) {
1622 if (!strcmp("--local", arg)) {
1626 if (!strcmp("--incremental", arg)) {
1630 if (!prefixcmp(arg, "--window=")) {
1632 window = strtoul(arg+9, &end, 0);
1633 if (!arg[9] || *end)
1637 if (!prefixcmp(arg, "--depth=")) {
1639 depth = strtoul(arg+8, &end, 0);
1640 if (!arg[8] || *end)
1644 if (!strcmp("--progress", arg)) {
1648 if (!strcmp("--all-progress", arg)) {
1652 if (!strcmp("-q", arg)) {
1656 if (!strcmp("--no-reuse-delta", arg)) {
1660 if (!strcmp("--no-reuse-object", arg)) {
1661 no_reuse_object = no_reuse_delta = 1;
1664 if (!strcmp("--delta-base-offset", arg)) {
1665 allow_ofs_delta = 1;
1668 if (!strcmp("--stdout", arg)) {
1672 if (!strcmp("--revs", arg)) {
1673 use_internal_rev_list = 1;
1676 if (!strcmp("--unpacked", arg) ||
1677 !prefixcmp(arg, "--unpacked=") ||
1678 !strcmp("--reflog", arg) ||
1679 !strcmp("--all", arg)) {
1680 use_internal_rev_list = 1;
1681 if (rp_ac >= rp_ac_alloc - 1) {
1682 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1683 rp_av = xrealloc(rp_av,
1684 rp_ac_alloc * sizeof(*rp_av));
1686 rp_av[rp_ac++] = arg;
1689 if (!strcmp("--thin", arg)) {
1690 use_internal_rev_list = 1;
1692 rp_av[1] = "--objects-edge";
1695 if (!prefixcmp(arg, "--index-version=")) {
1697 index_default_version = strtoul(arg + 16, &c, 10);
1698 if (index_default_version > 2)
1701 index_off32_limit = strtoul(c+1, &c, 0);
1702 if (*c || index_off32_limit & 0x80000000)
1709 /* Traditionally "pack-objects [options] base extra" failed;
1710 * we would however want to take refs parameter that would
1711 * have been given to upstream rev-list ourselves, which means
1712 * we somehow want to say what the base name is. So the
1715 * pack-objects [options] base <refs...>
1717 * in other words, we would treat the first non-option as the
1718 * base_name and send everything else to the internal revision
1722 if (!pack_to_stdout)
1723 base_name = argv[i++];
1725 if (pack_to_stdout != !base_name)
1728 if (!pack_to_stdout && thin)
1729 die("--thin cannot be used to build an indexable pack.");
1731 prepare_packed_git();
1734 start_progress(&progress_state, "Generating pack...",
1735 "Counting objects: ", 0);
1736 if (!use_internal_rev_list)
1737 read_object_list_from_stdin();
1739 rp_av[rp_ac] = NULL;
1740 get_object_list(rp_ac, rp_av);
1743 stop_progress(&progress_state);
1744 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1747 if (non_empty && !nr_result)
1749 if (progress && (nr_objects != nr_result))
1750 fprintf(stderr, "Result has %u objects.\n", nr_result);
1752 prepare_pack(window, depth);
1753 last_obj_offset = write_pack_file();
1754 if (!pack_to_stdout) {
1755 unsigned char object_list_sha1[20];
1756 mode_t mode = umask(0);
1759 mode = 0444 & ~mode;
1761 write_index_file(last_obj_offset, object_list_sha1);
1762 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
1763 base_name, sha1_to_hex(object_list_sha1));
1764 if (adjust_perm(pack_tmp_name, mode))
1765 die("unable to make temporary pack file readable: %s",
1767 if (rename(pack_tmp_name, tmpname))
1768 die("unable to rename temporary pack file: %s",
1770 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
1771 base_name, sha1_to_hex(object_list_sha1));
1772 if (adjust_perm(idx_tmp_name, mode))
1773 die("unable to make temporary index file readable: %s",
1775 if (rename(idx_tmp_name, tmpname))
1776 die("unable to rename temporary index file: %s",
1778 puts(sha1_to_hex(object_list_sha1));
1781 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1782 written, written_delta, reused, reused_delta);