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;
71 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
72 static int pack_compression_seen;
75 * The object names in objects array are hashed with this hashtable,
76 * to help looking up the entry by object name.
77 * This hashtable is built after all the objects are seen.
79 static int *object_ix;
80 static int object_ix_hashsz;
83 * Pack index for existing packs give us easy access to the offsets into
84 * corresponding pack file where each object's data starts, but the entries
85 * do not store the size of the compressed representation (uncompressed
86 * size is easily available by examining the pack entry header). It is
87 * also rather expensive to find the sha1 for an object given its offset.
89 * We build a hashtable of existing packs (pack_revindex), and keep reverse
90 * index here -- pack index file is sorted by object name mapping to offset;
91 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
92 * ordered by offset, so if you know the offset of an object, next offset
93 * is where its packed representation ends and the index_nr can be used to
94 * get the object sha1 from the main index.
96 struct revindex_entry {
100 struct pack_revindex {
101 struct packed_git *p;
102 struct revindex_entry *revindex;
104 static struct pack_revindex *pack_revindex;
105 static int pack_revindex_hashsz;
110 static uint32_t written, written_delta;
111 static uint32_t reused, reused_delta;
113 static int pack_revindex_ix(struct packed_git *p)
115 unsigned long ui = (unsigned long)p;
118 ui = ui ^ (ui >> 16); /* defeat structure alignment */
119 i = (int)(ui % pack_revindex_hashsz);
120 while (pack_revindex[i].p) {
121 if (pack_revindex[i].p == p)
123 if (++i == pack_revindex_hashsz)
129 static void prepare_pack_ix(void)
132 struct packed_git *p;
133 for (num = 0, p = packed_git; p; p = p->next)
137 pack_revindex_hashsz = num * 11;
138 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
139 for (p = packed_git; p; p = p->next) {
140 num = pack_revindex_ix(p);
142 pack_revindex[num].p = p;
144 /* revindex elements are lazily initialized */
147 static int cmp_offset(const void *a_, const void *b_)
149 const struct revindex_entry *a = a_;
150 const struct revindex_entry *b = b_;
151 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
155 * Ordered list of offsets of objects in the pack.
157 static void prepare_pack_revindex(struct pack_revindex *rix)
159 struct packed_git *p = rix->p;
160 int num_ent = p->num_objects;
162 const char *index = p->index_data;
164 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
167 if (p->index_version > 1) {
168 const uint32_t *off_32 =
169 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
170 const uint32_t *off_64 = off_32 + p->num_objects;
171 for (i = 0; i < num_ent; i++) {
172 uint32_t off = ntohl(*off_32++);
173 if (!(off & 0x80000000)) {
174 rix->revindex[i].offset = off;
176 rix->revindex[i].offset =
177 ((uint64_t)ntohl(*off_64++)) << 32;
178 rix->revindex[i].offset |=
181 rix->revindex[i].nr = i;
184 for (i = 0; i < num_ent; i++) {
185 uint32_t hl = *((uint32_t *)(index + 24 * i));
186 rix->revindex[i].offset = ntohl(hl);
187 rix->revindex[i].nr = i;
191 /* This knows the pack format -- the 20-byte trailer
192 * follows immediately after the last object data.
194 rix->revindex[num_ent].offset = p->pack_size - 20;
195 rix->revindex[num_ent].nr = -1;
196 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
199 static struct revindex_entry * find_packed_object(struct packed_git *p,
204 struct pack_revindex *rix;
205 struct revindex_entry *revindex;
206 num = pack_revindex_ix(p);
208 die("internal error: pack revindex uninitialized");
209 rix = &pack_revindex[num];
211 prepare_pack_revindex(rix);
212 revindex = rix->revindex;
214 hi = p->num_objects + 1;
216 int mi = (lo + hi) / 2;
217 if (revindex[mi].offset == ofs) {
218 return revindex + mi;
220 else if (ofs < revindex[mi].offset)
225 die("internal error: pack revindex corrupt");
228 static const unsigned char *find_packed_object_name(struct packed_git *p,
231 struct revindex_entry *entry = find_packed_object(p, ofs);
232 return nth_packed_object_sha1(p, entry->nr);
235 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
237 unsigned long othersize, delta_size;
238 enum object_type type;
239 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
243 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
244 delta_buf = diff_delta(otherbuf, othersize,
245 buf, size, &delta_size, 0);
246 if (!delta_buf || delta_size != entry->delta_size)
247 die("delta size changed");
254 * The per-object header is a pretty dense thing, which is
255 * - first byte: low four bits are "size", then three bits of "type",
256 * and the high bit is "size continues".
257 * - each byte afterwards: low seven bits are size continuation,
258 * with the high bit being "size continues"
260 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
265 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
266 die("bad type %d", type);
268 c = (type << 4) | (size & 15);
281 * we are going to reuse the existing object data as is. make
282 * sure it is not corrupt.
284 static int check_pack_inflate(struct packed_git *p,
285 struct pack_window **w_curs,
288 unsigned long expect)
291 unsigned char fakebuf[4096], *in;
294 memset(&stream, 0, sizeof(stream));
295 inflateInit(&stream);
297 in = use_pack(p, w_curs, offset, &stream.avail_in);
299 stream.next_out = fakebuf;
300 stream.avail_out = sizeof(fakebuf);
301 st = inflate(&stream, Z_FINISH);
302 offset += stream.next_in - in;
303 } while (st == Z_OK || st == Z_BUF_ERROR);
305 return (st == Z_STREAM_END &&
306 stream.total_out == expect &&
307 stream.total_in == len) ? 0 : -1;
310 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
311 off_t offset, off_t len, unsigned int nr)
313 const uint32_t *index_crc;
314 uint32_t data_crc = crc32(0, Z_NULL, 0);
318 void *data = use_pack(p, w_curs, offset, &avail);
321 data_crc = crc32(data_crc, data, avail);
326 index_crc = p->index_data;
327 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
329 return data_crc != ntohl(*index_crc);
332 static void copy_pack_data(struct sha1file *f,
333 struct packed_git *p,
334 struct pack_window **w_curs,
342 in = use_pack(p, w_curs, offset, &avail);
344 avail = (unsigned int)len;
345 sha1write(f, in, avail);
351 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
354 unsigned char fakebuf[4096];
357 memset(&stream, 0, sizeof(stream));
358 stream.next_in = data;
359 stream.avail_in = len;
360 stream.next_out = fakebuf;
361 stream.avail_out = sizeof(fakebuf);
362 inflateInit(&stream);
365 st = inflate(&stream, Z_FINISH);
366 if (st == Z_STREAM_END || st == Z_OK) {
367 st = (stream.total_out == expect &&
368 stream.total_in == len) ? 0 : -1;
371 if (st != Z_BUF_ERROR) {
375 stream.next_out = fakebuf;
376 stream.avail_out = sizeof(fakebuf);
382 static int revalidate_loose_object(struct object_entry *entry,
384 unsigned long mapsize)
386 /* we already know this is a loose object with new type header. */
387 enum object_type type;
388 unsigned long size, used;
393 used = unpack_object_header_gently(map, mapsize, &type, &size);
398 return check_loose_inflate(map, mapsize, size);
401 static unsigned long write_object(struct sha1file *f,
402 struct object_entry *entry)
405 enum object_type type;
407 unsigned char header[10];
410 enum object_type obj_type;
416 obj_type = entry->type;
418 to_reuse = 0; /* explicit */
419 else if (!entry->in_pack)
420 to_reuse = 0; /* can't reuse what we don't have */
421 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
422 to_reuse = 1; /* check_object() decided it for us */
423 else if (obj_type != entry->in_pack_type)
424 to_reuse = 0; /* pack has delta which is unusable */
425 else if (entry->delta)
426 to_reuse = 0; /* we want to pack afresh */
428 to_reuse = 1; /* we have it in-pack undeltified,
429 * and we do not need to deltify it.
432 /* differing core & pack compression when loose object -> must recompress */
433 if (!entry->in_pack && pack_compression_level != zlib_compression_level)
435 else if (!no_reuse_object && !entry->in_pack && !entry->delta) {
437 unsigned long mapsize;
438 map = map_sha1_file(entry->sha1, &mapsize);
439 if (map && !legacy_loose_object(map)) {
440 /* We can copy straight into the pack file */
441 if (revalidate_loose_object(entry, map, mapsize))
442 die("corrupt loose object %s",
443 sha1_to_hex(entry->sha1));
444 sha1write(f, map, mapsize);
445 munmap(map, mapsize);
451 munmap(map, mapsize);
455 buf = read_sha1_file(entry->sha1, &type, &size);
457 die("unable to read %s", sha1_to_hex(entry->sha1));
458 if (size != entry->size)
459 die("object %s size inconsistency (%lu vs %lu)",
460 sha1_to_hex(entry->sha1), size, entry->size);
462 buf = delta_against(buf, size, entry);
463 size = entry->delta_size;
464 obj_type = (allow_ofs_delta && entry->delta->offset) ?
465 OBJ_OFS_DELTA : OBJ_REF_DELTA;
468 * The object header is a byte of 'type' followed by zero or
469 * more bytes of length.
471 hdrlen = encode_header(obj_type, size, header);
472 sha1write(f, header, hdrlen);
474 if (obj_type == OBJ_OFS_DELTA) {
476 * Deltas with relative base contain an additional
477 * encoding of the relative offset for the delta
478 * base from this object's position in the pack.
480 off_t ofs = entry->offset - entry->delta->offset;
481 unsigned pos = sizeof(header) - 1;
482 header[pos] = ofs & 127;
484 header[--pos] = 128 | (--ofs & 127);
485 sha1write(f, header + pos, sizeof(header) - pos);
486 hdrlen += sizeof(header) - pos;
487 } else if (obj_type == OBJ_REF_DELTA) {
489 * Deltas with a base reference contain
490 * an additional 20 bytes for the base sha1.
492 sha1write(f, entry->delta->sha1, 20);
495 datalen = sha1write_compressed(f, buf, size, pack_compression_level);
499 struct packed_git *p = entry->in_pack;
500 struct pack_window *w_curs = NULL;
501 struct revindex_entry *revidx;
505 obj_type = (allow_ofs_delta && entry->delta->offset) ?
506 OBJ_OFS_DELTA : OBJ_REF_DELTA;
509 hdrlen = encode_header(obj_type, entry->size, header);
510 sha1write(f, header, hdrlen);
511 if (obj_type == OBJ_OFS_DELTA) {
512 off_t ofs = entry->offset - entry->delta->offset;
513 unsigned pos = sizeof(header) - 1;
514 header[pos] = ofs & 127;
516 header[--pos] = 128 | (--ofs & 127);
517 sha1write(f, header + pos, sizeof(header) - pos);
518 hdrlen += sizeof(header) - pos;
519 } else if (obj_type == OBJ_REF_DELTA) {
520 sha1write(f, entry->delta->sha1, 20);
524 offset = entry->in_pack_offset;
525 revidx = find_packed_object(p, offset);
526 datalen = revidx[1].offset - offset;
527 if (!pack_to_stdout && p->index_version > 1 &&
528 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
529 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
530 offset += entry->in_pack_header_size;
531 datalen -= entry->in_pack_header_size;
532 if (!pack_to_stdout && p->index_version == 1 &&
533 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
534 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
535 copy_pack_data(f, p, &w_curs, offset, datalen);
543 entry->crc32 = crc32_end(f);
544 return hdrlen + datalen;
547 static off_t write_one(struct sha1file *f,
548 struct object_entry *e,
553 /* offset is non zero if object is written already. */
554 if (e->offset || e->preferred_base)
557 /* if we are deltified, write out base object first. */
559 offset = write_one(f, e->delta, offset);
562 size = write_object(f, e);
564 /* make sure off_t is sufficiently large not to wrap */
565 if (offset > offset + size)
566 die("pack too large for current definition of off_t");
567 return offset + size;
570 static int open_object_dir_tmp(const char *path)
572 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
573 return mkstemp(tmpname);
576 static off_t write_pack_file(void)
580 off_t offset, last_obj_offset = 0;
581 struct pack_header hdr;
582 int do_progress = progress;
584 if (pack_to_stdout) {
585 f = sha1fd(1, "<stdout>");
588 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
590 die("unable to create %s: %s\n", tmpname, strerror(errno));
591 pack_tmp_name = xstrdup(tmpname);
592 f = sha1fd(fd, pack_tmp_name);
596 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
598 hdr.hdr_signature = htonl(PACK_SIGNATURE);
599 hdr.hdr_version = htonl(PACK_VERSION);
600 hdr.hdr_entries = htonl(nr_result);
601 sha1write(f, &hdr, sizeof(hdr));
602 offset = sizeof(hdr);
605 for (i = 0; i < nr_objects; i++) {
606 last_obj_offset = offset;
607 offset = write_one(f, objects + i, offset);
609 display_progress(&progress_state, written);
612 stop_progress(&progress_state);
614 if (written != nr_result)
615 die("wrote %u objects while expecting %u", written, nr_result);
616 sha1close(f, pack_file_sha1, 1);
618 return last_obj_offset;
621 static int sha1_sort(const void *_a, const void *_b)
623 const struct object_entry *a = *(struct object_entry **)_a;
624 const struct object_entry *b = *(struct object_entry **)_b;
625 return hashcmp(a->sha1, b->sha1);
628 static uint32_t index_default_version = 1;
629 static uint32_t index_off32_limit = 0x7fffffff;
631 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
634 struct object_entry **sorted_by_sha, **list, **last;
636 uint32_t i, index_version;
639 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
641 die("unable to create %s: %s\n", tmpname, strerror(errno));
642 idx_tmp_name = xstrdup(tmpname);
643 f = sha1fd(fd, idx_tmp_name);
648 xcalloc(nr_result, sizeof(struct object_entry *));
649 for (i = 0; i < nr_objects; i++)
650 if (!objects[i].preferred_base)
651 sorted_by_sha[j++] = objects + i;
653 die("listed %u objects while expecting %u", j, nr_result);
654 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
655 list = sorted_by_sha;
656 last = sorted_by_sha + nr_result;
658 sorted_by_sha = list = last = NULL;
660 /* if last object's offset is >= 2^31 we should use index V2 */
661 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
663 /* index versions 2 and above need a header */
664 if (index_version >= 2) {
665 struct pack_idx_header hdr;
666 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
667 hdr.idx_version = htonl(index_version);
668 sha1write(f, &hdr, sizeof(hdr));
672 * Write the first-level table (the list is sorted,
673 * but we use a 256-entry lookup to be able to avoid
674 * having to do eight extra binary search iterations).
676 for (i = 0; i < 256; i++) {
677 struct object_entry **next = list;
678 while (next < last) {
679 struct object_entry *entry = *next;
680 if (entry->sha1[0] != i)
684 array[i] = htonl(next - sorted_by_sha);
687 sha1write(f, array, 256 * 4);
689 /* Compute the SHA1 hash of sorted object names. */
692 /* Write the actual SHA1 entries. */
693 list = sorted_by_sha;
694 for (i = 0; i < nr_result; i++) {
695 struct object_entry *entry = *list++;
696 if (index_version < 2) {
697 uint32_t offset = htonl(entry->offset);
698 sha1write(f, &offset, 4);
700 sha1write(f, entry->sha1, 20);
701 SHA1_Update(&ctx, entry->sha1, 20);
704 if (index_version >= 2) {
705 unsigned int nr_large_offset = 0;
707 /* write the crc32 table */
708 list = sorted_by_sha;
709 for (i = 0; i < nr_objects; i++) {
710 struct object_entry *entry = *list++;
711 uint32_t crc32_val = htonl(entry->crc32);
712 sha1write(f, &crc32_val, 4);
715 /* write the 32-bit offset table */
716 list = sorted_by_sha;
717 for (i = 0; i < nr_objects; i++) {
718 struct object_entry *entry = *list++;
719 uint32_t offset = (entry->offset <= index_off32_limit) ?
720 entry->offset : (0x80000000 | nr_large_offset++);
721 offset = htonl(offset);
722 sha1write(f, &offset, 4);
725 /* write the large offset table */
726 list = sorted_by_sha;
727 while (nr_large_offset) {
728 struct object_entry *entry = *list++;
729 uint64_t offset = entry->offset;
730 if (offset > index_off32_limit) {
732 split[0] = htonl(offset >> 32);
733 split[1] = htonl(offset & 0xffffffff);
734 sha1write(f, split, 8);
740 sha1write(f, pack_file_sha1, 20);
741 sha1close(f, NULL, 1);
743 SHA1_Final(sha1, &ctx);
746 static int locate_object_entry_hash(const unsigned char *sha1)
750 memcpy(&ui, sha1, sizeof(unsigned int));
751 i = ui % object_ix_hashsz;
752 while (0 < object_ix[i]) {
753 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
755 if (++i == object_ix_hashsz)
761 static struct object_entry *locate_object_entry(const unsigned char *sha1)
765 if (!object_ix_hashsz)
768 i = locate_object_entry_hash(sha1);
770 return &objects[object_ix[i]-1];
774 static void rehash_objects(void)
777 struct object_entry *oe;
779 object_ix_hashsz = nr_objects * 3;
780 if (object_ix_hashsz < 1024)
781 object_ix_hashsz = 1024;
782 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
783 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
784 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
785 int ix = locate_object_entry_hash(oe->sha1);
789 object_ix[ix] = i + 1;
793 static unsigned name_hash(const char *name)
799 * This effectively just creates a sortable number from the
800 * last sixteen non-whitespace characters. Last characters
801 * count "most", so things that end in ".c" sort together.
803 while ((c = *name++) != 0) {
806 hash = (hash >> 2) + (c << 24);
811 static int add_object_entry(const unsigned char *sha1, enum object_type type,
812 unsigned hash, int exclude)
814 struct object_entry *entry;
815 struct packed_git *p, *found_pack = NULL;
816 off_t found_offset = 0;
819 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
822 entry = objects + object_ix[ix] - 1;
823 if (!entry->preferred_base)
825 entry->preferred_base = 1;
830 for (p = packed_git; p; p = p->next) {
831 off_t offset = find_pack_entry_one(sha1, p);
834 found_offset = offset;
841 if (local && !p->pack_local)
846 if (nr_objects >= nr_alloc) {
847 nr_alloc = (nr_alloc + 1024) * 3 / 2;
848 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
851 entry = objects + nr_objects++;
852 memset(entry, 0, sizeof(*entry));
853 hashcpy(entry->sha1, sha1);
858 entry->preferred_base = 1;
862 entry->in_pack = found_pack;
863 entry->in_pack_offset = found_offset;
866 if (object_ix_hashsz * 3 <= nr_objects * 4)
869 object_ix[-1 - ix] = nr_objects;
872 display_progress(&progress_state, nr_objects);
877 struct pbase_tree_cache {
878 unsigned char sha1[20];
882 unsigned long tree_size;
885 static struct pbase_tree_cache *(pbase_tree_cache[256]);
886 static int pbase_tree_cache_ix(const unsigned char *sha1)
888 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
890 static int pbase_tree_cache_ix_incr(int ix)
892 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
895 static struct pbase_tree {
896 struct pbase_tree *next;
897 /* This is a phony "cache" entry; we are not
898 * going to evict it nor find it through _get()
899 * mechanism -- this is for the toplevel node that
900 * would almost always change with any commit.
902 struct pbase_tree_cache pcache;
905 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
907 struct pbase_tree_cache *ent, *nent;
910 enum object_type type;
912 int my_ix = pbase_tree_cache_ix(sha1);
913 int available_ix = -1;
915 /* pbase-tree-cache acts as a limited hashtable.
916 * your object will be found at your index or within a few
917 * slots after that slot if it is cached.
919 for (neigh = 0; neigh < 8; neigh++) {
920 ent = pbase_tree_cache[my_ix];
921 if (ent && !hashcmp(ent->sha1, sha1)) {
925 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
926 ((0 <= available_ix) &&
927 (!ent && pbase_tree_cache[available_ix])))
928 available_ix = my_ix;
931 my_ix = pbase_tree_cache_ix_incr(my_ix);
934 /* Did not find one. Either we got a bogus request or
935 * we need to read and perhaps cache.
937 data = read_sha1_file(sha1, &type, &size);
940 if (type != OBJ_TREE) {
945 /* We need to either cache or return a throwaway copy */
947 if (available_ix < 0)
950 ent = pbase_tree_cache[available_ix];
951 my_ix = available_ix;
955 nent = xmalloc(sizeof(*nent));
956 nent->temporary = (available_ix < 0);
959 /* evict and reuse */
960 free(ent->tree_data);
963 hashcpy(nent->sha1, sha1);
964 nent->tree_data = data;
965 nent->tree_size = size;
967 if (!nent->temporary)
968 pbase_tree_cache[my_ix] = nent;
972 static void pbase_tree_put(struct pbase_tree_cache *cache)
974 if (!cache->temporary) {
978 free(cache->tree_data);
982 static int name_cmp_len(const char *name)
985 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
990 static void add_pbase_object(struct tree_desc *tree,
993 const char *fullname)
995 struct name_entry entry;
998 while (tree_entry(tree,&entry)) {
999 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1000 memcmp(name, entry.path, cmplen);
1005 if (name[cmplen] != '/') {
1006 unsigned hash = name_hash(fullname);
1007 add_object_entry(entry.sha1,
1008 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1012 if (S_ISDIR(entry.mode)) {
1013 struct tree_desc sub;
1014 struct pbase_tree_cache *tree;
1015 const char *down = name+cmplen+1;
1016 int downlen = name_cmp_len(down);
1018 tree = pbase_tree_get(entry.sha1);
1021 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1023 add_pbase_object(&sub, down, downlen, fullname);
1024 pbase_tree_put(tree);
1029 static unsigned *done_pbase_paths;
1030 static int done_pbase_paths_num;
1031 static int done_pbase_paths_alloc;
1032 static int done_pbase_path_pos(unsigned hash)
1035 int hi = done_pbase_paths_num;
1037 int mi = (hi + lo) / 2;
1038 if (done_pbase_paths[mi] == hash)
1040 if (done_pbase_paths[mi] < hash)
1048 static int check_pbase_path(unsigned hash)
1050 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1054 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1055 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1056 done_pbase_paths = xrealloc(done_pbase_paths,
1057 done_pbase_paths_alloc *
1060 done_pbase_paths_num++;
1061 if (pos < done_pbase_paths_num)
1062 memmove(done_pbase_paths + pos + 1,
1063 done_pbase_paths + pos,
1064 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1065 done_pbase_paths[pos] = hash;
1069 static void add_preferred_base_object(const char *name, unsigned hash)
1071 struct pbase_tree *it;
1074 if (!num_preferred_base || check_pbase_path(hash))
1077 cmplen = name_cmp_len(name);
1078 for (it = pbase_tree; it; it = it->next) {
1080 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
1083 struct tree_desc tree;
1084 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1085 add_pbase_object(&tree, name, cmplen, name);
1090 static void add_preferred_base(unsigned char *sha1)
1092 struct pbase_tree *it;
1095 unsigned char tree_sha1[20];
1097 if (window <= num_preferred_base++)
1100 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1104 for (it = pbase_tree; it; it = it->next) {
1105 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1111 it = xcalloc(1, sizeof(*it));
1112 it->next = pbase_tree;
1115 hashcpy(it->pcache.sha1, tree_sha1);
1116 it->pcache.tree_data = data;
1117 it->pcache.tree_size = size;
1120 static void check_object(struct object_entry *entry)
1122 if (entry->in_pack) {
1123 struct packed_git *p = entry->in_pack;
1124 struct pack_window *w_curs = NULL;
1125 const unsigned char *base_ref = NULL;
1126 struct object_entry *base_entry;
1127 unsigned long used, used_0;
1130 unsigned char *buf, c;
1132 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1135 * We want in_pack_type even if we do not reuse delta
1136 * since non-delta representations could still be reused.
1138 used = unpack_object_header_gently(buf, avail,
1139 &entry->in_pack_type,
1143 * Determine if this is a delta and if so whether we can
1144 * reuse it or not. Otherwise let's find out as cheaply as
1145 * possible what the actual type and size for this object is.
1147 switch (entry->in_pack_type) {
1149 /* Not a delta hence we've already got all we need. */
1150 entry->type = entry->in_pack_type;
1151 entry->in_pack_header_size = used;
1152 unuse_pack(&w_curs);
1155 if (!no_reuse_delta && !entry->preferred_base)
1156 base_ref = use_pack(p, &w_curs,
1157 entry->in_pack_offset + used, NULL);
1158 entry->in_pack_header_size = used + 20;
1161 buf = use_pack(p, &w_curs,
1162 entry->in_pack_offset + used, NULL);
1168 if (!ofs || MSB(ofs, 7))
1169 die("delta base offset overflow in pack for %s",
1170 sha1_to_hex(entry->sha1));
1172 ofs = (ofs << 7) + (c & 127);
1174 if (ofs >= entry->in_pack_offset)
1175 die("delta base offset out of bound for %s",
1176 sha1_to_hex(entry->sha1));
1177 ofs = entry->in_pack_offset - ofs;
1178 if (!no_reuse_delta && !entry->preferred_base)
1179 base_ref = find_packed_object_name(p, ofs);
1180 entry->in_pack_header_size = used + used_0;
1184 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1186 * If base_ref was set above that means we wish to
1187 * reuse delta data, and we even found that base
1188 * in the list of objects we want to pack. Goodie!
1190 * Depth value does not matter - find_deltas() will
1191 * never consider reused delta as the base object to
1192 * deltify other objects against, in order to avoid
1195 entry->type = entry->in_pack_type;
1196 entry->delta = base_entry;
1197 entry->delta_sibling = base_entry->delta_child;
1198 base_entry->delta_child = entry;
1199 unuse_pack(&w_curs);
1205 * This must be a delta and we already know what the
1206 * final object type is. Let's extract the actual
1207 * object size from the delta header.
1209 entry->size = get_size_from_delta(p, &w_curs,
1210 entry->in_pack_offset + entry->in_pack_header_size);
1211 unuse_pack(&w_curs);
1216 * No choice but to fall back to the recursive delta walk
1217 * with sha1_object_info() to find about the object type
1220 unuse_pack(&w_curs);
1223 entry->type = sha1_object_info(entry->sha1, &entry->size);
1224 if (entry->type < 0)
1225 die("unable to get type of object %s",
1226 sha1_to_hex(entry->sha1));
1229 static int pack_offset_sort(const void *_a, const void *_b)
1231 const struct object_entry *a = *(struct object_entry **)_a;
1232 const struct object_entry *b = *(struct object_entry **)_b;
1234 /* avoid filesystem trashing with loose objects */
1235 if (!a->in_pack && !b->in_pack)
1236 return hashcmp(a->sha1, b->sha1);
1238 if (a->in_pack < b->in_pack)
1240 if (a->in_pack > b->in_pack)
1242 return a->in_pack_offset < b->in_pack_offset ? -1 :
1243 (a->in_pack_offset > b->in_pack_offset);
1246 static void get_object_details(void)
1249 struct object_entry **sorted_by_offset;
1251 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1252 for (i = 0; i < nr_objects; i++)
1253 sorted_by_offset[i] = objects + i;
1254 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1257 for (i = 0; i < nr_objects; i++)
1258 check_object(sorted_by_offset[i]);
1259 free(sorted_by_offset);
1262 static int type_size_sort(const void *_a, const void *_b)
1264 const struct object_entry *a = *(struct object_entry **)_a;
1265 const struct object_entry *b = *(struct object_entry **)_b;
1267 if (a->type < b->type)
1269 if (a->type > b->type)
1271 if (a->hash < b->hash)
1273 if (a->hash > b->hash)
1275 if (a->preferred_base < b->preferred_base)
1277 if (a->preferred_base > b->preferred_base)
1279 if (a->size < b->size)
1281 if (a->size > b->size)
1283 return a > b ? -1 : (a < b); /* newest last */
1287 struct object_entry *entry;
1289 struct delta_index *index;
1293 * We search for deltas _backwards_ in a list sorted by type and
1294 * by size, so that we see progressively smaller and smaller files.
1295 * That's because we prefer deltas to be from the bigger file
1296 * to the smaller - deletes are potentially cheaper, but perhaps
1297 * more importantly, the bigger file is likely the more recent
1300 static int try_delta(struct unpacked *trg, struct unpacked *src,
1303 struct object_entry *trg_entry = trg->entry;
1304 struct object_entry *src_entry = src->entry;
1305 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1306 enum object_type type;
1309 /* Don't bother doing diffs between different types */
1310 if (trg_entry->type != src_entry->type)
1313 /* We do not compute delta to *create* objects we are not
1316 if (trg_entry->preferred_base)
1320 * We do not bother to try a delta that we discarded
1321 * on an earlier try, but only when reusing delta data.
1323 if (!no_reuse_delta && trg_entry->in_pack &&
1324 trg_entry->in_pack == src_entry->in_pack &&
1325 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1326 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1329 /* Let's not bust the allowed depth. */
1330 if (src_entry->depth >= max_depth)
1333 /* Now some size filtering heuristics. */
1334 trg_size = trg_entry->size;
1335 max_size = trg_size/2 - 20;
1336 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1339 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1340 max_size = trg_entry->delta_size-1;
1341 src_size = src_entry->size;
1342 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1343 if (sizediff >= max_size)
1346 /* Load data if not already done */
1348 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1350 die("object %s inconsistent object length (%lu vs %lu)",
1351 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1354 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1356 die("object %s inconsistent object length (%lu vs %lu)",
1357 sha1_to_hex(src_entry->sha1), sz, src_size);
1360 src->index = create_delta_index(src->data, src_size);
1362 die("out of memory");
1365 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1369 trg_entry->delta = src_entry;
1370 trg_entry->delta_size = delta_size;
1371 trg_entry->depth = src_entry->depth + 1;
1376 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1378 struct object_entry *child = me->delta_child;
1381 unsigned int c = check_delta_limit(child, n + 1);
1384 child = child->delta_sibling;
1389 static void find_deltas(struct object_entry **list, int window, int depth)
1391 uint32_t i = nr_objects, idx = 0, processed = 0;
1392 unsigned int array_size = window * sizeof(struct unpacked);
1393 struct unpacked *array;
1398 array = xmalloc(array_size);
1399 memset(array, 0, array_size);
1401 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1404 struct object_entry *entry = list[--i];
1405 struct unpacked *n = array + idx;
1408 if (!entry->preferred_base)
1412 display_progress(&progress_state, processed);
1415 /* This happens if we decided to reuse existing
1416 * delta from a pack. "!no_reuse_delta &&" is implied.
1420 if (entry->size < 50)
1422 free_delta_index(n->index);
1429 * If the current object is at pack edge, take the depth the
1430 * objects that depend on the current object into account
1431 * otherwise they would become too deep.
1434 if (entry->delta_child) {
1435 max_depth -= check_delta_limit(entry, 0);
1442 uint32_t other_idx = idx + j;
1444 if (other_idx >= window)
1445 other_idx -= window;
1446 m = array + other_idx;
1449 if (try_delta(n, m, max_depth) < 0)
1453 /* if we made n a delta, and if n is already at max
1454 * depth, leaving it in the window is pointless. we
1455 * should evict it first.
1457 if (entry->delta && depth <= entry->depth)
1467 stop_progress(&progress_state);
1469 for (i = 0; i < window; ++i) {
1470 free_delta_index(array[i].index);
1471 free(array[i].data);
1476 static void prepare_pack(int window, int depth)
1478 struct object_entry **delta_list;
1481 get_object_details();
1483 if (!window || !depth)
1486 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1487 for (i = 0; i < nr_objects; i++)
1488 delta_list[i] = objects + i;
1489 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1490 find_deltas(delta_list, window+1, depth);
1494 static int git_pack_config(const char *k, const char *v)
1496 if(!strcmp(k, "pack.window")) {
1497 window = git_config_int(k, v);
1500 if(!strcmp(k, "pack.depth")) {
1501 depth = git_config_int(k, v);
1504 if (!strcmp(k, "pack.compression")) {
1505 int level = git_config_int(k, v);
1507 level = Z_DEFAULT_COMPRESSION;
1508 else if (level < 0 || level > Z_BEST_COMPRESSION)
1509 die("bad pack compression level %d", level);
1510 pack_compression_level = level;
1511 pack_compression_seen = 1;
1514 return git_default_config(k, v);
1517 static void read_object_list_from_stdin(void)
1519 char line[40 + 1 + PATH_MAX + 2];
1520 unsigned char sha1[20];
1524 if (!fgets(line, sizeof(line), stdin)) {
1528 die("fgets returned NULL, not EOF, not error!");
1530 die("fgets: %s", strerror(errno));
1534 if (line[0] == '-') {
1535 if (get_sha1_hex(line+1, sha1))
1536 die("expected edge sha1, got garbage:\n %s",
1538 add_preferred_base(sha1);
1541 if (get_sha1_hex(line, sha1))
1542 die("expected sha1, got garbage:\n %s", line);
1544 hash = name_hash(line+41);
1545 add_preferred_base_object(line+41, hash);
1546 add_object_entry(sha1, 0, hash, 0);
1550 static void show_commit(struct commit *commit)
1552 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
1555 static void show_object(struct object_array_entry *p)
1557 unsigned hash = name_hash(p->name);
1558 add_preferred_base_object(p->name, hash);
1559 add_object_entry(p->item->sha1, p->item->type, hash, 0);
1562 static void show_edge(struct commit *commit)
1564 add_preferred_base(commit->object.sha1);
1567 static void get_object_list(int ac, const char **av)
1569 struct rev_info revs;
1573 init_revisions(&revs, NULL);
1574 save_commit_buffer = 0;
1575 track_object_refs = 0;
1576 setup_revisions(ac, av, &revs, NULL);
1578 while (fgets(line, sizeof(line), stdin) != NULL) {
1579 int len = strlen(line);
1580 if (line[len - 1] == '\n')
1585 if (!strcmp(line, "--not")) {
1586 flags ^= UNINTERESTING;
1589 die("not a rev '%s'", line);
1591 if (handle_revision_arg(line, &revs, flags, 1))
1592 die("bad revision '%s'", line);
1595 prepare_revision_walk(&revs);
1596 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1597 traverse_commit_list(&revs, show_commit, show_object);
1600 static int adjust_perm(const char *path, mode_t mode)
1602 if (chmod(path, mode))
1604 return adjust_shared_perm(path);
1607 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1609 int use_internal_rev_list = 0;
1612 off_t last_obj_offset;
1613 const char *base_name = NULL;
1615 int rp_ac_alloc = 64;
1618 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1620 rp_av[0] = "pack-objects";
1621 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1624 git_config(git_pack_config);
1625 if (!pack_compression_seen && core_compression_seen)
1626 pack_compression_level = core_compression_level;
1628 progress = isatty(2);
1629 for (i = 1; i < argc; i++) {
1630 const char *arg = argv[i];
1635 if (!strcmp("--non-empty", arg)) {
1639 if (!strcmp("--local", arg)) {
1643 if (!strcmp("--incremental", arg)) {
1647 if (!prefixcmp(arg, "--compression=")) {
1649 int level = strtoul(arg+14, &end, 0);
1650 if (!arg[14] || *end)
1653 level = Z_DEFAULT_COMPRESSION;
1654 else if (level < 0 || level > Z_BEST_COMPRESSION)
1655 die("bad pack compression level %d", level);
1656 pack_compression_level = level;
1659 if (!prefixcmp(arg, "--window=")) {
1661 window = strtoul(arg+9, &end, 0);
1662 if (!arg[9] || *end)
1666 if (!prefixcmp(arg, "--depth=")) {
1668 depth = strtoul(arg+8, &end, 0);
1669 if (!arg[8] || *end)
1673 if (!strcmp("--progress", arg)) {
1677 if (!strcmp("--all-progress", arg)) {
1681 if (!strcmp("-q", arg)) {
1685 if (!strcmp("--no-reuse-delta", arg)) {
1689 if (!strcmp("--no-reuse-object", arg)) {
1690 no_reuse_object = no_reuse_delta = 1;
1693 if (!strcmp("--delta-base-offset", arg)) {
1694 allow_ofs_delta = 1;
1697 if (!strcmp("--stdout", arg)) {
1701 if (!strcmp("--revs", arg)) {
1702 use_internal_rev_list = 1;
1705 if (!strcmp("--unpacked", arg) ||
1706 !prefixcmp(arg, "--unpacked=") ||
1707 !strcmp("--reflog", arg) ||
1708 !strcmp("--all", arg)) {
1709 use_internal_rev_list = 1;
1710 if (rp_ac >= rp_ac_alloc - 1) {
1711 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1712 rp_av = xrealloc(rp_av,
1713 rp_ac_alloc * sizeof(*rp_av));
1715 rp_av[rp_ac++] = arg;
1718 if (!strcmp("--thin", arg)) {
1719 use_internal_rev_list = 1;
1721 rp_av[1] = "--objects-edge";
1724 if (!prefixcmp(arg, "--index-version=")) {
1726 index_default_version = strtoul(arg + 16, &c, 10);
1727 if (index_default_version > 2)
1730 index_off32_limit = strtoul(c+1, &c, 0);
1731 if (*c || index_off32_limit & 0x80000000)
1738 /* Traditionally "pack-objects [options] base extra" failed;
1739 * we would however want to take refs parameter that would
1740 * have been given to upstream rev-list ourselves, which means
1741 * we somehow want to say what the base name is. So the
1744 * pack-objects [options] base <refs...>
1746 * in other words, we would treat the first non-option as the
1747 * base_name and send everything else to the internal revision
1751 if (!pack_to_stdout)
1752 base_name = argv[i++];
1754 if (pack_to_stdout != !base_name)
1757 if (!pack_to_stdout && thin)
1758 die("--thin cannot be used to build an indexable pack.");
1760 prepare_packed_git();
1763 start_progress(&progress_state, "Generating pack...",
1764 "Counting objects: ", 0);
1765 if (!use_internal_rev_list)
1766 read_object_list_from_stdin();
1768 rp_av[rp_ac] = NULL;
1769 get_object_list(rp_ac, rp_av);
1772 stop_progress(&progress_state);
1773 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1776 if (non_empty && !nr_result)
1778 if (progress && (nr_objects != nr_result))
1779 fprintf(stderr, "Result has %u objects.\n", nr_result);
1781 prepare_pack(window, depth);
1782 last_obj_offset = write_pack_file();
1783 if (!pack_to_stdout) {
1784 unsigned char object_list_sha1[20];
1785 mode_t mode = umask(0);
1788 mode = 0444 & ~mode;
1790 write_index_file(last_obj_offset, object_list_sha1);
1791 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
1792 base_name, sha1_to_hex(object_list_sha1));
1793 if (adjust_perm(pack_tmp_name, mode))
1794 die("unable to make temporary pack file readable: %s",
1796 if (rename(pack_tmp_name, tmpname))
1797 die("unable to rename temporary pack file: %s",
1799 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
1800 base_name, sha1_to_hex(object_list_sha1));
1801 if (adjust_perm(idx_tmp_name, mode))
1802 die("unable to make temporary index file readable: %s",
1804 if (rename(idx_tmp_name, tmpname))
1805 die("unable to rename temporary index file: %s",
1807 puts(sha1_to_hex(object_list_sha1));
1810 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1811 written, written_delta, reused, reused_delta);