10 #include "csum-file.h"
11 #include "tree-walk.h"
14 #include "list-objects.h"
16 static const char pack_usage[] = "\
17 git-pack-objects [{ -q | --progress | --all-progress }] \n\
18 [--local] [--incremental] [--window=N] [--depth=N] \n\
19 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
20 [--revs [--unpacked | --all]*] [--reflog] [--stdout | base-name] \n\
21 [<ref-list | <object-list]";
24 unsigned char sha1[20];
25 unsigned long size; /* uncompressed size */
26 off_t offset; /* offset into the final pack file;
27 * nonzero if already written.
29 unsigned int depth; /* delta depth */
30 unsigned int delta_limit; /* base adjustment for in-pack delta */
31 unsigned int hash; /* name hint hash */
32 enum object_type type;
33 enum object_type in_pack_type; /* could be delta */
34 unsigned long delta_size; /* delta data size (uncompressed) */
35 #define in_pack_header_size delta_size /* only when reusing pack data */
36 struct object_entry *delta; /* delta base object */
37 struct packed_git *in_pack; /* already in pack */
39 struct object_entry *delta_child; /* deltified objects who bases me */
40 struct object_entry *delta_sibling; /* other deltified objects who
41 * uses the same base as me
43 int preferred_base; /* we do not pack this, but is encouraged to
44 * be used as the base objectto delta huge
47 uint32_t crc32; /* crc of raw pack data for this object */
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 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
57 * elements in the objects array. The former is used to build the pack
58 * index (lists object names in the ascending order to help offset lookup),
59 * and the latter is used to group similar things together by try_delta()
63 static unsigned char object_list_sha1[20];
65 static int no_reuse_delta;
67 static int incremental;
68 static int allow_ofs_delta;
70 static struct object_entry **sorted_by_sha, **sorted_by_type;
71 static struct object_entry *objects;
72 static uint32_t nr_objects, nr_alloc, nr_result;
73 static const char *base_name;
74 static unsigned char pack_file_sha1[20];
75 static int progress = 1;
76 static volatile sig_atomic_t progress_update;
77 static int window = 10;
78 static int pack_to_stdout;
79 static int num_preferred_base;
82 * The object names in objects array are hashed with this hashtable,
83 * to help looking up the entry by object name. Binary search from
84 * sorted_by_sha is also possible but this was easier to code and faster.
85 * This hashtable is built after all the objects are seen.
87 static int *object_ix;
88 static int object_ix_hashsz;
91 * Pack index for existing packs give us easy access to the offsets into
92 * corresponding pack file where each object's data starts, but the entries
93 * do not store the size of the compressed representation (uncompressed
94 * size is easily available by examining the pack entry header). It is
95 * also rather expensive to find the sha1 for an object given its offset.
97 * We build a hashtable of existing packs (pack_revindex), and keep reverse
98 * index here -- pack index file is sorted by object name mapping to offset;
99 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
100 * ordered by offset, so if you know the offset of an object, next offset
101 * is where its packed representation ends and the index_nr can be used to
102 * get the object sha1 from the main index.
104 struct revindex_entry {
108 struct pack_revindex {
109 struct packed_git *p;
110 struct revindex_entry *revindex;
112 static struct pack_revindex *pack_revindex;
113 static int pack_revindex_hashsz;
118 static uint32_t written, written_delta;
119 static uint32_t reused, reused_delta;
121 static int pack_revindex_ix(struct packed_git *p)
123 unsigned long ui = (unsigned long)p;
126 ui = ui ^ (ui >> 16); /* defeat structure alignment */
127 i = (int)(ui % pack_revindex_hashsz);
128 while (pack_revindex[i].p) {
129 if (pack_revindex[i].p == p)
131 if (++i == pack_revindex_hashsz)
137 static void prepare_pack_ix(void)
140 struct packed_git *p;
141 for (num = 0, p = packed_git; p; p = p->next)
145 pack_revindex_hashsz = num * 11;
146 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
147 for (p = packed_git; p; p = p->next) {
148 num = pack_revindex_ix(p);
150 pack_revindex[num].p = p;
152 /* revindex elements are lazily initialized */
155 static int cmp_offset(const void *a_, const void *b_)
157 const struct revindex_entry *a = a_;
158 const struct revindex_entry *b = b_;
159 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
163 * Ordered list of offsets of objects in the pack.
165 static void prepare_pack_revindex(struct pack_revindex *rix)
167 struct packed_git *p = rix->p;
168 int num_ent = p->num_objects;
170 const char *index = p->index_data;
172 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
175 if (p->index_version > 1) {
176 const uint32_t *off_32 =
177 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
178 const uint32_t *off_64 = off_32 + p->num_objects;
179 for (i = 0; i < num_ent; i++) {
180 uint32_t off = ntohl(*off_32++);
181 if (!(off & 0x80000000)) {
182 rix->revindex[i].offset = off;
184 rix->revindex[i].offset =
185 ((uint64_t)ntohl(*off_64++)) << 32;
186 rix->revindex[i].offset |=
189 rix->revindex[i].nr = i;
192 for (i = 0; i < num_ent; i++) {
193 uint32_t hl = *((uint32_t *)(index + 24 * i));
194 rix->revindex[i].offset = ntohl(hl);
195 rix->revindex[i].nr = i;
199 /* This knows the pack format -- the 20-byte trailer
200 * follows immediately after the last object data.
202 rix->revindex[num_ent].offset = p->pack_size - 20;
203 rix->revindex[num_ent].nr = -1;
204 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
207 static struct revindex_entry * find_packed_object(struct packed_git *p,
212 struct pack_revindex *rix;
213 struct revindex_entry *revindex;
214 num = pack_revindex_ix(p);
216 die("internal error: pack revindex uninitialized");
217 rix = &pack_revindex[num];
219 prepare_pack_revindex(rix);
220 revindex = rix->revindex;
222 hi = p->num_objects + 1;
224 int mi = (lo + hi) / 2;
225 if (revindex[mi].offset == ofs) {
226 return revindex + mi;
228 else if (ofs < revindex[mi].offset)
233 die("internal error: pack revindex corrupt");
236 static const unsigned char *find_packed_object_name(struct packed_git *p,
239 struct revindex_entry *entry = find_packed_object(p, ofs);
240 return nth_packed_object_sha1(p, entry->nr);
243 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
245 unsigned long othersize, delta_size;
246 enum object_type type;
247 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
251 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
252 delta_buf = diff_delta(otherbuf, othersize,
253 buf, size, &delta_size, 0);
254 if (!delta_buf || delta_size != entry->delta_size)
255 die("delta size changed");
262 * The per-object header is a pretty dense thing, which is
263 * - first byte: low four bits are "size", then three bits of "type",
264 * and the high bit is "size continues".
265 * - each byte afterwards: low seven bits are size continuation,
266 * with the high bit being "size continues"
268 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
273 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
274 die("bad type %d", type);
276 c = (type << 4) | (size & 15);
289 * we are going to reuse the existing object data as is. make
290 * sure it is not corrupt.
292 static int check_pack_inflate(struct packed_git *p,
293 struct pack_window **w_curs,
296 unsigned long expect)
299 unsigned char fakebuf[4096], *in;
302 memset(&stream, 0, sizeof(stream));
303 inflateInit(&stream);
305 in = use_pack(p, w_curs, offset, &stream.avail_in);
307 stream.next_out = fakebuf;
308 stream.avail_out = sizeof(fakebuf);
309 st = inflate(&stream, Z_FINISH);
310 offset += stream.next_in - in;
311 } while (st == Z_OK || st == Z_BUF_ERROR);
313 return (st == Z_STREAM_END &&
314 stream.total_out == expect &&
315 stream.total_in == len) ? 0 : -1;
318 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
319 off_t offset, off_t len, unsigned int nr)
321 const uint32_t *index_crc;
322 uint32_t data_crc = crc32(0, Z_NULL, 0);
326 void *data = use_pack(p, w_curs, offset, &avail);
329 data_crc = crc32(data_crc, data, avail);
334 index_crc = p->index_data;
335 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
337 return data_crc != ntohl(*index_crc);
340 static void copy_pack_data(struct sha1file *f,
341 struct packed_git *p,
342 struct pack_window **w_curs,
350 in = use_pack(p, w_curs, offset, &avail);
352 avail = (unsigned int)len;
353 sha1write(f, in, avail);
359 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
362 unsigned char fakebuf[4096];
365 memset(&stream, 0, sizeof(stream));
366 stream.next_in = data;
367 stream.avail_in = len;
368 stream.next_out = fakebuf;
369 stream.avail_out = sizeof(fakebuf);
370 inflateInit(&stream);
373 st = inflate(&stream, Z_FINISH);
374 if (st == Z_STREAM_END || st == Z_OK) {
375 st = (stream.total_out == expect &&
376 stream.total_in == len) ? 0 : -1;
379 if (st != Z_BUF_ERROR) {
383 stream.next_out = fakebuf;
384 stream.avail_out = sizeof(fakebuf);
390 static int revalidate_loose_object(struct object_entry *entry,
392 unsigned long mapsize)
394 /* we already know this is a loose object with new type header. */
395 enum object_type type;
396 unsigned long size, used;
401 used = unpack_object_header_gently(map, mapsize, &type, &size);
406 return check_loose_inflate(map, mapsize, size);
409 static unsigned long write_object(struct sha1file *f,
410 struct object_entry *entry)
413 enum object_type type;
415 unsigned char header[10];
418 enum object_type obj_type;
424 obj_type = entry->type;
425 if (! entry->in_pack)
426 to_reuse = 0; /* can't reuse what we don't have */
427 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
428 to_reuse = 1; /* check_object() decided it for us */
429 else if (obj_type != entry->in_pack_type)
430 to_reuse = 0; /* pack has delta which is unusable */
431 else if (entry->delta)
432 to_reuse = 0; /* we want to pack afresh */
434 to_reuse = 1; /* we have it in-pack undeltified,
435 * and we do not need to deltify it.
438 if (!entry->in_pack && !entry->delta) {
440 unsigned long mapsize;
441 map = map_sha1_file(entry->sha1, &mapsize);
442 if (map && !legacy_loose_object(map)) {
443 /* We can copy straight into the pack file */
444 if (revalidate_loose_object(entry, map, mapsize))
445 die("corrupt loose object %s",
446 sha1_to_hex(entry->sha1));
447 sha1write(f, map, mapsize);
448 munmap(map, mapsize);
454 munmap(map, mapsize);
458 buf = read_sha1_file(entry->sha1, &type, &size);
460 die("unable to read %s", sha1_to_hex(entry->sha1));
461 if (size != entry->size)
462 die("object %s size inconsistency (%lu vs %lu)",
463 sha1_to_hex(entry->sha1), size, entry->size);
465 buf = delta_against(buf, size, entry);
466 size = entry->delta_size;
467 obj_type = (allow_ofs_delta && entry->delta->offset) ?
468 OBJ_OFS_DELTA : OBJ_REF_DELTA;
471 * The object header is a byte of 'type' followed by zero or
472 * more bytes of length.
474 hdrlen = encode_header(obj_type, size, header);
475 sha1write(f, header, hdrlen);
477 if (obj_type == OBJ_OFS_DELTA) {
479 * Deltas with relative base contain an additional
480 * encoding of the relative offset for the delta
481 * base from this object's position in the pack.
483 off_t ofs = entry->offset - entry->delta->offset;
484 unsigned pos = sizeof(header) - 1;
485 header[pos] = ofs & 127;
487 header[--pos] = 128 | (--ofs & 127);
488 sha1write(f, header + pos, sizeof(header) - pos);
489 hdrlen += sizeof(header) - pos;
490 } else if (obj_type == OBJ_REF_DELTA) {
492 * Deltas with a base reference contain
493 * an additional 20 bytes for the base sha1.
495 sha1write(f, entry->delta->sha1, 20);
498 datalen = sha1write_compressed(f, buf, size);
502 struct packed_git *p = entry->in_pack;
503 struct pack_window *w_curs = NULL;
504 struct revindex_entry *revidx;
508 obj_type = (allow_ofs_delta && entry->delta->offset) ?
509 OBJ_OFS_DELTA : OBJ_REF_DELTA;
512 hdrlen = encode_header(obj_type, entry->size, header);
513 sha1write(f, header, hdrlen);
514 if (obj_type == OBJ_OFS_DELTA) {
515 off_t ofs = entry->offset - entry->delta->offset;
516 unsigned pos = sizeof(header) - 1;
517 header[pos] = ofs & 127;
519 header[--pos] = 128 | (--ofs & 127);
520 sha1write(f, header + pos, sizeof(header) - pos);
521 hdrlen += sizeof(header) - pos;
522 } else if (obj_type == OBJ_REF_DELTA) {
523 sha1write(f, entry->delta->sha1, 20);
527 offset = entry->in_pack_offset;
528 revidx = find_packed_object(p, offset);
529 datalen = revidx[1].offset - offset;
530 if (!pack_to_stdout && p->index_version > 1 &&
531 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
532 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
533 offset += entry->in_pack_header_size;
534 datalen -= entry->in_pack_header_size;
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);
565 size = write_object(f, e);
567 /* make sure off_t is sufficiently large not to wrap */
568 if (offset > offset + size)
569 die("pack too large for current definition of off_t");
570 return offset + size;
573 static off_t write_pack_file(void)
577 off_t offset, last_obj_offset = 0;
578 struct pack_header hdr;
579 unsigned last_percent = 999;
580 int do_progress = progress;
583 f = sha1fd(1, "<stdout>");
587 f = sha1create("%s-%s.%s", base_name,
588 sha1_to_hex(object_list_sha1), "pack");
590 fprintf(stderr, "Writing %u objects.\n", nr_result);
592 hdr.hdr_signature = htonl(PACK_SIGNATURE);
593 hdr.hdr_version = htonl(PACK_VERSION);
594 hdr.hdr_entries = htonl(nr_result);
595 sha1write(f, &hdr, sizeof(hdr));
596 offset = sizeof(hdr);
599 for (i = 0; i < nr_objects; i++) {
600 last_obj_offset = offset;
601 offset = write_one(f, objects + i, offset);
603 unsigned percent = written * 100 / nr_result;
604 if (progress_update || percent != last_percent) {
605 fprintf(stderr, "%4u%% (%u/%u) done\r",
606 percent, written, nr_result);
608 last_percent = percent;
615 if (written != nr_result)
616 die("wrote %u objects while expecting %u", written, nr_result);
617 sha1close(f, pack_file_sha1, 1);
619 return last_obj_offset;
622 static uint32_t index_default_version = 1;
623 static uint32_t index_off32_limit = 0x7fffffff;
625 static void write_index_file(off_t last_obj_offset)
628 struct sha1file *f = sha1create("%s-%s.%s", base_name,
629 sha1_to_hex(object_list_sha1), "idx");
630 struct object_entry **list = sorted_by_sha;
631 struct object_entry **last = list + nr_result;
633 uint32_t index_version;
635 /* if last object's offset is >= 2^31 we should use index V2 */
636 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
638 /* index versions 2 and above need a header */
639 if (index_version >= 2) {
640 struct pack_idx_header hdr;
641 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
642 hdr.idx_version = htonl(index_version);
643 sha1write(f, &hdr, sizeof(hdr));
647 * Write the first-level table (the list is sorted,
648 * but we use a 256-entry lookup to be able to avoid
649 * having to do eight extra binary search iterations).
651 for (i = 0; i < 256; i++) {
652 struct object_entry **next = list;
653 while (next < last) {
654 struct object_entry *entry = *next;
655 if (entry->sha1[0] != i)
659 array[i] = htonl(next - sorted_by_sha);
662 sha1write(f, array, 256 * 4);
665 * Write the actual SHA1 entries..
667 list = sorted_by_sha;
668 for (i = 0; i < nr_result; i++) {
669 struct object_entry *entry = *list++;
670 if (index_version < 2) {
671 uint32_t offset = htonl(entry->offset);
672 sha1write(f, &offset, 4);
674 sha1write(f, entry->sha1, 20);
677 if (index_version >= 2) {
678 unsigned int nr_large_offset = 0;
680 /* write the crc32 table */
681 list = sorted_by_sha;
682 for (i = 0; i < nr_objects; i++) {
683 struct object_entry *entry = *list++;
684 uint32_t crc32_val = htonl(entry->crc32);
685 sha1write(f, &crc32_val, 4);
688 /* write the 32-bit offset table */
689 list = sorted_by_sha;
690 for (i = 0; i < nr_objects; i++) {
691 struct object_entry *entry = *list++;
692 uint32_t offset = (entry->offset <= index_off32_limit) ?
693 entry->offset : (0x80000000 | nr_large_offset++);
694 offset = htonl(offset);
695 sha1write(f, &offset, 4);
698 /* write the large offset table */
699 list = sorted_by_sha;
700 while (nr_large_offset) {
701 struct object_entry *entry = *list++;
702 uint64_t offset = entry->offset;
703 if (offset > index_off32_limit) {
705 split[0] = htonl(offset >> 32);
706 split[1] = htonl(offset & 0xffffffff);
707 sha1write(f, split, 8);
713 sha1write(f, pack_file_sha1, 20);
714 sha1close(f, NULL, 1);
717 static int locate_object_entry_hash(const unsigned char *sha1)
721 memcpy(&ui, sha1, sizeof(unsigned int));
722 i = ui % object_ix_hashsz;
723 while (0 < object_ix[i]) {
724 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
726 if (++i == object_ix_hashsz)
732 static struct object_entry *locate_object_entry(const unsigned char *sha1)
736 if (!object_ix_hashsz)
739 i = locate_object_entry_hash(sha1);
741 return &objects[object_ix[i]-1];
745 static void rehash_objects(void)
748 struct object_entry *oe;
750 object_ix_hashsz = nr_objects * 3;
751 if (object_ix_hashsz < 1024)
752 object_ix_hashsz = 1024;
753 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
754 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
755 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
756 int ix = locate_object_entry_hash(oe->sha1);
760 object_ix[ix] = i + 1;
764 static unsigned name_hash(const char *name)
770 * This effectively just creates a sortable number from the
771 * last sixteen non-whitespace characters. Last characters
772 * count "most", so things that end in ".c" sort together.
774 while ((c = *name++) != 0) {
777 hash = (hash >> 2) + (c << 24);
782 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
784 struct object_entry *entry;
785 struct packed_git *p, *found_pack = NULL;
786 off_t found_offset = 0;
789 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
792 entry = objects + object_ix[ix] - 1;
793 entry->preferred_base = 1;
799 for (p = packed_git; p; p = p->next) {
800 off_t offset = find_pack_entry_one(sha1, p);
804 if (local && !p->pack_local)
807 found_offset = offset;
814 if (nr_objects >= nr_alloc) {
815 nr_alloc = (nr_alloc + 1024) * 3 / 2;
816 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
819 entry = objects + nr_objects++;
820 memset(entry, 0, sizeof(*entry));
821 hashcpy(entry->sha1, sha1);
824 entry->preferred_base = 1;
826 entry->in_pack = found_pack;
827 entry->in_pack_offset = found_offset;
830 if (object_ix_hashsz * 3 <= nr_objects * 4)
833 object_ix[-1 - ix] = nr_objects;
835 if (progress_update) {
836 fprintf(stderr, "Counting objects...%u\r", nr_objects);
843 struct pbase_tree_cache {
844 unsigned char sha1[20];
848 unsigned long tree_size;
851 static struct pbase_tree_cache *(pbase_tree_cache[256]);
852 static int pbase_tree_cache_ix(const unsigned char *sha1)
854 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
856 static int pbase_tree_cache_ix_incr(int ix)
858 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
861 static struct pbase_tree {
862 struct pbase_tree *next;
863 /* This is a phony "cache" entry; we are not
864 * going to evict it nor find it through _get()
865 * mechanism -- this is for the toplevel node that
866 * would almost always change with any commit.
868 struct pbase_tree_cache pcache;
871 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
873 struct pbase_tree_cache *ent, *nent;
876 enum object_type type;
878 int my_ix = pbase_tree_cache_ix(sha1);
879 int available_ix = -1;
881 /* pbase-tree-cache acts as a limited hashtable.
882 * your object will be found at your index or within a few
883 * slots after that slot if it is cached.
885 for (neigh = 0; neigh < 8; neigh++) {
886 ent = pbase_tree_cache[my_ix];
887 if (ent && !hashcmp(ent->sha1, sha1)) {
891 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
892 ((0 <= available_ix) &&
893 (!ent && pbase_tree_cache[available_ix])))
894 available_ix = my_ix;
897 my_ix = pbase_tree_cache_ix_incr(my_ix);
900 /* Did not find one. Either we got a bogus request or
901 * we need to read and perhaps cache.
903 data = read_sha1_file(sha1, &type, &size);
906 if (type != OBJ_TREE) {
911 /* We need to either cache or return a throwaway copy */
913 if (available_ix < 0)
916 ent = pbase_tree_cache[available_ix];
917 my_ix = available_ix;
921 nent = xmalloc(sizeof(*nent));
922 nent->temporary = (available_ix < 0);
925 /* evict and reuse */
926 free(ent->tree_data);
929 hashcpy(nent->sha1, sha1);
930 nent->tree_data = data;
931 nent->tree_size = size;
933 if (!nent->temporary)
934 pbase_tree_cache[my_ix] = nent;
938 static void pbase_tree_put(struct pbase_tree_cache *cache)
940 if (!cache->temporary) {
944 free(cache->tree_data);
948 static int name_cmp_len(const char *name)
951 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
956 static void add_pbase_object(struct tree_desc *tree,
959 const char *fullname)
961 struct name_entry entry;
963 while (tree_entry(tree,&entry)) {
965 enum object_type type;
967 if (tree_entry_len(entry.path, entry.sha1) != cmplen ||
968 memcmp(entry.path, name, cmplen) ||
969 !has_sha1_file(entry.sha1) ||
970 (type = sha1_object_info(entry.sha1, &size)) < 0)
972 if (name[cmplen] != '/') {
973 unsigned hash = name_hash(fullname);
974 add_object_entry(entry.sha1, hash, 1);
977 if (type == OBJ_TREE) {
978 struct tree_desc sub;
979 struct pbase_tree_cache *tree;
980 const char *down = name+cmplen+1;
981 int downlen = name_cmp_len(down);
983 tree = pbase_tree_get(entry.sha1);
986 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
988 add_pbase_object(&sub, down, downlen, fullname);
989 pbase_tree_put(tree);
994 static unsigned *done_pbase_paths;
995 static int done_pbase_paths_num;
996 static int done_pbase_paths_alloc;
997 static int done_pbase_path_pos(unsigned hash)
1000 int hi = done_pbase_paths_num;
1002 int mi = (hi + lo) / 2;
1003 if (done_pbase_paths[mi] == hash)
1005 if (done_pbase_paths[mi] < hash)
1013 static int check_pbase_path(unsigned hash)
1015 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1019 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1020 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1021 done_pbase_paths = xrealloc(done_pbase_paths,
1022 done_pbase_paths_alloc *
1025 done_pbase_paths_num++;
1026 if (pos < done_pbase_paths_num)
1027 memmove(done_pbase_paths + pos + 1,
1028 done_pbase_paths + pos,
1029 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1030 done_pbase_paths[pos] = hash;
1034 static void add_preferred_base_object(const char *name, unsigned hash)
1036 struct pbase_tree *it;
1037 int cmplen = name_cmp_len(name);
1039 if (check_pbase_path(hash))
1042 for (it = pbase_tree; it; it = it->next) {
1044 hash = name_hash("");
1045 add_object_entry(it->pcache.sha1, hash, 1);
1048 struct tree_desc tree;
1049 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1050 add_pbase_object(&tree, name, cmplen, name);
1055 static void add_preferred_base(unsigned char *sha1)
1057 struct pbase_tree *it;
1060 unsigned char tree_sha1[20];
1062 if (window <= num_preferred_base++)
1065 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1069 for (it = pbase_tree; it; it = it->next) {
1070 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1076 it = xcalloc(1, sizeof(*it));
1077 it->next = pbase_tree;
1080 hashcpy(it->pcache.sha1, tree_sha1);
1081 it->pcache.tree_data = data;
1082 it->pcache.tree_size = size;
1085 static void check_object(struct object_entry *entry)
1087 if (entry->in_pack && !entry->preferred_base) {
1088 struct packed_git *p = entry->in_pack;
1089 struct pack_window *w_curs = NULL;
1090 unsigned long size, used;
1093 struct object_entry *base_entry = NULL;
1095 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1097 /* We want in_pack_type even if we do not reuse delta.
1098 * There is no point not reusing non-delta representations.
1100 used = unpack_object_header_gently(buf, avail,
1101 &entry->in_pack_type, &size);
1103 /* Check if it is delta, and the base is also an object
1104 * we are going to pack. If so we will reuse the existing
1107 if (!no_reuse_delta) {
1109 const unsigned char *base_name;
1111 unsigned long used_0;
1112 /* there is at least 20 bytes left in the pack */
1113 switch (entry->in_pack_type) {
1115 base_name = use_pack(p, &w_curs,
1116 entry->in_pack_offset + used, NULL);
1120 buf = use_pack(p, &w_curs,
1121 entry->in_pack_offset + used, NULL);
1127 if (!ofs || MSB(ofs, 7))
1128 die("delta base offset overflow in pack for %s",
1129 sha1_to_hex(entry->sha1));
1131 ofs = (ofs << 7) + (c & 127);
1133 if (ofs >= entry->in_pack_offset)
1134 die("delta base offset out of bound for %s",
1135 sha1_to_hex(entry->sha1));
1136 ofs = entry->in_pack_offset - ofs;
1137 base_name = find_packed_object_name(p, ofs);
1144 base_entry = locate_object_entry(base_name);
1146 unuse_pack(&w_curs);
1147 entry->in_pack_header_size = used;
1151 /* Depth value does not matter - find_deltas()
1152 * will never consider reused delta as the
1153 * base object to deltify other objects
1154 * against, in order to avoid circular deltas.
1157 /* uncompressed size of the delta data */
1159 entry->delta = base_entry;
1160 entry->type = entry->in_pack_type;
1162 entry->delta_sibling = base_entry->delta_child;
1163 base_entry->delta_child = entry;
1167 /* Otherwise we would do the usual */
1170 entry->type = sha1_object_info(entry->sha1, &entry->size);
1171 if (entry->type < 0)
1172 die("unable to get type of object %s",
1173 sha1_to_hex(entry->sha1));
1176 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1178 struct object_entry *child = me->delta_child;
1181 unsigned int c = check_delta_limit(child, n + 1);
1184 child = child->delta_sibling;
1189 static void get_object_details(void)
1192 struct object_entry *entry;
1195 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1196 check_object(entry);
1198 if (nr_objects == nr_result) {
1200 * Depth of objects that depend on the entry -- this
1201 * is subtracted from depth-max to break too deep
1202 * delta chain because of delta data reusing.
1203 * However, we loosen this restriction when we know we
1204 * are creating a thin pack -- it will have to be
1205 * expanded on the other end anyway, so do not
1206 * artificially cut the delta chain and let it go as
1209 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1210 if (!entry->delta && entry->delta_child)
1211 entry->delta_limit =
1212 check_delta_limit(entry, 1);
1216 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1218 static entry_sort_t current_sort;
1220 static int sort_comparator(const void *_a, const void *_b)
1222 struct object_entry *a = *(struct object_entry **)_a;
1223 struct object_entry *b = *(struct object_entry **)_b;
1224 return current_sort(a,b);
1227 static struct object_entry **create_sorted_list(entry_sort_t sort)
1229 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1232 for (i = 0; i < nr_objects; i++)
1233 list[i] = objects + i;
1234 current_sort = sort;
1235 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1239 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1241 return hashcmp(a->sha1, b->sha1);
1244 static struct object_entry **create_final_object_list(void)
1246 struct object_entry **list;
1249 for (i = nr_result = 0; i < nr_objects; i++)
1250 if (!objects[i].preferred_base)
1252 list = xmalloc(nr_result * sizeof(struct object_entry *));
1253 for (i = j = 0; i < nr_objects; i++) {
1254 if (!objects[i].preferred_base)
1255 list[j++] = objects + i;
1257 current_sort = sha1_sort;
1258 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1262 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1264 if (a->type < b->type)
1266 if (a->type > b->type)
1268 if (a->hash < b->hash)
1270 if (a->hash > b->hash)
1272 if (a->preferred_base < b->preferred_base)
1274 if (a->preferred_base > b->preferred_base)
1276 if (a->size < b->size)
1278 if (a->size > b->size)
1280 return a < b ? -1 : (a > b);
1284 struct object_entry *entry;
1286 struct delta_index *index;
1290 * We search for deltas _backwards_ in a list sorted by type and
1291 * by size, so that we see progressively smaller and smaller files.
1292 * That's because we prefer deltas to be from the bigger file
1293 * to the smaller - deletes are potentially cheaper, but perhaps
1294 * more importantly, the bigger file is likely the more recent
1297 static int try_delta(struct unpacked *trg, struct unpacked *src,
1300 struct object_entry *trg_entry = trg->entry;
1301 struct object_entry *src_entry = src->entry;
1302 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1303 enum object_type type;
1306 /* Don't bother doing diffs between different types */
1307 if (trg_entry->type != src_entry->type)
1310 /* We do not compute delta to *create* objects we are not
1313 if (trg_entry->preferred_base)
1317 * We do not bother to try a delta that we discarded
1318 * on an earlier try, but only when reusing delta data.
1320 if (!no_reuse_delta && trg_entry->in_pack &&
1321 trg_entry->in_pack == src_entry->in_pack &&
1322 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1323 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1327 * If the current object is at pack edge, take the depth the
1328 * objects that depend on the current object into account --
1329 * otherwise they would become too deep.
1331 if (trg_entry->delta_child) {
1332 if (max_depth <= trg_entry->delta_limit)
1334 max_depth -= trg_entry->delta_limit;
1336 if (src_entry->depth >= max_depth)
1339 /* Now some size filtering heuristics. */
1340 trg_size = trg_entry->size;
1341 max_size = trg_size/2 - 20;
1342 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1345 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1346 max_size = trg_entry->delta_size-1;
1347 src_size = src_entry->size;
1348 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1349 if (sizediff >= max_size)
1352 /* Load data if not already done */
1354 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1356 die("object %s inconsistent object length (%lu vs %lu)",
1357 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1360 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1362 die("object %s inconsistent object length (%lu vs %lu)",
1363 sha1_to_hex(src_entry->sha1), sz, src_size);
1366 src->index = create_delta_index(src->data, src_size);
1368 die("out of memory");
1371 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1375 trg_entry->delta = src_entry;
1376 trg_entry->delta_size = delta_size;
1377 trg_entry->depth = src_entry->depth + 1;
1382 static void progress_interval(int signum)
1384 progress_update = 1;
1387 static void find_deltas(struct object_entry **list, int window, int depth)
1389 uint32_t i = nr_objects, idx = 0, processed = 0;
1390 unsigned int array_size = window * sizeof(struct unpacked);
1391 struct unpacked *array;
1392 unsigned last_percent = 999;
1396 array = xmalloc(array_size);
1397 memset(array, 0, array_size);
1399 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
1402 struct object_entry *entry = list[--i];
1403 struct unpacked *n = array + idx;
1406 if (!entry->preferred_base)
1410 unsigned percent = processed * 100 / nr_result;
1411 if (percent != last_percent || progress_update) {
1412 fprintf(stderr, "%4u%% (%u/%u) done\r",
1413 percent, processed, nr_result);
1414 progress_update = 0;
1415 last_percent = percent;
1420 /* This happens if we decided to reuse existing
1421 * delta from a pack. "!no_reuse_delta &&" is implied.
1425 if (entry->size < 50)
1427 free_delta_index(n->index);
1435 uint32_t other_idx = idx + j;
1437 if (other_idx >= window)
1438 other_idx -= window;
1439 m = array + other_idx;
1442 if (try_delta(n, m, depth) < 0)
1445 /* if we made n a delta, and if n is already at max
1446 * depth, leaving it in the window is pointless. we
1447 * should evict it first.
1449 if (entry->delta && depth <= entry->depth)
1458 fputc('\n', stderr);
1460 for (i = 0; i < window; ++i) {
1461 free_delta_index(array[i].index);
1462 free(array[i].data);
1467 static void prepare_pack(int window, int depth)
1469 get_object_details();
1470 sorted_by_type = create_sorted_list(type_size_sort);
1471 if (window && depth)
1472 find_deltas(sorted_by_type, window+1, depth);
1475 static int reuse_cached_pack(unsigned char *sha1)
1477 static const char cache[] = "pack-cache/pack-%s.%s";
1478 char *cached_pack, *cached_idx;
1479 int ifd, ofd, ifd_ix = -1;
1481 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1482 ifd = open(cached_pack, O_RDONLY);
1486 if (!pack_to_stdout) {
1487 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1488 ifd_ix = open(cached_idx, O_RDONLY);
1496 fprintf(stderr, "Reusing %u objects pack %s\n", nr_objects,
1499 if (pack_to_stdout) {
1500 if (copy_fd(ifd, 1))
1505 char name[PATH_MAX];
1506 snprintf(name, sizeof(name),
1507 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1508 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1510 die("unable to open %s (%s)", name, strerror(errno));
1511 if (copy_fd(ifd, ofd))
1515 snprintf(name, sizeof(name),
1516 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1517 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1519 die("unable to open %s (%s)", name, strerror(errno));
1520 if (copy_fd(ifd_ix, ofd))
1523 puts(sha1_to_hex(sha1));
1529 static void setup_progress_signal(void)
1531 struct sigaction sa;
1534 memset(&sa, 0, sizeof(sa));
1535 sa.sa_handler = progress_interval;
1536 sigemptyset(&sa.sa_mask);
1537 sa.sa_flags = SA_RESTART;
1538 sigaction(SIGALRM, &sa, NULL);
1540 v.it_interval.tv_sec = 1;
1541 v.it_interval.tv_usec = 0;
1542 v.it_value = v.it_interval;
1543 setitimer(ITIMER_REAL, &v, NULL);
1546 static int git_pack_config(const char *k, const char *v)
1548 if(!strcmp(k, "pack.window")) {
1549 window = git_config_int(k, v);
1552 return git_default_config(k, v);
1555 static void read_object_list_from_stdin(void)
1557 char line[40 + 1 + PATH_MAX + 2];
1558 unsigned char sha1[20];
1562 if (!fgets(line, sizeof(line), stdin)) {
1566 die("fgets returned NULL, not EOF, not error!");
1568 die("fgets: %s", strerror(errno));
1572 if (line[0] == '-') {
1573 if (get_sha1_hex(line+1, sha1))
1574 die("expected edge sha1, got garbage:\n %s",
1576 add_preferred_base(sha1);
1579 if (get_sha1_hex(line, sha1))
1580 die("expected sha1, got garbage:\n %s", line);
1582 hash = name_hash(line+41);
1583 add_preferred_base_object(line+41, hash);
1584 add_object_entry(sha1, hash, 0);
1588 static void show_commit(struct commit *commit)
1590 unsigned hash = name_hash("");
1591 add_preferred_base_object("", hash);
1592 add_object_entry(commit->object.sha1, hash, 0);
1595 static void show_object(struct object_array_entry *p)
1597 unsigned hash = name_hash(p->name);
1598 add_preferred_base_object(p->name, hash);
1599 add_object_entry(p->item->sha1, hash, 0);
1602 static void show_edge(struct commit *commit)
1604 add_preferred_base(commit->object.sha1);
1607 static void get_object_list(int ac, const char **av)
1609 struct rev_info revs;
1613 init_revisions(&revs, NULL);
1614 save_commit_buffer = 0;
1615 track_object_refs = 0;
1616 setup_revisions(ac, av, &revs, NULL);
1618 while (fgets(line, sizeof(line), stdin) != NULL) {
1619 int len = strlen(line);
1620 if (line[len - 1] == '\n')
1625 if (!strcmp(line, "--not")) {
1626 flags ^= UNINTERESTING;
1629 die("not a rev '%s'", line);
1631 if (handle_revision_arg(line, &revs, flags, 1))
1632 die("bad revision '%s'", line);
1635 prepare_revision_walk(&revs);
1636 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1637 traverse_commit_list(&revs, show_commit, show_object);
1640 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1644 struct object_entry **list;
1645 int use_internal_rev_list = 0;
1649 int rp_ac_alloc = 64;
1652 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1654 rp_av[0] = "pack-objects";
1655 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1658 git_config(git_pack_config);
1660 progress = isatty(2);
1661 for (i = 1; i < argc; i++) {
1662 const char *arg = argv[i];
1667 if (!strcmp("--non-empty", arg)) {
1671 if (!strcmp("--local", arg)) {
1675 if (!strcmp("--incremental", arg)) {
1679 if (!prefixcmp(arg, "--window=")) {
1681 window = strtoul(arg+9, &end, 0);
1682 if (!arg[9] || *end)
1686 if (!prefixcmp(arg, "--depth=")) {
1688 depth = strtoul(arg+8, &end, 0);
1689 if (!arg[8] || *end)
1693 if (!strcmp("--progress", arg)) {
1697 if (!strcmp("--all-progress", arg)) {
1701 if (!strcmp("-q", arg)) {
1705 if (!strcmp("--no-reuse-delta", arg)) {
1709 if (!strcmp("--delta-base-offset", arg)) {
1710 allow_ofs_delta = 1;
1713 if (!strcmp("--stdout", arg)) {
1717 if (!strcmp("--revs", arg)) {
1718 use_internal_rev_list = 1;
1721 if (!strcmp("--unpacked", arg) ||
1722 !prefixcmp(arg, "--unpacked=") ||
1723 !strcmp("--reflog", arg) ||
1724 !strcmp("--all", arg)) {
1725 use_internal_rev_list = 1;
1726 if (rp_ac >= rp_ac_alloc - 1) {
1727 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1728 rp_av = xrealloc(rp_av,
1729 rp_ac_alloc * sizeof(*rp_av));
1731 rp_av[rp_ac++] = arg;
1734 if (!strcmp("--thin", arg)) {
1735 use_internal_rev_list = 1;
1737 rp_av[1] = "--objects-edge";
1740 if (!prefixcmp(arg, "--index-version=")) {
1742 index_default_version = strtoul(arg + 16, &c, 10);
1743 if (index_default_version > 2)
1746 index_off32_limit = strtoul(c+1, &c, 0);
1747 if (*c || index_off32_limit & 0x80000000)
1754 /* Traditionally "pack-objects [options] base extra" failed;
1755 * we would however want to take refs parameter that would
1756 * have been given to upstream rev-list ourselves, which means
1757 * we somehow want to say what the base name is. So the
1760 * pack-objects [options] base <refs...>
1762 * in other words, we would treat the first non-option as the
1763 * base_name and send everything else to the internal revision
1767 if (!pack_to_stdout)
1768 base_name = argv[i++];
1770 if (pack_to_stdout != !base_name)
1773 if (!pack_to_stdout && thin)
1774 die("--thin cannot be used to build an indexable pack.");
1776 prepare_packed_git();
1779 fprintf(stderr, "Generating pack...\n");
1780 setup_progress_signal();
1783 if (!use_internal_rev_list)
1784 read_object_list_from_stdin();
1786 rp_av[rp_ac] = NULL;
1787 get_object_list(rp_ac, rp_av);
1791 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1792 sorted_by_sha = create_final_object_list();
1793 if (non_empty && !nr_result)
1797 list = sorted_by_sha;
1798 for (i = 0; i < nr_result; i++) {
1799 struct object_entry *entry = *list++;
1800 SHA1_Update(&ctx, entry->sha1, 20);
1802 SHA1_Final(object_list_sha1, &ctx);
1803 if (progress && (nr_objects != nr_result))
1804 fprintf(stderr, "Result has %u objects.\n", nr_result);
1806 if (reuse_cached_pack(object_list_sha1))
1809 off_t last_obj_offset;
1811 prepare_pack(window, depth);
1812 if (progress == 1 && pack_to_stdout) {
1813 /* the other end usually displays progress itself */
1814 struct itimerval v = {{0,},};
1815 setitimer(ITIMER_REAL, &v, NULL);
1816 signal(SIGALRM, SIG_IGN );
1817 progress_update = 0;
1819 last_obj_offset = write_pack_file();
1820 if (!pack_to_stdout) {
1821 write_index_file(last_obj_offset);
1822 puts(sha1_to_hex(object_list_sha1));
1826 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1827 written, written_delta, reused, reused_delta);