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;
173 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
174 for (i = 0; i < num_ent; i++) {
175 uint32_t hl = *((uint32_t *)(index + 24 * i));
176 rix->revindex[i].offset = ntohl(hl);
177 rix->revindex[i].nr = i;
179 /* This knows the pack format -- the 20-byte trailer
180 * follows immediately after the last object data.
182 rix->revindex[num_ent].offset = p->pack_size - 20;
183 rix->revindex[num_ent].nr = -1;
184 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
187 static struct revindex_entry * find_packed_object(struct packed_git *p,
192 struct pack_revindex *rix;
193 struct revindex_entry *revindex;
194 num = pack_revindex_ix(p);
196 die("internal error: pack revindex uninitialized");
197 rix = &pack_revindex[num];
199 prepare_pack_revindex(rix);
200 revindex = rix->revindex;
202 hi = p->num_objects + 1;
204 int mi = (lo + hi) / 2;
205 if (revindex[mi].offset == ofs) {
206 return revindex + mi;
208 else if (ofs < revindex[mi].offset)
213 die("internal error: pack revindex corrupt");
216 static off_t find_packed_object_size(struct packed_git *p, off_t ofs)
218 struct revindex_entry *entry = find_packed_object(p, ofs);
219 return entry[1].offset - ofs;
222 static const unsigned char *find_packed_object_name(struct packed_git *p,
225 struct revindex_entry *entry = find_packed_object(p, ofs);
226 return nth_packed_object_sha1(p, entry->nr);
229 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
231 unsigned long othersize, delta_size;
232 enum object_type type;
233 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
237 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
238 delta_buf = diff_delta(otherbuf, othersize,
239 buf, size, &delta_size, 0);
240 if (!delta_buf || delta_size != entry->delta_size)
241 die("delta size changed");
248 * The per-object header is a pretty dense thing, which is
249 * - first byte: low four bits are "size", then three bits of "type",
250 * and the high bit is "size continues".
251 * - each byte afterwards: low seven bits are size continuation,
252 * with the high bit being "size continues"
254 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
259 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
260 die("bad type %d", type);
262 c = (type << 4) | (size & 15);
275 * we are going to reuse the existing object data as is. make
276 * sure it is not corrupt.
278 static int check_pack_inflate(struct packed_git *p,
279 struct pack_window **w_curs,
282 unsigned long expect)
285 unsigned char fakebuf[4096], *in;
288 memset(&stream, 0, sizeof(stream));
289 inflateInit(&stream);
291 in = use_pack(p, w_curs, offset, &stream.avail_in);
293 stream.next_out = fakebuf;
294 stream.avail_out = sizeof(fakebuf);
295 st = inflate(&stream, Z_FINISH);
296 offset += stream.next_in - in;
297 } while (st == Z_OK || st == Z_BUF_ERROR);
299 return (st == Z_STREAM_END &&
300 stream.total_out == expect &&
301 stream.total_in == len) ? 0 : -1;
304 static void copy_pack_data(struct sha1file *f,
305 struct packed_git *p,
306 struct pack_window **w_curs,
314 in = use_pack(p, w_curs, offset, &avail);
316 avail = (unsigned int)len;
317 sha1write(f, in, avail);
323 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
326 unsigned char fakebuf[4096];
329 memset(&stream, 0, sizeof(stream));
330 stream.next_in = data;
331 stream.avail_in = len;
332 stream.next_out = fakebuf;
333 stream.avail_out = sizeof(fakebuf);
334 inflateInit(&stream);
337 st = inflate(&stream, Z_FINISH);
338 if (st == Z_STREAM_END || st == Z_OK) {
339 st = (stream.total_out == expect &&
340 stream.total_in == len) ? 0 : -1;
343 if (st != Z_BUF_ERROR) {
347 stream.next_out = fakebuf;
348 stream.avail_out = sizeof(fakebuf);
354 static int revalidate_loose_object(struct object_entry *entry,
356 unsigned long mapsize)
358 /* we already know this is a loose object with new type header. */
359 enum object_type type;
360 unsigned long size, used;
365 used = unpack_object_header_gently(map, mapsize, &type, &size);
370 return check_loose_inflate(map, mapsize, size);
373 static unsigned long write_object(struct sha1file *f,
374 struct object_entry *entry)
377 enum object_type type;
379 unsigned char header[10];
382 enum object_type obj_type;
388 obj_type = entry->type;
389 if (! entry->in_pack)
390 to_reuse = 0; /* can't reuse what we don't have */
391 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
392 to_reuse = 1; /* check_object() decided it for us */
393 else if (obj_type != entry->in_pack_type)
394 to_reuse = 0; /* pack has delta which is unusable */
395 else if (entry->delta)
396 to_reuse = 0; /* we want to pack afresh */
398 to_reuse = 1; /* we have it in-pack undeltified,
399 * and we do not need to deltify it.
402 if (!entry->in_pack && !entry->delta) {
404 unsigned long mapsize;
405 map = map_sha1_file(entry->sha1, &mapsize);
406 if (map && !legacy_loose_object(map)) {
407 /* We can copy straight into the pack file */
408 if (revalidate_loose_object(entry, map, mapsize))
409 die("corrupt loose object %s",
410 sha1_to_hex(entry->sha1));
411 sha1write(f, map, mapsize);
412 munmap(map, mapsize);
418 munmap(map, mapsize);
422 buf = read_sha1_file(entry->sha1, &type, &size);
424 die("unable to read %s", sha1_to_hex(entry->sha1));
425 if (size != entry->size)
426 die("object %s size inconsistency (%lu vs %lu)",
427 sha1_to_hex(entry->sha1), size, entry->size);
429 buf = delta_against(buf, size, entry);
430 size = entry->delta_size;
431 obj_type = (allow_ofs_delta && entry->delta->offset) ?
432 OBJ_OFS_DELTA : OBJ_REF_DELTA;
435 * The object header is a byte of 'type' followed by zero or
436 * more bytes of length.
438 hdrlen = encode_header(obj_type, size, header);
439 sha1write(f, header, hdrlen);
441 if (obj_type == OBJ_OFS_DELTA) {
443 * Deltas with relative base contain an additional
444 * encoding of the relative offset for the delta
445 * base from this object's position in the pack.
447 off_t ofs = entry->offset - entry->delta->offset;
448 unsigned pos = sizeof(header) - 1;
449 header[pos] = ofs & 127;
451 header[--pos] = 128 | (--ofs & 127);
452 sha1write(f, header + pos, sizeof(header) - pos);
453 hdrlen += sizeof(header) - pos;
454 } else if (obj_type == OBJ_REF_DELTA) {
456 * Deltas with a base reference contain
457 * an additional 20 bytes for the base sha1.
459 sha1write(f, entry->delta->sha1, 20);
462 datalen = sha1write_compressed(f, buf, size);
466 struct packed_git *p = entry->in_pack;
467 struct pack_window *w_curs = NULL;
471 obj_type = (allow_ofs_delta && entry->delta->offset) ?
472 OBJ_OFS_DELTA : OBJ_REF_DELTA;
475 hdrlen = encode_header(obj_type, entry->size, header);
476 sha1write(f, header, hdrlen);
477 if (obj_type == OBJ_OFS_DELTA) {
478 off_t ofs = entry->offset - entry->delta->offset;
479 unsigned pos = sizeof(header) - 1;
480 header[pos] = ofs & 127;
482 header[--pos] = 128 | (--ofs & 127);
483 sha1write(f, header + pos, sizeof(header) - pos);
484 hdrlen += sizeof(header) - pos;
485 } else if (obj_type == OBJ_REF_DELTA) {
486 sha1write(f, entry->delta->sha1, 20);
490 offset = entry->in_pack_offset + entry->in_pack_header_size;
491 datalen = find_packed_object_size(p, entry->in_pack_offset)
492 - entry->in_pack_header_size;
493 if (!pack_to_stdout && check_pack_inflate(p, &w_curs,
494 offset, datalen, entry->size))
495 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
496 copy_pack_data(f, p, &w_curs, offset, datalen);
504 entry->crc32 = crc32_end(f);
505 return hdrlen + datalen;
508 static off_t write_one(struct sha1file *f,
509 struct object_entry *e,
514 /* offset is non zero if object is written already. */
515 if (e->offset || e->preferred_base)
518 /* if we are deltified, write out base object first. */
520 offset = write_one(f, e->delta, offset);
523 size = write_object(f, e);
525 /* make sure off_t is sufficiently large not to wrap */
526 if (offset > offset + size)
527 die("pack too large for current definition of off_t");
528 return offset + size;
531 static void write_pack_file(void)
536 struct pack_header hdr;
537 unsigned last_percent = 999;
538 int do_progress = progress;
541 f = sha1fd(1, "<stdout>");
545 f = sha1create("%s-%s.%s", base_name,
546 sha1_to_hex(object_list_sha1), "pack");
548 fprintf(stderr, "Writing %u objects.\n", nr_result);
550 hdr.hdr_signature = htonl(PACK_SIGNATURE);
551 hdr.hdr_version = htonl(PACK_VERSION);
552 hdr.hdr_entries = htonl(nr_result);
553 sha1write(f, &hdr, sizeof(hdr));
554 offset = sizeof(hdr);
557 for (i = 0; i < nr_objects; i++) {
558 offset = write_one(f, objects + i, offset);
560 unsigned percent = written * 100 / nr_result;
561 if (progress_update || percent != last_percent) {
562 fprintf(stderr, "%4u%% (%u/%u) done\r",
563 percent, written, nr_result);
565 last_percent = percent;
572 if (written != nr_result)
573 die("wrote %u objects while expecting %u", written, nr_result);
574 sha1close(f, pack_file_sha1, 1);
577 static void write_index_file(void)
580 struct sha1file *f = sha1create("%s-%s.%s", base_name,
581 sha1_to_hex(object_list_sha1), "idx");
582 struct object_entry **list = sorted_by_sha;
583 struct object_entry **last = list + nr_result;
587 * Write the first-level table (the list is sorted,
588 * but we use a 256-entry lookup to be able to avoid
589 * having to do eight extra binary search iterations).
591 for (i = 0; i < 256; i++) {
592 struct object_entry **next = list;
593 while (next < last) {
594 struct object_entry *entry = *next;
595 if (entry->sha1[0] != i)
599 array[i] = htonl(next - sorted_by_sha);
602 sha1write(f, array, 256 * 4);
605 * Write the actual SHA1 entries..
607 list = sorted_by_sha;
608 for (i = 0; i < nr_result; i++) {
609 struct object_entry *entry = *list++;
610 uint32_t offset = htonl(entry->offset);
611 sha1write(f, &offset, 4);
612 sha1write(f, entry->sha1, 20);
614 sha1write(f, pack_file_sha1, 20);
615 sha1close(f, NULL, 1);
618 static int locate_object_entry_hash(const unsigned char *sha1)
622 memcpy(&ui, sha1, sizeof(unsigned int));
623 i = ui % object_ix_hashsz;
624 while (0 < object_ix[i]) {
625 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
627 if (++i == object_ix_hashsz)
633 static struct object_entry *locate_object_entry(const unsigned char *sha1)
637 if (!object_ix_hashsz)
640 i = locate_object_entry_hash(sha1);
642 return &objects[object_ix[i]-1];
646 static void rehash_objects(void)
649 struct object_entry *oe;
651 object_ix_hashsz = nr_objects * 3;
652 if (object_ix_hashsz < 1024)
653 object_ix_hashsz = 1024;
654 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
655 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
656 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
657 int ix = locate_object_entry_hash(oe->sha1);
661 object_ix[ix] = i + 1;
665 static unsigned name_hash(const char *name)
671 * This effectively just creates a sortable number from the
672 * last sixteen non-whitespace characters. Last characters
673 * count "most", so things that end in ".c" sort together.
675 while ((c = *name++) != 0) {
678 hash = (hash >> 2) + (c << 24);
683 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
685 uint32_t idx = nr_objects;
686 struct object_entry *entry;
687 struct packed_git *p;
688 off_t found_offset = 0;
689 struct packed_git *found_pack = NULL;
693 for (p = packed_git; p; p = p->next) {
694 off_t offset = find_pack_entry_one(sha1, p);
698 if (local && !p->pack_local)
701 found_offset = offset;
707 if ((entry = locate_object_entry(sha1)) != NULL)
710 if (idx >= nr_alloc) {
711 nr_alloc = (idx + 1024) * 3 / 2;
712 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
714 entry = objects + idx;
715 nr_objects = idx + 1;
716 memset(entry, 0, sizeof(*entry));
717 hashcpy(entry->sha1, sha1);
720 if (object_ix_hashsz * 3 <= nr_objects * 4)
723 ix = locate_object_entry_hash(entry->sha1);
725 die("internal error in object hashing.");
726 object_ix[-1 - ix] = idx + 1;
731 if (progress_update) {
732 fprintf(stderr, "Counting objects...%u\r", nr_objects);
736 entry->preferred_base = 1;
739 entry->in_pack = found_pack;
740 entry->in_pack_offset = found_offset;
746 struct pbase_tree_cache {
747 unsigned char sha1[20];
751 unsigned long tree_size;
754 static struct pbase_tree_cache *(pbase_tree_cache[256]);
755 static int pbase_tree_cache_ix(const unsigned char *sha1)
757 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
759 static int pbase_tree_cache_ix_incr(int ix)
761 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
764 static struct pbase_tree {
765 struct pbase_tree *next;
766 /* This is a phony "cache" entry; we are not
767 * going to evict it nor find it through _get()
768 * mechanism -- this is for the toplevel node that
769 * would almost always change with any commit.
771 struct pbase_tree_cache pcache;
774 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
776 struct pbase_tree_cache *ent, *nent;
779 enum object_type type;
781 int my_ix = pbase_tree_cache_ix(sha1);
782 int available_ix = -1;
784 /* pbase-tree-cache acts as a limited hashtable.
785 * your object will be found at your index or within a few
786 * slots after that slot if it is cached.
788 for (neigh = 0; neigh < 8; neigh++) {
789 ent = pbase_tree_cache[my_ix];
790 if (ent && !hashcmp(ent->sha1, sha1)) {
794 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
795 ((0 <= available_ix) &&
796 (!ent && pbase_tree_cache[available_ix])))
797 available_ix = my_ix;
800 my_ix = pbase_tree_cache_ix_incr(my_ix);
803 /* Did not find one. Either we got a bogus request or
804 * we need to read and perhaps cache.
806 data = read_sha1_file(sha1, &type, &size);
809 if (type != OBJ_TREE) {
814 /* We need to either cache or return a throwaway copy */
816 if (available_ix < 0)
819 ent = pbase_tree_cache[available_ix];
820 my_ix = available_ix;
824 nent = xmalloc(sizeof(*nent));
825 nent->temporary = (available_ix < 0);
828 /* evict and reuse */
829 free(ent->tree_data);
832 hashcpy(nent->sha1, sha1);
833 nent->tree_data = data;
834 nent->tree_size = size;
836 if (!nent->temporary)
837 pbase_tree_cache[my_ix] = nent;
841 static void pbase_tree_put(struct pbase_tree_cache *cache)
843 if (!cache->temporary) {
847 free(cache->tree_data);
851 static int name_cmp_len(const char *name)
854 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
859 static void add_pbase_object(struct tree_desc *tree,
862 const char *fullname)
864 struct name_entry entry;
866 while (tree_entry(tree,&entry)) {
868 enum object_type type;
870 if (tree_entry_len(entry.path, entry.sha1) != cmplen ||
871 memcmp(entry.path, name, cmplen) ||
872 !has_sha1_file(entry.sha1) ||
873 (type = sha1_object_info(entry.sha1, &size)) < 0)
875 if (name[cmplen] != '/') {
876 unsigned hash = name_hash(fullname);
877 add_object_entry(entry.sha1, hash, 1);
880 if (type == OBJ_TREE) {
881 struct tree_desc sub;
882 struct pbase_tree_cache *tree;
883 const char *down = name+cmplen+1;
884 int downlen = name_cmp_len(down);
886 tree = pbase_tree_get(entry.sha1);
889 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
891 add_pbase_object(&sub, down, downlen, fullname);
892 pbase_tree_put(tree);
897 static unsigned *done_pbase_paths;
898 static int done_pbase_paths_num;
899 static int done_pbase_paths_alloc;
900 static int done_pbase_path_pos(unsigned hash)
903 int hi = done_pbase_paths_num;
905 int mi = (hi + lo) / 2;
906 if (done_pbase_paths[mi] == hash)
908 if (done_pbase_paths[mi] < hash)
916 static int check_pbase_path(unsigned hash)
918 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
922 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
923 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
924 done_pbase_paths = xrealloc(done_pbase_paths,
925 done_pbase_paths_alloc *
928 done_pbase_paths_num++;
929 if (pos < done_pbase_paths_num)
930 memmove(done_pbase_paths + pos + 1,
931 done_pbase_paths + pos,
932 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
933 done_pbase_paths[pos] = hash;
937 static void add_preferred_base_object(const char *name, unsigned hash)
939 struct pbase_tree *it;
940 int cmplen = name_cmp_len(name);
942 if (check_pbase_path(hash))
945 for (it = pbase_tree; it; it = it->next) {
947 hash = name_hash("");
948 add_object_entry(it->pcache.sha1, hash, 1);
951 struct tree_desc tree;
952 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
953 add_pbase_object(&tree, name, cmplen, name);
958 static void add_preferred_base(unsigned char *sha1)
960 struct pbase_tree *it;
963 unsigned char tree_sha1[20];
965 if (window <= num_preferred_base++)
968 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
972 for (it = pbase_tree; it; it = it->next) {
973 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
979 it = xcalloc(1, sizeof(*it));
980 it->next = pbase_tree;
983 hashcpy(it->pcache.sha1, tree_sha1);
984 it->pcache.tree_data = data;
985 it->pcache.tree_size = size;
988 static void check_object(struct object_entry *entry)
990 if (entry->in_pack && !entry->preferred_base) {
991 struct packed_git *p = entry->in_pack;
992 struct pack_window *w_curs = NULL;
993 unsigned long size, used;
996 struct object_entry *base_entry = NULL;
998 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1000 /* We want in_pack_type even if we do not reuse delta.
1001 * There is no point not reusing non-delta representations.
1003 used = unpack_object_header_gently(buf, avail,
1004 &entry->in_pack_type, &size);
1006 /* Check if it is delta, and the base is also an object
1007 * we are going to pack. If so we will reuse the existing
1010 if (!no_reuse_delta) {
1012 const unsigned char *base_name;
1014 unsigned long used_0;
1015 /* there is at least 20 bytes left in the pack */
1016 switch (entry->in_pack_type) {
1018 base_name = use_pack(p, &w_curs,
1019 entry->in_pack_offset + used, NULL);
1023 buf = use_pack(p, &w_curs,
1024 entry->in_pack_offset + used, NULL);
1030 if (!ofs || MSB(ofs, 7))
1031 die("delta base offset overflow in pack for %s",
1032 sha1_to_hex(entry->sha1));
1034 ofs = (ofs << 7) + (c & 127);
1036 if (ofs >= entry->in_pack_offset)
1037 die("delta base offset out of bound for %s",
1038 sha1_to_hex(entry->sha1));
1039 ofs = entry->in_pack_offset - ofs;
1040 base_name = find_packed_object_name(p, ofs);
1047 base_entry = locate_object_entry(base_name);
1049 unuse_pack(&w_curs);
1050 entry->in_pack_header_size = used;
1054 /* Depth value does not matter - find_deltas()
1055 * will never consider reused delta as the
1056 * base object to deltify other objects
1057 * against, in order to avoid circular deltas.
1060 /* uncompressed size of the delta data */
1062 entry->delta = base_entry;
1063 entry->type = entry->in_pack_type;
1065 entry->delta_sibling = base_entry->delta_child;
1066 base_entry->delta_child = entry;
1070 /* Otherwise we would do the usual */
1073 entry->type = sha1_object_info(entry->sha1, &entry->size);
1074 if (entry->type < 0)
1075 die("unable to get type of object %s",
1076 sha1_to_hex(entry->sha1));
1079 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1081 struct object_entry *child = me->delta_child;
1084 unsigned int c = check_delta_limit(child, n + 1);
1087 child = child->delta_sibling;
1092 static void get_object_details(void)
1095 struct object_entry *entry;
1098 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1099 check_object(entry);
1101 if (nr_objects == nr_result) {
1103 * Depth of objects that depend on the entry -- this
1104 * is subtracted from depth-max to break too deep
1105 * delta chain because of delta data reusing.
1106 * However, we loosen this restriction when we know we
1107 * are creating a thin pack -- it will have to be
1108 * expanded on the other end anyway, so do not
1109 * artificially cut the delta chain and let it go as
1112 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1113 if (!entry->delta && entry->delta_child)
1114 entry->delta_limit =
1115 check_delta_limit(entry, 1);
1119 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1121 static entry_sort_t current_sort;
1123 static int sort_comparator(const void *_a, const void *_b)
1125 struct object_entry *a = *(struct object_entry **)_a;
1126 struct object_entry *b = *(struct object_entry **)_b;
1127 return current_sort(a,b);
1130 static struct object_entry **create_sorted_list(entry_sort_t sort)
1132 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1135 for (i = 0; i < nr_objects; i++)
1136 list[i] = objects + i;
1137 current_sort = sort;
1138 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1142 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1144 return hashcmp(a->sha1, b->sha1);
1147 static struct object_entry **create_final_object_list(void)
1149 struct object_entry **list;
1152 for (i = nr_result = 0; i < nr_objects; i++)
1153 if (!objects[i].preferred_base)
1155 list = xmalloc(nr_result * sizeof(struct object_entry *));
1156 for (i = j = 0; i < nr_objects; i++) {
1157 if (!objects[i].preferred_base)
1158 list[j++] = objects + i;
1160 current_sort = sha1_sort;
1161 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1165 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1167 if (a->type < b->type)
1169 if (a->type > b->type)
1171 if (a->hash < b->hash)
1173 if (a->hash > b->hash)
1175 if (a->preferred_base < b->preferred_base)
1177 if (a->preferred_base > b->preferred_base)
1179 if (a->size < b->size)
1181 if (a->size > b->size)
1183 return a < b ? -1 : (a > b);
1187 struct object_entry *entry;
1189 struct delta_index *index;
1193 * We search for deltas _backwards_ in a list sorted by type and
1194 * by size, so that we see progressively smaller and smaller files.
1195 * That's because we prefer deltas to be from the bigger file
1196 * to the smaller - deletes are potentially cheaper, but perhaps
1197 * more importantly, the bigger file is likely the more recent
1200 static int try_delta(struct unpacked *trg, struct unpacked *src,
1203 struct object_entry *trg_entry = trg->entry;
1204 struct object_entry *src_entry = src->entry;
1205 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1206 enum object_type type;
1209 /* Don't bother doing diffs between different types */
1210 if (trg_entry->type != src_entry->type)
1213 /* We do not compute delta to *create* objects we are not
1216 if (trg_entry->preferred_base)
1220 * We do not bother to try a delta that we discarded
1221 * on an earlier try, but only when reusing delta data.
1223 if (!no_reuse_delta && trg_entry->in_pack &&
1224 trg_entry->in_pack == src_entry->in_pack &&
1225 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1226 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1230 * If the current object is at pack edge, take the depth the
1231 * objects that depend on the current object into account --
1232 * otherwise they would become too deep.
1234 if (trg_entry->delta_child) {
1235 if (max_depth <= trg_entry->delta_limit)
1237 max_depth -= trg_entry->delta_limit;
1239 if (src_entry->depth >= max_depth)
1242 /* Now some size filtering heuristics. */
1243 trg_size = trg_entry->size;
1244 max_size = trg_size/2 - 20;
1245 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1248 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1249 max_size = trg_entry->delta_size-1;
1250 src_size = src_entry->size;
1251 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1252 if (sizediff >= max_size)
1255 /* Load data if not already done */
1257 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1259 die("object %s inconsistent object length (%lu vs %lu)",
1260 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1263 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1265 die("object %s inconsistent object length (%lu vs %lu)",
1266 sha1_to_hex(src_entry->sha1), sz, src_size);
1269 src->index = create_delta_index(src->data, src_size);
1271 die("out of memory");
1274 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1278 trg_entry->delta = src_entry;
1279 trg_entry->delta_size = delta_size;
1280 trg_entry->depth = src_entry->depth + 1;
1285 static void progress_interval(int signum)
1287 progress_update = 1;
1290 static void find_deltas(struct object_entry **list, int window, int depth)
1292 uint32_t i = nr_objects, idx = 0, processed = 0;
1293 unsigned int array_size = window * sizeof(struct unpacked);
1294 struct unpacked *array;
1295 unsigned last_percent = 999;
1299 array = xmalloc(array_size);
1300 memset(array, 0, array_size);
1302 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
1305 struct object_entry *entry = list[--i];
1306 struct unpacked *n = array + idx;
1309 if (!entry->preferred_base)
1313 unsigned percent = processed * 100 / nr_result;
1314 if (percent != last_percent || progress_update) {
1315 fprintf(stderr, "%4u%% (%u/%u) done\r",
1316 percent, processed, nr_result);
1317 progress_update = 0;
1318 last_percent = percent;
1323 /* This happens if we decided to reuse existing
1324 * delta from a pack. "!no_reuse_delta &&" is implied.
1328 if (entry->size < 50)
1330 free_delta_index(n->index);
1338 uint32_t other_idx = idx + j;
1340 if (other_idx >= window)
1341 other_idx -= window;
1342 m = array + other_idx;
1345 if (try_delta(n, m, depth) < 0)
1348 /* if we made n a delta, and if n is already at max
1349 * depth, leaving it in the window is pointless. we
1350 * should evict it first.
1352 if (entry->delta && depth <= entry->depth)
1361 fputc('\n', stderr);
1363 for (i = 0; i < window; ++i) {
1364 free_delta_index(array[i].index);
1365 free(array[i].data);
1370 static void prepare_pack(int window, int depth)
1372 get_object_details();
1373 sorted_by_type = create_sorted_list(type_size_sort);
1374 if (window && depth)
1375 find_deltas(sorted_by_type, window+1, depth);
1378 static int reuse_cached_pack(unsigned char *sha1)
1380 static const char cache[] = "pack-cache/pack-%s.%s";
1381 char *cached_pack, *cached_idx;
1382 int ifd, ofd, ifd_ix = -1;
1384 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1385 ifd = open(cached_pack, O_RDONLY);
1389 if (!pack_to_stdout) {
1390 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1391 ifd_ix = open(cached_idx, O_RDONLY);
1399 fprintf(stderr, "Reusing %u objects pack %s\n", nr_objects,
1402 if (pack_to_stdout) {
1403 if (copy_fd(ifd, 1))
1408 char name[PATH_MAX];
1409 snprintf(name, sizeof(name),
1410 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1411 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1413 die("unable to open %s (%s)", name, strerror(errno));
1414 if (copy_fd(ifd, ofd))
1418 snprintf(name, sizeof(name),
1419 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1420 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1422 die("unable to open %s (%s)", name, strerror(errno));
1423 if (copy_fd(ifd_ix, ofd))
1426 puts(sha1_to_hex(sha1));
1432 static void setup_progress_signal(void)
1434 struct sigaction sa;
1437 memset(&sa, 0, sizeof(sa));
1438 sa.sa_handler = progress_interval;
1439 sigemptyset(&sa.sa_mask);
1440 sa.sa_flags = SA_RESTART;
1441 sigaction(SIGALRM, &sa, NULL);
1443 v.it_interval.tv_sec = 1;
1444 v.it_interval.tv_usec = 0;
1445 v.it_value = v.it_interval;
1446 setitimer(ITIMER_REAL, &v, NULL);
1449 static int git_pack_config(const char *k, const char *v)
1451 if(!strcmp(k, "pack.window")) {
1452 window = git_config_int(k, v);
1455 return git_default_config(k, v);
1458 static void read_object_list_from_stdin(void)
1460 char line[40 + 1 + PATH_MAX + 2];
1461 unsigned char sha1[20];
1465 if (!fgets(line, sizeof(line), stdin)) {
1469 die("fgets returned NULL, not EOF, not error!");
1471 die("fgets: %s", strerror(errno));
1475 if (line[0] == '-') {
1476 if (get_sha1_hex(line+1, sha1))
1477 die("expected edge sha1, got garbage:\n %s",
1479 add_preferred_base(sha1);
1482 if (get_sha1_hex(line, sha1))
1483 die("expected sha1, got garbage:\n %s", line);
1485 hash = name_hash(line+41);
1486 add_preferred_base_object(line+41, hash);
1487 add_object_entry(sha1, hash, 0);
1491 static void show_commit(struct commit *commit)
1493 unsigned hash = name_hash("");
1494 add_preferred_base_object("", hash);
1495 add_object_entry(commit->object.sha1, hash, 0);
1498 static void show_object(struct object_array_entry *p)
1500 unsigned hash = name_hash(p->name);
1501 add_preferred_base_object(p->name, hash);
1502 add_object_entry(p->item->sha1, hash, 0);
1505 static void show_edge(struct commit *commit)
1507 add_preferred_base(commit->object.sha1);
1510 static void get_object_list(int ac, const char **av)
1512 struct rev_info revs;
1516 init_revisions(&revs, NULL);
1517 save_commit_buffer = 0;
1518 track_object_refs = 0;
1519 setup_revisions(ac, av, &revs, NULL);
1521 while (fgets(line, sizeof(line), stdin) != NULL) {
1522 int len = strlen(line);
1523 if (line[len - 1] == '\n')
1528 if (!strcmp(line, "--not")) {
1529 flags ^= UNINTERESTING;
1532 die("not a rev '%s'", line);
1534 if (handle_revision_arg(line, &revs, flags, 1))
1535 die("bad revision '%s'", line);
1538 prepare_revision_walk(&revs);
1539 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1540 traverse_commit_list(&revs, show_commit, show_object);
1543 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1547 struct object_entry **list;
1548 int use_internal_rev_list = 0;
1552 int rp_ac_alloc = 64;
1555 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1557 rp_av[0] = "pack-objects";
1558 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1561 git_config(git_pack_config);
1563 progress = isatty(2);
1564 for (i = 1; i < argc; i++) {
1565 const char *arg = argv[i];
1570 if (!strcmp("--non-empty", arg)) {
1574 if (!strcmp("--local", arg)) {
1578 if (!strcmp("--incremental", arg)) {
1582 if (!prefixcmp(arg, "--window=")) {
1584 window = strtoul(arg+9, &end, 0);
1585 if (!arg[9] || *end)
1589 if (!prefixcmp(arg, "--depth=")) {
1591 depth = strtoul(arg+8, &end, 0);
1592 if (!arg[8] || *end)
1596 if (!strcmp("--progress", arg)) {
1600 if (!strcmp("--all-progress", arg)) {
1604 if (!strcmp("-q", arg)) {
1608 if (!strcmp("--no-reuse-delta", arg)) {
1612 if (!strcmp("--delta-base-offset", arg)) {
1613 allow_ofs_delta = 1;
1616 if (!strcmp("--stdout", arg)) {
1620 if (!strcmp("--revs", arg)) {
1621 use_internal_rev_list = 1;
1624 if (!strcmp("--unpacked", arg) ||
1625 !prefixcmp(arg, "--unpacked=") ||
1626 !strcmp("--reflog", arg) ||
1627 !strcmp("--all", arg)) {
1628 use_internal_rev_list = 1;
1629 if (rp_ac >= rp_ac_alloc - 1) {
1630 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1631 rp_av = xrealloc(rp_av,
1632 rp_ac_alloc * sizeof(*rp_av));
1634 rp_av[rp_ac++] = arg;
1637 if (!strcmp("--thin", arg)) {
1638 use_internal_rev_list = 1;
1640 rp_av[1] = "--objects-edge";
1646 /* Traditionally "pack-objects [options] base extra" failed;
1647 * we would however want to take refs parameter that would
1648 * have been given to upstream rev-list ourselves, which means
1649 * we somehow want to say what the base name is. So the
1652 * pack-objects [options] base <refs...>
1654 * in other words, we would treat the first non-option as the
1655 * base_name and send everything else to the internal revision
1659 if (!pack_to_stdout)
1660 base_name = argv[i++];
1662 if (pack_to_stdout != !base_name)
1665 if (!pack_to_stdout && thin)
1666 die("--thin cannot be used to build an indexable pack.");
1668 prepare_packed_git();
1671 fprintf(stderr, "Generating pack...\n");
1672 setup_progress_signal();
1675 if (!use_internal_rev_list)
1676 read_object_list_from_stdin();
1678 rp_av[rp_ac] = NULL;
1679 get_object_list(rp_ac, rp_av);
1683 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1684 sorted_by_sha = create_final_object_list();
1685 if (non_empty && !nr_result)
1689 list = sorted_by_sha;
1690 for (i = 0; i < nr_result; i++) {
1691 struct object_entry *entry = *list++;
1692 SHA1_Update(&ctx, entry->sha1, 20);
1694 SHA1_Final(object_list_sha1, &ctx);
1695 if (progress && (nr_objects != nr_result))
1696 fprintf(stderr, "Result has %u objects.\n", nr_result);
1698 if (reuse_cached_pack(object_list_sha1))
1702 prepare_pack(window, depth);
1703 if (progress == 1 && pack_to_stdout) {
1704 /* the other end usually displays progress itself */
1705 struct itimerval v = {{0,},};
1706 setitimer(ITIMER_REAL, &v, NULL);
1707 signal(SIGALRM, SIG_IGN );
1708 progress_update = 0;
1711 if (!pack_to_stdout) {
1713 puts(sha1_to_hex(object_list_sha1));
1717 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1718 written, written_delta, reused, reused_delta);