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 unsigned long write_object(struct sha1file *f,
352 struct object_entry *entry)
355 enum object_type type;
357 unsigned char header[10];
360 enum object_type obj_type;
366 obj_type = entry->type;
368 to_reuse = 0; /* explicit */
369 else if (!entry->in_pack)
370 to_reuse = 0; /* can't reuse what we don't have */
371 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
372 to_reuse = 1; /* check_object() decided it for us */
373 else if (obj_type != entry->in_pack_type)
374 to_reuse = 0; /* pack has delta which is unusable */
375 else if (entry->delta)
376 to_reuse = 0; /* we want to pack afresh */
378 to_reuse = 1; /* we have it in-pack undeltified,
379 * and we do not need to deltify it.
383 buf = read_sha1_file(entry->sha1, &type, &size);
385 die("unable to read %s", sha1_to_hex(entry->sha1));
386 if (size != entry->size)
387 die("object %s size inconsistency (%lu vs %lu)",
388 sha1_to_hex(entry->sha1), size, entry->size);
390 buf = delta_against(buf, size, entry);
391 size = entry->delta_size;
392 obj_type = (allow_ofs_delta && entry->delta->offset) ?
393 OBJ_OFS_DELTA : OBJ_REF_DELTA;
396 * The object header is a byte of 'type' followed by zero or
397 * more bytes of length.
399 hdrlen = encode_header(obj_type, size, header);
400 sha1write(f, header, hdrlen);
402 if (obj_type == OBJ_OFS_DELTA) {
404 * Deltas with relative base contain an additional
405 * encoding of the relative offset for the delta
406 * base from this object's position in the pack.
408 off_t ofs = entry->offset - entry->delta->offset;
409 unsigned pos = sizeof(header) - 1;
410 header[pos] = ofs & 127;
412 header[--pos] = 128 | (--ofs & 127);
413 sha1write(f, header + pos, sizeof(header) - pos);
414 hdrlen += sizeof(header) - pos;
415 } else if (obj_type == OBJ_REF_DELTA) {
417 * Deltas with a base reference contain
418 * an additional 20 bytes for the base sha1.
420 sha1write(f, entry->delta->sha1, 20);
423 datalen = sha1write_compressed(f, buf, size, pack_compression_level);
427 struct packed_git *p = entry->in_pack;
428 struct pack_window *w_curs = NULL;
429 struct revindex_entry *revidx;
433 obj_type = (allow_ofs_delta && entry->delta->offset) ?
434 OBJ_OFS_DELTA : OBJ_REF_DELTA;
437 hdrlen = encode_header(obj_type, entry->size, header);
438 sha1write(f, header, hdrlen);
439 if (obj_type == OBJ_OFS_DELTA) {
440 off_t ofs = entry->offset - entry->delta->offset;
441 unsigned pos = sizeof(header) - 1;
442 header[pos] = ofs & 127;
444 header[--pos] = 128 | (--ofs & 127);
445 sha1write(f, header + pos, sizeof(header) - pos);
446 hdrlen += sizeof(header) - pos;
447 } else if (obj_type == OBJ_REF_DELTA) {
448 sha1write(f, entry->delta->sha1, 20);
452 offset = entry->in_pack_offset;
453 revidx = find_packed_object(p, offset);
454 datalen = revidx[1].offset - offset;
455 if (!pack_to_stdout && p->index_version > 1 &&
456 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
457 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
458 offset += entry->in_pack_header_size;
459 datalen -= entry->in_pack_header_size;
460 if (!pack_to_stdout && p->index_version == 1 &&
461 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
462 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
463 copy_pack_data(f, p, &w_curs, offset, datalen);
471 entry->crc32 = crc32_end(f);
472 return hdrlen + datalen;
475 static off_t write_one(struct sha1file *f,
476 struct object_entry *e,
481 /* offset is non zero if object is written already. */
482 if (e->offset || e->preferred_base)
485 /* if we are deltified, write out base object first. */
487 offset = write_one(f, e->delta, offset);
490 size = write_object(f, e);
492 /* make sure off_t is sufficiently large not to wrap */
493 if (offset > offset + size)
494 die("pack too large for current definition of off_t");
495 return offset + size;
498 static int open_object_dir_tmp(const char *path)
500 snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
501 return mkstemp(tmpname);
504 static off_t write_pack_file(void)
508 off_t offset, last_obj_offset = 0;
509 struct pack_header hdr;
510 int do_progress = progress;
512 if (pack_to_stdout) {
513 f = sha1fd(1, "<stdout>");
516 int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
518 die("unable to create %s: %s\n", tmpname, strerror(errno));
519 pack_tmp_name = xstrdup(tmpname);
520 f = sha1fd(fd, pack_tmp_name);
524 start_progress(&progress_state, "Writing %u objects...", "", nr_result);
526 hdr.hdr_signature = htonl(PACK_SIGNATURE);
527 hdr.hdr_version = htonl(PACK_VERSION);
528 hdr.hdr_entries = htonl(nr_result);
529 sha1write(f, &hdr, sizeof(hdr));
530 offset = sizeof(hdr);
533 for (i = 0; i < nr_objects; i++) {
534 last_obj_offset = offset;
535 offset = write_one(f, objects + i, offset);
537 display_progress(&progress_state, written);
540 stop_progress(&progress_state);
542 if (written != nr_result)
543 die("wrote %u objects while expecting %u", written, nr_result);
544 sha1close(f, pack_file_sha1, 1);
546 return last_obj_offset;
549 static int sha1_sort(const void *_a, const void *_b)
551 const struct object_entry *a = *(struct object_entry **)_a;
552 const struct object_entry *b = *(struct object_entry **)_b;
553 return hashcmp(a->sha1, b->sha1);
556 static uint32_t index_default_version = 1;
557 static uint32_t index_off32_limit = 0x7fffffff;
559 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
562 struct object_entry **sorted_by_sha, **list, **last;
564 uint32_t i, index_version;
567 int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
569 die("unable to create %s: %s\n", tmpname, strerror(errno));
570 idx_tmp_name = xstrdup(tmpname);
571 f = sha1fd(fd, idx_tmp_name);
576 xcalloc(nr_result, sizeof(struct object_entry *));
577 for (i = 0; i < nr_objects; i++)
578 if (!objects[i].preferred_base)
579 sorted_by_sha[j++] = objects + i;
581 die("listed %u objects while expecting %u", j, nr_result);
582 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
583 list = sorted_by_sha;
584 last = sorted_by_sha + nr_result;
586 sorted_by_sha = list = last = NULL;
588 /* if last object's offset is >= 2^31 we should use index V2 */
589 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
591 /* index versions 2 and above need a header */
592 if (index_version >= 2) {
593 struct pack_idx_header hdr;
594 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
595 hdr.idx_version = htonl(index_version);
596 sha1write(f, &hdr, sizeof(hdr));
600 * Write the first-level table (the list is sorted,
601 * but we use a 256-entry lookup to be able to avoid
602 * having to do eight extra binary search iterations).
604 for (i = 0; i < 256; i++) {
605 struct object_entry **next = list;
606 while (next < last) {
607 struct object_entry *entry = *next;
608 if (entry->sha1[0] != i)
612 array[i] = htonl(next - sorted_by_sha);
615 sha1write(f, array, 256 * 4);
617 /* Compute the SHA1 hash of sorted object names. */
620 /* Write the actual SHA1 entries. */
621 list = sorted_by_sha;
622 for (i = 0; i < nr_result; i++) {
623 struct object_entry *entry = *list++;
624 if (index_version < 2) {
625 uint32_t offset = htonl(entry->offset);
626 sha1write(f, &offset, 4);
628 sha1write(f, entry->sha1, 20);
629 SHA1_Update(&ctx, entry->sha1, 20);
632 if (index_version >= 2) {
633 unsigned int nr_large_offset = 0;
635 /* write the crc32 table */
636 list = sorted_by_sha;
637 for (i = 0; i < nr_objects; i++) {
638 struct object_entry *entry = *list++;
639 uint32_t crc32_val = htonl(entry->crc32);
640 sha1write(f, &crc32_val, 4);
643 /* write the 32-bit offset table */
644 list = sorted_by_sha;
645 for (i = 0; i < nr_objects; i++) {
646 struct object_entry *entry = *list++;
647 uint32_t offset = (entry->offset <= index_off32_limit) ?
648 entry->offset : (0x80000000 | nr_large_offset++);
649 offset = htonl(offset);
650 sha1write(f, &offset, 4);
653 /* write the large offset table */
654 list = sorted_by_sha;
655 while (nr_large_offset) {
656 struct object_entry *entry = *list++;
657 uint64_t offset = entry->offset;
658 if (offset > index_off32_limit) {
660 split[0] = htonl(offset >> 32);
661 split[1] = htonl(offset & 0xffffffff);
662 sha1write(f, split, 8);
668 sha1write(f, pack_file_sha1, 20);
669 sha1close(f, NULL, 1);
671 SHA1_Final(sha1, &ctx);
674 static int locate_object_entry_hash(const unsigned char *sha1)
678 memcpy(&ui, sha1, sizeof(unsigned int));
679 i = ui % object_ix_hashsz;
680 while (0 < object_ix[i]) {
681 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
683 if (++i == object_ix_hashsz)
689 static struct object_entry *locate_object_entry(const unsigned char *sha1)
693 if (!object_ix_hashsz)
696 i = locate_object_entry_hash(sha1);
698 return &objects[object_ix[i]-1];
702 static void rehash_objects(void)
705 struct object_entry *oe;
707 object_ix_hashsz = nr_objects * 3;
708 if (object_ix_hashsz < 1024)
709 object_ix_hashsz = 1024;
710 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
711 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
712 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
713 int ix = locate_object_entry_hash(oe->sha1);
717 object_ix[ix] = i + 1;
721 static unsigned name_hash(const char *name)
730 * This effectively just creates a sortable number from the
731 * last sixteen non-whitespace characters. Last characters
732 * count "most", so things that end in ".c" sort together.
734 while ((c = *name++) != 0) {
737 hash = (hash >> 2) + (c << 24);
742 static int add_object_entry(const unsigned char *sha1, enum object_type type,
743 const char *name, int exclude)
745 struct object_entry *entry;
746 struct packed_git *p, *found_pack = NULL;
747 off_t found_offset = 0;
749 unsigned hash = name_hash(name);
751 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
754 entry = objects + object_ix[ix] - 1;
755 if (!entry->preferred_base)
757 entry->preferred_base = 1;
762 for (p = packed_git; p; p = p->next) {
763 off_t offset = find_pack_entry_one(sha1, p);
766 found_offset = offset;
773 if (local && !p->pack_local)
778 if (nr_objects >= nr_alloc) {
779 nr_alloc = (nr_alloc + 1024) * 3 / 2;
780 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
783 entry = objects + nr_objects++;
784 memset(entry, 0, sizeof(*entry));
785 hashcpy(entry->sha1, sha1);
790 entry->preferred_base = 1;
794 entry->in_pack = found_pack;
795 entry->in_pack_offset = found_offset;
798 if (object_ix_hashsz * 3 <= nr_objects * 4)
801 object_ix[-1 - ix] = nr_objects;
804 display_progress(&progress_state, nr_objects);
809 struct pbase_tree_cache {
810 unsigned char sha1[20];
814 unsigned long tree_size;
817 static struct pbase_tree_cache *(pbase_tree_cache[256]);
818 static int pbase_tree_cache_ix(const unsigned char *sha1)
820 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
822 static int pbase_tree_cache_ix_incr(int ix)
824 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
827 static struct pbase_tree {
828 struct pbase_tree *next;
829 /* This is a phony "cache" entry; we are not
830 * going to evict it nor find it through _get()
831 * mechanism -- this is for the toplevel node that
832 * would almost always change with any commit.
834 struct pbase_tree_cache pcache;
837 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
839 struct pbase_tree_cache *ent, *nent;
842 enum object_type type;
844 int my_ix = pbase_tree_cache_ix(sha1);
845 int available_ix = -1;
847 /* pbase-tree-cache acts as a limited hashtable.
848 * your object will be found at your index or within a few
849 * slots after that slot if it is cached.
851 for (neigh = 0; neigh < 8; neigh++) {
852 ent = pbase_tree_cache[my_ix];
853 if (ent && !hashcmp(ent->sha1, sha1)) {
857 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
858 ((0 <= available_ix) &&
859 (!ent && pbase_tree_cache[available_ix])))
860 available_ix = my_ix;
863 my_ix = pbase_tree_cache_ix_incr(my_ix);
866 /* Did not find one. Either we got a bogus request or
867 * we need to read and perhaps cache.
869 data = read_sha1_file(sha1, &type, &size);
872 if (type != OBJ_TREE) {
877 /* We need to either cache or return a throwaway copy */
879 if (available_ix < 0)
882 ent = pbase_tree_cache[available_ix];
883 my_ix = available_ix;
887 nent = xmalloc(sizeof(*nent));
888 nent->temporary = (available_ix < 0);
891 /* evict and reuse */
892 free(ent->tree_data);
895 hashcpy(nent->sha1, sha1);
896 nent->tree_data = data;
897 nent->tree_size = size;
899 if (!nent->temporary)
900 pbase_tree_cache[my_ix] = nent;
904 static void pbase_tree_put(struct pbase_tree_cache *cache)
906 if (!cache->temporary) {
910 free(cache->tree_data);
914 static int name_cmp_len(const char *name)
917 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
922 static void add_pbase_object(struct tree_desc *tree,
925 const char *fullname)
927 struct name_entry entry;
930 while (tree_entry(tree,&entry)) {
931 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
932 memcmp(name, entry.path, cmplen);
937 if (name[cmplen] != '/') {
938 add_object_entry(entry.sha1,
939 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
943 if (S_ISDIR(entry.mode)) {
944 struct tree_desc sub;
945 struct pbase_tree_cache *tree;
946 const char *down = name+cmplen+1;
947 int downlen = name_cmp_len(down);
949 tree = pbase_tree_get(entry.sha1);
952 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
954 add_pbase_object(&sub, down, downlen, fullname);
955 pbase_tree_put(tree);
960 static unsigned *done_pbase_paths;
961 static int done_pbase_paths_num;
962 static int done_pbase_paths_alloc;
963 static int done_pbase_path_pos(unsigned hash)
966 int hi = done_pbase_paths_num;
968 int mi = (hi + lo) / 2;
969 if (done_pbase_paths[mi] == hash)
971 if (done_pbase_paths[mi] < hash)
979 static int check_pbase_path(unsigned hash)
981 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
985 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
986 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
987 done_pbase_paths = xrealloc(done_pbase_paths,
988 done_pbase_paths_alloc *
991 done_pbase_paths_num++;
992 if (pos < done_pbase_paths_num)
993 memmove(done_pbase_paths + pos + 1,
994 done_pbase_paths + pos,
995 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
996 done_pbase_paths[pos] = hash;
1000 static void add_preferred_base_object(const char *name)
1002 struct pbase_tree *it;
1004 unsigned hash = name_hash(name);
1006 if (!num_preferred_base || check_pbase_path(hash))
1009 cmplen = name_cmp_len(name);
1010 for (it = pbase_tree; it; it = it->next) {
1012 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1015 struct tree_desc tree;
1016 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1017 add_pbase_object(&tree, name, cmplen, name);
1022 static void add_preferred_base(unsigned char *sha1)
1024 struct pbase_tree *it;
1027 unsigned char tree_sha1[20];
1029 if (window <= num_preferred_base++)
1032 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1036 for (it = pbase_tree; it; it = it->next) {
1037 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1043 it = xcalloc(1, sizeof(*it));
1044 it->next = pbase_tree;
1047 hashcpy(it->pcache.sha1, tree_sha1);
1048 it->pcache.tree_data = data;
1049 it->pcache.tree_size = size;
1052 static void check_object(struct object_entry *entry)
1054 if (entry->in_pack) {
1055 struct packed_git *p = entry->in_pack;
1056 struct pack_window *w_curs = NULL;
1057 const unsigned char *base_ref = NULL;
1058 struct object_entry *base_entry;
1059 unsigned long used, used_0;
1062 unsigned char *buf, c;
1064 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1067 * We want in_pack_type even if we do not reuse delta
1068 * since non-delta representations could still be reused.
1070 used = unpack_object_header_gently(buf, avail,
1071 &entry->in_pack_type,
1075 * Determine if this is a delta and if so whether we can
1076 * reuse it or not. Otherwise let's find out as cheaply as
1077 * possible what the actual type and size for this object is.
1079 switch (entry->in_pack_type) {
1081 /* Not a delta hence we've already got all we need. */
1082 entry->type = entry->in_pack_type;
1083 entry->in_pack_header_size = used;
1084 unuse_pack(&w_curs);
1087 if (!no_reuse_delta && !entry->preferred_base)
1088 base_ref = use_pack(p, &w_curs,
1089 entry->in_pack_offset + used, NULL);
1090 entry->in_pack_header_size = used + 20;
1093 buf = use_pack(p, &w_curs,
1094 entry->in_pack_offset + used, NULL);
1100 if (!ofs || MSB(ofs, 7))
1101 die("delta base offset overflow in pack for %s",
1102 sha1_to_hex(entry->sha1));
1104 ofs = (ofs << 7) + (c & 127);
1106 if (ofs >= entry->in_pack_offset)
1107 die("delta base offset out of bound for %s",
1108 sha1_to_hex(entry->sha1));
1109 ofs = entry->in_pack_offset - ofs;
1110 if (!no_reuse_delta && !entry->preferred_base)
1111 base_ref = find_packed_object_name(p, ofs);
1112 entry->in_pack_header_size = used + used_0;
1116 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1118 * If base_ref was set above that means we wish to
1119 * reuse delta data, and we even found that base
1120 * in the list of objects we want to pack. Goodie!
1122 * Depth value does not matter - find_deltas() will
1123 * never consider reused delta as the base object to
1124 * deltify other objects against, in order to avoid
1127 entry->type = entry->in_pack_type;
1128 entry->delta = base_entry;
1129 entry->delta_sibling = base_entry->delta_child;
1130 base_entry->delta_child = entry;
1131 unuse_pack(&w_curs);
1137 * This must be a delta and we already know what the
1138 * final object type is. Let's extract the actual
1139 * object size from the delta header.
1141 entry->size = get_size_from_delta(p, &w_curs,
1142 entry->in_pack_offset + entry->in_pack_header_size);
1143 unuse_pack(&w_curs);
1148 * No choice but to fall back to the recursive delta walk
1149 * with sha1_object_info() to find about the object type
1152 unuse_pack(&w_curs);
1155 entry->type = sha1_object_info(entry->sha1, &entry->size);
1156 if (entry->type < 0)
1157 die("unable to get type of object %s",
1158 sha1_to_hex(entry->sha1));
1161 static int pack_offset_sort(const void *_a, const void *_b)
1163 const struct object_entry *a = *(struct object_entry **)_a;
1164 const struct object_entry *b = *(struct object_entry **)_b;
1166 /* avoid filesystem trashing with loose objects */
1167 if (!a->in_pack && !b->in_pack)
1168 return hashcmp(a->sha1, b->sha1);
1170 if (a->in_pack < b->in_pack)
1172 if (a->in_pack > b->in_pack)
1174 return a->in_pack_offset < b->in_pack_offset ? -1 :
1175 (a->in_pack_offset > b->in_pack_offset);
1178 static void get_object_details(void)
1181 struct object_entry **sorted_by_offset;
1183 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1184 for (i = 0; i < nr_objects; i++)
1185 sorted_by_offset[i] = objects + i;
1186 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1189 for (i = 0; i < nr_objects; i++)
1190 check_object(sorted_by_offset[i]);
1191 free(sorted_by_offset);
1194 static int type_size_sort(const void *_a, const void *_b)
1196 const struct object_entry *a = *(struct object_entry **)_a;
1197 const struct object_entry *b = *(struct object_entry **)_b;
1199 if (a->type < b->type)
1201 if (a->type > b->type)
1203 if (a->hash < b->hash)
1205 if (a->hash > b->hash)
1207 if (a->preferred_base < b->preferred_base)
1209 if (a->preferred_base > b->preferred_base)
1211 if (a->size < b->size)
1213 if (a->size > b->size)
1215 return a > b ? -1 : (a < b); /* newest last */
1219 struct object_entry *entry;
1221 struct delta_index *index;
1225 * We search for deltas _backwards_ in a list sorted by type and
1226 * by size, so that we see progressively smaller and smaller files.
1227 * That's because we prefer deltas to be from the bigger file
1228 * to the smaller - deletes are potentially cheaper, but perhaps
1229 * more importantly, the bigger file is likely the more recent
1232 static int try_delta(struct unpacked *trg, struct unpacked *src,
1235 struct object_entry *trg_entry = trg->entry;
1236 struct object_entry *src_entry = src->entry;
1237 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1238 enum object_type type;
1241 /* Don't bother doing diffs between different types */
1242 if (trg_entry->type != src_entry->type)
1245 /* We do not compute delta to *create* objects we are not
1248 if (trg_entry->preferred_base)
1252 * We do not bother to try a delta that we discarded
1253 * on an earlier try, but only when reusing delta data.
1255 if (!no_reuse_delta && trg_entry->in_pack &&
1256 trg_entry->in_pack == src_entry->in_pack &&
1257 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1258 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1261 /* Let's not bust the allowed depth. */
1262 if (src_entry->depth >= max_depth)
1265 /* Now some size filtering heuristics. */
1266 trg_size = trg_entry->size;
1267 max_size = trg_size/2 - 20;
1268 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1271 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1272 max_size = trg_entry->delta_size-1;
1273 src_size = src_entry->size;
1274 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1275 if (sizediff >= max_size)
1278 /* Load data if not already done */
1280 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1282 die("object %s inconsistent object length (%lu vs %lu)",
1283 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1286 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1288 die("object %s inconsistent object length (%lu vs %lu)",
1289 sha1_to_hex(src_entry->sha1), sz, src_size);
1292 src->index = create_delta_index(src->data, src_size);
1294 die("out of memory");
1297 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1301 trg_entry->delta = src_entry;
1302 trg_entry->delta_size = delta_size;
1303 trg_entry->depth = src_entry->depth + 1;
1308 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1310 struct object_entry *child = me->delta_child;
1313 unsigned int c = check_delta_limit(child, n + 1);
1316 child = child->delta_sibling;
1321 static void find_deltas(struct object_entry **list, int window, int depth)
1323 uint32_t i = nr_objects, idx = 0, processed = 0;
1324 unsigned int array_size = window * sizeof(struct unpacked);
1325 struct unpacked *array;
1330 array = xmalloc(array_size);
1331 memset(array, 0, array_size);
1333 start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1336 struct object_entry *entry = list[--i];
1337 struct unpacked *n = array + idx;
1340 if (!entry->preferred_base)
1344 display_progress(&progress_state, processed);
1347 /* This happens if we decided to reuse existing
1348 * delta from a pack. "!no_reuse_delta &&" is implied.
1352 if (entry->size < 50)
1354 free_delta_index(n->index);
1361 * If the current object is at pack edge, take the depth the
1362 * objects that depend on the current object into account
1363 * otherwise they would become too deep.
1366 if (entry->delta_child) {
1367 max_depth -= check_delta_limit(entry, 0);
1374 uint32_t other_idx = idx + j;
1376 if (other_idx >= window)
1377 other_idx -= window;
1378 m = array + other_idx;
1381 if (try_delta(n, m, max_depth) < 0)
1385 /* if we made n a delta, and if n is already at max
1386 * depth, leaving it in the window is pointless. we
1387 * should evict it first.
1389 if (entry->delta && depth <= entry->depth)
1399 stop_progress(&progress_state);
1401 for (i = 0; i < window; ++i) {
1402 free_delta_index(array[i].index);
1403 free(array[i].data);
1408 static void prepare_pack(int window, int depth)
1410 struct object_entry **delta_list;
1413 get_object_details();
1415 if (!window || !depth)
1418 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1419 for (i = 0; i < nr_objects; i++)
1420 delta_list[i] = objects + i;
1421 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1422 find_deltas(delta_list, window+1, depth);
1426 static int git_pack_config(const char *k, const char *v)
1428 if(!strcmp(k, "pack.window")) {
1429 window = git_config_int(k, v);
1432 if(!strcmp(k, "pack.depth")) {
1433 depth = git_config_int(k, v);
1436 if (!strcmp(k, "pack.compression")) {
1437 int level = git_config_int(k, v);
1439 level = Z_DEFAULT_COMPRESSION;
1440 else if (level < 0 || level > Z_BEST_COMPRESSION)
1441 die("bad pack compression level %d", level);
1442 pack_compression_level = level;
1443 pack_compression_seen = 1;
1446 return git_default_config(k, v);
1449 static void read_object_list_from_stdin(void)
1451 char line[40 + 1 + PATH_MAX + 2];
1452 unsigned char sha1[20];
1455 if (!fgets(line, sizeof(line), stdin)) {
1459 die("fgets returned NULL, not EOF, not error!");
1461 die("fgets: %s", strerror(errno));
1465 if (line[0] == '-') {
1466 if (get_sha1_hex(line+1, sha1))
1467 die("expected edge sha1, got garbage:\n %s",
1469 add_preferred_base(sha1);
1472 if (get_sha1_hex(line, sha1))
1473 die("expected sha1, got garbage:\n %s", line);
1475 add_preferred_base_object(line+41);
1476 add_object_entry(sha1, 0, line+41, 0);
1480 static void show_commit(struct commit *commit)
1482 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1485 static void show_object(struct object_array_entry *p)
1487 add_preferred_base_object(p->name);
1488 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1491 static void show_edge(struct commit *commit)
1493 add_preferred_base(commit->object.sha1);
1496 static void get_object_list(int ac, const char **av)
1498 struct rev_info revs;
1502 init_revisions(&revs, NULL);
1503 save_commit_buffer = 0;
1504 track_object_refs = 0;
1505 setup_revisions(ac, av, &revs, NULL);
1507 while (fgets(line, sizeof(line), stdin) != NULL) {
1508 int len = strlen(line);
1509 if (line[len - 1] == '\n')
1514 if (!strcmp(line, "--not")) {
1515 flags ^= UNINTERESTING;
1518 die("not a rev '%s'", line);
1520 if (handle_revision_arg(line, &revs, flags, 1))
1521 die("bad revision '%s'", line);
1524 prepare_revision_walk(&revs);
1525 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1526 traverse_commit_list(&revs, show_commit, show_object);
1529 static int adjust_perm(const char *path, mode_t mode)
1531 if (chmod(path, mode))
1533 return adjust_shared_perm(path);
1536 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1538 int use_internal_rev_list = 0;
1541 off_t last_obj_offset;
1542 const char *base_name = NULL;
1544 int rp_ac_alloc = 64;
1547 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1549 rp_av[0] = "pack-objects";
1550 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1553 git_config(git_pack_config);
1554 if (!pack_compression_seen && core_compression_seen)
1555 pack_compression_level = core_compression_level;
1557 progress = isatty(2);
1558 for (i = 1; i < argc; i++) {
1559 const char *arg = argv[i];
1564 if (!strcmp("--non-empty", arg)) {
1568 if (!strcmp("--local", arg)) {
1572 if (!strcmp("--incremental", arg)) {
1576 if (!prefixcmp(arg, "--compression=")) {
1578 int level = strtoul(arg+14, &end, 0);
1579 if (!arg[14] || *end)
1582 level = Z_DEFAULT_COMPRESSION;
1583 else if (level < 0 || level > Z_BEST_COMPRESSION)
1584 die("bad pack compression level %d", level);
1585 pack_compression_level = level;
1588 if (!prefixcmp(arg, "--window=")) {
1590 window = strtoul(arg+9, &end, 0);
1591 if (!arg[9] || *end)
1595 if (!prefixcmp(arg, "--depth=")) {
1597 depth = strtoul(arg+8, &end, 0);
1598 if (!arg[8] || *end)
1602 if (!strcmp("--progress", arg)) {
1606 if (!strcmp("--all-progress", arg)) {
1610 if (!strcmp("-q", arg)) {
1614 if (!strcmp("--no-reuse-delta", arg)) {
1618 if (!strcmp("--no-reuse-object", arg)) {
1619 no_reuse_object = no_reuse_delta = 1;
1622 if (!strcmp("--delta-base-offset", arg)) {
1623 allow_ofs_delta = 1;
1626 if (!strcmp("--stdout", arg)) {
1630 if (!strcmp("--revs", arg)) {
1631 use_internal_rev_list = 1;
1634 if (!strcmp("--unpacked", arg) ||
1635 !prefixcmp(arg, "--unpacked=") ||
1636 !strcmp("--reflog", arg) ||
1637 !strcmp("--all", arg)) {
1638 use_internal_rev_list = 1;
1639 if (rp_ac >= rp_ac_alloc - 1) {
1640 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1641 rp_av = xrealloc(rp_av,
1642 rp_ac_alloc * sizeof(*rp_av));
1644 rp_av[rp_ac++] = arg;
1647 if (!strcmp("--thin", arg)) {
1648 use_internal_rev_list = 1;
1650 rp_av[1] = "--objects-edge";
1653 if (!prefixcmp(arg, "--index-version=")) {
1655 index_default_version = strtoul(arg + 16, &c, 10);
1656 if (index_default_version > 2)
1659 index_off32_limit = strtoul(c+1, &c, 0);
1660 if (*c || index_off32_limit & 0x80000000)
1667 /* Traditionally "pack-objects [options] base extra" failed;
1668 * we would however want to take refs parameter that would
1669 * have been given to upstream rev-list ourselves, which means
1670 * we somehow want to say what the base name is. So the
1673 * pack-objects [options] base <refs...>
1675 * in other words, we would treat the first non-option as the
1676 * base_name and send everything else to the internal revision
1680 if (!pack_to_stdout)
1681 base_name = argv[i++];
1683 if (pack_to_stdout != !base_name)
1686 if (!pack_to_stdout && thin)
1687 die("--thin cannot be used to build an indexable pack.");
1689 prepare_packed_git();
1692 start_progress(&progress_state, "Generating pack...",
1693 "Counting objects: ", 0);
1694 if (!use_internal_rev_list)
1695 read_object_list_from_stdin();
1697 rp_av[rp_ac] = NULL;
1698 get_object_list(rp_ac, rp_av);
1701 stop_progress(&progress_state);
1702 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1705 if (non_empty && !nr_result)
1707 if (progress && (nr_objects != nr_result))
1708 fprintf(stderr, "Result has %u objects.\n", nr_result);
1710 prepare_pack(window, depth);
1711 last_obj_offset = write_pack_file();
1712 if (!pack_to_stdout) {
1713 unsigned char object_list_sha1[20];
1714 mode_t mode = umask(0);
1717 mode = 0444 & ~mode;
1719 write_index_file(last_obj_offset, object_list_sha1);
1720 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
1721 base_name, sha1_to_hex(object_list_sha1));
1722 if (adjust_perm(pack_tmp_name, mode))
1723 die("unable to make temporary pack file readable: %s",
1725 if (rename(pack_tmp_name, tmpname))
1726 die("unable to rename temporary pack file: %s",
1728 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
1729 base_name, sha1_to_hex(object_list_sha1));
1730 if (adjust_perm(idx_tmp_name, mode))
1731 die("unable to make temporary index file readable: %s",
1733 if (rename(idx_tmp_name, tmpname))
1734 die("unable to rename temporary index file: %s",
1736 puts(sha1_to_hex(object_list_sha1));
1739 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1740 written, written_delta, reused, reused_delta);