11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
20 #ifdef THREADED_DELTA_SEARCH
21 #include "thread-utils.h"
25 static const char pack_usage[] = "\
26 git pack-objects [{ -q | --progress | --all-progress }] \n\
27 [--max-pack-size=N] [--local] [--incremental] \n\
28 [--window=N] [--window-memory=N] [--depth=N] \n\
29 [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
30 [--threads=N] [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
31 [--stdout | base-name] [--include-tag] \n\
32 [--keep-unreachable | --unpack-unreachable] \n\
33 [<ref-list | <object-list]";
36 struct pack_idx_entry idx;
37 unsigned long size; /* uncompressed size */
38 struct packed_git *in_pack; /* already in pack */
40 struct object_entry *delta; /* delta base object */
41 struct object_entry *delta_child; /* deltified objects who bases me */
42 struct object_entry *delta_sibling; /* other deltified objects who
43 * uses the same base as me
45 void *delta_data; /* cached delta (uncompressed) */
46 unsigned long delta_size; /* delta data size (uncompressed) */
47 unsigned long z_delta_size; /* delta data size (compressed) */
48 unsigned int hash; /* name hint hash */
49 enum object_type type;
50 enum object_type in_pack_type; /* could be delta */
51 unsigned char in_pack_header_size;
52 unsigned char preferred_base; /* we do not pack this, but is available
53 * to be used as the base object to delta
56 unsigned char no_try_delta;
60 * Objects we are going to pack are collected in objects array (dynamically
61 * expanded). nr_objects & nr_alloc controls this array. They are stored
62 * in the order we see -- typically rev-list --objects order that gives us
63 * nice "minimum seek" order.
65 static struct object_entry *objects;
66 static struct pack_idx_entry **written_list;
67 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
70 static int reuse_delta = 1, reuse_object = 1;
71 static int keep_unreachable, unpack_unreachable, include_tag;
73 static int incremental;
74 static int ignore_packed_keep;
75 static int allow_ofs_delta;
76 static const char *base_name;
77 static int progress = 1;
78 static int window = 10;
79 static uint32_t pack_size_limit, pack_size_limit_cfg;
80 static int depth = 50;
81 static int delta_search_threads = 1;
82 static int pack_to_stdout;
83 static int num_preferred_base;
84 static struct progress *progress_state;
85 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
86 static int pack_compression_seen;
88 static unsigned long delta_cache_size = 0;
89 static unsigned long max_delta_cache_size = 0;
90 static unsigned long cache_max_small_delta_size = 1000;
92 static unsigned long window_memory_limit = 0;
95 * The object names in objects array are hashed with this hashtable,
96 * to help looking up the entry by object name.
97 * This hashtable is built after all the objects are seen.
99 static int *object_ix;
100 static int object_ix_hashsz;
105 static uint32_t written, written_delta;
106 static uint32_t reused, reused_delta;
109 static void *get_delta(struct object_entry *entry)
111 unsigned long size, base_size, delta_size;
112 void *buf, *base_buf, *delta_buf;
113 enum object_type type;
115 buf = read_sha1_file(entry->idx.sha1, &type, &size);
117 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
118 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
120 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
121 delta_buf = diff_delta(base_buf, base_size,
122 buf, size, &delta_size, 0);
123 if (!delta_buf || delta_size != entry->delta_size)
124 die("delta size changed");
130 static unsigned long do_compress(void **pptr, unsigned long size)
134 unsigned long maxsize;
136 memset(&stream, 0, sizeof(stream));
137 deflateInit(&stream, pack_compression_level);
138 maxsize = deflateBound(&stream, size);
141 out = xmalloc(maxsize);
145 stream.avail_in = size;
146 stream.next_out = out;
147 stream.avail_out = maxsize;
148 while (deflate(&stream, Z_FINISH) == Z_OK)
153 return stream.total_out;
157 * The per-object header is a pretty dense thing, which is
158 * - first byte: low four bits are "size", then three bits of "type",
159 * and the high bit is "size continues".
160 * - each byte afterwards: low seven bits are size continuation,
161 * with the high bit being "size continues"
163 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
168 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
169 die("bad type %d", type);
171 c = (type << 4) | (size & 15);
184 * we are going to reuse the existing object data as is. make
185 * sure it is not corrupt.
187 static int check_pack_inflate(struct packed_git *p,
188 struct pack_window **w_curs,
191 unsigned long expect)
194 unsigned char fakebuf[4096], *in;
197 memset(&stream, 0, sizeof(stream));
198 git_inflate_init(&stream);
200 in = use_pack(p, w_curs, offset, &stream.avail_in);
202 stream.next_out = fakebuf;
203 stream.avail_out = sizeof(fakebuf);
204 st = git_inflate(&stream, Z_FINISH);
205 offset += stream.next_in - in;
206 } while (st == Z_OK || st == Z_BUF_ERROR);
207 git_inflate_end(&stream);
208 return (st == Z_STREAM_END &&
209 stream.total_out == expect &&
210 stream.total_in == len) ? 0 : -1;
213 static void copy_pack_data(struct sha1file *f,
214 struct packed_git *p,
215 struct pack_window **w_curs,
223 in = use_pack(p, w_curs, offset, &avail);
225 avail = (unsigned int)len;
226 sha1write(f, in, avail);
232 static unsigned long write_object(struct sha1file *f,
233 struct object_entry *entry,
236 unsigned long size, limit, datalen;
238 unsigned char header[10], dheader[10];
240 enum object_type type;
241 int usable_delta, to_reuse;
248 /* write limit if limited packsize and not first object */
249 if (!pack_size_limit || !nr_written)
251 else if (pack_size_limit <= write_offset)
253 * the earlier object did not fit the limit; avoid
254 * mistaking this with unlimited (i.e. limit = 0).
258 limit = pack_size_limit - write_offset;
261 usable_delta = 0; /* no delta */
262 else if (!pack_size_limit)
263 usable_delta = 1; /* unlimited packfile */
264 else if (entry->delta->idx.offset == (off_t)-1)
265 usable_delta = 0; /* base was written to another pack */
266 else if (entry->delta->idx.offset)
267 usable_delta = 1; /* base already exists in this pack */
269 usable_delta = 0; /* base could end up in another pack */
272 to_reuse = 0; /* explicit */
273 else if (!entry->in_pack)
274 to_reuse = 0; /* can't reuse what we don't have */
275 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
276 /* check_object() decided it for us ... */
277 to_reuse = usable_delta;
278 /* ... but pack split may override that */
279 else if (type != entry->in_pack_type)
280 to_reuse = 0; /* pack has delta which is unusable */
281 else if (entry->delta)
282 to_reuse = 0; /* we want to pack afresh */
284 to_reuse = 1; /* we have it in-pack undeltified,
285 * and we do not need to deltify it.
291 buf = read_sha1_file(entry->idx.sha1, &type, &size);
293 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
295 * make sure no cached delta data remains from a
296 * previous attempt before a pack split occured.
298 free(entry->delta_data);
299 entry->delta_data = NULL;
300 entry->z_delta_size = 0;
301 } else if (entry->delta_data) {
302 size = entry->delta_size;
303 buf = entry->delta_data;
304 entry->delta_data = NULL;
305 type = (allow_ofs_delta && entry->delta->idx.offset) ?
306 OBJ_OFS_DELTA : OBJ_REF_DELTA;
308 buf = get_delta(entry);
309 size = entry->delta_size;
310 type = (allow_ofs_delta && entry->delta->idx.offset) ?
311 OBJ_OFS_DELTA : OBJ_REF_DELTA;
314 if (entry->z_delta_size)
315 datalen = entry->z_delta_size;
317 datalen = do_compress(&buf, size);
320 * The object header is a byte of 'type' followed by zero or
321 * more bytes of length.
323 hdrlen = encode_header(type, size, header);
325 if (type == OBJ_OFS_DELTA) {
327 * Deltas with relative base contain an additional
328 * encoding of the relative offset for the delta
329 * base from this object's position in the pack.
331 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
332 unsigned pos = sizeof(dheader) - 1;
333 dheader[pos] = ofs & 127;
335 dheader[--pos] = 128 | (--ofs & 127);
336 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
340 sha1write(f, header, hdrlen);
341 sha1write(f, dheader + pos, sizeof(dheader) - pos);
342 hdrlen += sizeof(dheader) - pos;
343 } else if (type == OBJ_REF_DELTA) {
345 * Deltas with a base reference contain
346 * an additional 20 bytes for the base sha1.
348 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
352 sha1write(f, header, hdrlen);
353 sha1write(f, entry->delta->idx.sha1, 20);
356 if (limit && hdrlen + datalen + 20 >= limit) {
360 sha1write(f, header, hdrlen);
362 sha1write(f, buf, datalen);
366 struct packed_git *p = entry->in_pack;
367 struct pack_window *w_curs = NULL;
368 struct revindex_entry *revidx;
372 type = (allow_ofs_delta && entry->delta->idx.offset) ?
373 OBJ_OFS_DELTA : OBJ_REF_DELTA;
374 hdrlen = encode_header(type, entry->size, header);
376 offset = entry->in_pack_offset;
377 revidx = find_pack_revindex(p, offset);
378 datalen = revidx[1].offset - offset;
379 if (!pack_to_stdout && p->index_version > 1 &&
380 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
381 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
386 offset += entry->in_pack_header_size;
387 datalen -= entry->in_pack_header_size;
388 if (!pack_to_stdout && p->index_version == 1 &&
389 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
390 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
395 if (type == OBJ_OFS_DELTA) {
396 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
397 unsigned pos = sizeof(dheader) - 1;
398 dheader[pos] = ofs & 127;
400 dheader[--pos] = 128 | (--ofs & 127);
401 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
405 sha1write(f, header, hdrlen);
406 sha1write(f, dheader + pos, sizeof(dheader) - pos);
407 hdrlen += sizeof(dheader) - pos;
409 } else if (type == OBJ_REF_DELTA) {
410 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
414 sha1write(f, header, hdrlen);
415 sha1write(f, entry->delta->idx.sha1, 20);
419 if (limit && hdrlen + datalen + 20 >= limit) {
423 sha1write(f, header, hdrlen);
425 copy_pack_data(f, p, &w_curs, offset, datalen);
433 entry->idx.crc32 = crc32_end(f);
434 return hdrlen + datalen;
437 static int write_one(struct sha1file *f,
438 struct object_entry *e,
443 /* offset is non zero if object is written already. */
444 if (e->idx.offset || e->preferred_base)
447 /* if we are deltified, write out base object first. */
448 if (e->delta && !write_one(f, e->delta, offset))
451 e->idx.offset = *offset;
452 size = write_object(f, e, *offset);
457 written_list[nr_written++] = &e->idx;
459 /* make sure off_t is sufficiently large not to wrap */
460 if (*offset > *offset + size)
461 die("pack too large for current definition of off_t");
466 /* forward declaration for write_pack_file */
467 static int adjust_perm(const char *path, mode_t mode);
469 static void write_pack_file(void)
474 struct pack_header hdr;
475 uint32_t nr_remaining = nr_result;
476 time_t last_mtime = 0;
478 if (progress > pack_to_stdout)
479 progress_state = start_progress("Writing objects", nr_result);
480 written_list = xmalloc(nr_objects * sizeof(*written_list));
483 unsigned char sha1[20];
484 char *pack_tmp_name = NULL;
486 if (pack_to_stdout) {
487 f = sha1fd_throughput(1, "<stdout>", progress_state);
489 char tmpname[PATH_MAX];
491 fd = odb_mkstemp(tmpname, sizeof(tmpname),
492 "pack/tmp_pack_XXXXXX");
493 pack_tmp_name = xstrdup(tmpname);
494 f = sha1fd(fd, pack_tmp_name);
497 hdr.hdr_signature = htonl(PACK_SIGNATURE);
498 hdr.hdr_version = htonl(PACK_VERSION);
499 hdr.hdr_entries = htonl(nr_remaining);
500 sha1write(f, &hdr, sizeof(hdr));
501 offset = sizeof(hdr);
503 for (; i < nr_objects; i++) {
504 if (!write_one(f, objects + i, &offset))
506 display_progress(progress_state, written);
510 * Did we write the wrong # entries in the header?
511 * If so, rewrite it like in fast-import
513 if (pack_to_stdout) {
514 sha1close(f, sha1, CSUM_CLOSE);
515 } else if (nr_written == nr_remaining) {
516 sha1close(f, sha1, CSUM_FSYNC);
518 int fd = sha1close(f, sha1, 0);
519 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
520 nr_written, sha1, offset);
524 if (!pack_to_stdout) {
525 mode_t mode = umask(0);
527 char *idx_tmp_name, tmpname[PATH_MAX];
532 idx_tmp_name = write_idx_file(NULL, written_list,
535 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
536 base_name, sha1_to_hex(sha1));
537 free_pack_by_name(tmpname);
538 if (adjust_perm(pack_tmp_name, mode))
539 die("unable to make temporary pack file readable: %s",
541 if (rename(pack_tmp_name, tmpname))
542 die("unable to rename temporary pack file: %s",
546 * Packs are runtime accessed in their mtime
547 * order since newer packs are more likely to contain
548 * younger objects. So if we are creating multiple
549 * packs then we should modify the mtime of later ones
550 * to preserve this property.
552 if (stat(tmpname, &st) < 0) {
553 warning("failed to stat %s: %s",
554 tmpname, strerror(errno));
555 } else if (!last_mtime) {
556 last_mtime = st.st_mtime;
559 utb.actime = st.st_atime;
560 utb.modtime = --last_mtime;
561 if (utime(tmpname, &utb) < 0)
562 warning("failed utime() on %s: %s",
563 tmpname, strerror(errno));
566 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
567 base_name, sha1_to_hex(sha1));
568 if (adjust_perm(idx_tmp_name, mode))
569 die("unable to make temporary index file readable: %s",
571 if (rename(idx_tmp_name, tmpname))
572 die("unable to rename temporary index file: %s",
577 puts(sha1_to_hex(sha1));
580 /* mark written objects as written to previous pack */
581 for (j = 0; j < nr_written; j++) {
582 written_list[j]->offset = (off_t)-1;
584 nr_remaining -= nr_written;
585 } while (nr_remaining && i < nr_objects);
588 stop_progress(&progress_state);
589 if (written != nr_result)
590 die("wrote %"PRIu32" objects while expecting %"PRIu32,
593 * We have scanned through [0 ... i). Since we have written
594 * the correct number of objects, the remaining [i ... nr_objects)
595 * items must be either already written (due to out-of-order delta base)
596 * or a preferred base. Count those which are neither and complain if any.
598 for (j = 0; i < nr_objects; i++) {
599 struct object_entry *e = objects + i;
600 j += !e->idx.offset && !e->preferred_base;
603 die("wrote %"PRIu32" objects as expected but %"PRIu32
604 " unwritten", written, j);
607 static int locate_object_entry_hash(const unsigned char *sha1)
611 memcpy(&ui, sha1, sizeof(unsigned int));
612 i = ui % object_ix_hashsz;
613 while (0 < object_ix[i]) {
614 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
616 if (++i == object_ix_hashsz)
622 static struct object_entry *locate_object_entry(const unsigned char *sha1)
626 if (!object_ix_hashsz)
629 i = locate_object_entry_hash(sha1);
631 return &objects[object_ix[i]-1];
635 static void rehash_objects(void)
638 struct object_entry *oe;
640 object_ix_hashsz = nr_objects * 3;
641 if (object_ix_hashsz < 1024)
642 object_ix_hashsz = 1024;
643 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
644 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
645 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
646 int ix = locate_object_entry_hash(oe->idx.sha1);
650 object_ix[ix] = i + 1;
654 static unsigned name_hash(const char *name)
663 * This effectively just creates a sortable number from the
664 * last sixteen non-whitespace characters. Last characters
665 * count "most", so things that end in ".c" sort together.
667 while ((c = *name++) != 0) {
670 hash = (hash >> 2) + (c << 24);
675 static void setup_delta_attr_check(struct git_attr_check *check)
677 static struct git_attr *attr_delta;
680 attr_delta = git_attr("delta", 5);
682 check[0].attr = attr_delta;
685 static int no_try_delta(const char *path)
687 struct git_attr_check check[1];
689 setup_delta_attr_check(check);
690 if (git_checkattr(path, ARRAY_SIZE(check), check))
692 if (ATTR_FALSE(check->value))
697 static int add_object_entry(const unsigned char *sha1, enum object_type type,
698 const char *name, int exclude)
700 struct object_entry *entry;
701 struct packed_git *p, *found_pack = NULL;
702 off_t found_offset = 0;
704 unsigned hash = name_hash(name);
706 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
709 entry = objects + object_ix[ix] - 1;
710 if (!entry->preferred_base)
712 entry->preferred_base = 1;
717 if (!exclude && local && has_loose_object_nonlocal(sha1))
720 for (p = packed_git; p; p = p->next) {
721 off_t offset = find_pack_entry_one(sha1, p);
724 found_offset = offset;
731 if (local && !p->pack_local)
733 if (ignore_packed_keep && p->pack_local && p->pack_keep)
738 if (nr_objects >= nr_alloc) {
739 nr_alloc = (nr_alloc + 1024) * 3 / 2;
740 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
743 entry = objects + nr_objects++;
744 memset(entry, 0, sizeof(*entry));
745 hashcpy(entry->idx.sha1, sha1);
750 entry->preferred_base = 1;
754 entry->in_pack = found_pack;
755 entry->in_pack_offset = found_offset;
758 if (object_ix_hashsz * 3 <= nr_objects * 4)
761 object_ix[-1 - ix] = nr_objects;
763 display_progress(progress_state, nr_objects);
765 if (name && no_try_delta(name))
766 entry->no_try_delta = 1;
771 struct pbase_tree_cache {
772 unsigned char sha1[20];
776 unsigned long tree_size;
779 static struct pbase_tree_cache *(pbase_tree_cache[256]);
780 static int pbase_tree_cache_ix(const unsigned char *sha1)
782 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
784 static int pbase_tree_cache_ix_incr(int ix)
786 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
789 static struct pbase_tree {
790 struct pbase_tree *next;
791 /* This is a phony "cache" entry; we are not
792 * going to evict it nor find it through _get()
793 * mechanism -- this is for the toplevel node that
794 * would almost always change with any commit.
796 struct pbase_tree_cache pcache;
799 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
801 struct pbase_tree_cache *ent, *nent;
804 enum object_type type;
806 int my_ix = pbase_tree_cache_ix(sha1);
807 int available_ix = -1;
809 /* pbase-tree-cache acts as a limited hashtable.
810 * your object will be found at your index or within a few
811 * slots after that slot if it is cached.
813 for (neigh = 0; neigh < 8; neigh++) {
814 ent = pbase_tree_cache[my_ix];
815 if (ent && !hashcmp(ent->sha1, sha1)) {
819 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
820 ((0 <= available_ix) &&
821 (!ent && pbase_tree_cache[available_ix])))
822 available_ix = my_ix;
825 my_ix = pbase_tree_cache_ix_incr(my_ix);
828 /* Did not find one. Either we got a bogus request or
829 * we need to read and perhaps cache.
831 data = read_sha1_file(sha1, &type, &size);
834 if (type != OBJ_TREE) {
839 /* We need to either cache or return a throwaway copy */
841 if (available_ix < 0)
844 ent = pbase_tree_cache[available_ix];
845 my_ix = available_ix;
849 nent = xmalloc(sizeof(*nent));
850 nent->temporary = (available_ix < 0);
853 /* evict and reuse */
854 free(ent->tree_data);
857 hashcpy(nent->sha1, sha1);
858 nent->tree_data = data;
859 nent->tree_size = size;
861 if (!nent->temporary)
862 pbase_tree_cache[my_ix] = nent;
866 static void pbase_tree_put(struct pbase_tree_cache *cache)
868 if (!cache->temporary) {
872 free(cache->tree_data);
876 static int name_cmp_len(const char *name)
879 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
884 static void add_pbase_object(struct tree_desc *tree,
887 const char *fullname)
889 struct name_entry entry;
892 while (tree_entry(tree,&entry)) {
893 if (S_ISGITLINK(entry.mode))
895 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
896 memcmp(name, entry.path, cmplen);
901 if (name[cmplen] != '/') {
902 add_object_entry(entry.sha1,
903 object_type(entry.mode),
907 if (S_ISDIR(entry.mode)) {
908 struct tree_desc sub;
909 struct pbase_tree_cache *tree;
910 const char *down = name+cmplen+1;
911 int downlen = name_cmp_len(down);
913 tree = pbase_tree_get(entry.sha1);
916 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
918 add_pbase_object(&sub, down, downlen, fullname);
919 pbase_tree_put(tree);
924 static unsigned *done_pbase_paths;
925 static int done_pbase_paths_num;
926 static int done_pbase_paths_alloc;
927 static int done_pbase_path_pos(unsigned hash)
930 int hi = done_pbase_paths_num;
932 int mi = (hi + lo) / 2;
933 if (done_pbase_paths[mi] == hash)
935 if (done_pbase_paths[mi] < hash)
943 static int check_pbase_path(unsigned hash)
945 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
949 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
950 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
951 done_pbase_paths = xrealloc(done_pbase_paths,
952 done_pbase_paths_alloc *
955 done_pbase_paths_num++;
956 if (pos < done_pbase_paths_num)
957 memmove(done_pbase_paths + pos + 1,
958 done_pbase_paths + pos,
959 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
960 done_pbase_paths[pos] = hash;
964 static void add_preferred_base_object(const char *name)
966 struct pbase_tree *it;
968 unsigned hash = name_hash(name);
970 if (!num_preferred_base || check_pbase_path(hash))
973 cmplen = name_cmp_len(name);
974 for (it = pbase_tree; it; it = it->next) {
976 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
979 struct tree_desc tree;
980 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
981 add_pbase_object(&tree, name, cmplen, name);
986 static void add_preferred_base(unsigned char *sha1)
988 struct pbase_tree *it;
991 unsigned char tree_sha1[20];
993 if (window <= num_preferred_base++)
996 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1000 for (it = pbase_tree; it; it = it->next) {
1001 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1007 it = xcalloc(1, sizeof(*it));
1008 it->next = pbase_tree;
1011 hashcpy(it->pcache.sha1, tree_sha1);
1012 it->pcache.tree_data = data;
1013 it->pcache.tree_size = size;
1016 static void check_object(struct object_entry *entry)
1018 if (entry->in_pack) {
1019 struct packed_git *p = entry->in_pack;
1020 struct pack_window *w_curs = NULL;
1021 const unsigned char *base_ref = NULL;
1022 struct object_entry *base_entry;
1023 unsigned long used, used_0;
1026 unsigned char *buf, c;
1028 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1031 * We want in_pack_type even if we do not reuse delta
1032 * since non-delta representations could still be reused.
1034 used = unpack_object_header_buffer(buf, avail,
1035 &entry->in_pack_type,
1041 * Determine if this is a delta and if so whether we can
1042 * reuse it or not. Otherwise let's find out as cheaply as
1043 * possible what the actual type and size for this object is.
1045 switch (entry->in_pack_type) {
1047 /* Not a delta hence we've already got all we need. */
1048 entry->type = entry->in_pack_type;
1049 entry->in_pack_header_size = used;
1050 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1052 unuse_pack(&w_curs);
1055 if (reuse_delta && !entry->preferred_base)
1056 base_ref = use_pack(p, &w_curs,
1057 entry->in_pack_offset + used, NULL);
1058 entry->in_pack_header_size = used + 20;
1061 buf = use_pack(p, &w_curs,
1062 entry->in_pack_offset + used, NULL);
1068 if (!ofs || MSB(ofs, 7)) {
1069 error("delta base offset overflow in pack for %s",
1070 sha1_to_hex(entry->idx.sha1));
1074 ofs = (ofs << 7) + (c & 127);
1076 ofs = entry->in_pack_offset - ofs;
1077 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1078 error("delta base offset out of bound for %s",
1079 sha1_to_hex(entry->idx.sha1));
1082 if (reuse_delta && !entry->preferred_base) {
1083 struct revindex_entry *revidx;
1084 revidx = find_pack_revindex(p, ofs);
1087 base_ref = nth_packed_object_sha1(p, revidx->nr);
1089 entry->in_pack_header_size = used + used_0;
1093 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1095 * If base_ref was set above that means we wish to
1096 * reuse delta data, and we even found that base
1097 * in the list of objects we want to pack. Goodie!
1099 * Depth value does not matter - find_deltas() will
1100 * never consider reused delta as the base object to
1101 * deltify other objects against, in order to avoid
1104 entry->type = entry->in_pack_type;
1105 entry->delta = base_entry;
1106 entry->delta_size = entry->size;
1107 entry->delta_sibling = base_entry->delta_child;
1108 base_entry->delta_child = entry;
1109 unuse_pack(&w_curs);
1115 * This must be a delta and we already know what the
1116 * final object type is. Let's extract the actual
1117 * object size from the delta header.
1119 entry->size = get_size_from_delta(p, &w_curs,
1120 entry->in_pack_offset + entry->in_pack_header_size);
1121 if (entry->size == 0)
1123 unuse_pack(&w_curs);
1128 * No choice but to fall back to the recursive delta walk
1129 * with sha1_object_info() to find about the object type
1133 unuse_pack(&w_curs);
1136 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1138 * The error condition is checked in prepare_pack(). This is
1139 * to permit a missing preferred base object to be ignored
1140 * as a preferred base. Doing so can result in a larger
1141 * pack file, but the transfer will still take place.
1145 static int pack_offset_sort(const void *_a, const void *_b)
1147 const struct object_entry *a = *(struct object_entry **)_a;
1148 const struct object_entry *b = *(struct object_entry **)_b;
1150 /* avoid filesystem trashing with loose objects */
1151 if (!a->in_pack && !b->in_pack)
1152 return hashcmp(a->idx.sha1, b->idx.sha1);
1154 if (a->in_pack < b->in_pack)
1156 if (a->in_pack > b->in_pack)
1158 return a->in_pack_offset < b->in_pack_offset ? -1 :
1159 (a->in_pack_offset > b->in_pack_offset);
1162 static void get_object_details(void)
1165 struct object_entry **sorted_by_offset;
1167 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1168 for (i = 0; i < nr_objects; i++)
1169 sorted_by_offset[i] = objects + i;
1170 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1172 for (i = 0; i < nr_objects; i++)
1173 check_object(sorted_by_offset[i]);
1175 free(sorted_by_offset);
1179 * We search for deltas in a list sorted by type, by filename hash, and then
1180 * by size, so that we see progressively smaller and smaller files.
1181 * That's because we prefer deltas to be from the bigger file
1182 * to the smaller -- deletes are potentially cheaper, but perhaps
1183 * more importantly, the bigger file is likely the more recent
1184 * one. The deepest deltas are therefore the oldest objects which are
1185 * less susceptible to be accessed often.
1187 static int type_size_sort(const void *_a, const void *_b)
1189 const struct object_entry *a = *(struct object_entry **)_a;
1190 const struct object_entry *b = *(struct object_entry **)_b;
1192 if (a->type > b->type)
1194 if (a->type < b->type)
1196 if (a->hash > b->hash)
1198 if (a->hash < b->hash)
1200 if (a->preferred_base > b->preferred_base)
1202 if (a->preferred_base < b->preferred_base)
1204 if (a->size > b->size)
1206 if (a->size < b->size)
1208 return a < b ? -1 : (a > b); /* newest first */
1212 struct object_entry *entry;
1214 struct delta_index *index;
1218 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1219 unsigned long delta_size)
1221 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1224 if (delta_size < cache_max_small_delta_size)
1227 /* cache delta, if objects are large enough compared to delta size */
1228 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1234 #ifdef THREADED_DELTA_SEARCH
1236 static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1237 #define read_lock() pthread_mutex_lock(&read_mutex)
1238 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1240 static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1241 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1242 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1244 static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1245 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1246 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1250 #define read_lock() (void)0
1251 #define read_unlock() (void)0
1252 #define cache_lock() (void)0
1253 #define cache_unlock() (void)0
1254 #define progress_lock() (void)0
1255 #define progress_unlock() (void)0
1259 static int try_delta(struct unpacked *trg, struct unpacked *src,
1260 unsigned max_depth, unsigned long *mem_usage)
1262 struct object_entry *trg_entry = trg->entry;
1263 struct object_entry *src_entry = src->entry;
1264 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1266 enum object_type type;
1269 /* Don't bother doing diffs between different types */
1270 if (trg_entry->type != src_entry->type)
1274 * We do not bother to try a delta that we discarded
1275 * on an earlier try, but only when reusing delta data.
1277 if (reuse_delta && trg_entry->in_pack &&
1278 trg_entry->in_pack == src_entry->in_pack &&
1279 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1280 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1283 /* Let's not bust the allowed depth. */
1284 if (src->depth >= max_depth)
1287 /* Now some size filtering heuristics. */
1288 trg_size = trg_entry->size;
1289 if (!trg_entry->delta) {
1290 max_size = trg_size/2 - 20;
1293 max_size = trg_entry->delta_size;
1294 ref_depth = trg->depth;
1296 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1297 (max_depth - ref_depth + 1);
1300 src_size = src_entry->size;
1301 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1302 if (sizediff >= max_size)
1304 if (trg_size < src_size / 32)
1307 /* Load data if not already done */
1310 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1313 die("object %s cannot be read",
1314 sha1_to_hex(trg_entry->idx.sha1));
1316 die("object %s inconsistent object length (%lu vs %lu)",
1317 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1322 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1325 die("object %s cannot be read",
1326 sha1_to_hex(src_entry->idx.sha1));
1328 die("object %s inconsistent object length (%lu vs %lu)",
1329 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1333 src->index = create_delta_index(src->data, src_size);
1335 static int warned = 0;
1337 warning("suboptimal pack - out of memory");
1340 *mem_usage += sizeof_delta_index(src->index);
1343 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1347 if (trg_entry->delta) {
1348 /* Prefer only shallower same-sized deltas. */
1349 if (delta_size == trg_entry->delta_size &&
1350 src->depth + 1 >= trg->depth) {
1357 * Handle memory allocation outside of the cache
1358 * accounting lock. Compiler will optimize the strangeness
1359 * away when THREADED_DELTA_SEARCH is not defined.
1361 free(trg_entry->delta_data);
1363 if (trg_entry->delta_data) {
1364 delta_cache_size -= trg_entry->delta_size;
1365 trg_entry->delta_data = NULL;
1367 if (delta_cacheable(src_size, trg_size, delta_size)) {
1368 delta_cache_size += delta_size;
1370 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1376 trg_entry->delta = src_entry;
1377 trg_entry->delta_size = delta_size;
1378 trg->depth = src->depth + 1;
1383 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1385 struct object_entry *child = me->delta_child;
1388 unsigned int c = check_delta_limit(child, n + 1);
1391 child = child->delta_sibling;
1396 static unsigned long free_unpacked(struct unpacked *n)
1398 unsigned long freed_mem = sizeof_delta_index(n->index);
1399 free_delta_index(n->index);
1402 freed_mem += n->entry->size;
1411 static void find_deltas(struct object_entry **list, unsigned *list_size,
1412 int window, int depth, unsigned *processed)
1414 uint32_t i, idx = 0, count = 0;
1415 struct unpacked *array;
1416 unsigned long mem_usage = 0;
1418 array = xcalloc(window, sizeof(struct unpacked));
1421 struct object_entry *entry;
1422 struct unpacked *n = array + idx;
1423 int j, max_depth, best_base = -1;
1432 if (!entry->preferred_base) {
1434 display_progress(progress_state, *processed);
1438 mem_usage -= free_unpacked(n);
1441 while (window_memory_limit &&
1442 mem_usage > window_memory_limit &&
1444 uint32_t tail = (idx + window - count) % window;
1445 mem_usage -= free_unpacked(array + tail);
1449 /* We do not compute delta to *create* objects we are not
1452 if (entry->preferred_base)
1456 * If the current object is at pack edge, take the depth the
1457 * objects that depend on the current object into account
1458 * otherwise they would become too deep.
1461 if (entry->delta_child) {
1462 max_depth -= check_delta_limit(entry, 0);
1470 uint32_t other_idx = idx + j;
1472 if (other_idx >= window)
1473 other_idx -= window;
1474 m = array + other_idx;
1477 ret = try_delta(n, m, max_depth, &mem_usage);
1481 best_base = other_idx;
1485 * If we decided to cache the delta data, then it is best
1486 * to compress it right away. First because we have to do
1487 * it anyway, and doing it here while we're threaded will
1488 * save a lot of time in the non threaded write phase,
1489 * as well as allow for caching more deltas within
1490 * the same cache size limit.
1492 * But only if not writing to stdout, since in that case
1493 * the network is most likely throttling writes anyway,
1494 * and therefore it is best to go to the write phase ASAP
1495 * instead, as we can afford spending more time compressing
1496 * between writes at that moment.
1498 if (entry->delta_data && !pack_to_stdout) {
1499 entry->z_delta_size = do_compress(&entry->delta_data,
1502 delta_cache_size -= entry->delta_size;
1503 delta_cache_size += entry->z_delta_size;
1507 /* if we made n a delta, and if n is already at max
1508 * depth, leaving it in the window is pointless. we
1509 * should evict it first.
1511 if (entry->delta && max_depth <= n->depth)
1515 * Move the best delta base up in the window, after the
1516 * currently deltified object, to keep it longer. It will
1517 * be the first base object to be attempted next.
1520 struct unpacked swap = array[best_base];
1521 int dist = (window + idx - best_base) % window;
1522 int dst = best_base;
1524 int src = (dst + 1) % window;
1525 array[dst] = array[src];
1533 if (count + 1 < window)
1539 for (i = 0; i < window; ++i) {
1540 free_delta_index(array[i].index);
1541 free(array[i].data);
1546 #ifdef THREADED_DELTA_SEARCH
1549 * The main thread waits on the condition that (at least) one of the workers
1550 * has stopped working (which is indicated in the .working member of
1551 * struct thread_params).
1552 * When a work thread has completed its work, it sets .working to 0 and
1553 * signals the main thread and waits on the condition that .data_ready
1557 struct thread_params {
1559 struct object_entry **list;
1566 pthread_mutex_t mutex;
1567 pthread_cond_t cond;
1568 unsigned *processed;
1571 static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
1573 static void *threaded_find_deltas(void *arg)
1575 struct thread_params *me = arg;
1577 while (me->remaining) {
1578 find_deltas(me->list, &me->remaining,
1579 me->window, me->depth, me->processed);
1583 pthread_cond_signal(&progress_cond);
1587 * We must not set ->data_ready before we wait on the
1588 * condition because the main thread may have set it to 1
1589 * before we get here. In order to be sure that new
1590 * work is available if we see 1 in ->data_ready, it
1591 * was initialized to 0 before this thread was spawned
1592 * and we reset it to 0 right away.
1594 pthread_mutex_lock(&me->mutex);
1595 while (!me->data_ready)
1596 pthread_cond_wait(&me->cond, &me->mutex);
1598 pthread_mutex_unlock(&me->mutex);
1600 /* leave ->working 1 so that this doesn't get more work assigned */
1604 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1605 int window, int depth, unsigned *processed)
1607 struct thread_params p[delta_search_threads];
1608 int i, ret, active_threads = 0;
1610 if (delta_search_threads <= 1) {
1611 find_deltas(list, &list_size, window, depth, processed);
1615 /* Partition the work amongst work threads. */
1616 for (i = 0; i < delta_search_threads; i++) {
1617 unsigned sub_size = list_size / (delta_search_threads - i);
1619 p[i].window = window;
1621 p[i].processed = processed;
1623 p[i].data_ready = 0;
1625 /* try to split chunks on "path" boundaries */
1626 while (sub_size && sub_size < list_size &&
1627 list[sub_size]->hash &&
1628 list[sub_size]->hash == list[sub_size-1]->hash)
1632 p[i].list_size = sub_size;
1633 p[i].remaining = sub_size;
1636 list_size -= sub_size;
1639 /* Start work threads. */
1640 for (i = 0; i < delta_search_threads; i++) {
1641 if (!p[i].list_size)
1643 pthread_mutex_init(&p[i].mutex, NULL);
1644 pthread_cond_init(&p[i].cond, NULL);
1645 ret = pthread_create(&p[i].thread, NULL,
1646 threaded_find_deltas, &p[i]);
1648 die("unable to create thread: %s", strerror(ret));
1653 * Now let's wait for work completion. Each time a thread is done
1654 * with its work, we steal half of the remaining work from the
1655 * thread with the largest number of unprocessed objects and give
1656 * it to that newly idle thread. This ensure good load balancing
1657 * until the remaining object list segments are simply too short
1658 * to be worth splitting anymore.
1660 while (active_threads) {
1661 struct thread_params *target = NULL;
1662 struct thread_params *victim = NULL;
1663 unsigned sub_size = 0;
1667 for (i = 0; !target && i < delta_search_threads; i++)
1672 pthread_cond_wait(&progress_cond, &progress_mutex);
1675 for (i = 0; i < delta_search_threads; i++)
1676 if (p[i].remaining > 2*window &&
1677 (!victim || victim->remaining < p[i].remaining))
1680 sub_size = victim->remaining / 2;
1681 list = victim->list + victim->list_size - sub_size;
1682 while (sub_size && list[0]->hash &&
1683 list[0]->hash == list[-1]->hash) {
1689 * It is possible for some "paths" to have
1690 * so many objects that no hash boundary
1691 * might be found. Let's just steal the
1692 * exact half in that case.
1694 sub_size = victim->remaining / 2;
1697 target->list = list;
1698 victim->list_size -= sub_size;
1699 victim->remaining -= sub_size;
1701 target->list_size = sub_size;
1702 target->remaining = sub_size;
1703 target->working = 1;
1706 pthread_mutex_lock(&target->mutex);
1707 target->data_ready = 1;
1708 pthread_cond_signal(&target->cond);
1709 pthread_mutex_unlock(&target->mutex);
1712 pthread_join(target->thread, NULL);
1713 pthread_cond_destroy(&target->cond);
1714 pthread_mutex_destroy(&target->mutex);
1721 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1724 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1726 unsigned char peeled[20];
1728 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1729 !peel_ref(path, peeled) && /* peelable? */
1730 !is_null_sha1(peeled) && /* annotated tag? */
1731 locate_object_entry(peeled)) /* object packed? */
1732 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1736 static void prepare_pack(int window, int depth)
1738 struct object_entry **delta_list;
1739 uint32_t i, nr_deltas;
1742 get_object_details();
1745 * If we're locally repacking then we need to be doubly careful
1746 * from now on in order to make sure no stealth corruption gets
1747 * propagated to the new pack. Clients receiving streamed packs
1748 * should validate everything they get anyway so no need to incur
1749 * the additional cost here in that case.
1751 if (!pack_to_stdout)
1752 do_check_packed_object_crc = 1;
1754 if (!nr_objects || !window || !depth)
1757 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1760 for (i = 0; i < nr_objects; i++) {
1761 struct object_entry *entry = objects + i;
1764 /* This happens if we decided to reuse existing
1765 * delta from a pack. "reuse_delta &&" is implied.
1769 if (entry->size < 50)
1772 if (entry->no_try_delta)
1775 if (!entry->preferred_base) {
1777 if (entry->type < 0)
1778 die("unable to get type of object %s",
1779 sha1_to_hex(entry->idx.sha1));
1781 if (entry->type < 0) {
1783 * This object is not found, but we
1784 * don't have to include it anyway.
1790 delta_list[n++] = entry;
1793 if (nr_deltas && n > 1) {
1794 unsigned nr_done = 0;
1796 progress_state = start_progress("Compressing objects",
1798 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1799 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1800 stop_progress(&progress_state);
1801 if (nr_done != nr_deltas)
1802 die("inconsistency with delta count");
1807 static int git_pack_config(const char *k, const char *v, void *cb)
1809 if(!strcmp(k, "pack.window")) {
1810 window = git_config_int(k, v);
1813 if (!strcmp(k, "pack.windowmemory")) {
1814 window_memory_limit = git_config_ulong(k, v);
1817 if (!strcmp(k, "pack.depth")) {
1818 depth = git_config_int(k, v);
1821 if (!strcmp(k, "pack.compression")) {
1822 int level = git_config_int(k, v);
1824 level = Z_DEFAULT_COMPRESSION;
1825 else if (level < 0 || level > Z_BEST_COMPRESSION)
1826 die("bad pack compression level %d", level);
1827 pack_compression_level = level;
1828 pack_compression_seen = 1;
1831 if (!strcmp(k, "pack.deltacachesize")) {
1832 max_delta_cache_size = git_config_int(k, v);
1835 if (!strcmp(k, "pack.deltacachelimit")) {
1836 cache_max_small_delta_size = git_config_int(k, v);
1839 if (!strcmp(k, "pack.threads")) {
1840 delta_search_threads = git_config_int(k, v);
1841 if (delta_search_threads < 0)
1842 die("invalid number of threads specified (%d)",
1843 delta_search_threads);
1844 #ifndef THREADED_DELTA_SEARCH
1845 if (delta_search_threads != 1)
1846 warning("no threads support, ignoring %s", k);
1850 if (!strcmp(k, "pack.indexversion")) {
1851 pack_idx_default_version = git_config_int(k, v);
1852 if (pack_idx_default_version > 2)
1853 die("bad pack.indexversion=%"PRIu32,
1854 pack_idx_default_version);
1857 if (!strcmp(k, "pack.packsizelimit")) {
1858 pack_size_limit_cfg = git_config_ulong(k, v);
1861 return git_default_config(k, v, cb);
1864 static void read_object_list_from_stdin(void)
1866 char line[40 + 1 + PATH_MAX + 2];
1867 unsigned char sha1[20];
1870 if (!fgets(line, sizeof(line), stdin)) {
1874 die("fgets returned NULL, not EOF, not error!");
1876 die("fgets: %s", strerror(errno));
1880 if (line[0] == '-') {
1881 if (get_sha1_hex(line+1, sha1))
1882 die("expected edge sha1, got garbage:\n %s",
1884 add_preferred_base(sha1);
1887 if (get_sha1_hex(line, sha1))
1888 die("expected sha1, got garbage:\n %s", line);
1890 add_preferred_base_object(line+41);
1891 add_object_entry(sha1, 0, line+41, 0);
1895 #define OBJECT_ADDED (1u<<20)
1897 static void show_commit(struct commit *commit)
1899 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1900 commit->object.flags |= OBJECT_ADDED;
1903 static void show_object(struct object_array_entry *p)
1905 add_preferred_base_object(p->name);
1906 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1907 p->item->flags |= OBJECT_ADDED;
1908 free((char *)p->name);
1912 static void show_edge(struct commit *commit)
1914 add_preferred_base(commit->object.sha1);
1917 struct in_pack_object {
1919 struct object *object;
1925 struct in_pack_object *array;
1928 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1930 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1931 in_pack->array[in_pack->nr].object = object;
1936 * Compare the objects in the offset order, in order to emulate the
1937 * "git rev-list --objects" output that produced the pack originally.
1939 static int ofscmp(const void *a_, const void *b_)
1941 struct in_pack_object *a = (struct in_pack_object *)a_;
1942 struct in_pack_object *b = (struct in_pack_object *)b_;
1944 if (a->offset < b->offset)
1946 else if (a->offset > b->offset)
1949 return hashcmp(a->object->sha1, b->object->sha1);
1952 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1954 struct packed_git *p;
1955 struct in_pack in_pack;
1958 memset(&in_pack, 0, sizeof(in_pack));
1960 for (p = packed_git; p; p = p->next) {
1961 const unsigned char *sha1;
1964 if (!p->pack_local || p->pack_keep)
1966 if (open_pack_index(p))
1967 die("cannot open pack index");
1969 ALLOC_GROW(in_pack.array,
1970 in_pack.nr + p->num_objects,
1973 for (i = 0; i < p->num_objects; i++) {
1974 sha1 = nth_packed_object_sha1(p, i);
1975 o = lookup_unknown_object(sha1);
1976 if (!(o->flags & OBJECT_ADDED))
1977 mark_in_pack_object(o, p, &in_pack);
1978 o->flags |= OBJECT_ADDED;
1983 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1985 for (i = 0; i < in_pack.nr; i++) {
1986 struct object *o = in_pack.array[i].object;
1987 add_object_entry(o->sha1, o->type, "", 0);
1990 free(in_pack.array);
1993 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
1995 static struct packed_git *last_found = (void *)1;
1996 struct packed_git *p;
1998 p = (last_found != (void *)1) ? last_found : packed_git;
2001 if ((!p->pack_local || p->pack_keep) &&
2002 find_pack_entry_one(sha1, p)) {
2006 if (p == last_found)
2010 if (p == last_found)
2016 static void loosen_unused_packed_objects(struct rev_info *revs)
2018 struct packed_git *p;
2020 const unsigned char *sha1;
2022 for (p = packed_git; p; p = p->next) {
2023 if (!p->pack_local || p->pack_keep)
2026 if (open_pack_index(p))
2027 die("cannot open pack index");
2029 for (i = 0; i < p->num_objects; i++) {
2030 sha1 = nth_packed_object_sha1(p, i);
2031 if (!locate_object_entry(sha1) &&
2032 !has_sha1_pack_kept_or_nonlocal(sha1))
2033 if (force_object_loose(sha1, p->mtime))
2034 die("unable to force loose object");
2039 static void get_object_list(int ac, const char **av)
2041 struct rev_info revs;
2045 init_revisions(&revs, NULL);
2046 save_commit_buffer = 0;
2047 setup_revisions(ac, av, &revs, NULL);
2049 while (fgets(line, sizeof(line), stdin) != NULL) {
2050 int len = strlen(line);
2051 if (len && line[len - 1] == '\n')
2056 if (!strcmp(line, "--not")) {
2057 flags ^= UNINTERESTING;
2060 die("not a rev '%s'", line);
2062 if (handle_revision_arg(line, &revs, flags, 1))
2063 die("bad revision '%s'", line);
2066 if (prepare_revision_walk(&revs))
2067 die("revision walk setup failed");
2068 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2069 traverse_commit_list(&revs, show_commit, show_object);
2071 if (keep_unreachable)
2072 add_objects_in_unpacked_packs(&revs);
2073 if (unpack_unreachable)
2074 loosen_unused_packed_objects(&revs);
2077 static int adjust_perm(const char *path, mode_t mode)
2079 if (chmod(path, mode))
2081 return adjust_shared_perm(path);
2084 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2086 int use_internal_rev_list = 0;
2090 int rp_ac_alloc = 64;
2093 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2095 rp_av[0] = "pack-objects";
2096 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2099 git_config(git_pack_config, NULL);
2100 if (!pack_compression_seen && core_compression_seen)
2101 pack_compression_level = core_compression_level;
2103 progress = isatty(2);
2104 for (i = 1; i < argc; i++) {
2105 const char *arg = argv[i];
2110 if (!strcmp("--non-empty", arg)) {
2114 if (!strcmp("--local", arg)) {
2118 if (!strcmp("--incremental", arg)) {
2122 if (!strcmp("--honor-pack-keep", arg)) {
2123 ignore_packed_keep = 1;
2126 if (!prefixcmp(arg, "--compression=")) {
2128 int level = strtoul(arg+14, &end, 0);
2129 if (!arg[14] || *end)
2132 level = Z_DEFAULT_COMPRESSION;
2133 else if (level < 0 || level > Z_BEST_COMPRESSION)
2134 die("bad pack compression level %d", level);
2135 pack_compression_level = level;
2138 if (!prefixcmp(arg, "--max-pack-size=")) {
2140 pack_size_limit_cfg = 0;
2141 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2142 if (!arg[16] || *end)
2146 if (!prefixcmp(arg, "--window=")) {
2148 window = strtoul(arg+9, &end, 0);
2149 if (!arg[9] || *end)
2153 if (!prefixcmp(arg, "--window-memory=")) {
2154 if (!git_parse_ulong(arg+16, &window_memory_limit))
2158 if (!prefixcmp(arg, "--threads=")) {
2160 delta_search_threads = strtoul(arg+10, &end, 0);
2161 if (!arg[10] || *end || delta_search_threads < 0)
2163 #ifndef THREADED_DELTA_SEARCH
2164 if (delta_search_threads != 1)
2165 warning("no threads support, "
2166 "ignoring %s", arg);
2170 if (!prefixcmp(arg, "--depth=")) {
2172 depth = strtoul(arg+8, &end, 0);
2173 if (!arg[8] || *end)
2177 if (!strcmp("--progress", arg)) {
2181 if (!strcmp("--all-progress", arg)) {
2185 if (!strcmp("-q", arg)) {
2189 if (!strcmp("--no-reuse-delta", arg)) {
2193 if (!strcmp("--no-reuse-object", arg)) {
2194 reuse_object = reuse_delta = 0;
2197 if (!strcmp("--delta-base-offset", arg)) {
2198 allow_ofs_delta = 1;
2201 if (!strcmp("--stdout", arg)) {
2205 if (!strcmp("--revs", arg)) {
2206 use_internal_rev_list = 1;
2209 if (!strcmp("--keep-unreachable", arg)) {
2210 keep_unreachable = 1;
2213 if (!strcmp("--unpack-unreachable", arg)) {
2214 unpack_unreachable = 1;
2217 if (!strcmp("--include-tag", arg)) {
2221 if (!strcmp("--unpacked", arg) ||
2222 !strcmp("--reflog", arg) ||
2223 !strcmp("--all", arg)) {
2224 use_internal_rev_list = 1;
2225 if (rp_ac >= rp_ac_alloc - 1) {
2226 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2227 rp_av = xrealloc(rp_av,
2228 rp_ac_alloc * sizeof(*rp_av));
2230 rp_av[rp_ac++] = arg;
2233 if (!strcmp("--thin", arg)) {
2234 use_internal_rev_list = 1;
2236 rp_av[1] = "--objects-edge";
2239 if (!prefixcmp(arg, "--index-version=")) {
2241 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2242 if (pack_idx_default_version > 2)
2245 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2246 if (*c || pack_idx_off32_limit & 0x80000000)
2253 /* Traditionally "pack-objects [options] base extra" failed;
2254 * we would however want to take refs parameter that would
2255 * have been given to upstream rev-list ourselves, which means
2256 * we somehow want to say what the base name is. So the
2259 * pack-objects [options] base <refs...>
2261 * in other words, we would treat the first non-option as the
2262 * base_name and send everything else to the internal revision
2266 if (!pack_to_stdout)
2267 base_name = argv[i++];
2269 if (pack_to_stdout != !base_name)
2272 if (!pack_to_stdout && !pack_size_limit)
2273 pack_size_limit = pack_size_limit_cfg;
2275 if (pack_to_stdout && pack_size_limit)
2276 die("--max-pack-size cannot be used to build a pack for transfer.");
2278 if (!pack_to_stdout && thin)
2279 die("--thin cannot be used to build an indexable pack.");
2281 if (keep_unreachable && unpack_unreachable)
2282 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2284 #ifdef THREADED_DELTA_SEARCH
2285 if (!delta_search_threads) /* --threads=0 means autodetect */
2286 delta_search_threads = online_cpus();
2289 prepare_packed_git();
2292 progress_state = start_progress("Counting objects", 0);
2293 if (!use_internal_rev_list)
2294 read_object_list_from_stdin();
2296 rp_av[rp_ac] = NULL;
2297 get_object_list(rp_ac, rp_av);
2299 if (include_tag && nr_result)
2300 for_each_ref(add_ref_tag, NULL);
2301 stop_progress(&progress_state);
2303 if (non_empty && !nr_result)
2306 prepare_pack(window, depth);
2309 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2310 " reused %"PRIu32" (delta %"PRIu32")\n",
2311 written, written_delta, reused, reused_delta);