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 allow_ofs_delta;
75 static const char *base_name;
76 static int progress = 1;
77 static int window = 10;
78 static uint32_t pack_size_limit, pack_size_limit_cfg;
79 static int depth = 50;
80 static int delta_search_threads = 1;
81 static int pack_to_stdout;
82 static int num_preferred_base;
83 static struct progress *progress_state;
84 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
85 static int pack_compression_seen;
87 static unsigned long delta_cache_size = 0;
88 static unsigned long max_delta_cache_size = 0;
89 static unsigned long cache_max_small_delta_size = 1000;
91 static unsigned long window_memory_limit = 0;
94 * The object names in objects array are hashed with this hashtable,
95 * to help looking up the entry by object name.
96 * This hashtable is built after all the objects are seen.
98 static int *object_ix;
99 static int object_ix_hashsz;
104 static uint32_t written, written_delta;
105 static uint32_t reused, reused_delta;
108 static void *get_delta(struct object_entry *entry)
110 unsigned long size, base_size, delta_size;
111 void *buf, *base_buf, *delta_buf;
112 enum object_type type;
114 buf = read_sha1_file(entry->idx.sha1, &type, &size);
116 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
117 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
119 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
120 delta_buf = diff_delta(base_buf, base_size,
121 buf, size, &delta_size, 0);
122 if (!delta_buf || delta_size != entry->delta_size)
123 die("delta size changed");
129 static unsigned long do_compress(void **pptr, unsigned long size)
133 unsigned long maxsize;
135 memset(&stream, 0, sizeof(stream));
136 deflateInit(&stream, pack_compression_level);
137 maxsize = deflateBound(&stream, size);
140 out = xmalloc(maxsize);
144 stream.avail_in = size;
145 stream.next_out = out;
146 stream.avail_out = maxsize;
147 while (deflate(&stream, Z_FINISH) == Z_OK)
152 return stream.total_out;
156 * The per-object header is a pretty dense thing, which is
157 * - first byte: low four bits are "size", then three bits of "type",
158 * and the high bit is "size continues".
159 * - each byte afterwards: low seven bits are size continuation,
160 * with the high bit being "size continues"
162 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
167 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
168 die("bad type %d", type);
170 c = (type << 4) | (size & 15);
183 * we are going to reuse the existing object data as is. make
184 * sure it is not corrupt.
186 static int check_pack_inflate(struct packed_git *p,
187 struct pack_window **w_curs,
190 unsigned long expect)
193 unsigned char fakebuf[4096], *in;
196 memset(&stream, 0, sizeof(stream));
197 inflateInit(&stream);
199 in = use_pack(p, w_curs, offset, &stream.avail_in);
201 stream.next_out = fakebuf;
202 stream.avail_out = sizeof(fakebuf);
203 st = inflate(&stream, Z_FINISH);
204 offset += stream.next_in - in;
205 } while (st == Z_OK || st == Z_BUF_ERROR);
207 return (st == Z_STREAM_END &&
208 stream.total_out == expect &&
209 stream.total_in == len) ? 0 : -1;
212 static void copy_pack_data(struct sha1file *f,
213 struct packed_git *p,
214 struct pack_window **w_curs,
222 in = use_pack(p, w_curs, offset, &avail);
224 avail = (unsigned int)len;
225 sha1write(f, in, avail);
231 static unsigned long write_object(struct sha1file *f,
232 struct object_entry *entry,
235 unsigned long size, limit, datalen;
237 unsigned char header[10], dheader[10];
239 enum object_type type;
240 int usable_delta, to_reuse;
247 /* write limit if limited packsize and not first object */
248 limit = pack_size_limit && nr_written ?
249 pack_size_limit - write_offset : 0;
252 usable_delta = 0; /* no delta */
253 else if (!pack_size_limit)
254 usable_delta = 1; /* unlimited packfile */
255 else if (entry->delta->idx.offset == (off_t)-1)
256 usable_delta = 0; /* base was written to another pack */
257 else if (entry->delta->idx.offset)
258 usable_delta = 1; /* base already exists in this pack */
260 usable_delta = 0; /* base could end up in another pack */
263 to_reuse = 0; /* explicit */
264 else if (!entry->in_pack)
265 to_reuse = 0; /* can't reuse what we don't have */
266 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
267 /* check_object() decided it for us ... */
268 to_reuse = usable_delta;
269 /* ... but pack split may override that */
270 else if (type != entry->in_pack_type)
271 to_reuse = 0; /* pack has delta which is unusable */
272 else if (entry->delta)
273 to_reuse = 0; /* we want to pack afresh */
275 to_reuse = 1; /* we have it in-pack undeltified,
276 * and we do not need to deltify it.
281 buf = read_sha1_file(entry->idx.sha1, &type, &size);
283 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
285 * make sure no cached delta data remains from a
286 * previous attempt before a pack split occured.
288 free(entry->delta_data);
289 entry->delta_data = NULL;
290 entry->z_delta_size = 0;
291 } else if (entry->delta_data) {
292 size = entry->delta_size;
293 buf = entry->delta_data;
294 entry->delta_data = NULL;
295 type = (allow_ofs_delta && entry->delta->idx.offset) ?
296 OBJ_OFS_DELTA : OBJ_REF_DELTA;
298 buf = get_delta(entry);
299 size = entry->delta_size;
300 type = (allow_ofs_delta && entry->delta->idx.offset) ?
301 OBJ_OFS_DELTA : OBJ_REF_DELTA;
304 if (entry->z_delta_size)
305 datalen = entry->z_delta_size;
307 datalen = do_compress(&buf, size);
310 * The object header is a byte of 'type' followed by zero or
311 * more bytes of length.
313 hdrlen = encode_header(type, size, header);
315 if (type == OBJ_OFS_DELTA) {
317 * Deltas with relative base contain an additional
318 * encoding of the relative offset for the delta
319 * base from this object's position in the pack.
321 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
322 unsigned pos = sizeof(dheader) - 1;
323 dheader[pos] = ofs & 127;
325 dheader[--pos] = 128 | (--ofs & 127);
326 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
330 sha1write(f, header, hdrlen);
331 sha1write(f, dheader + pos, sizeof(dheader) - pos);
332 hdrlen += sizeof(dheader) - pos;
333 } else if (type == OBJ_REF_DELTA) {
335 * Deltas with a base reference contain
336 * an additional 20 bytes for the base sha1.
338 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
342 sha1write(f, header, hdrlen);
343 sha1write(f, entry->delta->idx.sha1, 20);
346 if (limit && hdrlen + datalen + 20 >= limit) {
350 sha1write(f, header, hdrlen);
352 sha1write(f, buf, datalen);
356 struct packed_git *p = entry->in_pack;
357 struct pack_window *w_curs = NULL;
358 struct revindex_entry *revidx;
362 type = (allow_ofs_delta && entry->delta->idx.offset) ?
363 OBJ_OFS_DELTA : OBJ_REF_DELTA;
366 hdrlen = encode_header(type, entry->size, header);
367 offset = entry->in_pack_offset;
368 revidx = find_pack_revindex(p, offset);
369 datalen = revidx[1].offset - offset;
370 if (!pack_to_stdout && p->index_version > 1 &&
371 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
372 die("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
373 offset += entry->in_pack_header_size;
374 datalen -= entry->in_pack_header_size;
375 if (type == OBJ_OFS_DELTA) {
376 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
377 unsigned pos = sizeof(dheader) - 1;
378 dheader[pos] = ofs & 127;
380 dheader[--pos] = 128 | (--ofs & 127);
381 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
383 sha1write(f, header, hdrlen);
384 sha1write(f, dheader + pos, sizeof(dheader) - pos);
385 hdrlen += sizeof(dheader) - pos;
386 } else if (type == OBJ_REF_DELTA) {
387 if (limit && hdrlen + 20 + datalen + 20 >= limit)
389 sha1write(f, header, hdrlen);
390 sha1write(f, entry->delta->idx.sha1, 20);
393 if (limit && hdrlen + datalen + 20 >= limit)
395 sha1write(f, header, hdrlen);
398 if (!pack_to_stdout && p->index_version == 1 &&
399 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
400 die("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
401 copy_pack_data(f, p, &w_curs, offset, datalen);
409 entry->idx.crc32 = crc32_end(f);
410 return hdrlen + datalen;
413 static off_t write_one(struct sha1file *f,
414 struct object_entry *e,
419 /* offset is non zero if object is written already. */
420 if (e->idx.offset || e->preferred_base)
423 /* if we are deltified, write out base object first. */
425 offset = write_one(f, e->delta, offset);
430 e->idx.offset = offset;
431 size = write_object(f, e, offset);
436 written_list[nr_written++] = &e->idx;
438 /* make sure off_t is sufficiently large not to wrap */
439 if (offset > offset + size)
440 die("pack too large for current definition of off_t");
441 return offset + size;
444 /* forward declaration for write_pack_file */
445 static int adjust_perm(const char *path, mode_t mode);
447 static void write_pack_file(void)
451 off_t offset, offset_one, last_obj_offset = 0;
452 struct pack_header hdr;
453 uint32_t nr_remaining = nr_result;
454 time_t last_mtime = 0;
456 if (progress > pack_to_stdout)
457 progress_state = start_progress("Writing objects", nr_result);
458 written_list = xmalloc(nr_objects * sizeof(*written_list));
461 unsigned char sha1[20];
462 char *pack_tmp_name = NULL;
464 if (pack_to_stdout) {
465 f = sha1fd_throughput(1, "<stdout>", progress_state);
467 char tmpname[PATH_MAX];
469 snprintf(tmpname, sizeof(tmpname),
470 "%s/tmp_pack_XXXXXX", get_object_directory());
471 fd = xmkstemp(tmpname);
472 pack_tmp_name = xstrdup(tmpname);
473 f = sha1fd(fd, pack_tmp_name);
476 hdr.hdr_signature = htonl(PACK_SIGNATURE);
477 hdr.hdr_version = htonl(PACK_VERSION);
478 hdr.hdr_entries = htonl(nr_remaining);
479 sha1write(f, &hdr, sizeof(hdr));
480 offset = sizeof(hdr);
482 for (; i < nr_objects; i++) {
483 last_obj_offset = offset;
484 offset_one = write_one(f, objects + i, offset);
488 display_progress(progress_state, written);
492 * Did we write the wrong # entries in the header?
493 * If so, rewrite it like in fast-import
495 if (pack_to_stdout) {
496 sha1close(f, sha1, CSUM_CLOSE);
497 } else if (nr_written == nr_remaining) {
498 sha1close(f, sha1, CSUM_FSYNC);
500 int fd = sha1close(f, NULL, 0);
501 fixup_pack_header_footer(fd, sha1, pack_tmp_name, nr_written);
505 if (!pack_to_stdout) {
506 mode_t mode = umask(0);
508 char *idx_tmp_name, tmpname[PATH_MAX];
513 idx_tmp_name = write_idx_file(NULL, written_list,
516 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
517 base_name, sha1_to_hex(sha1));
518 if (adjust_perm(pack_tmp_name, mode))
519 die("unable to make temporary pack file readable: %s",
521 if (rename(pack_tmp_name, tmpname))
522 die("unable to rename temporary pack file: %s",
526 * Packs are runtime accessed in their mtime
527 * order since newer packs are more likely to contain
528 * younger objects. So if we are creating multiple
529 * packs then we should modify the mtime of later ones
530 * to preserve this property.
532 if (stat(tmpname, &st) < 0) {
533 warning("failed to stat %s: %s",
534 tmpname, strerror(errno));
535 } else if (!last_mtime) {
536 last_mtime = st.st_mtime;
539 utb.actime = st.st_atime;
540 utb.modtime = --last_mtime;
541 if (utime(tmpname, &utb) < 0)
542 warning("failed utime() on %s: %s",
543 tmpname, strerror(errno));
546 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
547 base_name, sha1_to_hex(sha1));
548 if (adjust_perm(idx_tmp_name, mode))
549 die("unable to make temporary index file readable: %s",
551 if (rename(idx_tmp_name, tmpname))
552 die("unable to rename temporary index file: %s",
557 puts(sha1_to_hex(sha1));
560 /* mark written objects as written to previous pack */
561 for (j = 0; j < nr_written; j++) {
562 written_list[j]->offset = (off_t)-1;
564 nr_remaining -= nr_written;
565 } while (nr_remaining && i < nr_objects);
568 stop_progress(&progress_state);
569 if (written != nr_result)
570 die("wrote %"PRIu32" objects while expecting %"PRIu32,
573 * We have scanned through [0 ... i). Since we have written
574 * the correct number of objects, the remaining [i ... nr_objects)
575 * items must be either already written (due to out-of-order delta base)
576 * or a preferred base. Count those which are neither and complain if any.
578 for (j = 0; i < nr_objects; i++) {
579 struct object_entry *e = objects + i;
580 j += !e->idx.offset && !e->preferred_base;
583 die("wrote %"PRIu32" objects as expected but %"PRIu32
584 " unwritten", written, j);
587 static int locate_object_entry_hash(const unsigned char *sha1)
591 memcpy(&ui, sha1, sizeof(unsigned int));
592 i = ui % object_ix_hashsz;
593 while (0 < object_ix[i]) {
594 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
596 if (++i == object_ix_hashsz)
602 static struct object_entry *locate_object_entry(const unsigned char *sha1)
606 if (!object_ix_hashsz)
609 i = locate_object_entry_hash(sha1);
611 return &objects[object_ix[i]-1];
615 static void rehash_objects(void)
618 struct object_entry *oe;
620 object_ix_hashsz = nr_objects * 3;
621 if (object_ix_hashsz < 1024)
622 object_ix_hashsz = 1024;
623 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
624 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
625 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
626 int ix = locate_object_entry_hash(oe->idx.sha1);
630 object_ix[ix] = i + 1;
634 static unsigned name_hash(const char *name)
643 * This effectively just creates a sortable number from the
644 * last sixteen non-whitespace characters. Last characters
645 * count "most", so things that end in ".c" sort together.
647 while ((c = *name++) != 0) {
650 hash = (hash >> 2) + (c << 24);
655 static void setup_delta_attr_check(struct git_attr_check *check)
657 static struct git_attr *attr_delta;
660 attr_delta = git_attr("delta", 5);
662 check[0].attr = attr_delta;
665 static int no_try_delta(const char *path)
667 struct git_attr_check check[1];
669 setup_delta_attr_check(check);
670 if (git_checkattr(path, ARRAY_SIZE(check), check))
672 if (ATTR_FALSE(check->value))
677 static int add_object_entry(const unsigned char *sha1, enum object_type type,
678 const char *name, int exclude)
680 struct object_entry *entry;
681 struct packed_git *p, *found_pack = NULL;
682 off_t found_offset = 0;
684 unsigned hash = name_hash(name);
686 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
689 entry = objects + object_ix[ix] - 1;
690 if (!entry->preferred_base)
692 entry->preferred_base = 1;
697 for (p = packed_git; p; p = p->next) {
698 off_t offset = find_pack_entry_one(sha1, p);
701 found_offset = offset;
708 if (local && !p->pack_local)
713 if (nr_objects >= nr_alloc) {
714 nr_alloc = (nr_alloc + 1024) * 3 / 2;
715 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
718 entry = objects + nr_objects++;
719 memset(entry, 0, sizeof(*entry));
720 hashcpy(entry->idx.sha1, sha1);
725 entry->preferred_base = 1;
729 entry->in_pack = found_pack;
730 entry->in_pack_offset = found_offset;
733 if (object_ix_hashsz * 3 <= nr_objects * 4)
736 object_ix[-1 - ix] = nr_objects;
738 display_progress(progress_state, nr_objects);
740 if (name && no_try_delta(name))
741 entry->no_try_delta = 1;
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;
867 while (tree_entry(tree,&entry)) {
868 if (S_ISGITLINK(entry.mode))
870 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
871 memcmp(name, entry.path, cmplen);
876 if (name[cmplen] != '/') {
877 add_object_entry(entry.sha1,
878 object_type(entry.mode),
882 if (S_ISDIR(entry.mode)) {
883 struct tree_desc sub;
884 struct pbase_tree_cache *tree;
885 const char *down = name+cmplen+1;
886 int downlen = name_cmp_len(down);
888 tree = pbase_tree_get(entry.sha1);
891 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
893 add_pbase_object(&sub, down, downlen, fullname);
894 pbase_tree_put(tree);
899 static unsigned *done_pbase_paths;
900 static int done_pbase_paths_num;
901 static int done_pbase_paths_alloc;
902 static int done_pbase_path_pos(unsigned hash)
905 int hi = done_pbase_paths_num;
907 int mi = (hi + lo) / 2;
908 if (done_pbase_paths[mi] == hash)
910 if (done_pbase_paths[mi] < hash)
918 static int check_pbase_path(unsigned hash)
920 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
924 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
925 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
926 done_pbase_paths = xrealloc(done_pbase_paths,
927 done_pbase_paths_alloc *
930 done_pbase_paths_num++;
931 if (pos < done_pbase_paths_num)
932 memmove(done_pbase_paths + pos + 1,
933 done_pbase_paths + pos,
934 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
935 done_pbase_paths[pos] = hash;
939 static void add_preferred_base_object(const char *name)
941 struct pbase_tree *it;
943 unsigned hash = name_hash(name);
945 if (!num_preferred_base || check_pbase_path(hash))
948 cmplen = name_cmp_len(name);
949 for (it = pbase_tree; it; it = it->next) {
951 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
954 struct tree_desc tree;
955 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
956 add_pbase_object(&tree, name, cmplen, name);
961 static void add_preferred_base(unsigned char *sha1)
963 struct pbase_tree *it;
966 unsigned char tree_sha1[20];
968 if (window <= num_preferred_base++)
971 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
975 for (it = pbase_tree; it; it = it->next) {
976 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
982 it = xcalloc(1, sizeof(*it));
983 it->next = pbase_tree;
986 hashcpy(it->pcache.sha1, tree_sha1);
987 it->pcache.tree_data = data;
988 it->pcache.tree_size = size;
991 static void check_object(struct object_entry *entry)
993 if (entry->in_pack) {
994 struct packed_git *p = entry->in_pack;
995 struct pack_window *w_curs = NULL;
996 const unsigned char *base_ref = NULL;
997 struct object_entry *base_entry;
998 unsigned long used, used_0;
1001 unsigned char *buf, c;
1003 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1006 * We want in_pack_type even if we do not reuse delta
1007 * since non-delta representations could still be reused.
1009 used = unpack_object_header_gently(buf, avail,
1010 &entry->in_pack_type,
1014 * Determine if this is a delta and if so whether we can
1015 * reuse it or not. Otherwise let's find out as cheaply as
1016 * possible what the actual type and size for this object is.
1018 switch (entry->in_pack_type) {
1020 /* Not a delta hence we've already got all we need. */
1021 entry->type = entry->in_pack_type;
1022 entry->in_pack_header_size = used;
1023 unuse_pack(&w_curs);
1026 if (reuse_delta && !entry->preferred_base)
1027 base_ref = use_pack(p, &w_curs,
1028 entry->in_pack_offset + used, NULL);
1029 entry->in_pack_header_size = used + 20;
1032 buf = use_pack(p, &w_curs,
1033 entry->in_pack_offset + used, NULL);
1039 if (!ofs || MSB(ofs, 7))
1040 die("delta base offset overflow in pack for %s",
1041 sha1_to_hex(entry->idx.sha1));
1043 ofs = (ofs << 7) + (c & 127);
1045 if (ofs >= entry->in_pack_offset)
1046 die("delta base offset out of bound for %s",
1047 sha1_to_hex(entry->idx.sha1));
1048 ofs = entry->in_pack_offset - ofs;
1049 if (reuse_delta && !entry->preferred_base) {
1050 struct revindex_entry *revidx;
1051 revidx = find_pack_revindex(p, ofs);
1052 base_ref = nth_packed_object_sha1(p, revidx->nr);
1054 entry->in_pack_header_size = used + used_0;
1058 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1060 * If base_ref was set above that means we wish to
1061 * reuse delta data, and we even found that base
1062 * in the list of objects we want to pack. Goodie!
1064 * Depth value does not matter - find_deltas() will
1065 * never consider reused delta as the base object to
1066 * deltify other objects against, in order to avoid
1069 entry->type = entry->in_pack_type;
1070 entry->delta = base_entry;
1071 entry->delta_sibling = base_entry->delta_child;
1072 base_entry->delta_child = entry;
1073 unuse_pack(&w_curs);
1079 * This must be a delta and we already know what the
1080 * final object type is. Let's extract the actual
1081 * object size from the delta header.
1083 entry->size = get_size_from_delta(p, &w_curs,
1084 entry->in_pack_offset + entry->in_pack_header_size);
1085 unuse_pack(&w_curs);
1090 * No choice but to fall back to the recursive delta walk
1091 * with sha1_object_info() to find about the object type
1094 unuse_pack(&w_curs);
1097 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1099 * The error condition is checked in prepare_pack(). This is
1100 * to permit a missing preferred base object to be ignored
1101 * as a preferred base. Doing so can result in a larger
1102 * pack file, but the transfer will still take place.
1106 static int pack_offset_sort(const void *_a, const void *_b)
1108 const struct object_entry *a = *(struct object_entry **)_a;
1109 const struct object_entry *b = *(struct object_entry **)_b;
1111 /* avoid filesystem trashing with loose objects */
1112 if (!a->in_pack && !b->in_pack)
1113 return hashcmp(a->idx.sha1, b->idx.sha1);
1115 if (a->in_pack < b->in_pack)
1117 if (a->in_pack > b->in_pack)
1119 return a->in_pack_offset < b->in_pack_offset ? -1 :
1120 (a->in_pack_offset > b->in_pack_offset);
1123 static void get_object_details(void)
1126 struct object_entry **sorted_by_offset;
1128 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1129 for (i = 0; i < nr_objects; i++)
1130 sorted_by_offset[i] = objects + i;
1131 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1133 for (i = 0; i < nr_objects; i++)
1134 check_object(sorted_by_offset[i]);
1136 free(sorted_by_offset);
1140 * We search for deltas in a list sorted by type, by filename hash, and then
1141 * by size, so that we see progressively smaller and smaller files.
1142 * That's because we prefer deltas to be from the bigger file
1143 * to the smaller -- deletes are potentially cheaper, but perhaps
1144 * more importantly, the bigger file is likely the more recent
1145 * one. The deepest deltas are therefore the oldest objects which are
1146 * less susceptible to be accessed often.
1148 static int type_size_sort(const void *_a, const void *_b)
1150 const struct object_entry *a = *(struct object_entry **)_a;
1151 const struct object_entry *b = *(struct object_entry **)_b;
1153 if (a->type > b->type)
1155 if (a->type < b->type)
1157 if (a->hash > b->hash)
1159 if (a->hash < b->hash)
1161 if (a->preferred_base > b->preferred_base)
1163 if (a->preferred_base < b->preferred_base)
1165 if (a->size > b->size)
1167 if (a->size < b->size)
1169 return a < b ? -1 : (a > b); /* newest first */
1173 struct object_entry *entry;
1175 struct delta_index *index;
1179 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1180 unsigned long delta_size)
1182 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1185 if (delta_size < cache_max_small_delta_size)
1188 /* cache delta, if objects are large enough compared to delta size */
1189 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1195 #ifdef THREADED_DELTA_SEARCH
1197 static pthread_mutex_t read_mutex = PTHREAD_MUTEX_INITIALIZER;
1198 #define read_lock() pthread_mutex_lock(&read_mutex)
1199 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1201 static pthread_mutex_t cache_mutex = PTHREAD_MUTEX_INITIALIZER;
1202 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1203 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1205 static pthread_mutex_t progress_mutex = PTHREAD_MUTEX_INITIALIZER;
1206 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1207 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1211 #define read_lock() (void)0
1212 #define read_unlock() (void)0
1213 #define cache_lock() (void)0
1214 #define cache_unlock() (void)0
1215 #define progress_lock() (void)0
1216 #define progress_unlock() (void)0
1220 static int try_delta(struct unpacked *trg, struct unpacked *src,
1221 unsigned max_depth, unsigned long *mem_usage)
1223 struct object_entry *trg_entry = trg->entry;
1224 struct object_entry *src_entry = src->entry;
1225 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1227 enum object_type type;
1230 /* Don't bother doing diffs between different types */
1231 if (trg_entry->type != src_entry->type)
1235 * We do not bother to try a delta that we discarded
1236 * on an earlier try, but only when reusing delta data.
1238 if (reuse_delta && trg_entry->in_pack &&
1239 trg_entry->in_pack == src_entry->in_pack &&
1240 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1241 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1244 /* Let's not bust the allowed depth. */
1245 if (src->depth >= max_depth)
1248 /* Now some size filtering heuristics. */
1249 trg_size = trg_entry->size;
1250 if (!trg_entry->delta) {
1251 max_size = trg_size/2 - 20;
1254 max_size = trg_entry->delta_size;
1255 ref_depth = trg->depth;
1257 max_size = max_size * (max_depth - src->depth) /
1258 (max_depth - ref_depth + 1);
1261 src_size = src_entry->size;
1262 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1263 if (sizediff >= max_size)
1265 if (trg_size < src_size / 32)
1268 /* Load data if not already done */
1271 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1274 die("object %s cannot be read",
1275 sha1_to_hex(trg_entry->idx.sha1));
1277 die("object %s inconsistent object length (%lu vs %lu)",
1278 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1283 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1286 die("object %s cannot be read",
1287 sha1_to_hex(src_entry->idx.sha1));
1289 die("object %s inconsistent object length (%lu vs %lu)",
1290 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1294 src->index = create_delta_index(src->data, src_size);
1296 static int warned = 0;
1298 warning("suboptimal pack - out of memory");
1301 *mem_usage += sizeof_delta_index(src->index);
1304 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1308 if (trg_entry->delta) {
1309 /* Prefer only shallower same-sized deltas. */
1310 if (delta_size == trg_entry->delta_size &&
1311 src->depth + 1 >= trg->depth) {
1318 * Handle memory allocation outside of the cache
1319 * accounting lock. Compiler will optimize the strangeness
1320 * away when THREADED_DELTA_SEARCH is not defined.
1322 free(trg_entry->delta_data);
1324 if (trg_entry->delta_data) {
1325 delta_cache_size -= trg_entry->delta_size;
1326 trg_entry->delta_data = NULL;
1328 if (delta_cacheable(src_size, trg_size, delta_size)) {
1329 delta_cache_size += delta_size;
1331 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1337 trg_entry->delta = src_entry;
1338 trg_entry->delta_size = delta_size;
1339 trg->depth = src->depth + 1;
1344 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1346 struct object_entry *child = me->delta_child;
1349 unsigned int c = check_delta_limit(child, n + 1);
1352 child = child->delta_sibling;
1357 static unsigned long free_unpacked(struct unpacked *n)
1359 unsigned long freed_mem = sizeof_delta_index(n->index);
1360 free_delta_index(n->index);
1363 freed_mem += n->entry->size;
1372 static void find_deltas(struct object_entry **list, unsigned *list_size,
1373 int window, int depth, unsigned *processed)
1375 uint32_t i, idx = 0, count = 0;
1376 unsigned int array_size = window * sizeof(struct unpacked);
1377 struct unpacked *array;
1378 unsigned long mem_usage = 0;
1380 array = xmalloc(array_size);
1381 memset(array, 0, array_size);
1384 struct object_entry *entry = *list++;
1385 struct unpacked *n = array + idx;
1386 int j, max_depth, best_base = -1;
1394 if (!entry->preferred_base) {
1396 display_progress(progress_state, *processed);
1400 mem_usage -= free_unpacked(n);
1403 while (window_memory_limit &&
1404 mem_usage > window_memory_limit &&
1406 uint32_t tail = (idx + window - count) % window;
1407 mem_usage -= free_unpacked(array + tail);
1411 /* We do not compute delta to *create* objects we are not
1414 if (entry->preferred_base)
1418 * If the current object is at pack edge, take the depth the
1419 * objects that depend on the current object into account
1420 * otherwise they would become too deep.
1423 if (entry->delta_child) {
1424 max_depth -= check_delta_limit(entry, 0);
1432 uint32_t other_idx = idx + j;
1434 if (other_idx >= window)
1435 other_idx -= window;
1436 m = array + other_idx;
1439 ret = try_delta(n, m, max_depth, &mem_usage);
1443 best_base = other_idx;
1447 * If we decided to cache the delta data, then it is best
1448 * to compress it right away. First because we have to do
1449 * it anyway, and doing it here while we're threaded will
1450 * save a lot of time in the non threaded write phase,
1451 * as well as allow for caching more deltas within
1452 * the same cache size limit.
1454 * But only if not writing to stdout, since in that case
1455 * the network is most likely throttling writes anyway,
1456 * and therefore it is best to go to the write phase ASAP
1457 * instead, as we can afford spending more time compressing
1458 * between writes at that moment.
1460 if (entry->delta_data && !pack_to_stdout) {
1461 entry->z_delta_size = do_compress(&entry->delta_data,
1464 delta_cache_size -= entry->delta_size;
1465 delta_cache_size += entry->z_delta_size;
1469 /* if we made n a delta, and if n is already at max
1470 * depth, leaving it in the window is pointless. we
1471 * should evict it first.
1473 if (entry->delta && max_depth <= n->depth)
1477 * Move the best delta base up in the window, after the
1478 * currently deltified object, to keep it longer. It will
1479 * be the first base object to be attempted next.
1482 struct unpacked swap = array[best_base];
1483 int dist = (window + idx - best_base) % window;
1484 int dst = best_base;
1486 int src = (dst + 1) % window;
1487 array[dst] = array[src];
1495 if (count + 1 < window)
1501 for (i = 0; i < window; ++i) {
1502 free_delta_index(array[i].index);
1503 free(array[i].data);
1508 #ifdef THREADED_DELTA_SEARCH
1511 * The main thread waits on the condition that (at least) one of the workers
1512 * has stopped working (which is indicated in the .working member of
1513 * struct thread_params).
1514 * When a work thread has completed its work, it sets .working to 0 and
1515 * signals the main thread and waits on the condition that .data_ready
1519 struct thread_params {
1521 struct object_entry **list;
1528 pthread_mutex_t mutex;
1529 pthread_cond_t cond;
1530 unsigned *processed;
1533 static pthread_cond_t progress_cond = PTHREAD_COND_INITIALIZER;
1535 static void *threaded_find_deltas(void *arg)
1537 struct thread_params *me = arg;
1539 while (me->remaining) {
1540 find_deltas(me->list, &me->remaining,
1541 me->window, me->depth, me->processed);
1545 pthread_cond_signal(&progress_cond);
1549 * We must not set ->data_ready before we wait on the
1550 * condition because the main thread may have set it to 1
1551 * before we get here. In order to be sure that new
1552 * work is available if we see 1 in ->data_ready, it
1553 * was initialized to 0 before this thread was spawned
1554 * and we reset it to 0 right away.
1556 pthread_mutex_lock(&me->mutex);
1557 while (!me->data_ready)
1558 pthread_cond_wait(&me->cond, &me->mutex);
1560 pthread_mutex_unlock(&me->mutex);
1562 /* leave ->working 1 so that this doesn't get more work assigned */
1566 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1567 int window, int depth, unsigned *processed)
1569 struct thread_params p[delta_search_threads];
1570 int i, ret, active_threads = 0;
1572 if (delta_search_threads <= 1) {
1573 find_deltas(list, &list_size, window, depth, processed);
1577 /* Partition the work amongst work threads. */
1578 for (i = 0; i < delta_search_threads; i++) {
1579 unsigned sub_size = list_size / (delta_search_threads - i);
1581 p[i].window = window;
1583 p[i].processed = processed;
1585 p[i].data_ready = 0;
1587 /* try to split chunks on "path" boundaries */
1588 while (sub_size && sub_size < list_size &&
1589 list[sub_size]->hash &&
1590 list[sub_size]->hash == list[sub_size-1]->hash)
1594 p[i].list_size = sub_size;
1595 p[i].remaining = sub_size;
1598 list_size -= sub_size;
1601 /* Start work threads. */
1602 for (i = 0; i < delta_search_threads; i++) {
1603 if (!p[i].list_size)
1605 pthread_mutex_init(&p[i].mutex, NULL);
1606 pthread_cond_init(&p[i].cond, NULL);
1607 ret = pthread_create(&p[i].thread, NULL,
1608 threaded_find_deltas, &p[i]);
1610 die("unable to create thread: %s", strerror(ret));
1615 * Now let's wait for work completion. Each time a thread is done
1616 * with its work, we steal half of the remaining work from the
1617 * thread with the largest number of unprocessed objects and give
1618 * it to that newly idle thread. This ensure good load balancing
1619 * until the remaining object list segments are simply too short
1620 * to be worth splitting anymore.
1622 while (active_threads) {
1623 struct thread_params *target = NULL;
1624 struct thread_params *victim = NULL;
1625 unsigned sub_size = 0;
1629 for (i = 0; !target && i < delta_search_threads; i++)
1634 pthread_cond_wait(&progress_cond, &progress_mutex);
1637 for (i = 0; i < delta_search_threads; i++)
1638 if (p[i].remaining > 2*window &&
1639 (!victim || victim->remaining < p[i].remaining))
1642 sub_size = victim->remaining / 2;
1643 list = victim->list + victim->list_size - sub_size;
1644 while (sub_size && list[0]->hash &&
1645 list[0]->hash == list[-1]->hash) {
1651 * It is possible for some "paths" to have
1652 * so many objects that no hash boundary
1653 * might be found. Let's just steal the
1654 * exact half in that case.
1656 sub_size = victim->remaining / 2;
1659 target->list = list;
1660 victim->list_size -= sub_size;
1661 victim->remaining -= sub_size;
1663 target->list_size = sub_size;
1664 target->remaining = sub_size;
1665 target->working = 1;
1668 pthread_mutex_lock(&target->mutex);
1669 target->data_ready = 1;
1670 pthread_cond_signal(&target->cond);
1671 pthread_mutex_unlock(&target->mutex);
1674 pthread_join(target->thread, NULL);
1675 pthread_cond_destroy(&target->cond);
1676 pthread_mutex_destroy(&target->mutex);
1683 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1686 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1688 unsigned char peeled[20];
1690 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1691 !peel_ref(path, peeled) && /* peelable? */
1692 !is_null_sha1(peeled) && /* annotated tag? */
1693 locate_object_entry(peeled)) /* object packed? */
1694 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1698 static void prepare_pack(int window, int depth)
1700 struct object_entry **delta_list;
1701 uint32_t i, nr_deltas;
1704 get_object_details();
1706 if (!nr_objects || !window || !depth)
1709 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1712 for (i = 0; i < nr_objects; i++) {
1713 struct object_entry *entry = objects + i;
1716 /* This happens if we decided to reuse existing
1717 * delta from a pack. "reuse_delta &&" is implied.
1721 if (entry->size < 50)
1724 if (entry->no_try_delta)
1727 if (!entry->preferred_base) {
1729 if (entry->type < 0)
1730 die("unable to get type of object %s",
1731 sha1_to_hex(entry->idx.sha1));
1734 delta_list[n++] = entry;
1737 if (nr_deltas && n > 1) {
1738 unsigned nr_done = 0;
1740 progress_state = start_progress("Compressing objects",
1742 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1743 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1744 stop_progress(&progress_state);
1745 if (nr_done != nr_deltas)
1746 die("inconsistency with delta count");
1751 static int git_pack_config(const char *k, const char *v, void *cb)
1753 if(!strcmp(k, "pack.window")) {
1754 window = git_config_int(k, v);
1757 if (!strcmp(k, "pack.windowmemory")) {
1758 window_memory_limit = git_config_ulong(k, v);
1761 if (!strcmp(k, "pack.depth")) {
1762 depth = git_config_int(k, v);
1765 if (!strcmp(k, "pack.compression")) {
1766 int level = git_config_int(k, v);
1768 level = Z_DEFAULT_COMPRESSION;
1769 else if (level < 0 || level > Z_BEST_COMPRESSION)
1770 die("bad pack compression level %d", level);
1771 pack_compression_level = level;
1772 pack_compression_seen = 1;
1775 if (!strcmp(k, "pack.deltacachesize")) {
1776 max_delta_cache_size = git_config_int(k, v);
1779 if (!strcmp(k, "pack.deltacachelimit")) {
1780 cache_max_small_delta_size = git_config_int(k, v);
1783 if (!strcmp(k, "pack.threads")) {
1784 delta_search_threads = git_config_int(k, v);
1785 if (delta_search_threads < 0)
1786 die("invalid number of threads specified (%d)",
1787 delta_search_threads);
1788 #ifndef THREADED_DELTA_SEARCH
1789 if (delta_search_threads != 1)
1790 warning("no threads support, ignoring %s", k);
1794 if (!strcmp(k, "pack.indexversion")) {
1795 pack_idx_default_version = git_config_int(k, v);
1796 if (pack_idx_default_version > 2)
1797 die("bad pack.indexversion=%"PRIu32,
1798 pack_idx_default_version);
1801 if (!strcmp(k, "pack.packsizelimit")) {
1802 pack_size_limit_cfg = git_config_ulong(k, v);
1805 return git_default_config(k, v, cb);
1808 static void read_object_list_from_stdin(void)
1810 char line[40 + 1 + PATH_MAX + 2];
1811 unsigned char sha1[20];
1814 if (!fgets(line, sizeof(line), stdin)) {
1818 die("fgets returned NULL, not EOF, not error!");
1820 die("fgets: %s", strerror(errno));
1824 if (line[0] == '-') {
1825 if (get_sha1_hex(line+1, sha1))
1826 die("expected edge sha1, got garbage:\n %s",
1828 add_preferred_base(sha1);
1831 if (get_sha1_hex(line, sha1))
1832 die("expected sha1, got garbage:\n %s", line);
1834 add_preferred_base_object(line+41);
1835 add_object_entry(sha1, 0, line+41, 0);
1839 #define OBJECT_ADDED (1u<<20)
1841 static void show_commit(struct commit *commit)
1843 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1844 commit->object.flags |= OBJECT_ADDED;
1847 static void show_object(struct object_array_entry *p)
1849 add_preferred_base_object(p->name);
1850 add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1851 p->item->flags |= OBJECT_ADDED;
1854 static void show_edge(struct commit *commit)
1856 add_preferred_base(commit->object.sha1);
1859 struct in_pack_object {
1861 struct object *object;
1867 struct in_pack_object *array;
1870 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1872 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1873 in_pack->array[in_pack->nr].object = object;
1878 * Compare the objects in the offset order, in order to emulate the
1879 * "git-rev-list --objects" output that produced the pack originally.
1881 static int ofscmp(const void *a_, const void *b_)
1883 struct in_pack_object *a = (struct in_pack_object *)a_;
1884 struct in_pack_object *b = (struct in_pack_object *)b_;
1886 if (a->offset < b->offset)
1888 else if (a->offset > b->offset)
1891 return hashcmp(a->object->sha1, b->object->sha1);
1894 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1896 struct packed_git *p;
1897 struct in_pack in_pack;
1900 memset(&in_pack, 0, sizeof(in_pack));
1902 for (p = packed_git; p; p = p->next) {
1903 const unsigned char *sha1;
1906 for (i = 0; i < revs->num_ignore_packed; i++) {
1907 if (matches_pack_name(p, revs->ignore_packed[i]))
1910 if (revs->num_ignore_packed <= i)
1912 if (open_pack_index(p))
1913 die("cannot open pack index");
1915 ALLOC_GROW(in_pack.array,
1916 in_pack.nr + p->num_objects,
1919 for (i = 0; i < p->num_objects; i++) {
1920 sha1 = nth_packed_object_sha1(p, i);
1921 o = lookup_unknown_object(sha1);
1922 if (!(o->flags & OBJECT_ADDED))
1923 mark_in_pack_object(o, p, &in_pack);
1924 o->flags |= OBJECT_ADDED;
1929 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
1931 for (i = 0; i < in_pack.nr; i++) {
1932 struct object *o = in_pack.array[i].object;
1933 add_object_entry(o->sha1, o->type, "", 0);
1936 free(in_pack.array);
1939 static void loosen_unused_packed_objects(struct rev_info *revs)
1941 struct packed_git *p;
1943 const unsigned char *sha1;
1945 for (p = packed_git; p; p = p->next) {
1946 for (i = 0; i < revs->num_ignore_packed; i++) {
1947 if (matches_pack_name(p, revs->ignore_packed[i]))
1950 if (revs->num_ignore_packed <= i)
1953 if (open_pack_index(p))
1954 die("cannot open pack index");
1956 for (i = 0; i < p->num_objects; i++) {
1957 sha1 = nth_packed_object_sha1(p, i);
1958 if (!locate_object_entry(sha1))
1959 if (force_object_loose(sha1, p->mtime))
1960 die("unable to force loose object");
1965 static void get_object_list(int ac, const char **av)
1967 struct rev_info revs;
1971 init_revisions(&revs, NULL);
1972 save_commit_buffer = 0;
1973 setup_revisions(ac, av, &revs, NULL);
1975 while (fgets(line, sizeof(line), stdin) != NULL) {
1976 int len = strlen(line);
1977 if (len && line[len - 1] == '\n')
1982 if (!strcmp(line, "--not")) {
1983 flags ^= UNINTERESTING;
1986 die("not a rev '%s'", line);
1988 if (handle_revision_arg(line, &revs, flags, 1))
1989 die("bad revision '%s'", line);
1992 if (prepare_revision_walk(&revs))
1993 die("revision walk setup failed");
1994 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1995 traverse_commit_list(&revs, show_commit, show_object);
1997 if (keep_unreachable)
1998 add_objects_in_unpacked_packs(&revs);
1999 if (unpack_unreachable)
2000 loosen_unused_packed_objects(&revs);
2003 static int adjust_perm(const char *path, mode_t mode)
2005 if (chmod(path, mode))
2007 return adjust_shared_perm(path);
2010 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2012 int use_internal_rev_list = 0;
2016 int rp_ac_alloc = 64;
2019 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2021 rp_av[0] = "pack-objects";
2022 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2025 git_config(git_pack_config, NULL);
2026 if (!pack_compression_seen && core_compression_seen)
2027 pack_compression_level = core_compression_level;
2029 progress = isatty(2);
2030 for (i = 1; i < argc; i++) {
2031 const char *arg = argv[i];
2036 if (!strcmp("--non-empty", arg)) {
2040 if (!strcmp("--local", arg)) {
2044 if (!strcmp("--incremental", arg)) {
2048 if (!prefixcmp(arg, "--compression=")) {
2050 int level = strtoul(arg+14, &end, 0);
2051 if (!arg[14] || *end)
2054 level = Z_DEFAULT_COMPRESSION;
2055 else if (level < 0 || level > Z_BEST_COMPRESSION)
2056 die("bad pack compression level %d", level);
2057 pack_compression_level = level;
2060 if (!prefixcmp(arg, "--max-pack-size=")) {
2062 pack_size_limit_cfg = 0;
2063 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2064 if (!arg[16] || *end)
2068 if (!prefixcmp(arg, "--window=")) {
2070 window = strtoul(arg+9, &end, 0);
2071 if (!arg[9] || *end)
2075 if (!prefixcmp(arg, "--window-memory=")) {
2076 if (!git_parse_ulong(arg+16, &window_memory_limit))
2080 if (!prefixcmp(arg, "--threads=")) {
2082 delta_search_threads = strtoul(arg+10, &end, 0);
2083 if (!arg[10] || *end || delta_search_threads < 0)
2085 #ifndef THREADED_DELTA_SEARCH
2086 if (delta_search_threads != 1)
2087 warning("no threads support, "
2088 "ignoring %s", arg);
2092 if (!prefixcmp(arg, "--depth=")) {
2094 depth = strtoul(arg+8, &end, 0);
2095 if (!arg[8] || *end)
2099 if (!strcmp("--progress", arg)) {
2103 if (!strcmp("--all-progress", arg)) {
2107 if (!strcmp("-q", arg)) {
2111 if (!strcmp("--no-reuse-delta", arg)) {
2115 if (!strcmp("--no-reuse-object", arg)) {
2116 reuse_object = reuse_delta = 0;
2119 if (!strcmp("--delta-base-offset", arg)) {
2120 allow_ofs_delta = 1;
2123 if (!strcmp("--stdout", arg)) {
2127 if (!strcmp("--revs", arg)) {
2128 use_internal_rev_list = 1;
2131 if (!strcmp("--keep-unreachable", arg)) {
2132 keep_unreachable = 1;
2135 if (!strcmp("--unpack-unreachable", arg)) {
2136 unpack_unreachable = 1;
2139 if (!strcmp("--include-tag", arg)) {
2143 if (!strcmp("--unpacked", arg) ||
2144 !prefixcmp(arg, "--unpacked=") ||
2145 !strcmp("--reflog", arg) ||
2146 !strcmp("--all", arg)) {
2147 use_internal_rev_list = 1;
2148 if (rp_ac >= rp_ac_alloc - 1) {
2149 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2150 rp_av = xrealloc(rp_av,
2151 rp_ac_alloc * sizeof(*rp_av));
2153 rp_av[rp_ac++] = arg;
2156 if (!strcmp("--thin", arg)) {
2157 use_internal_rev_list = 1;
2159 rp_av[1] = "--objects-edge";
2162 if (!prefixcmp(arg, "--index-version=")) {
2164 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2165 if (pack_idx_default_version > 2)
2168 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2169 if (*c || pack_idx_off32_limit & 0x80000000)
2176 /* Traditionally "pack-objects [options] base extra" failed;
2177 * we would however want to take refs parameter that would
2178 * have been given to upstream rev-list ourselves, which means
2179 * we somehow want to say what the base name is. So the
2182 * pack-objects [options] base <refs...>
2184 * in other words, we would treat the first non-option as the
2185 * base_name and send everything else to the internal revision
2189 if (!pack_to_stdout)
2190 base_name = argv[i++];
2192 if (pack_to_stdout != !base_name)
2195 if (!pack_to_stdout && !pack_size_limit)
2196 pack_size_limit = pack_size_limit_cfg;
2198 if (pack_to_stdout && pack_size_limit)
2199 die("--max-pack-size cannot be used to build a pack for transfer.");
2201 if (!pack_to_stdout && thin)
2202 die("--thin cannot be used to build an indexable pack.");
2204 if (keep_unreachable && unpack_unreachable)
2205 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2207 #ifdef THREADED_DELTA_SEARCH
2208 if (!delta_search_threads) /* --threads=0 means autodetect */
2209 delta_search_threads = online_cpus();
2212 prepare_packed_git();
2215 progress_state = start_progress("Counting objects", 0);
2216 if (!use_internal_rev_list)
2217 read_object_list_from_stdin();
2219 rp_av[rp_ac] = NULL;
2220 get_object_list(rp_ac, rp_av);
2222 if (include_tag && nr_result)
2223 for_each_ref(add_ref_tag, NULL);
2224 stop_progress(&progress_state);
2226 if (non_empty && !nr_result)
2229 prepare_pack(window, depth);
2232 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2233 " reused %"PRIu32" (delta %"PRIu32")\n",
2234 written, written_delta, reused, reused_delta);