11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
21 #include "thread-utils.h"
25 static const char pack_usage[] =
26 "git pack-objects [{ -q | --progress | --all-progress }]\n"
27 " [--all-progress-implied]\n"
28 " [--max-pack-size=N] [--local] [--incremental]\n"
29 " [--window=N] [--window-memory=N] [--depth=N]\n"
30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
31 " [--threads=N] [--non-empty] [--revs [--unpacked | --all]*]\n"
32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
33 " [--keep-unreachable | --unpack-unreachable \n"
34 " [<ref-list | <object-list]";
37 struct pack_idx_entry idx;
38 unsigned long size; /* uncompressed size */
39 struct packed_git *in_pack; /* already in pack */
41 struct object_entry *delta; /* delta base object */
42 struct object_entry *delta_child; /* deltified objects who bases me */
43 struct object_entry *delta_sibling; /* other deltified objects who
44 * uses the same base as me
46 void *delta_data; /* cached delta (uncompressed) */
47 unsigned long delta_size; /* delta data size (uncompressed) */
48 unsigned long z_delta_size; /* delta data size (compressed) */
49 unsigned int hash; /* name hint hash */
50 enum object_type type;
51 enum object_type in_pack_type; /* could be delta */
52 unsigned char in_pack_header_size;
53 unsigned char preferred_base; /* we do not pack this, but is available
54 * to be used as the base object to delta
57 unsigned char no_try_delta;
61 * Objects we are going to pack are collected in objects array (dynamically
62 * expanded). nr_objects & nr_alloc controls this array. They are stored
63 * in the order we see -- typically rev-list --objects order that gives us
64 * nice "minimum seek" order.
66 static struct object_entry *objects;
67 static struct pack_idx_entry **written_list;
68 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
71 static int reuse_delta = 1, reuse_object = 1;
72 static int keep_unreachable, unpack_unreachable, include_tag;
74 static int incremental;
75 static int ignore_packed_keep;
76 static int allow_ofs_delta;
77 static const char *base_name;
78 static int progress = 1;
79 static int window = 10;
80 static unsigned long pack_size_limit, pack_size_limit_cfg;
81 static int depth = 50;
82 static int delta_search_threads;
83 static int pack_to_stdout;
84 static int num_preferred_base;
85 static struct progress *progress_state;
86 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
87 static int pack_compression_seen;
89 static unsigned long delta_cache_size = 0;
90 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
91 static unsigned long cache_max_small_delta_size = 1000;
93 static unsigned long window_memory_limit = 0;
96 * The object names in objects array are hashed with this hashtable,
97 * to help looking up the entry by object name.
98 * This hashtable is built after all the objects are seen.
100 static int *object_ix;
101 static int object_ix_hashsz;
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
110 static void *get_delta(struct object_entry *entry)
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
114 enum object_type type;
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
122 delta_buf = diff_delta(base_buf, base_size,
123 buf, size, &delta_size, 0);
124 if (!delta_buf || delta_size != entry->delta_size)
125 die("delta size changed");
131 static unsigned long do_compress(void **pptr, unsigned long size)
135 unsigned long maxsize;
137 memset(&stream, 0, sizeof(stream));
138 deflateInit(&stream, pack_compression_level);
139 maxsize = deflateBound(&stream, size);
142 out = xmalloc(maxsize);
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (deflate(&stream, Z_FINISH) == Z_OK)
154 return stream.total_out;
158 * we are going to reuse the existing object data as is. make
159 * sure it is not corrupt.
161 static int check_pack_inflate(struct packed_git *p,
162 struct pack_window **w_curs,
165 unsigned long expect)
168 unsigned char fakebuf[4096], *in;
171 memset(&stream, 0, sizeof(stream));
172 git_inflate_init(&stream);
174 in = use_pack(p, w_curs, offset, &stream.avail_in);
176 stream.next_out = fakebuf;
177 stream.avail_out = sizeof(fakebuf);
178 st = git_inflate(&stream, Z_FINISH);
179 offset += stream.next_in - in;
180 } while (st == Z_OK || st == Z_BUF_ERROR);
181 git_inflate_end(&stream);
182 return (st == Z_STREAM_END &&
183 stream.total_out == expect &&
184 stream.total_in == len) ? 0 : -1;
187 static void copy_pack_data(struct sha1file *f,
188 struct packed_git *p,
189 struct pack_window **w_curs,
197 in = use_pack(p, w_curs, offset, &avail);
199 avail = (unsigned int)len;
200 sha1write(f, in, avail);
206 static unsigned long write_object(struct sha1file *f,
207 struct object_entry *entry,
210 unsigned long size, limit, datalen;
212 unsigned char header[10], dheader[10];
214 enum object_type type;
215 int usable_delta, to_reuse;
222 /* apply size limit if limited packsize and not first object */
223 if (!pack_size_limit || !nr_written)
225 else if (pack_size_limit <= write_offset)
227 * the earlier object did not fit the limit; avoid
228 * mistaking this with unlimited (i.e. limit = 0).
232 limit = pack_size_limit - write_offset;
235 usable_delta = 0; /* no delta */
236 else if (!pack_size_limit)
237 usable_delta = 1; /* unlimited packfile */
238 else if (entry->delta->idx.offset == (off_t)-1)
239 usable_delta = 0; /* base was written to another pack */
240 else if (entry->delta->idx.offset)
241 usable_delta = 1; /* base already exists in this pack */
243 usable_delta = 0; /* base could end up in another pack */
246 to_reuse = 0; /* explicit */
247 else if (!entry->in_pack)
248 to_reuse = 0; /* can't reuse what we don't have */
249 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
250 /* check_object() decided it for us ... */
251 to_reuse = usable_delta;
252 /* ... but pack split may override that */
253 else if (type != entry->in_pack_type)
254 to_reuse = 0; /* pack has delta which is unusable */
255 else if (entry->delta)
256 to_reuse = 0; /* we want to pack afresh */
258 to_reuse = 1; /* we have it in-pack undeltified,
259 * and we do not need to deltify it.
265 buf = read_sha1_file(entry->idx.sha1, &type, &size);
267 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
269 * make sure no cached delta data remains from a
270 * previous attempt before a pack split occurred.
272 free(entry->delta_data);
273 entry->delta_data = NULL;
274 entry->z_delta_size = 0;
275 } else if (entry->delta_data) {
276 size = entry->delta_size;
277 buf = entry->delta_data;
278 entry->delta_data = NULL;
279 type = (allow_ofs_delta && entry->delta->idx.offset) ?
280 OBJ_OFS_DELTA : OBJ_REF_DELTA;
282 buf = get_delta(entry);
283 size = entry->delta_size;
284 type = (allow_ofs_delta && entry->delta->idx.offset) ?
285 OBJ_OFS_DELTA : OBJ_REF_DELTA;
288 if (entry->z_delta_size)
289 datalen = entry->z_delta_size;
291 datalen = do_compress(&buf, size);
294 * The object header is a byte of 'type' followed by zero or
295 * more bytes of length.
297 hdrlen = encode_in_pack_object_header(type, size, header);
299 if (type == OBJ_OFS_DELTA) {
301 * Deltas with relative base contain an additional
302 * encoding of the relative offset for the delta
303 * base from this object's position in the pack.
305 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
306 unsigned pos = sizeof(dheader) - 1;
307 dheader[pos] = ofs & 127;
309 dheader[--pos] = 128 | (--ofs & 127);
310 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
314 sha1write(f, header, hdrlen);
315 sha1write(f, dheader + pos, sizeof(dheader) - pos);
316 hdrlen += sizeof(dheader) - pos;
317 } else if (type == OBJ_REF_DELTA) {
319 * Deltas with a base reference contain
320 * an additional 20 bytes for the base sha1.
322 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
326 sha1write(f, header, hdrlen);
327 sha1write(f, entry->delta->idx.sha1, 20);
330 if (limit && hdrlen + datalen + 20 >= limit) {
334 sha1write(f, header, hdrlen);
336 sha1write(f, buf, datalen);
340 struct packed_git *p = entry->in_pack;
341 struct pack_window *w_curs = NULL;
342 struct revindex_entry *revidx;
346 type = (allow_ofs_delta && entry->delta->idx.offset) ?
347 OBJ_OFS_DELTA : OBJ_REF_DELTA;
348 hdrlen = encode_in_pack_object_header(type, entry->size, header);
350 offset = entry->in_pack_offset;
351 revidx = find_pack_revindex(p, offset);
352 datalen = revidx[1].offset - offset;
353 if (!pack_to_stdout && p->index_version > 1 &&
354 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
355 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
360 offset += entry->in_pack_header_size;
361 datalen -= entry->in_pack_header_size;
362 if (!pack_to_stdout && p->index_version == 1 &&
363 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
364 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
369 if (type == OBJ_OFS_DELTA) {
370 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
371 unsigned pos = sizeof(dheader) - 1;
372 dheader[pos] = ofs & 127;
374 dheader[--pos] = 128 | (--ofs & 127);
375 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
379 sha1write(f, header, hdrlen);
380 sha1write(f, dheader + pos, sizeof(dheader) - pos);
381 hdrlen += sizeof(dheader) - pos;
383 } else if (type == OBJ_REF_DELTA) {
384 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
388 sha1write(f, header, hdrlen);
389 sha1write(f, entry->delta->idx.sha1, 20);
393 if (limit && hdrlen + datalen + 20 >= limit) {
397 sha1write(f, header, hdrlen);
399 copy_pack_data(f, p, &w_curs, offset, datalen);
407 entry->idx.crc32 = crc32_end(f);
408 return hdrlen + datalen;
411 static int write_one(struct sha1file *f,
412 struct object_entry *e,
417 /* offset is non zero if object is written already. */
418 if (e->idx.offset || e->preferred_base)
421 /* if we are deltified, write out base object first. */
422 if (e->delta && !write_one(f, e->delta, offset))
425 e->idx.offset = *offset;
426 size = write_object(f, e, *offset);
431 written_list[nr_written++] = &e->idx;
433 /* make sure off_t is sufficiently large not to wrap */
434 if (*offset > *offset + size)
435 die("pack too large for current definition of off_t");
440 /* forward declaration for write_pack_file */
441 static int adjust_perm(const char *path, mode_t mode);
443 static void write_pack_file(void)
448 struct pack_header hdr;
449 uint32_t nr_remaining = nr_result;
450 time_t last_mtime = 0;
452 if (progress > pack_to_stdout)
453 progress_state = start_progress("Writing objects", nr_result);
454 written_list = xmalloc(nr_objects * sizeof(*written_list));
457 unsigned char sha1[20];
458 char *pack_tmp_name = NULL;
460 if (pack_to_stdout) {
461 f = sha1fd_throughput(1, "<stdout>", progress_state);
463 char tmpname[PATH_MAX];
465 fd = odb_mkstemp(tmpname, sizeof(tmpname),
466 "pack/tmp_pack_XXXXXX");
467 pack_tmp_name = xstrdup(tmpname);
468 f = sha1fd(fd, pack_tmp_name);
471 hdr.hdr_signature = htonl(PACK_SIGNATURE);
472 hdr.hdr_version = htonl(PACK_VERSION);
473 hdr.hdr_entries = htonl(nr_remaining);
474 sha1write(f, &hdr, sizeof(hdr));
475 offset = sizeof(hdr);
477 for (; i < nr_objects; i++) {
478 if (!write_one(f, objects + i, &offset))
480 display_progress(progress_state, written);
484 * Did we write the wrong # entries in the header?
485 * If so, rewrite it like in fast-import
487 if (pack_to_stdout) {
488 sha1close(f, sha1, CSUM_CLOSE);
489 } else if (nr_written == nr_remaining) {
490 sha1close(f, sha1, CSUM_FSYNC);
492 int fd = sha1close(f, sha1, 0);
493 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
494 nr_written, sha1, offset);
498 if (!pack_to_stdout) {
499 mode_t mode = umask(0);
501 const char *idx_tmp_name;
502 char tmpname[PATH_MAX];
507 idx_tmp_name = write_idx_file(NULL, written_list,
510 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
511 base_name, sha1_to_hex(sha1));
512 free_pack_by_name(tmpname);
513 if (adjust_perm(pack_tmp_name, mode))
514 die_errno("unable to make temporary pack file readable");
515 if (rename(pack_tmp_name, tmpname))
516 die_errno("unable to rename temporary pack file");
519 * Packs are runtime accessed in their mtime
520 * order since newer packs are more likely to contain
521 * younger objects. So if we are creating multiple
522 * packs then we should modify the mtime of later ones
523 * to preserve this property.
525 if (stat(tmpname, &st) < 0) {
526 warning("failed to stat %s: %s",
527 tmpname, strerror(errno));
528 } else if (!last_mtime) {
529 last_mtime = st.st_mtime;
532 utb.actime = st.st_atime;
533 utb.modtime = --last_mtime;
534 if (utime(tmpname, &utb) < 0)
535 warning("failed utime() on %s: %s",
536 tmpname, strerror(errno));
539 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
540 base_name, sha1_to_hex(sha1));
541 if (adjust_perm(idx_tmp_name, mode))
542 die_errno("unable to make temporary index file readable");
543 if (rename(idx_tmp_name, tmpname))
544 die_errno("unable to rename temporary index file");
546 free((void *) idx_tmp_name);
548 puts(sha1_to_hex(sha1));
551 /* mark written objects as written to previous pack */
552 for (j = 0; j < nr_written; j++) {
553 written_list[j]->offset = (off_t)-1;
555 nr_remaining -= nr_written;
556 } while (nr_remaining && i < nr_objects);
559 stop_progress(&progress_state);
560 if (written != nr_result)
561 die("wrote %"PRIu32" objects while expecting %"PRIu32,
565 static int locate_object_entry_hash(const unsigned char *sha1)
569 memcpy(&ui, sha1, sizeof(unsigned int));
570 i = ui % object_ix_hashsz;
571 while (0 < object_ix[i]) {
572 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
574 if (++i == object_ix_hashsz)
580 static struct object_entry *locate_object_entry(const unsigned char *sha1)
584 if (!object_ix_hashsz)
587 i = locate_object_entry_hash(sha1);
589 return &objects[object_ix[i]-1];
593 static void rehash_objects(void)
596 struct object_entry *oe;
598 object_ix_hashsz = nr_objects * 3;
599 if (object_ix_hashsz < 1024)
600 object_ix_hashsz = 1024;
601 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
602 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
603 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
604 int ix = locate_object_entry_hash(oe->idx.sha1);
608 object_ix[ix] = i + 1;
612 static unsigned name_hash(const char *name)
614 unsigned c, hash = 0;
620 * This effectively just creates a sortable number from the
621 * last sixteen non-whitespace characters. Last characters
622 * count "most", so things that end in ".c" sort together.
624 while ((c = *name++) != 0) {
627 hash = (hash >> 2) + (c << 24);
632 static void setup_delta_attr_check(struct git_attr_check *check)
634 static struct git_attr *attr_delta;
637 attr_delta = git_attr("delta");
639 check[0].attr = attr_delta;
642 static int no_try_delta(const char *path)
644 struct git_attr_check check[1];
646 setup_delta_attr_check(check);
647 if (git_checkattr(path, ARRAY_SIZE(check), check))
649 if (ATTR_FALSE(check->value))
654 static int add_object_entry(const unsigned char *sha1, enum object_type type,
655 const char *name, int exclude)
657 struct object_entry *entry;
658 struct packed_git *p, *found_pack = NULL;
659 off_t found_offset = 0;
661 unsigned hash = name_hash(name);
663 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
666 entry = objects + object_ix[ix] - 1;
667 if (!entry->preferred_base)
669 entry->preferred_base = 1;
674 if (!exclude && local && has_loose_object_nonlocal(sha1))
677 for (p = packed_git; p; p = p->next) {
678 off_t offset = find_pack_entry_one(sha1, p);
681 found_offset = offset;
688 if (local && !p->pack_local)
690 if (ignore_packed_keep && p->pack_local && p->pack_keep)
695 if (nr_objects >= nr_alloc) {
696 nr_alloc = (nr_alloc + 1024) * 3 / 2;
697 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
700 entry = objects + nr_objects++;
701 memset(entry, 0, sizeof(*entry));
702 hashcpy(entry->idx.sha1, sha1);
707 entry->preferred_base = 1;
711 entry->in_pack = found_pack;
712 entry->in_pack_offset = found_offset;
715 if (object_ix_hashsz * 3 <= nr_objects * 4)
718 object_ix[-1 - ix] = nr_objects;
720 display_progress(progress_state, nr_objects);
722 if (name && no_try_delta(name))
723 entry->no_try_delta = 1;
728 struct pbase_tree_cache {
729 unsigned char sha1[20];
733 unsigned long tree_size;
736 static struct pbase_tree_cache *(pbase_tree_cache[256]);
737 static int pbase_tree_cache_ix(const unsigned char *sha1)
739 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
741 static int pbase_tree_cache_ix_incr(int ix)
743 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
746 static struct pbase_tree {
747 struct pbase_tree *next;
748 /* This is a phony "cache" entry; we are not
749 * going to evict it nor find it through _get()
750 * mechanism -- this is for the toplevel node that
751 * would almost always change with any commit.
753 struct pbase_tree_cache pcache;
756 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
758 struct pbase_tree_cache *ent, *nent;
761 enum object_type type;
763 int my_ix = pbase_tree_cache_ix(sha1);
764 int available_ix = -1;
766 /* pbase-tree-cache acts as a limited hashtable.
767 * your object will be found at your index or within a few
768 * slots after that slot if it is cached.
770 for (neigh = 0; neigh < 8; neigh++) {
771 ent = pbase_tree_cache[my_ix];
772 if (ent && !hashcmp(ent->sha1, sha1)) {
776 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
777 ((0 <= available_ix) &&
778 (!ent && pbase_tree_cache[available_ix])))
779 available_ix = my_ix;
782 my_ix = pbase_tree_cache_ix_incr(my_ix);
785 /* Did not find one. Either we got a bogus request or
786 * we need to read and perhaps cache.
788 data = read_sha1_file(sha1, &type, &size);
791 if (type != OBJ_TREE) {
796 /* We need to either cache or return a throwaway copy */
798 if (available_ix < 0)
801 ent = pbase_tree_cache[available_ix];
802 my_ix = available_ix;
806 nent = xmalloc(sizeof(*nent));
807 nent->temporary = (available_ix < 0);
810 /* evict and reuse */
811 free(ent->tree_data);
814 hashcpy(nent->sha1, sha1);
815 nent->tree_data = data;
816 nent->tree_size = size;
818 if (!nent->temporary)
819 pbase_tree_cache[my_ix] = nent;
823 static void pbase_tree_put(struct pbase_tree_cache *cache)
825 if (!cache->temporary) {
829 free(cache->tree_data);
833 static int name_cmp_len(const char *name)
836 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
841 static void add_pbase_object(struct tree_desc *tree,
844 const char *fullname)
846 struct name_entry entry;
849 while (tree_entry(tree,&entry)) {
850 if (S_ISGITLINK(entry.mode))
852 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
853 memcmp(name, entry.path, cmplen);
858 if (name[cmplen] != '/') {
859 add_object_entry(entry.sha1,
860 object_type(entry.mode),
864 if (S_ISDIR(entry.mode)) {
865 struct tree_desc sub;
866 struct pbase_tree_cache *tree;
867 const char *down = name+cmplen+1;
868 int downlen = name_cmp_len(down);
870 tree = pbase_tree_get(entry.sha1);
873 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
875 add_pbase_object(&sub, down, downlen, fullname);
876 pbase_tree_put(tree);
881 static unsigned *done_pbase_paths;
882 static int done_pbase_paths_num;
883 static int done_pbase_paths_alloc;
884 static int done_pbase_path_pos(unsigned hash)
887 int hi = done_pbase_paths_num;
889 int mi = (hi + lo) / 2;
890 if (done_pbase_paths[mi] == hash)
892 if (done_pbase_paths[mi] < hash)
900 static int check_pbase_path(unsigned hash)
902 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
906 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
907 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
908 done_pbase_paths = xrealloc(done_pbase_paths,
909 done_pbase_paths_alloc *
912 done_pbase_paths_num++;
913 if (pos < done_pbase_paths_num)
914 memmove(done_pbase_paths + pos + 1,
915 done_pbase_paths + pos,
916 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
917 done_pbase_paths[pos] = hash;
921 static void add_preferred_base_object(const char *name)
923 struct pbase_tree *it;
925 unsigned hash = name_hash(name);
927 if (!num_preferred_base || check_pbase_path(hash))
930 cmplen = name_cmp_len(name);
931 for (it = pbase_tree; it; it = it->next) {
933 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
936 struct tree_desc tree;
937 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
938 add_pbase_object(&tree, name, cmplen, name);
943 static void add_preferred_base(unsigned char *sha1)
945 struct pbase_tree *it;
948 unsigned char tree_sha1[20];
950 if (window <= num_preferred_base++)
953 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
957 for (it = pbase_tree; it; it = it->next) {
958 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
964 it = xcalloc(1, sizeof(*it));
965 it->next = pbase_tree;
968 hashcpy(it->pcache.sha1, tree_sha1);
969 it->pcache.tree_data = data;
970 it->pcache.tree_size = size;
973 static void cleanup_preferred_base(void)
975 struct pbase_tree *it;
981 struct pbase_tree *this = it;
983 free(this->pcache.tree_data);
987 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
988 if (!pbase_tree_cache[i])
990 free(pbase_tree_cache[i]->tree_data);
991 free(pbase_tree_cache[i]);
992 pbase_tree_cache[i] = NULL;
995 free(done_pbase_paths);
996 done_pbase_paths = NULL;
997 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1000 static void check_object(struct object_entry *entry)
1002 if (entry->in_pack) {
1003 struct packed_git *p = entry->in_pack;
1004 struct pack_window *w_curs = NULL;
1005 const unsigned char *base_ref = NULL;
1006 struct object_entry *base_entry;
1007 unsigned long used, used_0;
1010 unsigned char *buf, c;
1012 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1015 * We want in_pack_type even if we do not reuse delta
1016 * since non-delta representations could still be reused.
1018 used = unpack_object_header_buffer(buf, avail,
1019 &entry->in_pack_type,
1025 * Determine if this is a delta and if so whether we can
1026 * reuse it or not. Otherwise let's find out as cheaply as
1027 * possible what the actual type and size for this object is.
1029 switch (entry->in_pack_type) {
1031 /* Not a delta hence we've already got all we need. */
1032 entry->type = entry->in_pack_type;
1033 entry->in_pack_header_size = used;
1034 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1036 unuse_pack(&w_curs);
1039 if (reuse_delta && !entry->preferred_base)
1040 base_ref = use_pack(p, &w_curs,
1041 entry->in_pack_offset + used, NULL);
1042 entry->in_pack_header_size = used + 20;
1045 buf = use_pack(p, &w_curs,
1046 entry->in_pack_offset + used, NULL);
1052 if (!ofs || MSB(ofs, 7)) {
1053 error("delta base offset overflow in pack for %s",
1054 sha1_to_hex(entry->idx.sha1));
1058 ofs = (ofs << 7) + (c & 127);
1060 ofs = entry->in_pack_offset - ofs;
1061 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1062 error("delta base offset out of bound for %s",
1063 sha1_to_hex(entry->idx.sha1));
1066 if (reuse_delta && !entry->preferred_base) {
1067 struct revindex_entry *revidx;
1068 revidx = find_pack_revindex(p, ofs);
1071 base_ref = nth_packed_object_sha1(p, revidx->nr);
1073 entry->in_pack_header_size = used + used_0;
1077 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1079 * If base_ref was set above that means we wish to
1080 * reuse delta data, and we even found that base
1081 * in the list of objects we want to pack. Goodie!
1083 * Depth value does not matter - find_deltas() will
1084 * never consider reused delta as the base object to
1085 * deltify other objects against, in order to avoid
1088 entry->type = entry->in_pack_type;
1089 entry->delta = base_entry;
1090 entry->delta_size = entry->size;
1091 entry->delta_sibling = base_entry->delta_child;
1092 base_entry->delta_child = entry;
1093 unuse_pack(&w_curs);
1099 * This must be a delta and we already know what the
1100 * final object type is. Let's extract the actual
1101 * object size from the delta header.
1103 entry->size = get_size_from_delta(p, &w_curs,
1104 entry->in_pack_offset + entry->in_pack_header_size);
1105 if (entry->size == 0)
1107 unuse_pack(&w_curs);
1112 * No choice but to fall back to the recursive delta walk
1113 * with sha1_object_info() to find about the object type
1117 unuse_pack(&w_curs);
1120 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1122 * The error condition is checked in prepare_pack(). This is
1123 * to permit a missing preferred base object to be ignored
1124 * as a preferred base. Doing so can result in a larger
1125 * pack file, but the transfer will still take place.
1129 static int pack_offset_sort(const void *_a, const void *_b)
1131 const struct object_entry *a = *(struct object_entry **)_a;
1132 const struct object_entry *b = *(struct object_entry **)_b;
1134 /* avoid filesystem trashing with loose objects */
1135 if (!a->in_pack && !b->in_pack)
1136 return hashcmp(a->idx.sha1, b->idx.sha1);
1138 if (a->in_pack < b->in_pack)
1140 if (a->in_pack > b->in_pack)
1142 return a->in_pack_offset < b->in_pack_offset ? -1 :
1143 (a->in_pack_offset > b->in_pack_offset);
1146 static void get_object_details(void)
1149 struct object_entry **sorted_by_offset;
1151 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1152 for (i = 0; i < nr_objects; i++)
1153 sorted_by_offset[i] = objects + i;
1154 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1156 for (i = 0; i < nr_objects; i++)
1157 check_object(sorted_by_offset[i]);
1159 free(sorted_by_offset);
1163 * We search for deltas in a list sorted by type, by filename hash, and then
1164 * by size, so that we see progressively smaller and smaller files.
1165 * That's because we prefer deltas to be from the bigger file
1166 * to the smaller -- deletes are potentially cheaper, but perhaps
1167 * more importantly, the bigger file is likely the more recent
1168 * one. The deepest deltas are therefore the oldest objects which are
1169 * less susceptible to be accessed often.
1171 static int type_size_sort(const void *_a, const void *_b)
1173 const struct object_entry *a = *(struct object_entry **)_a;
1174 const struct object_entry *b = *(struct object_entry **)_b;
1176 if (a->type > b->type)
1178 if (a->type < b->type)
1180 if (a->hash > b->hash)
1182 if (a->hash < b->hash)
1184 if (a->preferred_base > b->preferred_base)
1186 if (a->preferred_base < b->preferred_base)
1188 if (a->size > b->size)
1190 if (a->size < b->size)
1192 return a < b ? -1 : (a > b); /* newest first */
1196 struct object_entry *entry;
1198 struct delta_index *index;
1202 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1203 unsigned long delta_size)
1205 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1208 if (delta_size < cache_max_small_delta_size)
1211 /* cache delta, if objects are large enough compared to delta size */
1212 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1220 static pthread_mutex_t read_mutex;
1221 #define read_lock() pthread_mutex_lock(&read_mutex)
1222 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1224 static pthread_mutex_t cache_mutex;
1225 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1226 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1228 static pthread_mutex_t progress_mutex;
1229 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1230 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1234 #define read_lock() (void)0
1235 #define read_unlock() (void)0
1236 #define cache_lock() (void)0
1237 #define cache_unlock() (void)0
1238 #define progress_lock() (void)0
1239 #define progress_unlock() (void)0
1243 static int try_delta(struct unpacked *trg, struct unpacked *src,
1244 unsigned max_depth, unsigned long *mem_usage)
1246 struct object_entry *trg_entry = trg->entry;
1247 struct object_entry *src_entry = src->entry;
1248 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1250 enum object_type type;
1253 /* Don't bother doing diffs between different types */
1254 if (trg_entry->type != src_entry->type)
1258 * We do not bother to try a delta that we discarded
1259 * on an earlier try, but only when reusing delta data.
1261 if (reuse_delta && trg_entry->in_pack &&
1262 trg_entry->in_pack == src_entry->in_pack &&
1263 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1264 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1267 /* Let's not bust the allowed depth. */
1268 if (src->depth >= max_depth)
1271 /* Now some size filtering heuristics. */
1272 trg_size = trg_entry->size;
1273 if (!trg_entry->delta) {
1274 max_size = trg_size/2 - 20;
1277 max_size = trg_entry->delta_size;
1278 ref_depth = trg->depth;
1280 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1281 (max_depth - ref_depth + 1);
1284 src_size = src_entry->size;
1285 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1286 if (sizediff >= max_size)
1288 if (trg_size < src_size / 32)
1291 /* Load data if not already done */
1294 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1297 die("object %s cannot be read",
1298 sha1_to_hex(trg_entry->idx.sha1));
1300 die("object %s inconsistent object length (%lu vs %lu)",
1301 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1306 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1309 die("object %s cannot be read",
1310 sha1_to_hex(src_entry->idx.sha1));
1312 die("object %s inconsistent object length (%lu vs %lu)",
1313 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1317 src->index = create_delta_index(src->data, src_size);
1319 static int warned = 0;
1321 warning("suboptimal pack - out of memory");
1324 *mem_usage += sizeof_delta_index(src->index);
1327 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1331 if (trg_entry->delta) {
1332 /* Prefer only shallower same-sized deltas. */
1333 if (delta_size == trg_entry->delta_size &&
1334 src->depth + 1 >= trg->depth) {
1341 * Handle memory allocation outside of the cache
1342 * accounting lock. Compiler will optimize the strangeness
1343 * away when NO_PTHREADS is defined.
1345 free(trg_entry->delta_data);
1347 if (trg_entry->delta_data) {
1348 delta_cache_size -= trg_entry->delta_size;
1349 trg_entry->delta_data = NULL;
1351 if (delta_cacheable(src_size, trg_size, delta_size)) {
1352 delta_cache_size += delta_size;
1354 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1360 trg_entry->delta = src_entry;
1361 trg_entry->delta_size = delta_size;
1362 trg->depth = src->depth + 1;
1367 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1369 struct object_entry *child = me->delta_child;
1372 unsigned int c = check_delta_limit(child, n + 1);
1375 child = child->delta_sibling;
1380 static unsigned long free_unpacked(struct unpacked *n)
1382 unsigned long freed_mem = sizeof_delta_index(n->index);
1383 free_delta_index(n->index);
1386 freed_mem += n->entry->size;
1395 static void find_deltas(struct object_entry **list, unsigned *list_size,
1396 int window, int depth, unsigned *processed)
1398 uint32_t i, idx = 0, count = 0;
1399 struct unpacked *array;
1400 unsigned long mem_usage = 0;
1402 array = xcalloc(window, sizeof(struct unpacked));
1405 struct object_entry *entry;
1406 struct unpacked *n = array + idx;
1407 int j, max_depth, best_base = -1;
1416 if (!entry->preferred_base) {
1418 display_progress(progress_state, *processed);
1422 mem_usage -= free_unpacked(n);
1425 while (window_memory_limit &&
1426 mem_usage > window_memory_limit &&
1428 uint32_t tail = (idx + window - count) % window;
1429 mem_usage -= free_unpacked(array + tail);
1433 /* We do not compute delta to *create* objects we are not
1436 if (entry->preferred_base)
1440 * If the current object is at pack edge, take the depth the
1441 * objects that depend on the current object into account
1442 * otherwise they would become too deep.
1445 if (entry->delta_child) {
1446 max_depth -= check_delta_limit(entry, 0);
1454 uint32_t other_idx = idx + j;
1456 if (other_idx >= window)
1457 other_idx -= window;
1458 m = array + other_idx;
1461 ret = try_delta(n, m, max_depth, &mem_usage);
1465 best_base = other_idx;
1469 * If we decided to cache the delta data, then it is best
1470 * to compress it right away. First because we have to do
1471 * it anyway, and doing it here while we're threaded will
1472 * save a lot of time in the non threaded write phase,
1473 * as well as allow for caching more deltas within
1474 * the same cache size limit.
1476 * But only if not writing to stdout, since in that case
1477 * the network is most likely throttling writes anyway,
1478 * and therefore it is best to go to the write phase ASAP
1479 * instead, as we can afford spending more time compressing
1480 * between writes at that moment.
1482 if (entry->delta_data && !pack_to_stdout) {
1483 entry->z_delta_size = do_compress(&entry->delta_data,
1486 delta_cache_size -= entry->delta_size;
1487 delta_cache_size += entry->z_delta_size;
1491 /* if we made n a delta, and if n is already at max
1492 * depth, leaving it in the window is pointless. we
1493 * should evict it first.
1495 if (entry->delta && max_depth <= n->depth)
1499 * Move the best delta base up in the window, after the
1500 * currently deltified object, to keep it longer. It will
1501 * be the first base object to be attempted next.
1504 struct unpacked swap = array[best_base];
1505 int dist = (window + idx - best_base) % window;
1506 int dst = best_base;
1508 int src = (dst + 1) % window;
1509 array[dst] = array[src];
1517 if (count + 1 < window)
1523 for (i = 0; i < window; ++i) {
1524 free_delta_index(array[i].index);
1525 free(array[i].data);
1533 * The main thread waits on the condition that (at least) one of the workers
1534 * has stopped working (which is indicated in the .working member of
1535 * struct thread_params).
1536 * When a work thread has completed its work, it sets .working to 0 and
1537 * signals the main thread and waits on the condition that .data_ready
1541 struct thread_params {
1543 struct object_entry **list;
1550 pthread_mutex_t mutex;
1551 pthread_cond_t cond;
1552 unsigned *processed;
1555 static pthread_cond_t progress_cond;
1558 * Mutex and conditional variable can't be statically-initialized on Windows.
1560 static void init_threaded_search(void)
1562 pthread_mutex_init(&read_mutex, NULL);
1563 pthread_mutex_init(&cache_mutex, NULL);
1564 pthread_mutex_init(&progress_mutex, NULL);
1565 pthread_cond_init(&progress_cond, NULL);
1568 static void cleanup_threaded_search(void)
1570 pthread_cond_destroy(&progress_cond);
1571 pthread_mutex_destroy(&read_mutex);
1572 pthread_mutex_destroy(&cache_mutex);
1573 pthread_mutex_destroy(&progress_mutex);
1576 static void *threaded_find_deltas(void *arg)
1578 struct thread_params *me = arg;
1580 while (me->remaining) {
1581 find_deltas(me->list, &me->remaining,
1582 me->window, me->depth, me->processed);
1586 pthread_cond_signal(&progress_cond);
1590 * We must not set ->data_ready before we wait on the
1591 * condition because the main thread may have set it to 1
1592 * before we get here. In order to be sure that new
1593 * work is available if we see 1 in ->data_ready, it
1594 * was initialized to 0 before this thread was spawned
1595 * and we reset it to 0 right away.
1597 pthread_mutex_lock(&me->mutex);
1598 while (!me->data_ready)
1599 pthread_cond_wait(&me->cond, &me->mutex);
1601 pthread_mutex_unlock(&me->mutex);
1603 /* leave ->working 1 so that this doesn't get more work assigned */
1607 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1608 int window, int depth, unsigned *processed)
1610 struct thread_params *p;
1611 int i, ret, active_threads = 0;
1613 init_threaded_search();
1615 if (!delta_search_threads) /* --threads=0 means autodetect */
1616 delta_search_threads = online_cpus();
1617 if (delta_search_threads <= 1) {
1618 find_deltas(list, &list_size, window, depth, processed);
1619 cleanup_threaded_search();
1622 if (progress > pack_to_stdout)
1623 fprintf(stderr, "Delta compression using up to %d threads.\n",
1624 delta_search_threads);
1625 p = xcalloc(delta_search_threads, sizeof(*p));
1627 /* Partition the work amongst work threads. */
1628 for (i = 0; i < delta_search_threads; i++) {
1629 unsigned sub_size = list_size / (delta_search_threads - i);
1631 /* don't use too small segments or no deltas will be found */
1632 if (sub_size < 2*window && i+1 < delta_search_threads)
1635 p[i].window = window;
1637 p[i].processed = processed;
1639 p[i].data_ready = 0;
1641 /* try to split chunks on "path" boundaries */
1642 while (sub_size && sub_size < list_size &&
1643 list[sub_size]->hash &&
1644 list[sub_size]->hash == list[sub_size-1]->hash)
1648 p[i].list_size = sub_size;
1649 p[i].remaining = sub_size;
1652 list_size -= sub_size;
1655 /* Start work threads. */
1656 for (i = 0; i < delta_search_threads; i++) {
1657 if (!p[i].list_size)
1659 pthread_mutex_init(&p[i].mutex, NULL);
1660 pthread_cond_init(&p[i].cond, NULL);
1661 ret = pthread_create(&p[i].thread, NULL,
1662 threaded_find_deltas, &p[i]);
1664 die("unable to create thread: %s", strerror(ret));
1669 * Now let's wait for work completion. Each time a thread is done
1670 * with its work, we steal half of the remaining work from the
1671 * thread with the largest number of unprocessed objects and give
1672 * it to that newly idle thread. This ensure good load balancing
1673 * until the remaining object list segments are simply too short
1674 * to be worth splitting anymore.
1676 while (active_threads) {
1677 struct thread_params *target = NULL;
1678 struct thread_params *victim = NULL;
1679 unsigned sub_size = 0;
1683 for (i = 0; !target && i < delta_search_threads; i++)
1688 pthread_cond_wait(&progress_cond, &progress_mutex);
1691 for (i = 0; i < delta_search_threads; i++)
1692 if (p[i].remaining > 2*window &&
1693 (!victim || victim->remaining < p[i].remaining))
1696 sub_size = victim->remaining / 2;
1697 list = victim->list + victim->list_size - sub_size;
1698 while (sub_size && list[0]->hash &&
1699 list[0]->hash == list[-1]->hash) {
1705 * It is possible for some "paths" to have
1706 * so many objects that no hash boundary
1707 * might be found. Let's just steal the
1708 * exact half in that case.
1710 sub_size = victim->remaining / 2;
1713 target->list = list;
1714 victim->list_size -= sub_size;
1715 victim->remaining -= sub_size;
1717 target->list_size = sub_size;
1718 target->remaining = sub_size;
1719 target->working = 1;
1722 pthread_mutex_lock(&target->mutex);
1723 target->data_ready = 1;
1724 pthread_cond_signal(&target->cond);
1725 pthread_mutex_unlock(&target->mutex);
1728 pthread_join(target->thread, NULL);
1729 pthread_cond_destroy(&target->cond);
1730 pthread_mutex_destroy(&target->mutex);
1734 cleanup_threaded_search();
1739 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1742 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1744 unsigned char peeled[20];
1746 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1747 !peel_ref(path, peeled) && /* peelable? */
1748 !is_null_sha1(peeled) && /* annotated tag? */
1749 locate_object_entry(peeled)) /* object packed? */
1750 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1754 static void prepare_pack(int window, int depth)
1756 struct object_entry **delta_list;
1757 uint32_t i, nr_deltas;
1760 get_object_details();
1763 * If we're locally repacking then we need to be doubly careful
1764 * from now on in order to make sure no stealth corruption gets
1765 * propagated to the new pack. Clients receiving streamed packs
1766 * should validate everything they get anyway so no need to incur
1767 * the additional cost here in that case.
1769 if (!pack_to_stdout)
1770 do_check_packed_object_crc = 1;
1772 if (!nr_objects || !window || !depth)
1775 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1778 for (i = 0; i < nr_objects; i++) {
1779 struct object_entry *entry = objects + i;
1782 /* This happens if we decided to reuse existing
1783 * delta from a pack. "reuse_delta &&" is implied.
1787 if (entry->size < 50)
1790 if (entry->no_try_delta)
1793 if (!entry->preferred_base) {
1795 if (entry->type < 0)
1796 die("unable to get type of object %s",
1797 sha1_to_hex(entry->idx.sha1));
1799 if (entry->type < 0) {
1801 * This object is not found, but we
1802 * don't have to include it anyway.
1808 delta_list[n++] = entry;
1811 if (nr_deltas && n > 1) {
1812 unsigned nr_done = 0;
1814 progress_state = start_progress("Compressing objects",
1816 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1817 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1818 stop_progress(&progress_state);
1819 if (nr_done != nr_deltas)
1820 die("inconsistency with delta count");
1825 static int git_pack_config(const char *k, const char *v, void *cb)
1827 if (!strcmp(k, "pack.window")) {
1828 window = git_config_int(k, v);
1831 if (!strcmp(k, "pack.windowmemory")) {
1832 window_memory_limit = git_config_ulong(k, v);
1835 if (!strcmp(k, "pack.depth")) {
1836 depth = git_config_int(k, v);
1839 if (!strcmp(k, "pack.compression")) {
1840 int level = git_config_int(k, v);
1842 level = Z_DEFAULT_COMPRESSION;
1843 else if (level < 0 || level > Z_BEST_COMPRESSION)
1844 die("bad pack compression level %d", level);
1845 pack_compression_level = level;
1846 pack_compression_seen = 1;
1849 if (!strcmp(k, "pack.deltacachesize")) {
1850 max_delta_cache_size = git_config_int(k, v);
1853 if (!strcmp(k, "pack.deltacachelimit")) {
1854 cache_max_small_delta_size = git_config_int(k, v);
1857 if (!strcmp(k, "pack.threads")) {
1858 delta_search_threads = git_config_int(k, v);
1859 if (delta_search_threads < 0)
1860 die("invalid number of threads specified (%d)",
1861 delta_search_threads);
1863 if (delta_search_threads != 1)
1864 warning("no threads support, ignoring %s", k);
1868 if (!strcmp(k, "pack.indexversion")) {
1869 pack_idx_default_version = git_config_int(k, v);
1870 if (pack_idx_default_version > 2)
1871 die("bad pack.indexversion=%"PRIu32,
1872 pack_idx_default_version);
1875 if (!strcmp(k, "pack.packsizelimit")) {
1876 pack_size_limit_cfg = git_config_ulong(k, v);
1879 return git_default_config(k, v, cb);
1882 static void read_object_list_from_stdin(void)
1884 char line[40 + 1 + PATH_MAX + 2];
1885 unsigned char sha1[20];
1888 if (!fgets(line, sizeof(line), stdin)) {
1892 die("fgets returned NULL, not EOF, not error!");
1898 if (line[0] == '-') {
1899 if (get_sha1_hex(line+1, sha1))
1900 die("expected edge sha1, got garbage:\n %s",
1902 add_preferred_base(sha1);
1905 if (get_sha1_hex(line, sha1))
1906 die("expected sha1, got garbage:\n %s", line);
1908 add_preferred_base_object(line+41);
1909 add_object_entry(sha1, 0, line+41, 0);
1913 #define OBJECT_ADDED (1u<<20)
1915 static void show_commit(struct commit *commit, void *data)
1917 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1918 commit->object.flags |= OBJECT_ADDED;
1921 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1923 char *name = path_name(path, last);
1925 add_preferred_base_object(name);
1926 add_object_entry(obj->sha1, obj->type, name, 0);
1927 obj->flags |= OBJECT_ADDED;
1930 * We will have generated the hash from the name,
1931 * but not saved a pointer to it - we can free it
1936 static void show_edge(struct commit *commit)
1938 add_preferred_base(commit->object.sha1);
1941 struct in_pack_object {
1943 struct object *object;
1949 struct in_pack_object *array;
1952 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1954 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1955 in_pack->array[in_pack->nr].object = object;
1960 * Compare the objects in the offset order, in order to emulate the
1961 * "git rev-list --objects" output that produced the pack originally.
1963 static int ofscmp(const void *a_, const void *b_)
1965 struct in_pack_object *a = (struct in_pack_object *)a_;
1966 struct in_pack_object *b = (struct in_pack_object *)b_;
1968 if (a->offset < b->offset)
1970 else if (a->offset > b->offset)
1973 return hashcmp(a->object->sha1, b->object->sha1);
1976 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1978 struct packed_git *p;
1979 struct in_pack in_pack;
1982 memset(&in_pack, 0, sizeof(in_pack));
1984 for (p = packed_git; p; p = p->next) {
1985 const unsigned char *sha1;
1988 if (!p->pack_local || p->pack_keep)
1990 if (open_pack_index(p))
1991 die("cannot open pack index");
1993 ALLOC_GROW(in_pack.array,
1994 in_pack.nr + p->num_objects,
1997 for (i = 0; i < p->num_objects; i++) {
1998 sha1 = nth_packed_object_sha1(p, i);
1999 o = lookup_unknown_object(sha1);
2000 if (!(o->flags & OBJECT_ADDED))
2001 mark_in_pack_object(o, p, &in_pack);
2002 o->flags |= OBJECT_ADDED;
2007 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2009 for (i = 0; i < in_pack.nr; i++) {
2010 struct object *o = in_pack.array[i].object;
2011 add_object_entry(o->sha1, o->type, "", 0);
2014 free(in_pack.array);
2017 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2019 static struct packed_git *last_found = (void *)1;
2020 struct packed_git *p;
2022 p = (last_found != (void *)1) ? last_found : packed_git;
2025 if ((!p->pack_local || p->pack_keep) &&
2026 find_pack_entry_one(sha1, p)) {
2030 if (p == last_found)
2034 if (p == last_found)
2040 static void loosen_unused_packed_objects(struct rev_info *revs)
2042 struct packed_git *p;
2044 const unsigned char *sha1;
2046 for (p = packed_git; p; p = p->next) {
2047 if (!p->pack_local || p->pack_keep)
2050 if (open_pack_index(p))
2051 die("cannot open pack index");
2053 for (i = 0; i < p->num_objects; i++) {
2054 sha1 = nth_packed_object_sha1(p, i);
2055 if (!locate_object_entry(sha1) &&
2056 !has_sha1_pack_kept_or_nonlocal(sha1))
2057 if (force_object_loose(sha1, p->mtime))
2058 die("unable to force loose object");
2063 static void get_object_list(int ac, const char **av)
2065 struct rev_info revs;
2069 init_revisions(&revs, NULL);
2070 save_commit_buffer = 0;
2071 setup_revisions(ac, av, &revs, NULL);
2073 while (fgets(line, sizeof(line), stdin) != NULL) {
2074 int len = strlen(line);
2075 if (len && line[len - 1] == '\n')
2080 if (!strcmp(line, "--not")) {
2081 flags ^= UNINTERESTING;
2084 die("not a rev '%s'", line);
2086 if (handle_revision_arg(line, &revs, flags, 1))
2087 die("bad revision '%s'", line);
2090 if (prepare_revision_walk(&revs))
2091 die("revision walk setup failed");
2092 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2093 traverse_commit_list(&revs, show_commit, show_object, NULL);
2095 if (keep_unreachable)
2096 add_objects_in_unpacked_packs(&revs);
2097 if (unpack_unreachable)
2098 loosen_unused_packed_objects(&revs);
2101 static int adjust_perm(const char *path, mode_t mode)
2103 if (chmod(path, mode))
2105 return adjust_shared_perm(path);
2108 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2110 int use_internal_rev_list = 0;
2112 int all_progress_implied = 0;
2115 int rp_ac_alloc = 64;
2118 read_replace_refs = 0;
2120 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2122 rp_av[0] = "pack-objects";
2123 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2126 git_config(git_pack_config, NULL);
2127 if (!pack_compression_seen && core_compression_seen)
2128 pack_compression_level = core_compression_level;
2130 progress = isatty(2);
2131 for (i = 1; i < argc; i++) {
2132 const char *arg = argv[i];
2137 if (!strcmp("--non-empty", arg)) {
2141 if (!strcmp("--local", arg)) {
2145 if (!strcmp("--incremental", arg)) {
2149 if (!strcmp("--honor-pack-keep", arg)) {
2150 ignore_packed_keep = 1;
2153 if (!prefixcmp(arg, "--compression=")) {
2155 int level = strtoul(arg+14, &end, 0);
2156 if (!arg[14] || *end)
2159 level = Z_DEFAULT_COMPRESSION;
2160 else if (level < 0 || level > Z_BEST_COMPRESSION)
2161 die("bad pack compression level %d", level);
2162 pack_compression_level = level;
2165 if (!prefixcmp(arg, "--max-pack-size=")) {
2166 pack_size_limit_cfg = 0;
2167 if (!git_parse_ulong(arg+16, &pack_size_limit))
2171 if (!prefixcmp(arg, "--window=")) {
2173 window = strtoul(arg+9, &end, 0);
2174 if (!arg[9] || *end)
2178 if (!prefixcmp(arg, "--window-memory=")) {
2179 if (!git_parse_ulong(arg+16, &window_memory_limit))
2183 if (!prefixcmp(arg, "--threads=")) {
2185 delta_search_threads = strtoul(arg+10, &end, 0);
2186 if (!arg[10] || *end || delta_search_threads < 0)
2189 if (delta_search_threads != 1)
2190 warning("no threads support, "
2191 "ignoring %s", arg);
2195 if (!prefixcmp(arg, "--depth=")) {
2197 depth = strtoul(arg+8, &end, 0);
2198 if (!arg[8] || *end)
2202 if (!strcmp("--progress", arg)) {
2206 if (!strcmp("--all-progress", arg)) {
2210 if (!strcmp("--all-progress-implied", arg)) {
2211 all_progress_implied = 1;
2214 if (!strcmp("-q", arg)) {
2218 if (!strcmp("--no-reuse-delta", arg)) {
2222 if (!strcmp("--no-reuse-object", arg)) {
2223 reuse_object = reuse_delta = 0;
2226 if (!strcmp("--delta-base-offset", arg)) {
2227 allow_ofs_delta = 1;
2230 if (!strcmp("--stdout", arg)) {
2234 if (!strcmp("--revs", arg)) {
2235 use_internal_rev_list = 1;
2238 if (!strcmp("--keep-unreachable", arg)) {
2239 keep_unreachable = 1;
2242 if (!strcmp("--unpack-unreachable", arg)) {
2243 unpack_unreachable = 1;
2246 if (!strcmp("--include-tag", arg)) {
2250 if (!strcmp("--unpacked", arg) ||
2251 !strcmp("--reflog", arg) ||
2252 !strcmp("--all", arg)) {
2253 use_internal_rev_list = 1;
2254 if (rp_ac >= rp_ac_alloc - 1) {
2255 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2256 rp_av = xrealloc(rp_av,
2257 rp_ac_alloc * sizeof(*rp_av));
2259 rp_av[rp_ac++] = arg;
2262 if (!strcmp("--thin", arg)) {
2263 use_internal_rev_list = 1;
2265 rp_av[1] = "--objects-edge";
2268 if (!prefixcmp(arg, "--index-version=")) {
2270 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2271 if (pack_idx_default_version > 2)
2274 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2275 if (*c || pack_idx_off32_limit & 0x80000000)
2279 if (!strcmp(arg, "--keep-true-parents")) {
2280 grafts_replace_parents = 0;
2286 /* Traditionally "pack-objects [options] base extra" failed;
2287 * we would however want to take refs parameter that would
2288 * have been given to upstream rev-list ourselves, which means
2289 * we somehow want to say what the base name is. So the
2292 * pack-objects [options] base <refs...>
2294 * in other words, we would treat the first non-option as the
2295 * base_name and send everything else to the internal revision
2299 if (!pack_to_stdout)
2300 base_name = argv[i++];
2302 if (pack_to_stdout != !base_name)
2305 if (!pack_to_stdout && !pack_size_limit)
2306 pack_size_limit = pack_size_limit_cfg;
2307 if (pack_to_stdout && pack_size_limit)
2308 die("--max-pack-size cannot be used to build a pack for transfer.");
2309 if (pack_size_limit && pack_size_limit < 1024*1024) {
2310 warning("minimum pack size limit is 1 MiB");
2311 pack_size_limit = 1024*1024;
2314 if (!pack_to_stdout && thin)
2315 die("--thin cannot be used to build an indexable pack.");
2317 if (keep_unreachable && unpack_unreachable)
2318 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2320 if (progress && all_progress_implied)
2323 prepare_packed_git();
2326 progress_state = start_progress("Counting objects", 0);
2327 if (!use_internal_rev_list)
2328 read_object_list_from_stdin();
2330 rp_av[rp_ac] = NULL;
2331 get_object_list(rp_ac, rp_av);
2333 cleanup_preferred_base();
2334 if (include_tag && nr_result)
2335 for_each_ref(add_ref_tag, NULL);
2336 stop_progress(&progress_state);
2338 if (non_empty && !nr_result)
2341 prepare_pack(window, depth);
2344 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2345 " reused %"PRIu32" (delta %"PRIu32")\n",
2346 written, written_delta, reused, reused_delta);