11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
16 #include "list-objects.h"
19 #include "thread-utils.h"
21 static const char pack_usage[] =
22 "git pack-objects [ -q | --progress | --all-progress ]\n"
23 " [--all-progress-implied]\n"
24 " [--max-pack-size=<n>] [--local] [--incremental]\n"
25 " [--window=<n>] [--window-memory=<n>] [--depth=<n>]\n"
26 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
27 " [--threads=<n>] [--non-empty] [--revs [--unpacked | --all]]\n"
28 " [--reflog] [--stdout | base-name] [--include-tag]\n"
29 " [--keep-unreachable | --unpack-unreachable]\n"
30 " [< ref-list | < object-list]";
33 struct pack_idx_entry idx;
34 unsigned long size; /* uncompressed size */
35 struct packed_git *in_pack; /* already in pack */
37 struct object_entry *delta; /* delta base object */
38 struct object_entry *delta_child; /* deltified objects who bases me */
39 struct object_entry *delta_sibling; /* other deltified objects who
40 * uses the same base as me
42 void *delta_data; /* cached delta (uncompressed) */
43 unsigned long delta_size; /* delta data size (uncompressed) */
44 unsigned long z_delta_size; /* delta data size (compressed) */
45 unsigned int hash; /* name hint hash */
46 enum object_type type;
47 enum object_type in_pack_type; /* could be delta */
48 unsigned char in_pack_header_size;
49 unsigned char preferred_base; /* we do not pack this, but is available
50 * to be used as the base object to delta
53 unsigned char no_try_delta;
57 * Objects we are going to pack are collected in objects array (dynamically
58 * expanded). nr_objects & nr_alloc controls this array. They are stored
59 * in the order we see -- typically rev-list --objects order that gives us
60 * nice "minimum seek" order.
62 static struct object_entry *objects;
63 static struct pack_idx_entry **written_list;
64 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
67 static int reuse_delta = 1, reuse_object = 1;
68 static int keep_unreachable, unpack_unreachable, include_tag;
70 static int incremental;
71 static int ignore_packed_keep;
72 static int allow_ofs_delta;
73 static struct pack_idx_option pack_idx_opts;
74 static const char *base_name;
75 static int progress = 1;
76 static int window = 10;
77 static unsigned long pack_size_limit, pack_size_limit_cfg;
78 static int depth = 50;
79 static int delta_search_threads;
80 static int pack_to_stdout;
81 static int num_preferred_base;
82 static struct progress *progress_state;
83 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
84 static int pack_compression_seen;
86 static unsigned long delta_cache_size = 0;
87 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
88 static unsigned long cache_max_small_delta_size = 1000;
90 static unsigned long window_memory_limit = 0;
93 * The object names in objects array are hashed with this hashtable,
94 * to help looking up the entry by object name.
95 * This hashtable is built after all the objects are seen.
97 static int *object_ix;
98 static int object_ix_hashsz;
103 static uint32_t written, written_delta;
104 static uint32_t reused, reused_delta;
107 static void *get_delta(struct object_entry *entry)
109 unsigned long size, base_size, delta_size;
110 void *buf, *base_buf, *delta_buf;
111 enum object_type type;
113 buf = read_sha1_file(entry->idx.sha1, &type, &size);
115 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
116 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
118 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
119 delta_buf = diff_delta(base_buf, base_size,
120 buf, size, &delta_size, 0);
121 if (!delta_buf || delta_size != entry->delta_size)
122 die("delta size changed");
128 static unsigned long do_compress(void **pptr, unsigned long size)
132 unsigned long maxsize;
134 memset(&stream, 0, sizeof(stream));
135 git_deflate_init(&stream, pack_compression_level);
136 maxsize = git_deflate_bound(&stream, size);
139 out = xmalloc(maxsize);
143 stream.avail_in = size;
144 stream.next_out = out;
145 stream.avail_out = maxsize;
146 while (git_deflate(&stream, Z_FINISH) == Z_OK)
148 git_deflate_end(&stream);
151 return stream.total_out;
155 * we are going to reuse the existing object data as is. make
156 * sure it is not corrupt.
158 static int check_pack_inflate(struct packed_git *p,
159 struct pack_window **w_curs,
162 unsigned long expect)
165 unsigned char fakebuf[4096], *in;
168 memset(&stream, 0, sizeof(stream));
169 git_inflate_init(&stream);
171 in = use_pack(p, w_curs, offset, &stream.avail_in);
173 stream.next_out = fakebuf;
174 stream.avail_out = sizeof(fakebuf);
175 st = git_inflate(&stream, Z_FINISH);
176 offset += stream.next_in - in;
177 } while (st == Z_OK || st == Z_BUF_ERROR);
178 git_inflate_end(&stream);
179 return (st == Z_STREAM_END &&
180 stream.total_out == expect &&
181 stream.total_in == len) ? 0 : -1;
184 static void copy_pack_data(struct sha1file *f,
185 struct packed_git *p,
186 struct pack_window **w_curs,
194 in = use_pack(p, w_curs, offset, &avail);
196 avail = (unsigned long)len;
197 sha1write(f, in, avail);
203 static unsigned long write_object(struct sha1file *f,
204 struct object_entry *entry,
207 unsigned long size, limit, datalen;
209 unsigned char header[10], dheader[10];
211 enum object_type type;
212 int usable_delta, to_reuse;
219 /* apply size limit if limited packsize and not first object */
220 if (!pack_size_limit || !nr_written)
222 else if (pack_size_limit <= write_offset)
224 * the earlier object did not fit the limit; avoid
225 * mistaking this with unlimited (i.e. limit = 0).
229 limit = pack_size_limit - write_offset;
232 usable_delta = 0; /* no delta */
233 else if (!pack_size_limit)
234 usable_delta = 1; /* unlimited packfile */
235 else if (entry->delta->idx.offset == (off_t)-1)
236 usable_delta = 0; /* base was written to another pack */
237 else if (entry->delta->idx.offset)
238 usable_delta = 1; /* base already exists in this pack */
240 usable_delta = 0; /* base could end up in another pack */
243 to_reuse = 0; /* explicit */
244 else if (!entry->in_pack)
245 to_reuse = 0; /* can't reuse what we don't have */
246 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
247 /* check_object() decided it for us ... */
248 to_reuse = usable_delta;
249 /* ... but pack split may override that */
250 else if (type != entry->in_pack_type)
251 to_reuse = 0; /* pack has delta which is unusable */
252 else if (entry->delta)
253 to_reuse = 0; /* we want to pack afresh */
255 to_reuse = 1; /* we have it in-pack undeltified,
256 * and we do not need to deltify it.
262 buf = read_sha1_file(entry->idx.sha1, &type, &size);
264 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
266 * make sure no cached delta data remains from a
267 * previous attempt before a pack split occurred.
269 free(entry->delta_data);
270 entry->delta_data = NULL;
271 entry->z_delta_size = 0;
272 } else if (entry->delta_data) {
273 size = entry->delta_size;
274 buf = entry->delta_data;
275 entry->delta_data = NULL;
276 type = (allow_ofs_delta && entry->delta->idx.offset) ?
277 OBJ_OFS_DELTA : OBJ_REF_DELTA;
279 buf = get_delta(entry);
280 size = entry->delta_size;
281 type = (allow_ofs_delta && entry->delta->idx.offset) ?
282 OBJ_OFS_DELTA : OBJ_REF_DELTA;
285 if (entry->z_delta_size)
286 datalen = entry->z_delta_size;
288 datalen = do_compress(&buf, size);
291 * The object header is a byte of 'type' followed by zero or
292 * more bytes of length.
294 hdrlen = encode_in_pack_object_header(type, size, header);
296 if (type == OBJ_OFS_DELTA) {
298 * Deltas with relative base contain an additional
299 * encoding of the relative offset for the delta
300 * base from this object's position in the pack.
302 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
303 unsigned pos = sizeof(dheader) - 1;
304 dheader[pos] = ofs & 127;
306 dheader[--pos] = 128 | (--ofs & 127);
307 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
311 sha1write(f, header, hdrlen);
312 sha1write(f, dheader + pos, sizeof(dheader) - pos);
313 hdrlen += sizeof(dheader) - pos;
314 } else if (type == OBJ_REF_DELTA) {
316 * Deltas with a base reference contain
317 * an additional 20 bytes for the base sha1.
319 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
323 sha1write(f, header, hdrlen);
324 sha1write(f, entry->delta->idx.sha1, 20);
327 if (limit && hdrlen + datalen + 20 >= limit) {
331 sha1write(f, header, hdrlen);
333 sha1write(f, buf, datalen);
337 struct packed_git *p = entry->in_pack;
338 struct pack_window *w_curs = NULL;
339 struct revindex_entry *revidx;
343 type = (allow_ofs_delta && entry->delta->idx.offset) ?
344 OBJ_OFS_DELTA : OBJ_REF_DELTA;
345 hdrlen = encode_in_pack_object_header(type, entry->size, header);
347 offset = entry->in_pack_offset;
348 revidx = find_pack_revindex(p, offset);
349 datalen = revidx[1].offset - offset;
350 if (!pack_to_stdout && p->index_version > 1 &&
351 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
352 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
357 offset += entry->in_pack_header_size;
358 datalen -= entry->in_pack_header_size;
359 if (!pack_to_stdout && p->index_version == 1 &&
360 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
361 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
366 if (type == OBJ_OFS_DELTA) {
367 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
368 unsigned pos = sizeof(dheader) - 1;
369 dheader[pos] = ofs & 127;
371 dheader[--pos] = 128 | (--ofs & 127);
372 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
376 sha1write(f, header, hdrlen);
377 sha1write(f, dheader + pos, sizeof(dheader) - pos);
378 hdrlen += sizeof(dheader) - pos;
380 } else if (type == OBJ_REF_DELTA) {
381 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
385 sha1write(f, header, hdrlen);
386 sha1write(f, entry->delta->idx.sha1, 20);
390 if (limit && hdrlen + datalen + 20 >= limit) {
394 sha1write(f, header, hdrlen);
396 copy_pack_data(f, p, &w_curs, offset, datalen);
404 entry->idx.crc32 = crc32_end(f);
405 return hdrlen + datalen;
408 static int write_one(struct sha1file *f,
409 struct object_entry *e,
414 /* offset is non zero if object is written already. */
415 if (e->idx.offset || e->preferred_base)
418 /* if we are deltified, write out base object first. */
419 if (e->delta && !write_one(f, e->delta, offset))
422 e->idx.offset = *offset;
423 size = write_object(f, e, *offset);
428 written_list[nr_written++] = &e->idx;
430 /* make sure off_t is sufficiently large not to wrap */
431 if (signed_add_overflows(*offset, size))
432 die("pack too large for current definition of off_t");
437 static void write_pack_file(void)
442 struct pack_header hdr;
443 uint32_t nr_remaining = nr_result;
444 time_t last_mtime = 0;
446 if (progress > pack_to_stdout)
447 progress_state = start_progress("Writing objects", nr_result);
448 written_list = xmalloc(nr_objects * sizeof(*written_list));
451 unsigned char sha1[20];
452 char *pack_tmp_name = NULL;
454 if (pack_to_stdout) {
455 f = sha1fd_throughput(1, "<stdout>", progress_state);
457 char tmpname[PATH_MAX];
459 fd = odb_mkstemp(tmpname, sizeof(tmpname),
460 "pack/tmp_pack_XXXXXX");
461 pack_tmp_name = xstrdup(tmpname);
462 f = sha1fd(fd, pack_tmp_name);
465 hdr.hdr_signature = htonl(PACK_SIGNATURE);
466 hdr.hdr_version = htonl(PACK_VERSION);
467 hdr.hdr_entries = htonl(nr_remaining);
468 sha1write(f, &hdr, sizeof(hdr));
469 offset = sizeof(hdr);
471 for (; i < nr_objects; i++) {
472 if (!write_one(f, objects + i, &offset))
474 display_progress(progress_state, written);
478 * Did we write the wrong # entries in the header?
479 * If so, rewrite it like in fast-import
481 if (pack_to_stdout) {
482 sha1close(f, sha1, CSUM_CLOSE);
483 } else if (nr_written == nr_remaining) {
484 sha1close(f, sha1, CSUM_FSYNC);
486 int fd = sha1close(f, sha1, 0);
487 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
488 nr_written, sha1, offset);
492 if (!pack_to_stdout) {
494 const char *idx_tmp_name;
495 char tmpname[PATH_MAX];
497 idx_tmp_name = write_idx_file(NULL, written_list, nr_written,
498 &pack_idx_opts, sha1);
500 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
501 base_name, sha1_to_hex(sha1));
502 free_pack_by_name(tmpname);
503 if (adjust_shared_perm(pack_tmp_name))
504 die_errno("unable to make temporary pack file readable");
505 if (rename(pack_tmp_name, tmpname))
506 die_errno("unable to rename temporary pack file");
509 * Packs are runtime accessed in their mtime
510 * order since newer packs are more likely to contain
511 * younger objects. So if we are creating multiple
512 * packs then we should modify the mtime of later ones
513 * to preserve this property.
515 if (stat(tmpname, &st) < 0) {
516 warning("failed to stat %s: %s",
517 tmpname, strerror(errno));
518 } else if (!last_mtime) {
519 last_mtime = st.st_mtime;
522 utb.actime = st.st_atime;
523 utb.modtime = --last_mtime;
524 if (utime(tmpname, &utb) < 0)
525 warning("failed utime() on %s: %s",
526 tmpname, strerror(errno));
529 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
530 base_name, sha1_to_hex(sha1));
531 if (adjust_shared_perm(idx_tmp_name))
532 die_errno("unable to make temporary index file readable");
533 if (rename(idx_tmp_name, tmpname))
534 die_errno("unable to rename temporary index file");
536 free((void *) idx_tmp_name);
538 puts(sha1_to_hex(sha1));
541 /* mark written objects as written to previous pack */
542 for (j = 0; j < nr_written; j++) {
543 written_list[j]->offset = (off_t)-1;
545 nr_remaining -= nr_written;
546 } while (nr_remaining && i < nr_objects);
549 stop_progress(&progress_state);
550 if (written != nr_result)
551 die("wrote %"PRIu32" objects while expecting %"PRIu32,
555 static int locate_object_entry_hash(const unsigned char *sha1)
559 memcpy(&ui, sha1, sizeof(unsigned int));
560 i = ui % object_ix_hashsz;
561 while (0 < object_ix[i]) {
562 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
564 if (++i == object_ix_hashsz)
570 static struct object_entry *locate_object_entry(const unsigned char *sha1)
574 if (!object_ix_hashsz)
577 i = locate_object_entry_hash(sha1);
579 return &objects[object_ix[i]-1];
583 static void rehash_objects(void)
586 struct object_entry *oe;
588 object_ix_hashsz = nr_objects * 3;
589 if (object_ix_hashsz < 1024)
590 object_ix_hashsz = 1024;
591 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
592 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
593 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
594 int ix = locate_object_entry_hash(oe->idx.sha1);
598 object_ix[ix] = i + 1;
602 static unsigned name_hash(const char *name)
604 unsigned c, hash = 0;
610 * This effectively just creates a sortable number from the
611 * last sixteen non-whitespace characters. Last characters
612 * count "most", so things that end in ".c" sort together.
614 while ((c = *name++) != 0) {
617 hash = (hash >> 2) + (c << 24);
622 static void setup_delta_attr_check(struct git_attr_check *check)
624 static struct git_attr *attr_delta;
627 attr_delta = git_attr("delta");
629 check[0].attr = attr_delta;
632 static int no_try_delta(const char *path)
634 struct git_attr_check check[1];
636 setup_delta_attr_check(check);
637 if (git_checkattr(path, ARRAY_SIZE(check), check))
639 if (ATTR_FALSE(check->value))
644 static int add_object_entry(const unsigned char *sha1, enum object_type type,
645 const char *name, int exclude)
647 struct object_entry *entry;
648 struct packed_git *p, *found_pack = NULL;
649 off_t found_offset = 0;
651 unsigned hash = name_hash(name);
653 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
656 entry = objects + object_ix[ix] - 1;
657 if (!entry->preferred_base)
659 entry->preferred_base = 1;
664 if (!exclude && local && has_loose_object_nonlocal(sha1))
667 for (p = packed_git; p; p = p->next) {
668 off_t offset = find_pack_entry_one(sha1, p);
671 found_offset = offset;
678 if (local && !p->pack_local)
680 if (ignore_packed_keep && p->pack_local && p->pack_keep)
685 if (nr_objects >= nr_alloc) {
686 nr_alloc = (nr_alloc + 1024) * 3 / 2;
687 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
690 entry = objects + nr_objects++;
691 memset(entry, 0, sizeof(*entry));
692 hashcpy(entry->idx.sha1, sha1);
697 entry->preferred_base = 1;
701 entry->in_pack = found_pack;
702 entry->in_pack_offset = found_offset;
705 if (object_ix_hashsz * 3 <= nr_objects * 4)
708 object_ix[-1 - ix] = nr_objects;
710 display_progress(progress_state, nr_objects);
712 if (name && no_try_delta(name))
713 entry->no_try_delta = 1;
718 struct pbase_tree_cache {
719 unsigned char sha1[20];
723 unsigned long tree_size;
726 static struct pbase_tree_cache *(pbase_tree_cache[256]);
727 static int pbase_tree_cache_ix(const unsigned char *sha1)
729 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
731 static int pbase_tree_cache_ix_incr(int ix)
733 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
736 static struct pbase_tree {
737 struct pbase_tree *next;
738 /* This is a phony "cache" entry; we are not
739 * going to evict it nor find it through _get()
740 * mechanism -- this is for the toplevel node that
741 * would almost always change with any commit.
743 struct pbase_tree_cache pcache;
746 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
748 struct pbase_tree_cache *ent, *nent;
751 enum object_type type;
753 int my_ix = pbase_tree_cache_ix(sha1);
754 int available_ix = -1;
756 /* pbase-tree-cache acts as a limited hashtable.
757 * your object will be found at your index or within a few
758 * slots after that slot if it is cached.
760 for (neigh = 0; neigh < 8; neigh++) {
761 ent = pbase_tree_cache[my_ix];
762 if (ent && !hashcmp(ent->sha1, sha1)) {
766 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
767 ((0 <= available_ix) &&
768 (!ent && pbase_tree_cache[available_ix])))
769 available_ix = my_ix;
772 my_ix = pbase_tree_cache_ix_incr(my_ix);
775 /* Did not find one. Either we got a bogus request or
776 * we need to read and perhaps cache.
778 data = read_sha1_file(sha1, &type, &size);
781 if (type != OBJ_TREE) {
786 /* We need to either cache or return a throwaway copy */
788 if (available_ix < 0)
791 ent = pbase_tree_cache[available_ix];
792 my_ix = available_ix;
796 nent = xmalloc(sizeof(*nent));
797 nent->temporary = (available_ix < 0);
800 /* evict and reuse */
801 free(ent->tree_data);
804 hashcpy(nent->sha1, sha1);
805 nent->tree_data = data;
806 nent->tree_size = size;
808 if (!nent->temporary)
809 pbase_tree_cache[my_ix] = nent;
813 static void pbase_tree_put(struct pbase_tree_cache *cache)
815 if (!cache->temporary) {
819 free(cache->tree_data);
823 static int name_cmp_len(const char *name)
826 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
831 static void add_pbase_object(struct tree_desc *tree,
834 const char *fullname)
836 struct name_entry entry;
839 while (tree_entry(tree,&entry)) {
840 if (S_ISGITLINK(entry.mode))
842 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
843 memcmp(name, entry.path, cmplen);
848 if (name[cmplen] != '/') {
849 add_object_entry(entry.sha1,
850 object_type(entry.mode),
854 if (S_ISDIR(entry.mode)) {
855 struct tree_desc sub;
856 struct pbase_tree_cache *tree;
857 const char *down = name+cmplen+1;
858 int downlen = name_cmp_len(down);
860 tree = pbase_tree_get(entry.sha1);
863 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
865 add_pbase_object(&sub, down, downlen, fullname);
866 pbase_tree_put(tree);
871 static unsigned *done_pbase_paths;
872 static int done_pbase_paths_num;
873 static int done_pbase_paths_alloc;
874 static int done_pbase_path_pos(unsigned hash)
877 int hi = done_pbase_paths_num;
879 int mi = (hi + lo) / 2;
880 if (done_pbase_paths[mi] == hash)
882 if (done_pbase_paths[mi] < hash)
890 static int check_pbase_path(unsigned hash)
892 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
896 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
897 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
898 done_pbase_paths = xrealloc(done_pbase_paths,
899 done_pbase_paths_alloc *
902 done_pbase_paths_num++;
903 if (pos < done_pbase_paths_num)
904 memmove(done_pbase_paths + pos + 1,
905 done_pbase_paths + pos,
906 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
907 done_pbase_paths[pos] = hash;
911 static void add_preferred_base_object(const char *name)
913 struct pbase_tree *it;
915 unsigned hash = name_hash(name);
917 if (!num_preferred_base || check_pbase_path(hash))
920 cmplen = name_cmp_len(name);
921 for (it = pbase_tree; it; it = it->next) {
923 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
926 struct tree_desc tree;
927 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
928 add_pbase_object(&tree, name, cmplen, name);
933 static void add_preferred_base(unsigned char *sha1)
935 struct pbase_tree *it;
938 unsigned char tree_sha1[20];
940 if (window <= num_preferred_base++)
943 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
947 for (it = pbase_tree; it; it = it->next) {
948 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
954 it = xcalloc(1, sizeof(*it));
955 it->next = pbase_tree;
958 hashcpy(it->pcache.sha1, tree_sha1);
959 it->pcache.tree_data = data;
960 it->pcache.tree_size = size;
963 static void cleanup_preferred_base(void)
965 struct pbase_tree *it;
971 struct pbase_tree *this = it;
973 free(this->pcache.tree_data);
977 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
978 if (!pbase_tree_cache[i])
980 free(pbase_tree_cache[i]->tree_data);
981 free(pbase_tree_cache[i]);
982 pbase_tree_cache[i] = NULL;
985 free(done_pbase_paths);
986 done_pbase_paths = NULL;
987 done_pbase_paths_num = done_pbase_paths_alloc = 0;
990 static void check_object(struct object_entry *entry)
992 if (entry->in_pack) {
993 struct packed_git *p = entry->in_pack;
994 struct pack_window *w_curs = NULL;
995 const unsigned char *base_ref = NULL;
996 struct object_entry *base_entry;
997 unsigned long used, used_0;
1000 unsigned char *buf, c;
1002 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1005 * We want in_pack_type even if we do not reuse delta
1006 * since non-delta representations could still be reused.
1008 used = unpack_object_header_buffer(buf, avail,
1009 &entry->in_pack_type,
1015 * Determine if this is a delta and if so whether we can
1016 * reuse it or not. Otherwise let's find out as cheaply as
1017 * possible what the actual type and size for this object is.
1019 switch (entry->in_pack_type) {
1021 /* Not a delta hence we've already got all we need. */
1022 entry->type = entry->in_pack_type;
1023 entry->in_pack_header_size = used;
1024 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1026 unuse_pack(&w_curs);
1029 if (reuse_delta && !entry->preferred_base)
1030 base_ref = use_pack(p, &w_curs,
1031 entry->in_pack_offset + used, NULL);
1032 entry->in_pack_header_size = used + 20;
1035 buf = use_pack(p, &w_curs,
1036 entry->in_pack_offset + used, NULL);
1042 if (!ofs || MSB(ofs, 7)) {
1043 error("delta base offset overflow in pack for %s",
1044 sha1_to_hex(entry->idx.sha1));
1048 ofs = (ofs << 7) + (c & 127);
1050 ofs = entry->in_pack_offset - ofs;
1051 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1052 error("delta base offset out of bound for %s",
1053 sha1_to_hex(entry->idx.sha1));
1056 if (reuse_delta && !entry->preferred_base) {
1057 struct revindex_entry *revidx;
1058 revidx = find_pack_revindex(p, ofs);
1061 base_ref = nth_packed_object_sha1(p, revidx->nr);
1063 entry->in_pack_header_size = used + used_0;
1067 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1069 * If base_ref was set above that means we wish to
1070 * reuse delta data, and we even found that base
1071 * in the list of objects we want to pack. Goodie!
1073 * Depth value does not matter - find_deltas() will
1074 * never consider reused delta as the base object to
1075 * deltify other objects against, in order to avoid
1078 entry->type = entry->in_pack_type;
1079 entry->delta = base_entry;
1080 entry->delta_size = entry->size;
1081 entry->delta_sibling = base_entry->delta_child;
1082 base_entry->delta_child = entry;
1083 unuse_pack(&w_curs);
1089 * This must be a delta and we already know what the
1090 * final object type is. Let's extract the actual
1091 * object size from the delta header.
1093 entry->size = get_size_from_delta(p, &w_curs,
1094 entry->in_pack_offset + entry->in_pack_header_size);
1095 if (entry->size == 0)
1097 unuse_pack(&w_curs);
1102 * No choice but to fall back to the recursive delta walk
1103 * with sha1_object_info() to find about the object type
1107 unuse_pack(&w_curs);
1110 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1112 * The error condition is checked in prepare_pack(). This is
1113 * to permit a missing preferred base object to be ignored
1114 * as a preferred base. Doing so can result in a larger
1115 * pack file, but the transfer will still take place.
1119 static int pack_offset_sort(const void *_a, const void *_b)
1121 const struct object_entry *a = *(struct object_entry **)_a;
1122 const struct object_entry *b = *(struct object_entry **)_b;
1124 /* avoid filesystem trashing with loose objects */
1125 if (!a->in_pack && !b->in_pack)
1126 return hashcmp(a->idx.sha1, b->idx.sha1);
1128 if (a->in_pack < b->in_pack)
1130 if (a->in_pack > b->in_pack)
1132 return a->in_pack_offset < b->in_pack_offset ? -1 :
1133 (a->in_pack_offset > b->in_pack_offset);
1136 static void get_object_details(void)
1139 struct object_entry **sorted_by_offset;
1141 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1142 for (i = 0; i < nr_objects; i++)
1143 sorted_by_offset[i] = objects + i;
1144 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1146 for (i = 0; i < nr_objects; i++) {
1147 struct object_entry *entry = sorted_by_offset[i];
1148 check_object(entry);
1149 if (big_file_threshold <= entry->size)
1150 entry->no_try_delta = 1;
1153 free(sorted_by_offset);
1157 * We search for deltas in a list sorted by type, by filename hash, and then
1158 * by size, so that we see progressively smaller and smaller files.
1159 * That's because we prefer deltas to be from the bigger file
1160 * to the smaller -- deletes are potentially cheaper, but perhaps
1161 * more importantly, the bigger file is likely the more recent
1162 * one. The deepest deltas are therefore the oldest objects which are
1163 * less susceptible to be accessed often.
1165 static int type_size_sort(const void *_a, const void *_b)
1167 const struct object_entry *a = *(struct object_entry **)_a;
1168 const struct object_entry *b = *(struct object_entry **)_b;
1170 if (a->type > b->type)
1172 if (a->type < b->type)
1174 if (a->hash > b->hash)
1176 if (a->hash < b->hash)
1178 if (a->preferred_base > b->preferred_base)
1180 if (a->preferred_base < b->preferred_base)
1182 if (a->size > b->size)
1184 if (a->size < b->size)
1186 return a < b ? -1 : (a > b); /* newest first */
1190 struct object_entry *entry;
1192 struct delta_index *index;
1196 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1197 unsigned long delta_size)
1199 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1202 if (delta_size < cache_max_small_delta_size)
1205 /* cache delta, if objects are large enough compared to delta size */
1206 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1214 static pthread_mutex_t read_mutex;
1215 #define read_lock() pthread_mutex_lock(&read_mutex)
1216 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1218 static pthread_mutex_t cache_mutex;
1219 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1220 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1222 static pthread_mutex_t progress_mutex;
1223 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1224 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1228 #define read_lock() (void)0
1229 #define read_unlock() (void)0
1230 #define cache_lock() (void)0
1231 #define cache_unlock() (void)0
1232 #define progress_lock() (void)0
1233 #define progress_unlock() (void)0
1237 static int try_delta(struct unpacked *trg, struct unpacked *src,
1238 unsigned max_depth, unsigned long *mem_usage)
1240 struct object_entry *trg_entry = trg->entry;
1241 struct object_entry *src_entry = src->entry;
1242 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1244 enum object_type type;
1247 /* Don't bother doing diffs between different types */
1248 if (trg_entry->type != src_entry->type)
1252 * We do not bother to try a delta that we discarded
1253 * on an earlier try, but only when reusing delta data.
1255 if (reuse_delta && trg_entry->in_pack &&
1256 trg_entry->in_pack == src_entry->in_pack &&
1257 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1258 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1261 /* Let's not bust the allowed depth. */
1262 if (src->depth >= max_depth)
1265 /* Now some size filtering heuristics. */
1266 trg_size = trg_entry->size;
1267 if (!trg_entry->delta) {
1268 max_size = trg_size/2 - 20;
1271 max_size = trg_entry->delta_size;
1272 ref_depth = trg->depth;
1274 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1275 (max_depth - ref_depth + 1);
1278 src_size = src_entry->size;
1279 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1280 if (sizediff >= max_size)
1282 if (trg_size < src_size / 32)
1285 /* Load data if not already done */
1288 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1291 die("object %s cannot be read",
1292 sha1_to_hex(trg_entry->idx.sha1));
1294 die("object %s inconsistent object length (%lu vs %lu)",
1295 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1300 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1303 if (src_entry->preferred_base) {
1304 static int warned = 0;
1306 warning("object %s cannot be read",
1307 sha1_to_hex(src_entry->idx.sha1));
1309 * Those objects are not included in the
1310 * resulting pack. Be resilient and ignore
1311 * them if they can't be read, in case the
1312 * pack could be created nevertheless.
1316 die("object %s cannot be read",
1317 sha1_to_hex(src_entry->idx.sha1));
1320 die("object %s inconsistent object length (%lu vs %lu)",
1321 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1325 src->index = create_delta_index(src->data, src_size);
1327 static int warned = 0;
1329 warning("suboptimal pack - out of memory");
1332 *mem_usage += sizeof_delta_index(src->index);
1335 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1339 if (trg_entry->delta) {
1340 /* Prefer only shallower same-sized deltas. */
1341 if (delta_size == trg_entry->delta_size &&
1342 src->depth + 1 >= trg->depth) {
1349 * Handle memory allocation outside of the cache
1350 * accounting lock. Compiler will optimize the strangeness
1351 * away when NO_PTHREADS is defined.
1353 free(trg_entry->delta_data);
1355 if (trg_entry->delta_data) {
1356 delta_cache_size -= trg_entry->delta_size;
1357 trg_entry->delta_data = NULL;
1359 if (delta_cacheable(src_size, trg_size, delta_size)) {
1360 delta_cache_size += delta_size;
1362 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1368 trg_entry->delta = src_entry;
1369 trg_entry->delta_size = delta_size;
1370 trg->depth = src->depth + 1;
1375 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1377 struct object_entry *child = me->delta_child;
1380 unsigned int c = check_delta_limit(child, n + 1);
1383 child = child->delta_sibling;
1388 static unsigned long free_unpacked(struct unpacked *n)
1390 unsigned long freed_mem = sizeof_delta_index(n->index);
1391 free_delta_index(n->index);
1394 freed_mem += n->entry->size;
1403 static void find_deltas(struct object_entry **list, unsigned *list_size,
1404 int window, int depth, unsigned *processed)
1406 uint32_t i, idx = 0, count = 0;
1407 struct unpacked *array;
1408 unsigned long mem_usage = 0;
1410 array = xcalloc(window, sizeof(struct unpacked));
1413 struct object_entry *entry;
1414 struct unpacked *n = array + idx;
1415 int j, max_depth, best_base = -1;
1424 if (!entry->preferred_base) {
1426 display_progress(progress_state, *processed);
1430 mem_usage -= free_unpacked(n);
1433 while (window_memory_limit &&
1434 mem_usage > window_memory_limit &&
1436 uint32_t tail = (idx + window - count) % window;
1437 mem_usage -= free_unpacked(array + tail);
1441 /* We do not compute delta to *create* objects we are not
1444 if (entry->preferred_base)
1448 * If the current object is at pack edge, take the depth the
1449 * objects that depend on the current object into account
1450 * otherwise they would become too deep.
1453 if (entry->delta_child) {
1454 max_depth -= check_delta_limit(entry, 0);
1462 uint32_t other_idx = idx + j;
1464 if (other_idx >= window)
1465 other_idx -= window;
1466 m = array + other_idx;
1469 ret = try_delta(n, m, max_depth, &mem_usage);
1473 best_base = other_idx;
1477 * If we decided to cache the delta data, then it is best
1478 * to compress it right away. First because we have to do
1479 * it anyway, and doing it here while we're threaded will
1480 * save a lot of time in the non threaded write phase,
1481 * as well as allow for caching more deltas within
1482 * the same cache size limit.
1484 * But only if not writing to stdout, since in that case
1485 * the network is most likely throttling writes anyway,
1486 * and therefore it is best to go to the write phase ASAP
1487 * instead, as we can afford spending more time compressing
1488 * between writes at that moment.
1490 if (entry->delta_data && !pack_to_stdout) {
1491 entry->z_delta_size = do_compress(&entry->delta_data,
1494 delta_cache_size -= entry->delta_size;
1495 delta_cache_size += entry->z_delta_size;
1499 /* if we made n a delta, and if n is already at max
1500 * depth, leaving it in the window is pointless. we
1501 * should evict it first.
1503 if (entry->delta && max_depth <= n->depth)
1507 * Move the best delta base up in the window, after the
1508 * currently deltified object, to keep it longer. It will
1509 * be the first base object to be attempted next.
1512 struct unpacked swap = array[best_base];
1513 int dist = (window + idx - best_base) % window;
1514 int dst = best_base;
1516 int src = (dst + 1) % window;
1517 array[dst] = array[src];
1525 if (count + 1 < window)
1531 for (i = 0; i < window; ++i) {
1532 free_delta_index(array[i].index);
1533 free(array[i].data);
1540 static void try_to_free_from_threads(size_t size)
1543 release_pack_memory(size, -1);
1547 static try_to_free_t old_try_to_free_routine;
1550 * The main thread waits on the condition that (at least) one of the workers
1551 * has stopped working (which is indicated in the .working member of
1552 * struct thread_params).
1553 * When a work thread has completed its work, it sets .working to 0 and
1554 * signals the main thread and waits on the condition that .data_ready
1558 struct thread_params {
1560 struct object_entry **list;
1567 pthread_mutex_t mutex;
1568 pthread_cond_t cond;
1569 unsigned *processed;
1572 static pthread_cond_t progress_cond;
1575 * Mutex and conditional variable can't be statically-initialized on Windows.
1577 static void init_threaded_search(void)
1579 init_recursive_mutex(&read_mutex);
1580 pthread_mutex_init(&cache_mutex, NULL);
1581 pthread_mutex_init(&progress_mutex, NULL);
1582 pthread_cond_init(&progress_cond, NULL);
1583 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
1586 static void cleanup_threaded_search(void)
1588 set_try_to_free_routine(old_try_to_free_routine);
1589 pthread_cond_destroy(&progress_cond);
1590 pthread_mutex_destroy(&read_mutex);
1591 pthread_mutex_destroy(&cache_mutex);
1592 pthread_mutex_destroy(&progress_mutex);
1595 static void *threaded_find_deltas(void *arg)
1597 struct thread_params *me = arg;
1599 while (me->remaining) {
1600 find_deltas(me->list, &me->remaining,
1601 me->window, me->depth, me->processed);
1605 pthread_cond_signal(&progress_cond);
1609 * We must not set ->data_ready before we wait on the
1610 * condition because the main thread may have set it to 1
1611 * before we get here. In order to be sure that new
1612 * work is available if we see 1 in ->data_ready, it
1613 * was initialized to 0 before this thread was spawned
1614 * and we reset it to 0 right away.
1616 pthread_mutex_lock(&me->mutex);
1617 while (!me->data_ready)
1618 pthread_cond_wait(&me->cond, &me->mutex);
1620 pthread_mutex_unlock(&me->mutex);
1622 /* leave ->working 1 so that this doesn't get more work assigned */
1626 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1627 int window, int depth, unsigned *processed)
1629 struct thread_params *p;
1630 int i, ret, active_threads = 0;
1632 init_threaded_search();
1634 if (!delta_search_threads) /* --threads=0 means autodetect */
1635 delta_search_threads = online_cpus();
1636 if (delta_search_threads <= 1) {
1637 find_deltas(list, &list_size, window, depth, processed);
1638 cleanup_threaded_search();
1641 if (progress > pack_to_stdout)
1642 fprintf(stderr, "Delta compression using up to %d threads.\n",
1643 delta_search_threads);
1644 p = xcalloc(delta_search_threads, sizeof(*p));
1646 /* Partition the work amongst work threads. */
1647 for (i = 0; i < delta_search_threads; i++) {
1648 unsigned sub_size = list_size / (delta_search_threads - i);
1650 /* don't use too small segments or no deltas will be found */
1651 if (sub_size < 2*window && i+1 < delta_search_threads)
1654 p[i].window = window;
1656 p[i].processed = processed;
1658 p[i].data_ready = 0;
1660 /* try to split chunks on "path" boundaries */
1661 while (sub_size && sub_size < list_size &&
1662 list[sub_size]->hash &&
1663 list[sub_size]->hash == list[sub_size-1]->hash)
1667 p[i].list_size = sub_size;
1668 p[i].remaining = sub_size;
1671 list_size -= sub_size;
1674 /* Start work threads. */
1675 for (i = 0; i < delta_search_threads; i++) {
1676 if (!p[i].list_size)
1678 pthread_mutex_init(&p[i].mutex, NULL);
1679 pthread_cond_init(&p[i].cond, NULL);
1680 ret = pthread_create(&p[i].thread, NULL,
1681 threaded_find_deltas, &p[i]);
1683 die("unable to create thread: %s", strerror(ret));
1688 * Now let's wait for work completion. Each time a thread is done
1689 * with its work, we steal half of the remaining work from the
1690 * thread with the largest number of unprocessed objects and give
1691 * it to that newly idle thread. This ensure good load balancing
1692 * until the remaining object list segments are simply too short
1693 * to be worth splitting anymore.
1695 while (active_threads) {
1696 struct thread_params *target = NULL;
1697 struct thread_params *victim = NULL;
1698 unsigned sub_size = 0;
1702 for (i = 0; !target && i < delta_search_threads; i++)
1707 pthread_cond_wait(&progress_cond, &progress_mutex);
1710 for (i = 0; i < delta_search_threads; i++)
1711 if (p[i].remaining > 2*window &&
1712 (!victim || victim->remaining < p[i].remaining))
1715 sub_size = victim->remaining / 2;
1716 list = victim->list + victim->list_size - sub_size;
1717 while (sub_size && list[0]->hash &&
1718 list[0]->hash == list[-1]->hash) {
1724 * It is possible for some "paths" to have
1725 * so many objects that no hash boundary
1726 * might be found. Let's just steal the
1727 * exact half in that case.
1729 sub_size = victim->remaining / 2;
1732 target->list = list;
1733 victim->list_size -= sub_size;
1734 victim->remaining -= sub_size;
1736 target->list_size = sub_size;
1737 target->remaining = sub_size;
1738 target->working = 1;
1741 pthread_mutex_lock(&target->mutex);
1742 target->data_ready = 1;
1743 pthread_cond_signal(&target->cond);
1744 pthread_mutex_unlock(&target->mutex);
1747 pthread_join(target->thread, NULL);
1748 pthread_cond_destroy(&target->cond);
1749 pthread_mutex_destroy(&target->mutex);
1753 cleanup_threaded_search();
1758 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1761 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1763 unsigned char peeled[20];
1765 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1766 !peel_ref(path, peeled) && /* peelable? */
1767 !is_null_sha1(peeled) && /* annotated tag? */
1768 locate_object_entry(peeled)) /* object packed? */
1769 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1773 static void prepare_pack(int window, int depth)
1775 struct object_entry **delta_list;
1776 uint32_t i, nr_deltas;
1779 get_object_details();
1782 * If we're locally repacking then we need to be doubly careful
1783 * from now on in order to make sure no stealth corruption gets
1784 * propagated to the new pack. Clients receiving streamed packs
1785 * should validate everything they get anyway so no need to incur
1786 * the additional cost here in that case.
1788 if (!pack_to_stdout)
1789 do_check_packed_object_crc = 1;
1791 if (!nr_objects || !window || !depth)
1794 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1797 for (i = 0; i < nr_objects; i++) {
1798 struct object_entry *entry = objects + i;
1801 /* This happens if we decided to reuse existing
1802 * delta from a pack. "reuse_delta &&" is implied.
1806 if (entry->size < 50)
1809 if (entry->no_try_delta)
1812 if (!entry->preferred_base) {
1814 if (entry->type < 0)
1815 die("unable to get type of object %s",
1816 sha1_to_hex(entry->idx.sha1));
1818 if (entry->type < 0) {
1820 * This object is not found, but we
1821 * don't have to include it anyway.
1827 delta_list[n++] = entry;
1830 if (nr_deltas && n > 1) {
1831 unsigned nr_done = 0;
1833 progress_state = start_progress("Compressing objects",
1835 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1836 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1837 stop_progress(&progress_state);
1838 if (nr_done != nr_deltas)
1839 die("inconsistency with delta count");
1844 static int git_pack_config(const char *k, const char *v, void *cb)
1846 if (!strcmp(k, "pack.window")) {
1847 window = git_config_int(k, v);
1850 if (!strcmp(k, "pack.windowmemory")) {
1851 window_memory_limit = git_config_ulong(k, v);
1854 if (!strcmp(k, "pack.depth")) {
1855 depth = git_config_int(k, v);
1858 if (!strcmp(k, "pack.compression")) {
1859 int level = git_config_int(k, v);
1861 level = Z_DEFAULT_COMPRESSION;
1862 else if (level < 0 || level > Z_BEST_COMPRESSION)
1863 die("bad pack compression level %d", level);
1864 pack_compression_level = level;
1865 pack_compression_seen = 1;
1868 if (!strcmp(k, "pack.deltacachesize")) {
1869 max_delta_cache_size = git_config_int(k, v);
1872 if (!strcmp(k, "pack.deltacachelimit")) {
1873 cache_max_small_delta_size = git_config_int(k, v);
1876 if (!strcmp(k, "pack.threads")) {
1877 delta_search_threads = git_config_int(k, v);
1878 if (delta_search_threads < 0)
1879 die("invalid number of threads specified (%d)",
1880 delta_search_threads);
1882 if (delta_search_threads != 1)
1883 warning("no threads support, ignoring %s", k);
1887 if (!strcmp(k, "pack.indexversion")) {
1888 pack_idx_opts.version = git_config_int(k, v);
1889 if (pack_idx_opts.version > 2)
1890 die("bad pack.indexversion=%"PRIu32,
1891 pack_idx_opts.version);
1894 if (!strcmp(k, "pack.packsizelimit")) {
1895 pack_size_limit_cfg = git_config_ulong(k, v);
1898 return git_default_config(k, v, cb);
1901 static void read_object_list_from_stdin(void)
1903 char line[40 + 1 + PATH_MAX + 2];
1904 unsigned char sha1[20];
1907 if (!fgets(line, sizeof(line), stdin)) {
1911 die("fgets returned NULL, not EOF, not error!");
1917 if (line[0] == '-') {
1918 if (get_sha1_hex(line+1, sha1))
1919 die("expected edge sha1, got garbage:\n %s",
1921 add_preferred_base(sha1);
1924 if (get_sha1_hex(line, sha1))
1925 die("expected sha1, got garbage:\n %s", line);
1927 add_preferred_base_object(line+41);
1928 add_object_entry(sha1, 0, line+41, 0);
1932 #define OBJECT_ADDED (1u<<20)
1934 static void show_commit(struct commit *commit, void *data)
1936 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1937 commit->object.flags |= OBJECT_ADDED;
1940 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1942 char *name = path_name(path, last);
1944 add_preferred_base_object(name);
1945 add_object_entry(obj->sha1, obj->type, name, 0);
1946 obj->flags |= OBJECT_ADDED;
1949 * We will have generated the hash from the name,
1950 * but not saved a pointer to it - we can free it
1955 static void show_edge(struct commit *commit)
1957 add_preferred_base(commit->object.sha1);
1960 struct in_pack_object {
1962 struct object *object;
1968 struct in_pack_object *array;
1971 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1973 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1974 in_pack->array[in_pack->nr].object = object;
1979 * Compare the objects in the offset order, in order to emulate the
1980 * "git rev-list --objects" output that produced the pack originally.
1982 static int ofscmp(const void *a_, const void *b_)
1984 struct in_pack_object *a = (struct in_pack_object *)a_;
1985 struct in_pack_object *b = (struct in_pack_object *)b_;
1987 if (a->offset < b->offset)
1989 else if (a->offset > b->offset)
1992 return hashcmp(a->object->sha1, b->object->sha1);
1995 static void add_objects_in_unpacked_packs(struct rev_info *revs)
1997 struct packed_git *p;
1998 struct in_pack in_pack;
2001 memset(&in_pack, 0, sizeof(in_pack));
2003 for (p = packed_git; p; p = p->next) {
2004 const unsigned char *sha1;
2007 if (!p->pack_local || p->pack_keep)
2009 if (open_pack_index(p))
2010 die("cannot open pack index");
2012 ALLOC_GROW(in_pack.array,
2013 in_pack.nr + p->num_objects,
2016 for (i = 0; i < p->num_objects; i++) {
2017 sha1 = nth_packed_object_sha1(p, i);
2018 o = lookup_unknown_object(sha1);
2019 if (!(o->flags & OBJECT_ADDED))
2020 mark_in_pack_object(o, p, &in_pack);
2021 o->flags |= OBJECT_ADDED;
2026 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2028 for (i = 0; i < in_pack.nr; i++) {
2029 struct object *o = in_pack.array[i].object;
2030 add_object_entry(o->sha1, o->type, "", 0);
2033 free(in_pack.array);
2036 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2038 static struct packed_git *last_found = (void *)1;
2039 struct packed_git *p;
2041 p = (last_found != (void *)1) ? last_found : packed_git;
2044 if ((!p->pack_local || p->pack_keep) &&
2045 find_pack_entry_one(sha1, p)) {
2049 if (p == last_found)
2053 if (p == last_found)
2059 static void loosen_unused_packed_objects(struct rev_info *revs)
2061 struct packed_git *p;
2063 const unsigned char *sha1;
2065 for (p = packed_git; p; p = p->next) {
2066 if (!p->pack_local || p->pack_keep)
2069 if (open_pack_index(p))
2070 die("cannot open pack index");
2072 for (i = 0; i < p->num_objects; i++) {
2073 sha1 = nth_packed_object_sha1(p, i);
2074 if (!locate_object_entry(sha1) &&
2075 !has_sha1_pack_kept_or_nonlocal(sha1))
2076 if (force_object_loose(sha1, p->mtime))
2077 die("unable to force loose object");
2082 static void get_object_list(int ac, const char **av)
2084 struct rev_info revs;
2088 init_revisions(&revs, NULL);
2089 save_commit_buffer = 0;
2090 setup_revisions(ac, av, &revs, NULL);
2092 while (fgets(line, sizeof(line), stdin) != NULL) {
2093 int len = strlen(line);
2094 if (len && line[len - 1] == '\n')
2099 if (!strcmp(line, "--not")) {
2100 flags ^= UNINTERESTING;
2103 die("not a rev '%s'", line);
2105 if (handle_revision_arg(line, &revs, flags, 1))
2106 die("bad revision '%s'", line);
2109 if (prepare_revision_walk(&revs))
2110 die("revision walk setup failed");
2111 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2112 traverse_commit_list(&revs, show_commit, show_object, NULL);
2114 if (keep_unreachable)
2115 add_objects_in_unpacked_packs(&revs);
2116 if (unpack_unreachable)
2117 loosen_unused_packed_objects(&revs);
2120 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2122 int use_internal_rev_list = 0;
2124 int all_progress_implied = 0;
2127 int rp_ac_alloc = 64;
2130 read_replace_refs = 0;
2132 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2134 rp_av[0] = "pack-objects";
2135 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2138 reset_pack_idx_option(&pack_idx_opts);
2139 git_config(git_pack_config, NULL);
2140 if (!pack_compression_seen && core_compression_seen)
2141 pack_compression_level = core_compression_level;
2143 progress = isatty(2);
2144 for (i = 1; i < argc; i++) {
2145 const char *arg = argv[i];
2150 if (!strcmp("--non-empty", arg)) {
2154 if (!strcmp("--local", arg)) {
2158 if (!strcmp("--incremental", arg)) {
2162 if (!strcmp("--honor-pack-keep", arg)) {
2163 ignore_packed_keep = 1;
2166 if (!prefixcmp(arg, "--compression=")) {
2168 int level = strtoul(arg+14, &end, 0);
2169 if (!arg[14] || *end)
2172 level = Z_DEFAULT_COMPRESSION;
2173 else if (level < 0 || level > Z_BEST_COMPRESSION)
2174 die("bad pack compression level %d", level);
2175 pack_compression_level = level;
2178 if (!prefixcmp(arg, "--max-pack-size=")) {
2179 pack_size_limit_cfg = 0;
2180 if (!git_parse_ulong(arg+16, &pack_size_limit))
2184 if (!prefixcmp(arg, "--window=")) {
2186 window = strtoul(arg+9, &end, 0);
2187 if (!arg[9] || *end)
2191 if (!prefixcmp(arg, "--window-memory=")) {
2192 if (!git_parse_ulong(arg+16, &window_memory_limit))
2196 if (!prefixcmp(arg, "--threads=")) {
2198 delta_search_threads = strtoul(arg+10, &end, 0);
2199 if (!arg[10] || *end || delta_search_threads < 0)
2202 if (delta_search_threads != 1)
2203 warning("no threads support, "
2204 "ignoring %s", arg);
2208 if (!prefixcmp(arg, "--depth=")) {
2210 depth = strtoul(arg+8, &end, 0);
2211 if (!arg[8] || *end)
2215 if (!strcmp("--progress", arg)) {
2219 if (!strcmp("--all-progress", arg)) {
2223 if (!strcmp("--all-progress-implied", arg)) {
2224 all_progress_implied = 1;
2227 if (!strcmp("-q", arg)) {
2231 if (!strcmp("--no-reuse-delta", arg)) {
2235 if (!strcmp("--no-reuse-object", arg)) {
2236 reuse_object = reuse_delta = 0;
2239 if (!strcmp("--delta-base-offset", arg)) {
2240 allow_ofs_delta = 1;
2243 if (!strcmp("--stdout", arg)) {
2247 if (!strcmp("--revs", arg)) {
2248 use_internal_rev_list = 1;
2251 if (!strcmp("--keep-unreachable", arg)) {
2252 keep_unreachable = 1;
2255 if (!strcmp("--unpack-unreachable", arg)) {
2256 unpack_unreachable = 1;
2259 if (!strcmp("--include-tag", arg)) {
2263 if (!strcmp("--unpacked", arg) ||
2264 !strcmp("--reflog", arg) ||
2265 !strcmp("--all", arg)) {
2266 use_internal_rev_list = 1;
2267 if (rp_ac >= rp_ac_alloc - 1) {
2268 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2269 rp_av = xrealloc(rp_av,
2270 rp_ac_alloc * sizeof(*rp_av));
2272 rp_av[rp_ac++] = arg;
2275 if (!strcmp("--thin", arg)) {
2276 use_internal_rev_list = 1;
2278 rp_av[1] = "--objects-edge";
2281 if (!prefixcmp(arg, "--index-version=")) {
2283 pack_idx_opts.version = strtoul(arg + 16, &c, 10);
2284 if (pack_idx_opts.version > 2)
2287 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2288 if (*c || pack_idx_opts.off32_limit & 0x80000000)
2292 if (!strcmp(arg, "--keep-true-parents")) {
2293 grafts_replace_parents = 0;
2299 /* Traditionally "pack-objects [options] base extra" failed;
2300 * we would however want to take refs parameter that would
2301 * have been given to upstream rev-list ourselves, which means
2302 * we somehow want to say what the base name is. So the
2305 * pack-objects [options] base <refs...>
2307 * in other words, we would treat the first non-option as the
2308 * base_name and send everything else to the internal revision
2312 if (!pack_to_stdout)
2313 base_name = argv[i++];
2315 if (pack_to_stdout != !base_name)
2318 if (!pack_to_stdout && !pack_size_limit)
2319 pack_size_limit = pack_size_limit_cfg;
2320 if (pack_to_stdout && pack_size_limit)
2321 die("--max-pack-size cannot be used to build a pack for transfer.");
2322 if (pack_size_limit && pack_size_limit < 1024*1024) {
2323 warning("minimum pack size limit is 1 MiB");
2324 pack_size_limit = 1024*1024;
2327 if (!pack_to_stdout && thin)
2328 die("--thin cannot be used to build an indexable pack.");
2330 if (keep_unreachable && unpack_unreachable)
2331 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2333 if (progress && all_progress_implied)
2336 prepare_packed_git();
2339 progress_state = start_progress("Counting objects", 0);
2340 if (!use_internal_rev_list)
2341 read_object_list_from_stdin();
2343 rp_av[rp_ac] = NULL;
2344 get_object_list(rp_ac, rp_av);
2346 cleanup_preferred_base();
2347 if (include_tag && nr_result)
2348 for_each_ref(add_ref_tag, NULL);
2349 stop_progress(&progress_state);
2351 if (non_empty && !nr_result)
2354 prepare_pack(window, depth);
2357 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2358 " reused %"PRIu32" (delta %"PRIu32")\n",
2359 written, written_delta, reused, reused_delta);