9 static const char pack_usage[] = "git-pack-objects [-q] [--no-reuse-delta] [--non-empty] [--local] [--incremental] [--window=N] [--depth=N] {--stdout | base-name} < object-list";
12 unsigned char sha1[20];
13 unsigned long size; /* uncompressed size */
14 unsigned long offset; /* offset into the final pack file;
15 * nonzero if already written.
17 unsigned int depth; /* delta depth */
18 unsigned int delta_limit; /* base adjustment for in-pack delta */
19 unsigned int hash; /* name hint hash */
20 enum object_type type;
21 enum object_type in_pack_type; /* could be delta */
22 unsigned long delta_size; /* delta data size (uncompressed) */
23 struct object_entry *delta; /* delta base object */
24 struct packed_git *in_pack; /* already in pack */
25 unsigned int in_pack_offset;
26 struct object_entry *delta_child; /* delitified objects who bases me */
27 struct object_entry *delta_sibling; /* other deltified objects who
28 * uses the same base as me
33 * Objects we are going to pack are colected in objects array (dynamically
34 * expanded). nr_objects & nr_alloc controls this array. They are stored
35 * in the order we see -- typically rev-list --objects order that gives us
36 * nice "minimum seek" order.
38 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
39 * elements in the objects array. The former is used to build the pack
40 * index (lists object names in the ascending order to help offset lookup),
41 * and the latter is used to group similar things together by try_delta()
45 static unsigned char object_list_sha1[20];
46 static int non_empty = 0;
47 static int no_reuse_delta = 0;
49 static int incremental = 0;
50 static struct object_entry **sorted_by_sha, **sorted_by_type;
51 static struct object_entry *objects = NULL;
52 static int nr_objects = 0, nr_alloc = 0;
53 static const char *base_name;
54 static unsigned char pack_file_sha1[20];
55 static int progress = 1;
58 * The object names in objects array are hashed with this hashtable,
59 * to help looking up the entry by object name. Binary search from
60 * sorted_by_sha is also possible but this was easier to code and faster.
61 * This hashtable is built after all the objects are seen.
63 static int *object_ix = NULL;
64 static int object_ix_hashsz = 0;
67 * Pack index for existing packs give us easy access to the offsets into
68 * corresponding pack file where each object's data starts, but the entries
69 * do not store the size of the compressed representation (uncompressed
70 * size is easily available by examining the pack entry header). We build
71 * a hashtable of existing packs (pack_revindex), and keep reverse index
72 * here -- pack index file is sorted by object name mapping to offset; this
73 * pack_revindex[].revindex array is an ordered list of offsets, so if you
74 * know the offset of an object, next offset is where its packed
75 * representation ends.
77 struct pack_revindex {
79 unsigned long *revindex;
80 } *pack_revindex = NULL;
81 static int pack_revindex_hashsz = 0;
86 static int written = 0;
87 static int written_delta = 0;
88 static int reused = 0;
89 static int reused_delta = 0;
91 static int pack_revindex_ix(struct packed_git *p)
93 unsigned int ui = (unsigned int) p;
96 ui = ui ^ (ui >> 16); /* defeat structure alignment */
97 i = (int)(ui % pack_revindex_hashsz);
98 while (pack_revindex[i].p) {
99 if (pack_revindex[i].p == p)
101 if (++i == pack_revindex_hashsz)
107 static void prepare_pack_ix(void)
110 struct packed_git *p;
111 for (num = 0, p = packed_git; p; p = p->next)
115 pack_revindex_hashsz = num * 11;
116 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
117 for (p = packed_git; p; p = p->next) {
118 num = pack_revindex_ix(p);
120 pack_revindex[num].p = p;
122 /* revindex elements are lazily initialized */
125 static int cmp_offset(const void *a_, const void *b_)
127 unsigned long a = *(unsigned long *) a_;
128 unsigned long b = *(unsigned long *) b_;
138 * Ordered list of offsets of objects in the pack.
140 static void prepare_pack_revindex(struct pack_revindex *rix)
142 struct packed_git *p = rix->p;
143 int num_ent = num_packed_objects(p);
145 void *index = p->index_base + 256;
147 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
148 for (i = 0; i < num_ent; i++) {
149 long hl = *((long *)(index + 24 * i));
150 rix->revindex[i] = ntohl(hl);
152 /* This knows the pack format -- the 20-byte trailer
153 * follows immediately after the last object data.
155 rix->revindex[num_ent] = p->pack_size - 20;
156 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
159 static unsigned long find_packed_object_size(struct packed_git *p,
164 struct pack_revindex *rix;
165 unsigned long *revindex;
166 num = pack_revindex_ix(p);
168 die("internal error: pack revindex uninitialized");
169 rix = &pack_revindex[num];
171 prepare_pack_revindex(rix);
172 revindex = rix->revindex;
174 hi = num_packed_objects(p) + 1;
176 int mi = (lo + hi) / 2;
177 if (revindex[mi] == ofs) {
178 return revindex[mi+1] - ofs;
180 else if (ofs < revindex[mi])
185 die("internal error: pack revindex corrupt");
188 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
190 unsigned long othersize, delta_size;
192 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
196 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
197 delta_buf = diff_delta(otherbuf, othersize,
198 buf, size, &delta_size, 0);
199 if (!delta_buf || delta_size != entry->delta_size)
200 die("delta size changed");
207 * The per-object header is a pretty dense thing, which is
208 * - first byte: low four bits are "size", then three bits of "type",
209 * and the high bit is "size continues".
210 * - each byte afterwards: low seven bits are size continuation,
211 * with the high bit being "size continues"
213 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
218 if (type < OBJ_COMMIT || type > OBJ_DELTA)
219 die("bad type %d", type);
221 c = (type << 4) | (size & 15);
233 static unsigned long write_object(struct sha1file *f, struct object_entry *entry)
238 unsigned char header[10];
239 unsigned hdrlen, datalen;
240 enum object_type obj_type;
243 obj_type = entry->type;
244 if (! entry->in_pack)
245 to_reuse = 0; /* can't reuse what we don't have */
246 else if (obj_type == OBJ_DELTA)
247 to_reuse = 1; /* check_object() decided it for us */
248 else if (obj_type != entry->in_pack_type)
249 to_reuse = 0; /* pack has delta which is unusable */
250 else if (entry->delta)
251 to_reuse = 0; /* we want to pack afresh */
253 to_reuse = 1; /* we have it in-pack undeltified,
254 * and we do not need to deltify it.
258 buf = read_sha1_file(entry->sha1, type, &size);
260 die("unable to read %s", sha1_to_hex(entry->sha1));
261 if (size != entry->size)
262 die("object %s size inconsistency (%lu vs %lu)",
263 sha1_to_hex(entry->sha1), size, entry->size);
265 buf = delta_against(buf, size, entry);
266 size = entry->delta_size;
267 obj_type = OBJ_DELTA;
270 * The object header is a byte of 'type' followed by zero or
271 * more bytes of length. For deltas, the 20 bytes of delta
274 hdrlen = encode_header(obj_type, size, header);
275 sha1write(f, header, hdrlen);
278 sha1write(f, entry->delta, 20);
281 datalen = sha1write_compressed(f, buf, size);
285 struct packed_git *p = entry->in_pack;
288 datalen = find_packed_object_size(p, entry->in_pack_offset);
289 buf = p->pack_base + entry->in_pack_offset;
290 sha1write(f, buf, datalen);
292 hdrlen = 0; /* not really */
293 if (obj_type == OBJ_DELTA)
297 if (obj_type == OBJ_DELTA)
300 return hdrlen + datalen;
303 static unsigned long write_one(struct sha1file *f,
304 struct object_entry *e,
305 unsigned long offset)
308 /* offset starts from header size and cannot be zero
309 * if it is written already.
313 offset += write_object(f, e);
314 /* if we are deltified, write out its base object. */
316 offset = write_one(f, e->delta, offset);
320 static void write_pack_file(void)
324 unsigned long offset;
325 struct pack_header hdr;
328 f = sha1fd(1, "<stdout>");
330 f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "pack");
331 hdr.hdr_signature = htonl(PACK_SIGNATURE);
332 hdr.hdr_version = htonl(PACK_VERSION);
333 hdr.hdr_entries = htonl(nr_objects);
334 sha1write(f, &hdr, sizeof(hdr));
335 offset = sizeof(hdr);
336 for (i = 0; i < nr_objects; i++)
337 offset = write_one(f, objects + i, offset);
339 sha1close(f, pack_file_sha1, 1);
342 static void write_index_file(void)
345 struct sha1file *f = sha1create("%s-%s.%s", base_name, sha1_to_hex(object_list_sha1), "idx");
346 struct object_entry **list = sorted_by_sha;
347 struct object_entry **last = list + nr_objects;
348 unsigned int array[256];
351 * Write the first-level table (the list is sorted,
352 * but we use a 256-entry lookup to be able to avoid
353 * having to do eight extra binary search iterations).
355 for (i = 0; i < 256; i++) {
356 struct object_entry **next = list;
357 while (next < last) {
358 struct object_entry *entry = *next;
359 if (entry->sha1[0] != i)
363 array[i] = htonl(next - sorted_by_sha);
366 sha1write(f, array, 256 * sizeof(int));
369 * Write the actual SHA1 entries..
371 list = sorted_by_sha;
372 for (i = 0; i < nr_objects; i++) {
373 struct object_entry *entry = *list++;
374 unsigned int offset = htonl(entry->offset);
375 sha1write(f, &offset, 4);
376 sha1write(f, entry->sha1, 20);
378 sha1write(f, pack_file_sha1, 20);
379 sha1close(f, NULL, 1);
382 static int add_object_entry(unsigned char *sha1, unsigned int hash)
384 unsigned int idx = nr_objects;
385 struct object_entry *entry;
386 struct packed_git *p;
387 unsigned int found_offset = 0;
388 struct packed_git *found_pack = NULL;
390 for (p = packed_git; p; p = p->next) {
392 if (find_pack_entry_one(sha1, &e, p)) {
395 if (local && !p->pack_local)
398 found_offset = e.offset;
404 if (idx >= nr_alloc) {
405 unsigned int needed = (idx + 1024) * 3 / 2;
406 objects = xrealloc(objects, needed * sizeof(*entry));
409 entry = objects + idx;
410 memset(entry, 0, sizeof(*entry));
411 memcpy(entry->sha1, sha1, 20);
414 entry->in_pack = found_pack;
415 entry->in_pack_offset = found_offset;
421 static int locate_object_entry_hash(unsigned char *sha1)
425 memcpy(&ui, sha1, sizeof(unsigned int));
426 i = ui % object_ix_hashsz;
427 while (0 < object_ix[i]) {
428 if (!memcmp(sha1, objects[object_ix[i]-1].sha1, 20))
430 if (++i == object_ix_hashsz)
436 static struct object_entry *locate_object_entry(unsigned char *sha1)
438 int i = locate_object_entry_hash(sha1);
440 return &objects[object_ix[i]-1];
444 static void check_object(struct object_entry *entry)
448 if (entry->in_pack) {
449 unsigned char base[20];
451 struct object_entry *base_entry;
453 /* We want in_pack_type even if we do not reuse delta.
454 * There is no point not reusing non-delta representations.
456 check_reuse_pack_delta(entry->in_pack,
457 entry->in_pack_offset,
459 &entry->in_pack_type);
461 /* Check if it is delta, and the base is also an object
462 * we are going to pack. If so we will reuse the existing
465 if (!no_reuse_delta &&
466 entry->in_pack_type == OBJ_DELTA &&
467 (base_entry = locate_object_entry(base))) {
469 /* Depth value does not matter - find_deltas()
470 * will never consider reused delta as the
471 * base object to deltify other objects
472 * against, in order to avoid circular deltas.
475 /* uncompressed size of the delta data */
476 entry->size = entry->delta_size = size;
477 entry->delta = base_entry;
478 entry->type = OBJ_DELTA;
480 entry->delta_sibling = base_entry->delta_child;
481 base_entry->delta_child = entry;
485 /* Otherwise we would do the usual */
488 if (sha1_object_info(entry->sha1, type, &entry->size))
489 die("unable to get type of object %s",
490 sha1_to_hex(entry->sha1));
492 if (!strcmp(type, "commit")) {
493 entry->type = OBJ_COMMIT;
494 } else if (!strcmp(type, "tree")) {
495 entry->type = OBJ_TREE;
496 } else if (!strcmp(type, "blob")) {
497 entry->type = OBJ_BLOB;
498 } else if (!strcmp(type, "tag")) {
499 entry->type = OBJ_TAG;
501 die("unable to pack object %s of type %s",
502 sha1_to_hex(entry->sha1), type);
505 static void hash_objects(void)
508 struct object_entry *oe;
510 object_ix_hashsz = nr_objects * 2;
511 object_ix = xcalloc(sizeof(int), object_ix_hashsz);
512 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
513 int ix = locate_object_entry_hash(oe->sha1);
515 error("the same object '%s' added twice",
516 sha1_to_hex(oe->sha1));
520 object_ix[ix] = i + 1;
524 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
526 struct object_entry *child = me->delta_child;
529 unsigned int c = check_delta_limit(child, n + 1);
532 child = child->delta_sibling;
537 static void get_object_details(void)
540 struct object_entry *entry;
544 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
546 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
547 if (!entry->delta && entry->delta_child)
549 check_delta_limit(entry, 1);
552 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
554 static entry_sort_t current_sort;
556 static int sort_comparator(const void *_a, const void *_b)
558 struct object_entry *a = *(struct object_entry **)_a;
559 struct object_entry *b = *(struct object_entry **)_b;
560 return current_sort(a,b);
563 static struct object_entry **create_sorted_list(entry_sort_t sort)
565 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
568 for (i = 0; i < nr_objects; i++)
569 list[i] = objects + i;
571 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
575 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
577 return memcmp(a->sha1, b->sha1, 20);
580 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
582 if (a->type < b->type)
584 if (a->type > b->type)
586 if (a->hash < b->hash)
588 if (a->hash > b->hash)
590 if (a->size < b->size)
592 if (a->size > b->size)
594 return a < b ? -1 : (a > b);
598 struct object_entry *entry;
603 * We search for deltas _backwards_ in a list sorted by type and
604 * by size, so that we see progressively smaller and smaller files.
605 * That's because we prefer deltas to be from the bigger file
606 * to the smaller - deletes are potentially cheaper, but perhaps
607 * more importantly, the bigger file is likely the more recent
610 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
612 struct object_entry *cur_entry = cur->entry;
613 struct object_entry *old_entry = old->entry;
614 unsigned long size, oldsize, delta_size, sizediff;
618 /* Don't bother doing diffs between different types */
619 if (cur_entry->type != old_entry->type)
622 /* If the current object is at edge, take the depth the objects
623 * that depend on the current object into account -- otherwise
624 * they would become too deep.
626 if (cur_entry->delta_child) {
627 if (max_depth <= cur_entry->delta_limit)
629 max_depth -= cur_entry->delta_limit;
632 size = cur_entry->size;
635 oldsize = old_entry->size;
636 sizediff = oldsize > size ? oldsize - size : size - oldsize;
637 if (sizediff > size / 8)
639 if (old_entry->depth >= max_depth)
645 * We always delta from the bigger to the smaller, since that's
646 * more space-efficient (deletes don't have to say _what_ they
649 max_size = size / 2 - 20;
650 if (cur_entry->delta)
651 max_size = cur_entry->delta_size-1;
652 if (sizediff >= max_size)
654 delta_buf = diff_delta(old->data, oldsize,
655 cur->data, size, &delta_size, max_size);
658 cur_entry->delta = old_entry;
659 cur_entry->delta_size = delta_size;
660 cur_entry->depth = old_entry->depth + 1;
665 static volatile int progress_update = 0;
666 static void progress_interval(int signum)
668 signal(SIGALRM, progress_interval);
672 static void find_deltas(struct object_entry **list, int window, int depth)
675 unsigned int array_size = window * sizeof(struct unpacked);
676 struct unpacked *array = xmalloc(array_size);
678 memset(array, 0, array_size);
683 struct object_entry *entry = list[i];
684 struct unpacked *n = array + idx;
689 if (progress_update || i == 0) {
690 fprintf(stderr, "Deltifying (%d %d%%)\r",
691 nr_objects-i, (nr_objects-i) * 100/nr_objects);
696 /* This happens if we decided to reuse existing
697 * delta from a pack. "!no_reuse_delta &&" is implied.
703 n->data = read_sha1_file(entry->sha1, type, &size);
704 if (size != entry->size)
705 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
709 unsigned int other_idx = idx + j;
711 if (other_idx >= window)
713 m = array + other_idx;
716 if (try_delta(n, m, depth) < 0)
727 for (i = 0; i < window; ++i)
732 static void prepare_pack(int window, int depth)
734 get_object_details();
735 sorted_by_type = create_sorted_list(type_size_sort);
737 find_deltas(sorted_by_type, window+1, depth);
741 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
743 static const char cache[] = "pack-cache/pack-%s.%s";
744 char *cached_pack, *cached_idx;
745 int ifd, ofd, ifd_ix = -1;
747 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
748 ifd = open(cached_pack, O_RDONLY);
752 if (!pack_to_stdout) {
753 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
754 ifd_ix = open(cached_idx, O_RDONLY);
762 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
765 if (pack_to_stdout) {
772 snprintf(name, sizeof(name),
773 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
774 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
776 die("unable to open %s (%s)", name, strerror(errno));
777 if (copy_fd(ifd, ofd))
781 snprintf(name, sizeof(name),
782 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
783 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
785 die("unable to open %s (%s)", name, strerror(errno));
786 if (copy_fd(ifd_ix, ofd))
789 puts(sha1_to_hex(sha1));
795 int main(int argc, char **argv)
798 char line[PATH_MAX + 20];
799 int window = 10, depth = 10, pack_to_stdout = 0;
800 struct object_entry **list;
803 setup_git_directory();
805 for (i = 1; i < argc; i++) {
806 const char *arg = argv[i];
809 if (!strcmp("--non-empty", arg)) {
813 if (!strcmp("--local", arg)) {
817 if (!strcmp("--incremental", arg)) {
821 if (!strncmp("--window=", arg, 9)) {
823 window = strtoul(arg+9, &end, 0);
828 if (!strncmp("--depth=", arg, 8)) {
830 depth = strtoul(arg+8, &end, 0);
835 if (!strcmp("-q", arg)) {
839 if (!strcmp("--no-reuse-delta", arg)) {
843 if (!strcmp("--stdout", arg)) {
854 if (pack_to_stdout != !base_name)
857 prepare_packed_git();
861 v.it_interval.tv_sec = 1;
862 v.it_interval.tv_usec = 0;
863 v.it_value = v.it_interval;
864 signal(SIGALRM, progress_interval);
865 setitimer(ITIMER_REAL, &v, NULL);
866 fprintf(stderr, "Generating pack...\n");
869 while (fgets(line, sizeof(line), stdin) != NULL) {
872 unsigned char sha1[20];
874 if (progress_update) {
875 fprintf(stderr, "Counting objects...%d\r", nr_objects);
878 if (get_sha1_hex(line, sha1))
879 die("expected sha1, got garbage:\n %s", line);
883 unsigned char c = *p++;
886 hash = hash * 11 + c;
888 add_object_entry(sha1, hash);
891 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
892 if (non_empty && !nr_objects)
895 sorted_by_sha = create_sorted_list(sha1_sort);
897 list = sorted_by_sha;
898 for (i = 0; i < nr_objects; i++) {
899 struct object_entry *entry = *list++;
900 SHA1_Update(&ctx, entry->sha1, 20);
902 SHA1_Final(object_list_sha1, &ctx);
904 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
907 prepare_pack(window, depth);
908 if (!pack_to_stdout) {
910 puts(sha1_to_hex(object_list_sha1));
914 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
915 nr_objects, written, written_delta, reused, reused_delta);