10 #include "tree-walk.h"
14 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";
17 unsigned char sha1[20];
18 unsigned long size; /* uncompressed size */
19 unsigned long offset; /* offset into the final pack file;
20 * nonzero if already written.
22 unsigned int depth; /* delta depth */
23 unsigned int delta_limit; /* base adjustment for in-pack delta */
24 unsigned int hash; /* name hint hash */
25 enum object_type type;
26 enum object_type in_pack_type; /* could be delta */
27 unsigned long delta_size; /* delta data size (uncompressed) */
28 struct object_entry *delta; /* delta base object */
29 struct packed_git *in_pack; /* already in pack */
30 unsigned int in_pack_offset;
31 struct object_entry *delta_child; /* deltified objects who bases me */
32 struct object_entry *delta_sibling; /* other deltified objects who
33 * uses the same base as me
35 int preferred_base; /* we do not pack this, but is encouraged to
36 * be used as the base objectto delta huge
42 * Objects we are going to pack are collected in objects array (dynamically
43 * expanded). nr_objects & nr_alloc controls this array. They are stored
44 * in the order we see -- typically rev-list --objects order that gives us
45 * nice "minimum seek" order.
47 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
48 * elements in the objects array. The former is used to build the pack
49 * index (lists object names in the ascending order to help offset lookup),
50 * and the latter is used to group similar things together by try_delta()
54 static unsigned char object_list_sha1[20];
55 static int non_empty = 0;
56 static int no_reuse_delta = 0;
58 static int incremental = 0;
59 static struct object_entry **sorted_by_sha, **sorted_by_type;
60 static struct object_entry *objects = NULL;
61 static int nr_objects = 0, nr_alloc = 0, nr_result = 0;
62 static const char *base_name;
63 static unsigned char pack_file_sha1[20];
64 static int progress = 1;
65 static volatile sig_atomic_t progress_update = 0;
66 static int window = 10;
69 * The object names in objects array are hashed with this hashtable,
70 * to help looking up the entry by object name. Binary search from
71 * sorted_by_sha is also possible but this was easier to code and faster.
72 * This hashtable is built after all the objects are seen.
74 static int *object_ix = NULL;
75 static int object_ix_hashsz = 0;
78 * Pack index for existing packs give us easy access to the offsets into
79 * corresponding pack file where each object's data starts, but the entries
80 * do not store the size of the compressed representation (uncompressed
81 * size is easily available by examining the pack entry header). We build
82 * a hashtable of existing packs (pack_revindex), and keep reverse index
83 * here -- pack index file is sorted by object name mapping to offset; this
84 * pack_revindex[].revindex array is an ordered list of offsets, so if you
85 * know the offset of an object, next offset is where its packed
86 * representation ends.
88 struct pack_revindex {
90 unsigned long *revindex;
91 } *pack_revindex = NULL;
92 static int pack_revindex_hashsz = 0;
97 static int written = 0;
98 static int written_delta = 0;
99 static int reused = 0;
100 static int reused_delta = 0;
102 static int pack_revindex_ix(struct packed_git *p)
104 unsigned long ui = (unsigned long)p;
107 ui = ui ^ (ui >> 16); /* defeat structure alignment */
108 i = (int)(ui % pack_revindex_hashsz);
109 while (pack_revindex[i].p) {
110 if (pack_revindex[i].p == p)
112 if (++i == pack_revindex_hashsz)
118 static void prepare_pack_ix(void)
121 struct packed_git *p;
122 for (num = 0, p = packed_git; p; p = p->next)
126 pack_revindex_hashsz = num * 11;
127 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
128 for (p = packed_git; p; p = p->next) {
129 num = pack_revindex_ix(p);
131 pack_revindex[num].p = p;
133 /* revindex elements are lazily initialized */
136 static int cmp_offset(const void *a_, const void *b_)
138 unsigned long a = *(unsigned long *) a_;
139 unsigned long b = *(unsigned long *) b_;
149 * Ordered list of offsets of objects in the pack.
151 static void prepare_pack_revindex(struct pack_revindex *rix)
153 struct packed_git *p = rix->p;
154 int num_ent = num_packed_objects(p);
156 void *index = p->index_base + 256;
158 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
159 for (i = 0; i < num_ent; i++) {
160 unsigned int hl = *((unsigned int *)((char *) index + 24*i));
161 rix->revindex[i] = ntohl(hl);
163 /* This knows the pack format -- the 20-byte trailer
164 * follows immediately after the last object data.
166 rix->revindex[num_ent] = p->pack_size - 20;
167 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
170 static unsigned long find_packed_object_size(struct packed_git *p,
175 struct pack_revindex *rix;
176 unsigned long *revindex;
177 num = pack_revindex_ix(p);
179 die("internal error: pack revindex uninitialized");
180 rix = &pack_revindex[num];
182 prepare_pack_revindex(rix);
183 revindex = rix->revindex;
185 hi = num_packed_objects(p) + 1;
187 int mi = (lo + hi) / 2;
188 if (revindex[mi] == ofs) {
189 return revindex[mi+1] - ofs;
191 else if (ofs < revindex[mi])
196 die("internal error: pack revindex corrupt");
199 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
201 unsigned long othersize, delta_size;
203 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
207 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
208 delta_buf = diff_delta(otherbuf, othersize,
209 buf, size, &delta_size, 0);
210 if (!delta_buf || delta_size != entry->delta_size)
211 die("delta size changed");
218 * The per-object header is a pretty dense thing, which is
219 * - first byte: low four bits are "size", then three bits of "type",
220 * and the high bit is "size continues".
221 * - each byte afterwards: low seven bits are size continuation,
222 * with the high bit being "size continues"
224 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
229 if (type < OBJ_COMMIT || type > OBJ_DELTA)
230 die("bad type %d", type);
232 c = (type << 4) | (size & 15);
244 static unsigned long write_object(struct sha1file *f,
245 struct object_entry *entry)
250 unsigned char header[10];
251 unsigned hdrlen, datalen;
252 enum object_type obj_type;
255 if (entry->preferred_base)
258 obj_type = entry->type;
259 if (! entry->in_pack)
260 to_reuse = 0; /* can't reuse what we don't have */
261 else if (obj_type == OBJ_DELTA)
262 to_reuse = 1; /* check_object() decided it for us */
263 else if (obj_type != entry->in_pack_type)
264 to_reuse = 0; /* pack has delta which is unusable */
265 else if (entry->delta)
266 to_reuse = 0; /* we want to pack afresh */
268 to_reuse = 1; /* we have it in-pack undeltified,
269 * and we do not need to deltify it.
272 if (!entry->in_pack && !entry->delta) {
274 unsigned long mapsize;
275 map = map_sha1_file(entry->sha1, &mapsize);
276 if (map && !legacy_loose_object(map)) {
277 /* We can copy straight into the pack file */
278 sha1write(f, map, mapsize);
279 munmap(map, mapsize);
285 munmap(map, mapsize);
289 buf = read_sha1_file(entry->sha1, type, &size);
291 die("unable to read %s", sha1_to_hex(entry->sha1));
292 if (size != entry->size)
293 die("object %s size inconsistency (%lu vs %lu)",
294 sha1_to_hex(entry->sha1), size, entry->size);
296 buf = delta_against(buf, size, entry);
297 size = entry->delta_size;
298 obj_type = OBJ_DELTA;
301 * The object header is a byte of 'type' followed by zero or
302 * more bytes of length. For deltas, the 20 bytes of delta
305 hdrlen = encode_header(obj_type, size, header);
306 sha1write(f, header, hdrlen);
309 sha1write(f, entry->delta, 20);
312 datalen = sha1write_compressed(f, buf, size);
316 struct packed_git *p = entry->in_pack;
319 datalen = find_packed_object_size(p, entry->in_pack_offset);
320 buf = (char *) p->pack_base + entry->in_pack_offset;
321 sha1write(f, buf, datalen);
323 hdrlen = 0; /* not really */
324 if (obj_type == OBJ_DELTA)
328 if (obj_type == OBJ_DELTA)
331 return hdrlen + datalen;
334 static unsigned long write_one(struct sha1file *f,
335 struct object_entry *e,
336 unsigned long offset)
339 /* offset starts from header size and cannot be zero
340 * if it is written already.
344 offset += write_object(f, e);
345 /* if we are deltified, write out its base object. */
347 offset = write_one(f, e->delta, offset);
351 static void write_pack_file(void)
355 unsigned long offset;
356 struct pack_header hdr;
357 unsigned last_percent = 999;
361 f = sha1fd(1, "<stdout>");
363 f = sha1create("%s-%s.%s", base_name,
364 sha1_to_hex(object_list_sha1), "pack");
365 do_progress = progress;
368 fprintf(stderr, "Writing %d objects.\n", nr_result);
370 hdr.hdr_signature = htonl(PACK_SIGNATURE);
371 hdr.hdr_version = htonl(PACK_VERSION);
372 hdr.hdr_entries = htonl(nr_result);
373 sha1write(f, &hdr, sizeof(hdr));
374 offset = sizeof(hdr);
377 for (i = 0; i < nr_objects; i++) {
378 offset = write_one(f, objects + i, offset);
380 unsigned percent = written * 100 / nr_result;
381 if (progress_update || percent != last_percent) {
382 fprintf(stderr, "%4u%% (%u/%u) done\r",
383 percent, written, nr_result);
385 last_percent = percent;
392 sha1close(f, pack_file_sha1, 1);
395 static void write_index_file(void)
398 struct sha1file *f = sha1create("%s-%s.%s", base_name,
399 sha1_to_hex(object_list_sha1), "idx");
400 struct object_entry **list = sorted_by_sha;
401 struct object_entry **last = list + nr_result;
402 unsigned int array[256];
405 * Write the first-level table (the list is sorted,
406 * but we use a 256-entry lookup to be able to avoid
407 * having to do eight extra binary search iterations).
409 for (i = 0; i < 256; i++) {
410 struct object_entry **next = list;
411 while (next < last) {
412 struct object_entry *entry = *next;
413 if (entry->sha1[0] != i)
417 array[i] = htonl(next - sorted_by_sha);
420 sha1write(f, array, 256 * sizeof(int));
423 * Write the actual SHA1 entries..
425 list = sorted_by_sha;
426 for (i = 0; i < nr_result; i++) {
427 struct object_entry *entry = *list++;
428 unsigned int offset = htonl(entry->offset);
429 sha1write(f, &offset, 4);
430 sha1write(f, entry->sha1, 20);
432 sha1write(f, pack_file_sha1, 20);
433 sha1close(f, NULL, 1);
436 static int locate_object_entry_hash(const unsigned char *sha1)
440 memcpy(&ui, sha1, sizeof(unsigned int));
441 i = ui % object_ix_hashsz;
442 while (0 < object_ix[i]) {
443 if (!memcmp(sha1, objects[object_ix[i]-1].sha1, 20))
445 if (++i == object_ix_hashsz)
451 static struct object_entry *locate_object_entry(const unsigned char *sha1)
455 if (!object_ix_hashsz)
458 i = locate_object_entry_hash(sha1);
460 return &objects[object_ix[i]-1];
464 static void rehash_objects(void)
467 struct object_entry *oe;
469 object_ix_hashsz = nr_objects * 3;
470 if (object_ix_hashsz < 1024)
471 object_ix_hashsz = 1024;
472 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
473 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
474 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
475 int ix = locate_object_entry_hash(oe->sha1);
479 object_ix[ix] = i + 1;
483 static unsigned name_hash(const char *name)
489 * This effectively just creates a sortable number from the
490 * last sixteen non-whitespace characters. Last characters
491 * count "most", so things that end in ".c" sort together.
493 while ((c = *name++) != 0) {
496 hash = (hash >> 2) + (c << 24);
501 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
503 unsigned int idx = nr_objects;
504 struct object_entry *entry;
505 struct packed_git *p;
506 unsigned int found_offset = 0;
507 struct packed_git *found_pack = NULL;
511 for (p = packed_git; p; p = p->next) {
513 if (find_pack_entry_one(sha1, &e, p)) {
516 if (local && !p->pack_local)
519 found_offset = e.offset;
525 if ((entry = locate_object_entry(sha1)) != NULL)
528 if (idx >= nr_alloc) {
529 unsigned int needed = (idx + 1024) * 3 / 2;
530 objects = xrealloc(objects, needed * sizeof(*entry));
533 entry = objects + idx;
534 nr_objects = idx + 1;
535 memset(entry, 0, sizeof(*entry));
536 memcpy(entry->sha1, sha1, 20);
539 if (object_ix_hashsz * 3 <= nr_objects * 4)
542 ix = locate_object_entry_hash(entry->sha1);
544 die("internal error in object hashing.");
545 object_ix[-1 - ix] = idx + 1;
550 if (progress_update) {
551 fprintf(stderr, "Counting objects...%d\r", nr_objects);
555 entry->preferred_base = 1;
558 entry->in_pack = found_pack;
559 entry->in_pack_offset = found_offset;
565 struct pbase_tree_cache {
566 unsigned char sha1[20];
570 unsigned long tree_size;
573 static struct pbase_tree_cache *(pbase_tree_cache[256]);
574 static int pbase_tree_cache_ix(const unsigned char *sha1)
576 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
578 static int pbase_tree_cache_ix_incr(int ix)
580 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
583 static struct pbase_tree {
584 struct pbase_tree *next;
585 /* This is a phony "cache" entry; we are not
586 * going to evict it nor find it through _get()
587 * mechanism -- this is for the toplevel node that
588 * would almost always change with any commit.
590 struct pbase_tree_cache pcache;
593 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
595 struct pbase_tree_cache *ent, *nent;
600 int my_ix = pbase_tree_cache_ix(sha1);
601 int available_ix = -1;
603 /* pbase-tree-cache acts as a limited hashtable.
604 * your object will be found at your index or within a few
605 * slots after that slot if it is cached.
607 for (neigh = 0; neigh < 8; neigh++) {
608 ent = pbase_tree_cache[my_ix];
609 if (ent && !memcmp(ent->sha1, sha1, 20)) {
613 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
614 ((0 <= available_ix) &&
615 (!ent && pbase_tree_cache[available_ix])))
616 available_ix = my_ix;
619 my_ix = pbase_tree_cache_ix_incr(my_ix);
622 /* Did not find one. Either we got a bogus request or
623 * we need to read and perhaps cache.
625 data = read_sha1_file(sha1, type, &size);
628 if (strcmp(type, tree_type)) {
633 /* We need to either cache or return a throwaway copy */
635 if (available_ix < 0)
638 ent = pbase_tree_cache[available_ix];
639 my_ix = available_ix;
643 nent = xmalloc(sizeof(*nent));
644 nent->temporary = (available_ix < 0);
647 /* evict and reuse */
648 free(ent->tree_data);
651 memcpy(nent->sha1, sha1, 20);
652 nent->tree_data = data;
653 nent->tree_size = size;
655 if (!nent->temporary)
656 pbase_tree_cache[my_ix] = nent;
660 static void pbase_tree_put(struct pbase_tree_cache *cache)
662 if (!cache->temporary) {
666 free(cache->tree_data);
670 static int name_cmp_len(const char *name)
673 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
678 static void add_pbase_object(struct tree_desc *tree,
681 const char *fullname)
683 struct name_entry entry;
685 while (tree_entry(tree,&entry)) {
689 if (entry.pathlen != cmplen ||
690 memcmp(entry.path, name, cmplen) ||
691 !has_sha1_file(entry.sha1) ||
692 sha1_object_info(entry.sha1, type, &size))
694 if (name[cmplen] != '/') {
695 unsigned hash = name_hash(fullname);
696 add_object_entry(entry.sha1, hash, 1);
699 if (!strcmp(type, tree_type)) {
700 struct tree_desc sub;
701 struct pbase_tree_cache *tree;
702 const char *down = name+cmplen+1;
703 int downlen = name_cmp_len(down);
705 tree = pbase_tree_get(entry.sha1);
708 sub.buf = tree->tree_data;
709 sub.size = tree->tree_size;
711 add_pbase_object(&sub, down, downlen, fullname);
712 pbase_tree_put(tree);
717 static unsigned *done_pbase_paths;
718 static int done_pbase_paths_num;
719 static int done_pbase_paths_alloc;
720 static int done_pbase_path_pos(unsigned hash)
723 int hi = done_pbase_paths_num;
725 int mi = (hi + lo) / 2;
726 if (done_pbase_paths[mi] == hash)
728 if (done_pbase_paths[mi] < hash)
736 static int check_pbase_path(unsigned hash)
738 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
742 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
743 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
744 done_pbase_paths = xrealloc(done_pbase_paths,
745 done_pbase_paths_alloc *
748 done_pbase_paths_num++;
749 if (pos < done_pbase_paths_num)
750 memmove(done_pbase_paths + pos + 1,
751 done_pbase_paths + pos,
752 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
753 done_pbase_paths[pos] = hash;
757 static void add_preferred_base_object(char *name, unsigned hash)
759 struct pbase_tree *it;
760 int cmplen = name_cmp_len(name);
762 if (check_pbase_path(hash))
765 for (it = pbase_tree; it; it = it->next) {
767 hash = name_hash("");
768 add_object_entry(it->pcache.sha1, hash, 1);
771 struct tree_desc tree;
772 tree.buf = it->pcache.tree_data;
773 tree.size = it->pcache.tree_size;
774 add_pbase_object(&tree, name, cmplen, name);
779 static void add_preferred_base(unsigned char *sha1)
781 struct pbase_tree *it;
784 unsigned char tree_sha1[20];
786 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
790 for (it = pbase_tree; it; it = it->next) {
791 if (!memcmp(it->pcache.sha1, tree_sha1, 20)) {
797 it = xcalloc(1, sizeof(*it));
798 it->next = pbase_tree;
801 memcpy(it->pcache.sha1, tree_sha1, 20);
802 it->pcache.tree_data = data;
803 it->pcache.tree_size = size;
806 static void check_object(struct object_entry *entry)
810 if (entry->in_pack && !entry->preferred_base) {
811 unsigned char base[20];
813 struct object_entry *base_entry;
815 /* We want in_pack_type even if we do not reuse delta.
816 * There is no point not reusing non-delta representations.
818 check_reuse_pack_delta(entry->in_pack,
819 entry->in_pack_offset,
821 &entry->in_pack_type);
823 /* Check if it is delta, and the base is also an object
824 * we are going to pack. If so we will reuse the existing
827 if (!no_reuse_delta &&
828 entry->in_pack_type == OBJ_DELTA &&
829 (base_entry = locate_object_entry(base)) &&
830 (!base_entry->preferred_base)) {
832 /* Depth value does not matter - find_deltas()
833 * will never consider reused delta as the
834 * base object to deltify other objects
835 * against, in order to avoid circular deltas.
838 /* uncompressed size of the delta data */
839 entry->size = entry->delta_size = size;
840 entry->delta = base_entry;
841 entry->type = OBJ_DELTA;
843 entry->delta_sibling = base_entry->delta_child;
844 base_entry->delta_child = entry;
848 /* Otherwise we would do the usual */
851 if (sha1_object_info(entry->sha1, type, &entry->size))
852 die("unable to get type of object %s",
853 sha1_to_hex(entry->sha1));
855 if (!strcmp(type, commit_type)) {
856 entry->type = OBJ_COMMIT;
857 } else if (!strcmp(type, tree_type)) {
858 entry->type = OBJ_TREE;
859 } else if (!strcmp(type, blob_type)) {
860 entry->type = OBJ_BLOB;
861 } else if (!strcmp(type, tag_type)) {
862 entry->type = OBJ_TAG;
864 die("unable to pack object %s of type %s",
865 sha1_to_hex(entry->sha1), type);
868 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
870 struct object_entry *child = me->delta_child;
873 unsigned int c = check_delta_limit(child, n + 1);
876 child = child->delta_sibling;
881 static void get_object_details(void)
884 struct object_entry *entry;
887 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
890 if (nr_objects == nr_result) {
892 * Depth of objects that depend on the entry -- this
893 * is subtracted from depth-max to break too deep
894 * delta chain because of delta data reusing.
895 * However, we loosen this restriction when we know we
896 * are creating a thin pack -- it will have to be
897 * expanded on the other end anyway, so do not
898 * artificially cut the delta chain and let it go as
901 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
902 if (!entry->delta && entry->delta_child)
904 check_delta_limit(entry, 1);
908 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
910 static entry_sort_t current_sort;
912 static int sort_comparator(const void *_a, const void *_b)
914 struct object_entry *a = *(struct object_entry **)_a;
915 struct object_entry *b = *(struct object_entry **)_b;
916 return current_sort(a,b);
919 static struct object_entry **create_sorted_list(entry_sort_t sort)
921 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
924 for (i = 0; i < nr_objects; i++)
925 list[i] = objects + i;
927 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
931 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
933 return memcmp(a->sha1, b->sha1, 20);
936 static struct object_entry **create_final_object_list(void)
938 struct object_entry **list;
941 for (i = nr_result = 0; i < nr_objects; i++)
942 if (!objects[i].preferred_base)
944 list = xmalloc(nr_result * sizeof(struct object_entry *));
945 for (i = j = 0; i < nr_objects; i++) {
946 if (!objects[i].preferred_base)
947 list[j++] = objects + i;
949 current_sort = sha1_sort;
950 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
954 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
956 if (a->type < b->type)
958 if (a->type > b->type)
960 if (a->hash < b->hash)
962 if (a->hash > b->hash)
964 if (a->preferred_base < b->preferred_base)
966 if (a->preferred_base > b->preferred_base)
968 if (a->size < b->size)
970 if (a->size > b->size)
972 return a < b ? -1 : (a > b);
976 struct object_entry *entry;
978 struct delta_index *index;
982 * We search for deltas _backwards_ in a list sorted by type and
983 * by size, so that we see progressively smaller and smaller files.
984 * That's because we prefer deltas to be from the bigger file
985 * to the smaller - deletes are potentially cheaper, but perhaps
986 * more importantly, the bigger file is likely the more recent
989 static int try_delta(struct unpacked *trg, struct unpacked *src,
992 struct object_entry *trg_entry = trg->entry;
993 struct object_entry *src_entry = src->entry;
994 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
998 /* Don't bother doing diffs between different types */
999 if (trg_entry->type != src_entry->type)
1002 /* We do not compute delta to *create* objects we are not
1005 if (trg_entry->preferred_base)
1009 * We do not bother to try a delta that we discarded
1010 * on an earlier try, but only when reusing delta data.
1012 if (!no_reuse_delta && trg_entry->in_pack &&
1013 trg_entry->in_pack == src_entry->in_pack)
1017 * If the current object is at pack edge, take the depth the
1018 * objects that depend on the current object into account --
1019 * otherwise they would become too deep.
1021 if (trg_entry->delta_child) {
1022 if (max_depth <= trg_entry->delta_limit)
1024 max_depth -= trg_entry->delta_limit;
1026 if (src_entry->depth >= max_depth)
1029 /* Now some size filtering heuristics. */
1030 trg_size = trg_entry->size;
1031 max_size = trg_size/2 - 20;
1032 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1035 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1036 max_size = trg_entry->delta_size-1;
1037 src_size = src_entry->size;
1038 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1039 if (sizediff >= max_size)
1042 /* Load data if not already done */
1044 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1046 die("object %s inconsistent object length (%lu vs %lu)",
1047 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1050 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1052 die("object %s inconsistent object length (%lu vs %lu)",
1053 sha1_to_hex(src_entry->sha1), sz, src_size);
1056 src->index = create_delta_index(src->data, src_size);
1058 die("out of memory");
1061 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1065 trg_entry->delta = src_entry;
1066 trg_entry->delta_size = delta_size;
1067 trg_entry->depth = src_entry->depth + 1;
1072 static void progress_interval(int signum)
1074 progress_update = 1;
1077 static void find_deltas(struct object_entry **list, int window, int depth)
1080 unsigned int array_size = window * sizeof(struct unpacked);
1081 struct unpacked *array = xmalloc(array_size);
1082 unsigned processed = 0;
1083 unsigned last_percent = 999;
1085 memset(array, 0, array_size);
1089 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1092 struct object_entry *entry = list[i];
1093 struct unpacked *n = array + idx;
1096 if (!entry->preferred_base)
1100 unsigned percent = processed * 100 / nr_result;
1101 if (percent != last_percent || progress_update) {
1102 fprintf(stderr, "%4u%% (%u/%u) done\r",
1103 percent, processed, nr_result);
1104 progress_update = 0;
1105 last_percent = percent;
1110 /* This happens if we decided to reuse existing
1111 * delta from a pack. "!no_reuse_delta &&" is implied.
1115 if (entry->size < 50)
1117 free_delta_index(n->index);
1125 unsigned int other_idx = idx + j;
1127 if (other_idx >= window)
1128 other_idx -= window;
1129 m = array + other_idx;
1132 if (try_delta(n, m, depth) < 0)
1135 /* if we made n a delta, and if n is already at max
1136 * depth, leaving it in the window is pointless. we
1137 * should evict it first.
1139 if (entry->delta && depth <= entry->depth)
1148 fputc('\n', stderr);
1150 for (i = 0; i < window; ++i) {
1151 free_delta_index(array[i].index);
1152 free(array[i].data);
1157 static void prepare_pack(int window, int depth)
1159 get_object_details();
1160 sorted_by_type = create_sorted_list(type_size_sort);
1161 if (window && depth)
1162 find_deltas(sorted_by_type, window+1, depth);
1165 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
1167 static const char cache[] = "pack-cache/pack-%s.%s";
1168 char *cached_pack, *cached_idx;
1169 int ifd, ofd, ifd_ix = -1;
1171 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1172 ifd = open(cached_pack, O_RDONLY);
1176 if (!pack_to_stdout) {
1177 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1178 ifd_ix = open(cached_idx, O_RDONLY);
1186 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1189 if (pack_to_stdout) {
1190 if (copy_fd(ifd, 1))
1195 char name[PATH_MAX];
1196 snprintf(name, sizeof(name),
1197 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1198 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1200 die("unable to open %s (%s)", name, strerror(errno));
1201 if (copy_fd(ifd, ofd))
1205 snprintf(name, sizeof(name),
1206 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1207 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1209 die("unable to open %s (%s)", name, strerror(errno));
1210 if (copy_fd(ifd_ix, ofd))
1213 puts(sha1_to_hex(sha1));
1219 static void setup_progress_signal(void)
1221 struct sigaction sa;
1224 memset(&sa, 0, sizeof(sa));
1225 sa.sa_handler = progress_interval;
1226 sigemptyset(&sa.sa_mask);
1227 sa.sa_flags = SA_RESTART;
1228 sigaction(SIGALRM, &sa, NULL);
1230 v.it_interval.tv_sec = 1;
1231 v.it_interval.tv_usec = 0;
1232 v.it_value = v.it_interval;
1233 setitimer(ITIMER_REAL, &v, NULL);
1236 static int git_pack_config(const char *k, const char *v)
1238 if(!strcmp(k, "pack.window")) {
1239 window = git_config_int(k, v);
1242 return git_default_config(k, v);
1245 int main(int argc, char **argv)
1248 char line[40 + 1 + PATH_MAX + 2];
1249 int depth = 10, pack_to_stdout = 0;
1250 struct object_entry **list;
1251 int num_preferred_base = 0;
1254 setup_git_directory();
1255 git_config(git_pack_config);
1257 progress = isatty(2);
1258 for (i = 1; i < argc; i++) {
1259 const char *arg = argv[i];
1262 if (!strcmp("--non-empty", arg)) {
1266 if (!strcmp("--local", arg)) {
1270 if (!strcmp("--progress", arg)) {
1274 if (!strcmp("--incremental", arg)) {
1278 if (!strncmp("--window=", arg, 9)) {
1280 window = strtoul(arg+9, &end, 0);
1281 if (!arg[9] || *end)
1285 if (!strncmp("--depth=", arg, 8)) {
1287 depth = strtoul(arg+8, &end, 0);
1288 if (!arg[8] || *end)
1292 if (!strcmp("--progress", arg)) {
1296 if (!strcmp("-q", arg)) {
1300 if (!strcmp("--no-reuse-delta", arg)) {
1304 if (!strcmp("--stdout", arg)) {
1315 if (pack_to_stdout != !base_name)
1318 prepare_packed_git();
1321 fprintf(stderr, "Generating pack...\n");
1322 setup_progress_signal();
1326 unsigned char sha1[20];
1329 if (!fgets(line, sizeof(line), stdin)) {
1333 die("fgets returned NULL, not EOF, not error!");
1335 die("fgets: %s", strerror(errno));
1340 if (line[0] == '-') {
1341 if (get_sha1_hex(line+1, sha1))
1342 die("expected edge sha1, got garbage:\n %s",
1344 if (num_preferred_base++ < window)
1345 add_preferred_base(sha1);
1348 if (get_sha1_hex(line, sha1))
1349 die("expected sha1, got garbage:\n %s", line);
1350 hash = name_hash(line+41);
1351 add_preferred_base_object(line+41, hash);
1352 add_object_entry(sha1, hash, 0);
1355 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1356 sorted_by_sha = create_final_object_list();
1357 if (non_empty && !nr_result)
1361 list = sorted_by_sha;
1362 for (i = 0; i < nr_result; i++) {
1363 struct object_entry *entry = *list++;
1364 SHA1_Update(&ctx, entry->sha1, 20);
1366 SHA1_Final(object_list_sha1, &ctx);
1367 if (progress && (nr_objects != nr_result))
1368 fprintf(stderr, "Result has %d objects.\n", nr_result);
1370 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1374 prepare_pack(window, depth);
1375 if (progress && pack_to_stdout) {
1376 /* the other end usually displays progress itself */
1377 struct itimerval v = {{0,},};
1378 setitimer(ITIMER_REAL, &v, NULL);
1379 signal(SIGALRM, SIG_IGN );
1380 progress_update = 0;
1383 if (!pack_to_stdout) {
1385 puts(sha1_to_hex(object_list_sha1));
1389 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1390 nr_result, written, written_delta, reused, reused_delta);