pack-objects: re-validate data we copy from elsewhere.
[git.git] / builtin-pack-objects.c
blob11cc3c89f5789366100f80267d0f9c4d87f3745d
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "blob.h"
5 #include "commit.h"
6 #include "tag.h"
7 #include "tree.h"
8 #include "delta.h"
9 #include "pack.h"
10 #include "csum-file.h"
11 #include "tree-walk.h"
12 #include <sys/time.h>
13 #include <signal.h>
15 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 struct object_entry {
18 unsigned char sha1[20];
19 unsigned long size; /* uncompressed size */
20 unsigned long offset; /* offset into the final pack file;
21 * nonzero if already written.
23 unsigned int depth; /* delta depth */
24 unsigned int delta_limit; /* base adjustment for in-pack delta */
25 unsigned int hash; /* name hint hash */
26 enum object_type type;
27 enum object_type in_pack_type; /* could be delta */
28 unsigned long delta_size; /* delta data size (uncompressed) */
29 struct object_entry *delta; /* delta base object */
30 struct packed_git *in_pack; /* already in pack */
31 unsigned int in_pack_offset;
32 struct object_entry *delta_child; /* deltified objects who bases me */
33 struct object_entry *delta_sibling; /* other deltified objects who
34 * uses the same base as me
36 int preferred_base; /* we do not pack this, but is encouraged to
37 * be used as the base objectto delta huge
38 * objects against.
43 * Objects we are going to pack are collected in objects array (dynamically
44 * expanded). nr_objects & nr_alloc controls this array. They are stored
45 * in the order we see -- typically rev-list --objects order that gives us
46 * nice "minimum seek" order.
48 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
49 * elements in the objects array. The former is used to build the pack
50 * index (lists object names in the ascending order to help offset lookup),
51 * and the latter is used to group similar things together by try_delta()
52 * heuristics.
55 static unsigned char object_list_sha1[20];
56 static int non_empty;
57 static int no_reuse_delta;
58 static int local;
59 static int incremental;
60 static struct object_entry **sorted_by_sha, **sorted_by_type;
61 static struct object_entry *objects;
62 static int nr_objects, nr_alloc, nr_result;
63 static const char *base_name;
64 static unsigned char pack_file_sha1[20];
65 static int progress = 1;
66 static volatile sig_atomic_t progress_update;
67 static int window = 10;
68 static int pack_to_stdout;
71 * The object names in objects array are hashed with this hashtable,
72 * to help looking up the entry by object name. Binary search from
73 * sorted_by_sha is also possible but this was easier to code and faster.
74 * This hashtable is built after all the objects are seen.
76 static int *object_ix;
77 static int object_ix_hashsz;
80 * Pack index for existing packs give us easy access to the offsets into
81 * corresponding pack file where each object's data starts, but the entries
82 * do not store the size of the compressed representation (uncompressed
83 * size is easily available by examining the pack entry header). We build
84 * a hashtable of existing packs (pack_revindex), and keep reverse index
85 * here -- pack index file is sorted by object name mapping to offset; this
86 * pack_revindex[].revindex array is an ordered list of offsets, so if you
87 * know the offset of an object, next offset is where its packed
88 * representation ends.
90 struct pack_revindex {
91 struct packed_git *p;
92 unsigned long *revindex;
93 } *pack_revindex = NULL;
94 static int pack_revindex_hashsz;
97 * stats
99 static int written;
100 static int written_delta;
101 static int reused;
102 static int reused_delta;
104 static int pack_revindex_ix(struct packed_git *p)
106 unsigned long ui = (unsigned long)p;
107 int i;
109 ui = ui ^ (ui >> 16); /* defeat structure alignment */
110 i = (int)(ui % pack_revindex_hashsz);
111 while (pack_revindex[i].p) {
112 if (pack_revindex[i].p == p)
113 return i;
114 if (++i == pack_revindex_hashsz)
115 i = 0;
117 return -1 - i;
120 static void prepare_pack_ix(void)
122 int num;
123 struct packed_git *p;
124 for (num = 0, p = packed_git; p; p = p->next)
125 num++;
126 if (!num)
127 return;
128 pack_revindex_hashsz = num * 11;
129 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
130 for (p = packed_git; p; p = p->next) {
131 num = pack_revindex_ix(p);
132 num = - 1 - num;
133 pack_revindex[num].p = p;
135 /* revindex elements are lazily initialized */
138 static int cmp_offset(const void *a_, const void *b_)
140 unsigned long a = *(unsigned long *) a_;
141 unsigned long b = *(unsigned long *) b_;
142 if (a < b)
143 return -1;
144 else if (a == b)
145 return 0;
146 else
147 return 1;
151 * Ordered list of offsets of objects in the pack.
153 static void prepare_pack_revindex(struct pack_revindex *rix)
155 struct packed_git *p = rix->p;
156 int num_ent = num_packed_objects(p);
157 int i;
158 void *index = p->index_base + 256;
160 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
161 for (i = 0; i < num_ent; i++) {
162 unsigned int hl = *((unsigned int *)((char *) index + 24*i));
163 rix->revindex[i] = ntohl(hl);
165 /* This knows the pack format -- the 20-byte trailer
166 * follows immediately after the last object data.
168 rix->revindex[num_ent] = p->pack_size - 20;
169 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
172 static unsigned long find_packed_object_size(struct packed_git *p,
173 unsigned long ofs)
175 int num;
176 int lo, hi;
177 struct pack_revindex *rix;
178 unsigned long *revindex;
179 num = pack_revindex_ix(p);
180 if (num < 0)
181 die("internal error: pack revindex uninitialized");
182 rix = &pack_revindex[num];
183 if (!rix->revindex)
184 prepare_pack_revindex(rix);
185 revindex = rix->revindex;
186 lo = 0;
187 hi = num_packed_objects(p) + 1;
188 do {
189 int mi = (lo + hi) / 2;
190 if (revindex[mi] == ofs) {
191 return revindex[mi+1] - ofs;
193 else if (ofs < revindex[mi])
194 hi = mi;
195 else
196 lo = mi + 1;
197 } while (lo < hi);
198 die("internal error: pack revindex corrupt");
201 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
203 unsigned long othersize, delta_size;
204 char type[10];
205 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
206 void *delta_buf;
208 if (!otherbuf)
209 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
210 delta_buf = diff_delta(otherbuf, othersize,
211 buf, size, &delta_size, 0);
212 if (!delta_buf || delta_size != entry->delta_size)
213 die("delta size changed");
214 free(buf);
215 free(otherbuf);
216 return delta_buf;
220 * The per-object header is a pretty dense thing, which is
221 * - first byte: low four bits are "size", then three bits of "type",
222 * and the high bit is "size continues".
223 * - each byte afterwards: low seven bits are size continuation,
224 * with the high bit being "size continues"
226 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
228 int n = 1;
229 unsigned char c;
231 if (type < OBJ_COMMIT || type > OBJ_DELTA)
232 die("bad type %d", type);
234 c = (type << 4) | (size & 15);
235 size >>= 4;
236 while (size) {
237 *hdr++ = c | 0x80;
238 c = size & 0x7f;
239 size >>= 7;
240 n++;
242 *hdr = c;
243 return n;
246 static int revalidate_one(struct object_entry *entry,
247 void *data, char *type, unsigned long size)
249 int err;
250 if (!data)
251 return -1;
252 if (size != entry->size)
253 return -1;
254 err = check_sha1_signature(entry->sha1, data, size,
255 type_names[entry->type]);
256 free(data);
257 return err;
261 * we are going to reuse the existing pack entry data. make
262 * sure it is not corrupt.
264 static int revalidate_pack_entry(struct object_entry *entry)
266 void *data;
267 char type[20];
268 unsigned long size;
269 struct pack_entry e;
271 if (pack_to_stdout)
272 return 0;
274 e.p = entry->in_pack;
275 e.offset = entry->in_pack_offset;
277 /* the caller has already called use_packed_git() for us */
278 data = unpack_entry_gently(&e, type, &size);
279 return revalidate_one(entry, data, type, size);
282 static int revalidate_loose_object(struct object_entry *entry,
283 unsigned char *map,
284 unsigned long mapsize)
286 /* we already know this is a loose object with new type header. */
287 void *data;
288 char type[20];
289 unsigned long size;
291 if (pack_to_stdout)
292 return 0;
294 data = unpack_sha1_file(map, mapsize, type, &size);
295 return revalidate_one(entry, data, type, size);
298 static unsigned long write_object(struct sha1file *f,
299 struct object_entry *entry)
301 unsigned long size;
302 char type[10];
303 void *buf;
304 unsigned char header[10];
305 unsigned hdrlen, datalen;
306 enum object_type obj_type;
307 int to_reuse = 0;
309 if (entry->preferred_base)
310 return 0;
312 obj_type = entry->type;
313 if (! entry->in_pack)
314 to_reuse = 0; /* can't reuse what we don't have */
315 else if (obj_type == OBJ_DELTA)
316 to_reuse = 1; /* check_object() decided it for us */
317 else if (obj_type != entry->in_pack_type)
318 to_reuse = 0; /* pack has delta which is unusable */
319 else if (entry->delta)
320 to_reuse = 0; /* we want to pack afresh */
321 else
322 to_reuse = 1; /* we have it in-pack undeltified,
323 * and we do not need to deltify it.
326 if (!entry->in_pack && !entry->delta) {
327 unsigned char *map;
328 unsigned long mapsize;
329 map = map_sha1_file(entry->sha1, &mapsize);
330 if (map && !legacy_loose_object(map)) {
331 /* We can copy straight into the pack file */
332 if (revalidate_loose_object(entry, map, mapsize))
333 die("corrupt loose object %s",
334 sha1_to_hex(entry->sha1));
335 sha1write(f, map, mapsize);
336 munmap(map, mapsize);
337 written++;
338 reused++;
339 return mapsize;
341 if (map)
342 munmap(map, mapsize);
345 if (!to_reuse) {
346 buf = read_sha1_file(entry->sha1, type, &size);
347 if (!buf)
348 die("unable to read %s", sha1_to_hex(entry->sha1));
349 if (size != entry->size)
350 die("object %s size inconsistency (%lu vs %lu)",
351 sha1_to_hex(entry->sha1), size, entry->size);
352 if (entry->delta) {
353 buf = delta_against(buf, size, entry);
354 size = entry->delta_size;
355 obj_type = OBJ_DELTA;
358 * The object header is a byte of 'type' followed by zero or
359 * more bytes of length. For deltas, the 20 bytes of delta
360 * sha1 follows that.
362 hdrlen = encode_header(obj_type, size, header);
363 sha1write(f, header, hdrlen);
365 if (entry->delta) {
366 sha1write(f, entry->delta, 20);
367 hdrlen += 20;
369 datalen = sha1write_compressed(f, buf, size);
370 free(buf);
372 else {
373 struct packed_git *p = entry->in_pack;
374 use_packed_git(p);
376 datalen = find_packed_object_size(p, entry->in_pack_offset);
377 buf = (char *) p->pack_base + entry->in_pack_offset;
379 if (revalidate_pack_entry(entry))
380 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
381 sha1write(f, buf, datalen);
382 unuse_packed_git(p);
383 hdrlen = 0; /* not really */
384 if (obj_type == OBJ_DELTA)
385 reused_delta++;
386 reused++;
388 if (obj_type == OBJ_DELTA)
389 written_delta++;
390 written++;
391 return hdrlen + datalen;
394 static unsigned long write_one(struct sha1file *f,
395 struct object_entry *e,
396 unsigned long offset)
398 if (e->offset)
399 /* offset starts from header size and cannot be zero
400 * if it is written already.
402 return offset;
403 e->offset = offset;
404 offset += write_object(f, e);
405 /* if we are deltified, write out its base object. */
406 if (e->delta)
407 offset = write_one(f, e->delta, offset);
408 return offset;
411 static void write_pack_file(void)
413 int i;
414 struct sha1file *f;
415 unsigned long offset;
416 struct pack_header hdr;
417 unsigned last_percent = 999;
418 int do_progress = 0;
420 if (!base_name)
421 f = sha1fd(1, "<stdout>");
422 else {
423 f = sha1create("%s-%s.%s", base_name,
424 sha1_to_hex(object_list_sha1), "pack");
425 do_progress = progress;
427 if (do_progress)
428 fprintf(stderr, "Writing %d objects.\n", nr_result);
430 hdr.hdr_signature = htonl(PACK_SIGNATURE);
431 hdr.hdr_version = htonl(PACK_VERSION);
432 hdr.hdr_entries = htonl(nr_result);
433 sha1write(f, &hdr, sizeof(hdr));
434 offset = sizeof(hdr);
435 if (!nr_result)
436 goto done;
437 for (i = 0; i < nr_objects; i++) {
438 offset = write_one(f, objects + i, offset);
439 if (do_progress) {
440 unsigned percent = written * 100 / nr_result;
441 if (progress_update || percent != last_percent) {
442 fprintf(stderr, "%4u%% (%u/%u) done\r",
443 percent, written, nr_result);
444 progress_update = 0;
445 last_percent = percent;
449 if (do_progress)
450 fputc('\n', stderr);
451 done:
452 sha1close(f, pack_file_sha1, 1);
455 static void write_index_file(void)
457 int i;
458 struct sha1file *f = sha1create("%s-%s.%s", base_name,
459 sha1_to_hex(object_list_sha1), "idx");
460 struct object_entry **list = sorted_by_sha;
461 struct object_entry **last = list + nr_result;
462 unsigned int array[256];
465 * Write the first-level table (the list is sorted,
466 * but we use a 256-entry lookup to be able to avoid
467 * having to do eight extra binary search iterations).
469 for (i = 0; i < 256; i++) {
470 struct object_entry **next = list;
471 while (next < last) {
472 struct object_entry *entry = *next;
473 if (entry->sha1[0] != i)
474 break;
475 next++;
477 array[i] = htonl(next - sorted_by_sha);
478 list = next;
480 sha1write(f, array, 256 * sizeof(int));
483 * Write the actual SHA1 entries..
485 list = sorted_by_sha;
486 for (i = 0; i < nr_result; i++) {
487 struct object_entry *entry = *list++;
488 unsigned int offset = htonl(entry->offset);
489 sha1write(f, &offset, 4);
490 sha1write(f, entry->sha1, 20);
492 sha1write(f, pack_file_sha1, 20);
493 sha1close(f, NULL, 1);
496 static int locate_object_entry_hash(const unsigned char *sha1)
498 int i;
499 unsigned int ui;
500 memcpy(&ui, sha1, sizeof(unsigned int));
501 i = ui % object_ix_hashsz;
502 while (0 < object_ix[i]) {
503 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
504 return i;
505 if (++i == object_ix_hashsz)
506 i = 0;
508 return -1 - i;
511 static struct object_entry *locate_object_entry(const unsigned char *sha1)
513 int i;
515 if (!object_ix_hashsz)
516 return NULL;
518 i = locate_object_entry_hash(sha1);
519 if (0 <= i)
520 return &objects[object_ix[i]-1];
521 return NULL;
524 static void rehash_objects(void)
526 int i;
527 struct object_entry *oe;
529 object_ix_hashsz = nr_objects * 3;
530 if (object_ix_hashsz < 1024)
531 object_ix_hashsz = 1024;
532 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
533 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
534 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
535 int ix = locate_object_entry_hash(oe->sha1);
536 if (0 <= ix)
537 continue;
538 ix = -1 - ix;
539 object_ix[ix] = i + 1;
543 static unsigned name_hash(const char *name)
545 unsigned char c;
546 unsigned hash = 0;
549 * This effectively just creates a sortable number from the
550 * last sixteen non-whitespace characters. Last characters
551 * count "most", so things that end in ".c" sort together.
553 while ((c = *name++) != 0) {
554 if (isspace(c))
555 continue;
556 hash = (hash >> 2) + (c << 24);
558 return hash;
561 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
563 unsigned int idx = nr_objects;
564 struct object_entry *entry;
565 struct packed_git *p;
566 unsigned int found_offset = 0;
567 struct packed_git *found_pack = NULL;
568 int ix, status = 0;
570 if (!exclude) {
571 for (p = packed_git; p; p = p->next) {
572 struct pack_entry e;
573 if (find_pack_entry_one(sha1, &e, p)) {
574 if (incremental)
575 return 0;
576 if (local && !p->pack_local)
577 return 0;
578 if (!found_pack) {
579 found_offset = e.offset;
580 found_pack = e.p;
585 if ((entry = locate_object_entry(sha1)) != NULL)
586 goto already_added;
588 if (idx >= nr_alloc) {
589 unsigned int needed = (idx + 1024) * 3 / 2;
590 objects = xrealloc(objects, needed * sizeof(*entry));
591 nr_alloc = needed;
593 entry = objects + idx;
594 nr_objects = idx + 1;
595 memset(entry, 0, sizeof(*entry));
596 hashcpy(entry->sha1, sha1);
597 entry->hash = hash;
599 if (object_ix_hashsz * 3 <= nr_objects * 4)
600 rehash_objects();
601 else {
602 ix = locate_object_entry_hash(entry->sha1);
603 if (0 <= ix)
604 die("internal error in object hashing.");
605 object_ix[-1 - ix] = idx + 1;
607 status = 1;
609 already_added:
610 if (progress_update) {
611 fprintf(stderr, "Counting objects...%d\r", nr_objects);
612 progress_update = 0;
614 if (exclude)
615 entry->preferred_base = 1;
616 else {
617 if (found_pack) {
618 entry->in_pack = found_pack;
619 entry->in_pack_offset = found_offset;
622 return status;
625 struct pbase_tree_cache {
626 unsigned char sha1[20];
627 int ref;
628 int temporary;
629 void *tree_data;
630 unsigned long tree_size;
633 static struct pbase_tree_cache *(pbase_tree_cache[256]);
634 static int pbase_tree_cache_ix(const unsigned char *sha1)
636 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
638 static int pbase_tree_cache_ix_incr(int ix)
640 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
643 static struct pbase_tree {
644 struct pbase_tree *next;
645 /* This is a phony "cache" entry; we are not
646 * going to evict it nor find it through _get()
647 * mechanism -- this is for the toplevel node that
648 * would almost always change with any commit.
650 struct pbase_tree_cache pcache;
651 } *pbase_tree;
653 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
655 struct pbase_tree_cache *ent, *nent;
656 void *data;
657 unsigned long size;
658 char type[20];
659 int neigh;
660 int my_ix = pbase_tree_cache_ix(sha1);
661 int available_ix = -1;
663 /* pbase-tree-cache acts as a limited hashtable.
664 * your object will be found at your index or within a few
665 * slots after that slot if it is cached.
667 for (neigh = 0; neigh < 8; neigh++) {
668 ent = pbase_tree_cache[my_ix];
669 if (ent && !hashcmp(ent->sha1, sha1)) {
670 ent->ref++;
671 return ent;
673 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
674 ((0 <= available_ix) &&
675 (!ent && pbase_tree_cache[available_ix])))
676 available_ix = my_ix;
677 if (!ent)
678 break;
679 my_ix = pbase_tree_cache_ix_incr(my_ix);
682 /* Did not find one. Either we got a bogus request or
683 * we need to read and perhaps cache.
685 data = read_sha1_file(sha1, type, &size);
686 if (!data)
687 return NULL;
688 if (strcmp(type, tree_type)) {
689 free(data);
690 return NULL;
693 /* We need to either cache or return a throwaway copy */
695 if (available_ix < 0)
696 ent = NULL;
697 else {
698 ent = pbase_tree_cache[available_ix];
699 my_ix = available_ix;
702 if (!ent) {
703 nent = xmalloc(sizeof(*nent));
704 nent->temporary = (available_ix < 0);
706 else {
707 /* evict and reuse */
708 free(ent->tree_data);
709 nent = ent;
711 hashcpy(nent->sha1, sha1);
712 nent->tree_data = data;
713 nent->tree_size = size;
714 nent->ref = 1;
715 if (!nent->temporary)
716 pbase_tree_cache[my_ix] = nent;
717 return nent;
720 static void pbase_tree_put(struct pbase_tree_cache *cache)
722 if (!cache->temporary) {
723 cache->ref--;
724 return;
726 free(cache->tree_data);
727 free(cache);
730 static int name_cmp_len(const char *name)
732 int i;
733 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
735 return i;
738 static void add_pbase_object(struct tree_desc *tree,
739 const char *name,
740 int cmplen,
741 const char *fullname)
743 struct name_entry entry;
745 while (tree_entry(tree,&entry)) {
746 unsigned long size;
747 char type[20];
749 if (entry.pathlen != cmplen ||
750 memcmp(entry.path, name, cmplen) ||
751 !has_sha1_file(entry.sha1) ||
752 sha1_object_info(entry.sha1, type, &size))
753 continue;
754 if (name[cmplen] != '/') {
755 unsigned hash = name_hash(fullname);
756 add_object_entry(entry.sha1, hash, 1);
757 return;
759 if (!strcmp(type, tree_type)) {
760 struct tree_desc sub;
761 struct pbase_tree_cache *tree;
762 const char *down = name+cmplen+1;
763 int downlen = name_cmp_len(down);
765 tree = pbase_tree_get(entry.sha1);
766 if (!tree)
767 return;
768 sub.buf = tree->tree_data;
769 sub.size = tree->tree_size;
771 add_pbase_object(&sub, down, downlen, fullname);
772 pbase_tree_put(tree);
777 static unsigned *done_pbase_paths;
778 static int done_pbase_paths_num;
779 static int done_pbase_paths_alloc;
780 static int done_pbase_path_pos(unsigned hash)
782 int lo = 0;
783 int hi = done_pbase_paths_num;
784 while (lo < hi) {
785 int mi = (hi + lo) / 2;
786 if (done_pbase_paths[mi] == hash)
787 return mi;
788 if (done_pbase_paths[mi] < hash)
789 hi = mi;
790 else
791 lo = mi + 1;
793 return -lo-1;
796 static int check_pbase_path(unsigned hash)
798 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
799 if (0 <= pos)
800 return 1;
801 pos = -pos - 1;
802 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
803 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
804 done_pbase_paths = xrealloc(done_pbase_paths,
805 done_pbase_paths_alloc *
806 sizeof(unsigned));
808 done_pbase_paths_num++;
809 if (pos < done_pbase_paths_num)
810 memmove(done_pbase_paths + pos + 1,
811 done_pbase_paths + pos,
812 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
813 done_pbase_paths[pos] = hash;
814 return 0;
817 static void add_preferred_base_object(char *name, unsigned hash)
819 struct pbase_tree *it;
820 int cmplen = name_cmp_len(name);
822 if (check_pbase_path(hash))
823 return;
825 for (it = pbase_tree; it; it = it->next) {
826 if (cmplen == 0) {
827 hash = name_hash("");
828 add_object_entry(it->pcache.sha1, hash, 1);
830 else {
831 struct tree_desc tree;
832 tree.buf = it->pcache.tree_data;
833 tree.size = it->pcache.tree_size;
834 add_pbase_object(&tree, name, cmplen, name);
839 static void add_preferred_base(unsigned char *sha1)
841 struct pbase_tree *it;
842 void *data;
843 unsigned long size;
844 unsigned char tree_sha1[20];
846 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
847 if (!data)
848 return;
850 for (it = pbase_tree; it; it = it->next) {
851 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
852 free(data);
853 return;
857 it = xcalloc(1, sizeof(*it));
858 it->next = pbase_tree;
859 pbase_tree = it;
861 hashcpy(it->pcache.sha1, tree_sha1);
862 it->pcache.tree_data = data;
863 it->pcache.tree_size = size;
866 static void check_object(struct object_entry *entry)
868 char type[20];
870 if (entry->in_pack && !entry->preferred_base) {
871 unsigned char base[20];
872 unsigned long size;
873 struct object_entry *base_entry;
875 /* We want in_pack_type even if we do not reuse delta.
876 * There is no point not reusing non-delta representations.
878 check_reuse_pack_delta(entry->in_pack,
879 entry->in_pack_offset,
880 base, &size,
881 &entry->in_pack_type);
883 /* Check if it is delta, and the base is also an object
884 * we are going to pack. If so we will reuse the existing
885 * delta.
887 if (!no_reuse_delta &&
888 entry->in_pack_type == OBJ_DELTA &&
889 (base_entry = locate_object_entry(base)) &&
890 (!base_entry->preferred_base)) {
892 /* Depth value does not matter - find_deltas()
893 * will never consider reused delta as the
894 * base object to deltify other objects
895 * against, in order to avoid circular deltas.
898 /* uncompressed size of the delta data */
899 entry->size = entry->delta_size = size;
900 entry->delta = base_entry;
901 entry->type = OBJ_DELTA;
903 entry->delta_sibling = base_entry->delta_child;
904 base_entry->delta_child = entry;
906 return;
908 /* Otherwise we would do the usual */
911 if (sha1_object_info(entry->sha1, type, &entry->size))
912 die("unable to get type of object %s",
913 sha1_to_hex(entry->sha1));
915 if (!strcmp(type, commit_type)) {
916 entry->type = OBJ_COMMIT;
917 } else if (!strcmp(type, tree_type)) {
918 entry->type = OBJ_TREE;
919 } else if (!strcmp(type, blob_type)) {
920 entry->type = OBJ_BLOB;
921 } else if (!strcmp(type, tag_type)) {
922 entry->type = OBJ_TAG;
923 } else
924 die("unable to pack object %s of type %s",
925 sha1_to_hex(entry->sha1), type);
928 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
930 struct object_entry *child = me->delta_child;
931 unsigned int m = n;
932 while (child) {
933 unsigned int c = check_delta_limit(child, n + 1);
934 if (m < c)
935 m = c;
936 child = child->delta_sibling;
938 return m;
941 static void get_object_details(void)
943 int i;
944 struct object_entry *entry;
946 prepare_pack_ix();
947 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
948 check_object(entry);
950 if (nr_objects == nr_result) {
952 * Depth of objects that depend on the entry -- this
953 * is subtracted from depth-max to break too deep
954 * delta chain because of delta data reusing.
955 * However, we loosen this restriction when we know we
956 * are creating a thin pack -- it will have to be
957 * expanded on the other end anyway, so do not
958 * artificially cut the delta chain and let it go as
959 * deep as it wants.
961 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
962 if (!entry->delta && entry->delta_child)
963 entry->delta_limit =
964 check_delta_limit(entry, 1);
968 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
970 static entry_sort_t current_sort;
972 static int sort_comparator(const void *_a, const void *_b)
974 struct object_entry *a = *(struct object_entry **)_a;
975 struct object_entry *b = *(struct object_entry **)_b;
976 return current_sort(a,b);
979 static struct object_entry **create_sorted_list(entry_sort_t sort)
981 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
982 int i;
984 for (i = 0; i < nr_objects; i++)
985 list[i] = objects + i;
986 current_sort = sort;
987 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
988 return list;
991 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
993 return hashcmp(a->sha1, b->sha1);
996 static struct object_entry **create_final_object_list(void)
998 struct object_entry **list;
999 int i, j;
1001 for (i = nr_result = 0; i < nr_objects; i++)
1002 if (!objects[i].preferred_base)
1003 nr_result++;
1004 list = xmalloc(nr_result * sizeof(struct object_entry *));
1005 for (i = j = 0; i < nr_objects; i++) {
1006 if (!objects[i].preferred_base)
1007 list[j++] = objects + i;
1009 current_sort = sha1_sort;
1010 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1011 return list;
1014 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1016 if (a->type < b->type)
1017 return -1;
1018 if (a->type > b->type)
1019 return 1;
1020 if (a->hash < b->hash)
1021 return -1;
1022 if (a->hash > b->hash)
1023 return 1;
1024 if (a->preferred_base < b->preferred_base)
1025 return -1;
1026 if (a->preferred_base > b->preferred_base)
1027 return 1;
1028 if (a->size < b->size)
1029 return -1;
1030 if (a->size > b->size)
1031 return 1;
1032 return a < b ? -1 : (a > b);
1035 struct unpacked {
1036 struct object_entry *entry;
1037 void *data;
1038 struct delta_index *index;
1042 * We search for deltas _backwards_ in a list sorted by type and
1043 * by size, so that we see progressively smaller and smaller files.
1044 * That's because we prefer deltas to be from the bigger file
1045 * to the smaller - deletes are potentially cheaper, but perhaps
1046 * more importantly, the bigger file is likely the more recent
1047 * one.
1049 static int try_delta(struct unpacked *trg, struct unpacked *src,
1050 unsigned max_depth)
1052 struct object_entry *trg_entry = trg->entry;
1053 struct object_entry *src_entry = src->entry;
1054 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1055 char type[10];
1056 void *delta_buf;
1058 /* Don't bother doing diffs between different types */
1059 if (trg_entry->type != src_entry->type)
1060 return -1;
1062 /* We do not compute delta to *create* objects we are not
1063 * going to pack.
1065 if (trg_entry->preferred_base)
1066 return -1;
1069 * We do not bother to try a delta that we discarded
1070 * on an earlier try, but only when reusing delta data.
1072 if (!no_reuse_delta && trg_entry->in_pack &&
1073 trg_entry->in_pack == src_entry->in_pack)
1074 return 0;
1077 * If the current object is at pack edge, take the depth the
1078 * objects that depend on the current object into account --
1079 * otherwise they would become too deep.
1081 if (trg_entry->delta_child) {
1082 if (max_depth <= trg_entry->delta_limit)
1083 return 0;
1084 max_depth -= trg_entry->delta_limit;
1086 if (src_entry->depth >= max_depth)
1087 return 0;
1089 /* Now some size filtering heuristics. */
1090 trg_size = trg_entry->size;
1091 max_size = trg_size/2 - 20;
1092 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1093 if (max_size == 0)
1094 return 0;
1095 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1096 max_size = trg_entry->delta_size-1;
1097 src_size = src_entry->size;
1098 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1099 if (sizediff >= max_size)
1100 return 0;
1102 /* Load data if not already done */
1103 if (!trg->data) {
1104 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1105 if (sz != trg_size)
1106 die("object %s inconsistent object length (%lu vs %lu)",
1107 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1109 if (!src->data) {
1110 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1111 if (sz != src_size)
1112 die("object %s inconsistent object length (%lu vs %lu)",
1113 sha1_to_hex(src_entry->sha1), sz, src_size);
1115 if (!src->index) {
1116 src->index = create_delta_index(src->data, src_size);
1117 if (!src->index)
1118 die("out of memory");
1121 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1122 if (!delta_buf)
1123 return 0;
1125 trg_entry->delta = src_entry;
1126 trg_entry->delta_size = delta_size;
1127 trg_entry->depth = src_entry->depth + 1;
1128 free(delta_buf);
1129 return 1;
1132 static void progress_interval(int signum)
1134 progress_update = 1;
1137 static void find_deltas(struct object_entry **list, int window, int depth)
1139 int i, idx;
1140 unsigned int array_size = window * sizeof(struct unpacked);
1141 struct unpacked *array = xmalloc(array_size);
1142 unsigned processed = 0;
1143 unsigned last_percent = 999;
1145 memset(array, 0, array_size);
1146 i = nr_objects;
1147 idx = 0;
1148 if (progress)
1149 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1151 while (--i >= 0) {
1152 struct object_entry *entry = list[i];
1153 struct unpacked *n = array + idx;
1154 int j;
1156 if (!entry->preferred_base)
1157 processed++;
1159 if (progress) {
1160 unsigned percent = processed * 100 / nr_result;
1161 if (percent != last_percent || progress_update) {
1162 fprintf(stderr, "%4u%% (%u/%u) done\r",
1163 percent, processed, nr_result);
1164 progress_update = 0;
1165 last_percent = percent;
1169 if (entry->delta)
1170 /* This happens if we decided to reuse existing
1171 * delta from a pack. "!no_reuse_delta &&" is implied.
1173 continue;
1175 if (entry->size < 50)
1176 continue;
1177 free_delta_index(n->index);
1178 n->index = NULL;
1179 free(n->data);
1180 n->data = NULL;
1181 n->entry = entry;
1183 j = window;
1184 while (--j > 0) {
1185 unsigned int other_idx = idx + j;
1186 struct unpacked *m;
1187 if (other_idx >= window)
1188 other_idx -= window;
1189 m = array + other_idx;
1190 if (!m->entry)
1191 break;
1192 if (try_delta(n, m, depth) < 0)
1193 break;
1195 /* if we made n a delta, and if n is already at max
1196 * depth, leaving it in the window is pointless. we
1197 * should evict it first.
1199 if (entry->delta && depth <= entry->depth)
1200 continue;
1202 idx++;
1203 if (idx >= window)
1204 idx = 0;
1207 if (progress)
1208 fputc('\n', stderr);
1210 for (i = 0; i < window; ++i) {
1211 free_delta_index(array[i].index);
1212 free(array[i].data);
1214 free(array);
1217 static void prepare_pack(int window, int depth)
1219 get_object_details();
1220 sorted_by_type = create_sorted_list(type_size_sort);
1221 if (window && depth)
1222 find_deltas(sorted_by_type, window+1, depth);
1225 static int reuse_cached_pack(unsigned char *sha1)
1227 static const char cache[] = "pack-cache/pack-%s.%s";
1228 char *cached_pack, *cached_idx;
1229 int ifd, ofd, ifd_ix = -1;
1231 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1232 ifd = open(cached_pack, O_RDONLY);
1233 if (ifd < 0)
1234 return 0;
1236 if (!pack_to_stdout) {
1237 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1238 ifd_ix = open(cached_idx, O_RDONLY);
1239 if (ifd_ix < 0) {
1240 close(ifd);
1241 return 0;
1245 if (progress)
1246 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1247 sha1_to_hex(sha1));
1249 if (pack_to_stdout) {
1250 if (copy_fd(ifd, 1))
1251 exit(1);
1252 close(ifd);
1254 else {
1255 char name[PATH_MAX];
1256 snprintf(name, sizeof(name),
1257 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1258 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1259 if (ofd < 0)
1260 die("unable to open %s (%s)", name, strerror(errno));
1261 if (copy_fd(ifd, ofd))
1262 exit(1);
1263 close(ifd);
1265 snprintf(name, sizeof(name),
1266 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1267 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1268 if (ofd < 0)
1269 die("unable to open %s (%s)", name, strerror(errno));
1270 if (copy_fd(ifd_ix, ofd))
1271 exit(1);
1272 close(ifd_ix);
1273 puts(sha1_to_hex(sha1));
1276 return 1;
1279 static void setup_progress_signal(void)
1281 struct sigaction sa;
1282 struct itimerval v;
1284 memset(&sa, 0, sizeof(sa));
1285 sa.sa_handler = progress_interval;
1286 sigemptyset(&sa.sa_mask);
1287 sa.sa_flags = SA_RESTART;
1288 sigaction(SIGALRM, &sa, NULL);
1290 v.it_interval.tv_sec = 1;
1291 v.it_interval.tv_usec = 0;
1292 v.it_value = v.it_interval;
1293 setitimer(ITIMER_REAL, &v, NULL);
1296 static int git_pack_config(const char *k, const char *v)
1298 if(!strcmp(k, "pack.window")) {
1299 window = git_config_int(k, v);
1300 return 0;
1302 return git_default_config(k, v);
1305 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1307 SHA_CTX ctx;
1308 char line[40 + 1 + PATH_MAX + 2];
1309 int depth = 10;
1310 struct object_entry **list;
1311 int num_preferred_base = 0;
1312 int i;
1314 git_config(git_pack_config);
1316 progress = isatty(2);
1317 for (i = 1; i < argc; i++) {
1318 const char *arg = argv[i];
1320 if (*arg == '-') {
1321 if (!strcmp("--non-empty", arg)) {
1322 non_empty = 1;
1323 continue;
1325 if (!strcmp("--local", arg)) {
1326 local = 1;
1327 continue;
1329 if (!strcmp("--progress", arg)) {
1330 progress = 1;
1331 continue;
1333 if (!strcmp("--incremental", arg)) {
1334 incremental = 1;
1335 continue;
1337 if (!strncmp("--window=", arg, 9)) {
1338 char *end;
1339 window = strtoul(arg+9, &end, 0);
1340 if (!arg[9] || *end)
1341 usage(pack_usage);
1342 continue;
1344 if (!strncmp("--depth=", arg, 8)) {
1345 char *end;
1346 depth = strtoul(arg+8, &end, 0);
1347 if (!arg[8] || *end)
1348 usage(pack_usage);
1349 continue;
1351 if (!strcmp("--progress", arg)) {
1352 progress = 1;
1353 continue;
1355 if (!strcmp("-q", arg)) {
1356 progress = 0;
1357 continue;
1359 if (!strcmp("--no-reuse-delta", arg)) {
1360 no_reuse_delta = 1;
1361 continue;
1363 if (!strcmp("--stdout", arg)) {
1364 pack_to_stdout = 1;
1365 continue;
1367 usage(pack_usage);
1369 if (base_name)
1370 usage(pack_usage);
1371 base_name = arg;
1374 if (pack_to_stdout != !base_name)
1375 usage(pack_usage);
1377 prepare_packed_git();
1379 if (progress) {
1380 fprintf(stderr, "Generating pack...\n");
1381 setup_progress_signal();
1384 for (;;) {
1385 unsigned char sha1[20];
1386 unsigned hash;
1388 if (!fgets(line, sizeof(line), stdin)) {
1389 if (feof(stdin))
1390 break;
1391 if (!ferror(stdin))
1392 die("fgets returned NULL, not EOF, not error!");
1393 if (errno != EINTR)
1394 die("fgets: %s", strerror(errno));
1395 clearerr(stdin);
1396 continue;
1399 if (line[0] == '-') {
1400 if (get_sha1_hex(line+1, sha1))
1401 die("expected edge sha1, got garbage:\n %s",
1402 line+1);
1403 if (num_preferred_base++ < window)
1404 add_preferred_base(sha1);
1405 continue;
1407 if (get_sha1_hex(line, sha1))
1408 die("expected sha1, got garbage:\n %s", line);
1409 hash = name_hash(line+41);
1410 add_preferred_base_object(line+41, hash);
1411 add_object_entry(sha1, hash, 0);
1413 if (progress)
1414 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1415 sorted_by_sha = create_final_object_list();
1416 if (non_empty && !nr_result)
1417 return 0;
1419 SHA1_Init(&ctx);
1420 list = sorted_by_sha;
1421 for (i = 0; i < nr_result; i++) {
1422 struct object_entry *entry = *list++;
1423 SHA1_Update(&ctx, entry->sha1, 20);
1425 SHA1_Final(object_list_sha1, &ctx);
1426 if (progress && (nr_objects != nr_result))
1427 fprintf(stderr, "Result has %d objects.\n", nr_result);
1429 if (reuse_cached_pack(object_list_sha1))
1431 else {
1432 if (nr_result)
1433 prepare_pack(window, depth);
1434 if (progress && pack_to_stdout) {
1435 /* the other end usually displays progress itself */
1436 struct itimerval v = {{0,},};
1437 setitimer(ITIMER_REAL, &v, NULL);
1438 signal(SIGALRM, SIG_IGN );
1439 progress_update = 0;
1441 write_pack_file();
1442 if (!pack_to_stdout) {
1443 write_index_file();
1444 puts(sha1_to_hex(object_list_sha1));
1447 if (progress)
1448 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1449 nr_result, written, written_delta, reused, reused_delta);
1450 return 0;