![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
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.
22 */
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
35 */
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.
39 */
40 };
42 /*
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.
47 *
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.
53 */
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;
70 /*
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.
75 */
76 static int *object_ix;
77 static int object_ix_hashsz;
79 /*
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.
89 */
90 struct pack_revindex {
91 struct packed_git *p;
92 unsigned long *revindex;
93 } *pack_revindex = NULL;
94 static int pack_revindex_hashsz;
96 /*
97 * stats
98 */
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)
105 {
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;
116 }
117 return -1 - i;
118 }
120 static void prepare_pack_ix(void)
121 {
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;
134 }
135 /* revindex elements are lazily initialized */
136 }
138 static int cmp_offset(const void *a_, const void *b_)
139 {
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;
148 }
150 /*
151 * Ordered list of offsets of objects in the pack.
152 */
153 static void prepare_pack_revindex(struct pack_revindex *rix)
154 {
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);
164 }
165 /* This knows the pack format -- the 20-byte trailer
166 * follows immediately after the last object data.
167 */
168 rix->revindex[num_ent] = p->pack_size - 20;
169 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
170 }
172 static unsigned long find_packed_object_size(struct packed_git *p,
173 unsigned long ofs)
174 {
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;
192 }
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");
199 }
201 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
202 {
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;
217 }
219 /*
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"
225 */
226 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
227 {
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++;
241 }
242 *hdr = c;
243 return n;
244 }
246 static int check_inflate(unsigned char *data, unsigned long len, unsigned long expect)
247 {
248 z_stream stream;
249 unsigned char fakebuf[4096];
250 int st;
252 memset(&stream, 0, sizeof(stream));
253 stream.next_in = data;
254 stream.avail_in = len;
255 stream.next_out = fakebuf;
256 stream.avail_out = sizeof(fakebuf);
257 inflateInit(&stream);
259 while (1) {
260 st = inflate(&stream, Z_FINISH);
261 if (st == Z_STREAM_END || st == Z_OK) {
262 st = (stream.total_out == expect &&
263 stream.total_in == len) ? 0 : -1;
264 break;
265 }
266 if (st != Z_BUF_ERROR) {
267 st = -1;
268 break;
269 }
270 stream.next_out = fakebuf;
271 stream.avail_out = sizeof(fakebuf);
272 }
273 inflateEnd(&stream);
274 return st;
275 }
277 /*
278 * we are going to reuse the existing pack entry data. make
279 * sure it is not corrupt.
280 */
281 static int revalidate_pack_entry(struct object_entry *entry, unsigned char *data, unsigned long len)
282 {
283 enum object_type type;
284 unsigned long size, used;
286 if (pack_to_stdout)
287 return 0;
289 /* the caller has already called use_packed_git() for us,
290 * so it is safe to access the pack data from mmapped location.
291 * make sure the entry inflates correctly.
292 */
293 used = unpack_object_header_gently(data, len, &type, &size);
294 if (!used)
295 return -1;
296 if (type == OBJ_DELTA)
297 used += 20; /* skip base object name */
298 data += used;
299 len -= used;
300 return check_inflate(data, len, entry->size);
301 }
303 static int revalidate_loose_object(struct object_entry *entry,
304 unsigned char *map,
305 unsigned long mapsize)
306 {
307 /* we already know this is a loose object with new type header. */
308 enum object_type type;
309 unsigned long size, used;
311 if (pack_to_stdout)
312 return 0;
314 used = unpack_object_header_gently(map, mapsize, &type, &size);
315 if (!used)
316 return -1;
317 map += used;
318 mapsize -= used;
319 return check_inflate(map, mapsize, size);
320 }
322 static unsigned long write_object(struct sha1file *f,
323 struct object_entry *entry)
324 {
325 unsigned long size;
326 char type[10];
327 void *buf;
328 unsigned char header[10];
329 unsigned hdrlen, datalen;
330 enum object_type obj_type;
331 int to_reuse = 0;
333 if (entry->preferred_base)
334 return 0;
336 obj_type = entry->type;
337 if (! entry->in_pack)
338 to_reuse = 0; /* can't reuse what we don't have */
339 else if (obj_type == OBJ_DELTA)
340 to_reuse = 1; /* check_object() decided it for us */
341 else if (obj_type != entry->in_pack_type)
342 to_reuse = 0; /* pack has delta which is unusable */
343 else if (entry->delta)
344 to_reuse = 0; /* we want to pack afresh */
345 else
346 to_reuse = 1; /* we have it in-pack undeltified,
347 * and we do not need to deltify it.
348 */
350 if (!entry->in_pack && !entry->delta) {
351 unsigned char *map;
352 unsigned long mapsize;
353 map = map_sha1_file(entry->sha1, &mapsize);
354 if (map && !legacy_loose_object(map)) {
355 /* We can copy straight into the pack file */
356 if (revalidate_loose_object(entry, map, mapsize))
357 die("corrupt loose object %s",
358 sha1_to_hex(entry->sha1));
359 sha1write(f, map, mapsize);
360 munmap(map, mapsize);
361 written++;
362 reused++;
363 return mapsize;
364 }
365 if (map)
366 munmap(map, mapsize);
367 }
369 if (!to_reuse) {
370 buf = read_sha1_file(entry->sha1, type, &size);
371 if (!buf)
372 die("unable to read %s", sha1_to_hex(entry->sha1));
373 if (size != entry->size)
374 die("object %s size inconsistency (%lu vs %lu)",
375 sha1_to_hex(entry->sha1), size, entry->size);
376 if (entry->delta) {
377 buf = delta_against(buf, size, entry);
378 size = entry->delta_size;
379 obj_type = OBJ_DELTA;
380 }
381 /*
382 * The object header is a byte of 'type' followed by zero or
383 * more bytes of length. For deltas, the 20 bytes of delta
384 * sha1 follows that.
385 */
386 hdrlen = encode_header(obj_type, size, header);
387 sha1write(f, header, hdrlen);
389 if (entry->delta) {
390 sha1write(f, entry->delta, 20);
391 hdrlen += 20;
392 }
393 datalen = sha1write_compressed(f, buf, size);
394 free(buf);
395 }
396 else {
397 struct packed_git *p = entry->in_pack;
398 use_packed_git(p);
400 datalen = find_packed_object_size(p, entry->in_pack_offset);
401 buf = (char *) p->pack_base + entry->in_pack_offset;
403 if (revalidate_pack_entry(entry, buf, datalen))
404 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
405 sha1write(f, buf, datalen);
406 unuse_packed_git(p);
407 hdrlen = 0; /* not really */
408 if (obj_type == OBJ_DELTA)
409 reused_delta++;
410 reused++;
411 }
412 if (obj_type == OBJ_DELTA)
413 written_delta++;
414 written++;
415 return hdrlen + datalen;
416 }
418 static unsigned long write_one(struct sha1file *f,
419 struct object_entry *e,
420 unsigned long offset)
421 {
422 if (e->offset)
423 /* offset starts from header size and cannot be zero
424 * if it is written already.
425 */
426 return offset;
427 e->offset = offset;
428 offset += write_object(f, e);
429 /* if we are deltified, write out its base object. */
430 if (e->delta)
431 offset = write_one(f, e->delta, offset);
432 return offset;
433 }
435 static void write_pack_file(void)
436 {
437 int i;
438 struct sha1file *f;
439 unsigned long offset;
440 struct pack_header hdr;
441 unsigned last_percent = 999;
442 int do_progress = 0;
444 if (!base_name)
445 f = sha1fd(1, "<stdout>");
446 else {
447 f = sha1create("%s-%s.%s", base_name,
448 sha1_to_hex(object_list_sha1), "pack");
449 do_progress = progress;
450 }
451 if (do_progress)
452 fprintf(stderr, "Writing %d objects.\n", nr_result);
454 hdr.hdr_signature = htonl(PACK_SIGNATURE);
455 hdr.hdr_version = htonl(PACK_VERSION);
456 hdr.hdr_entries = htonl(nr_result);
457 sha1write(f, &hdr, sizeof(hdr));
458 offset = sizeof(hdr);
459 if (!nr_result)
460 goto done;
461 for (i = 0; i < nr_objects; i++) {
462 offset = write_one(f, objects + i, offset);
463 if (do_progress) {
464 unsigned percent = written * 100 / nr_result;
465 if (progress_update || percent != last_percent) {
466 fprintf(stderr, "%4u%% (%u/%u) done\r",
467 percent, written, nr_result);
468 progress_update = 0;
469 last_percent = percent;
470 }
471 }
472 }
473 if (do_progress)
474 fputc('\n', stderr);
475 done:
476 sha1close(f, pack_file_sha1, 1);
477 }
479 static void write_index_file(void)
480 {
481 int i;
482 struct sha1file *f = sha1create("%s-%s.%s", base_name,
483 sha1_to_hex(object_list_sha1), "idx");
484 struct object_entry **list = sorted_by_sha;
485 struct object_entry **last = list + nr_result;
486 unsigned int array[256];
488 /*
489 * Write the first-level table (the list is sorted,
490 * but we use a 256-entry lookup to be able to avoid
491 * having to do eight extra binary search iterations).
492 */
493 for (i = 0; i < 256; i++) {
494 struct object_entry **next = list;
495 while (next < last) {
496 struct object_entry *entry = *next;
497 if (entry->sha1[0] != i)
498 break;
499 next++;
500 }
501 array[i] = htonl(next - sorted_by_sha);
502 list = next;
503 }
504 sha1write(f, array, 256 * sizeof(int));
506 /*
507 * Write the actual SHA1 entries..
508 */
509 list = sorted_by_sha;
510 for (i = 0; i < nr_result; i++) {
511 struct object_entry *entry = *list++;
512 unsigned int offset = htonl(entry->offset);
513 sha1write(f, &offset, 4);
514 sha1write(f, entry->sha1, 20);
515 }
516 sha1write(f, pack_file_sha1, 20);
517 sha1close(f, NULL, 1);
518 }
520 static int locate_object_entry_hash(const unsigned char *sha1)
521 {
522 int i;
523 unsigned int ui;
524 memcpy(&ui, sha1, sizeof(unsigned int));
525 i = ui % object_ix_hashsz;
526 while (0 < object_ix[i]) {
527 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
528 return i;
529 if (++i == object_ix_hashsz)
530 i = 0;
531 }
532 return -1 - i;
533 }
535 static struct object_entry *locate_object_entry(const unsigned char *sha1)
536 {
537 int i;
539 if (!object_ix_hashsz)
540 return NULL;
542 i = locate_object_entry_hash(sha1);
543 if (0 <= i)
544 return &objects[object_ix[i]-1];
545 return NULL;
546 }
548 static void rehash_objects(void)
549 {
550 int i;
551 struct object_entry *oe;
553 object_ix_hashsz = nr_objects * 3;
554 if (object_ix_hashsz < 1024)
555 object_ix_hashsz = 1024;
556 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
557 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
558 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
559 int ix = locate_object_entry_hash(oe->sha1);
560 if (0 <= ix)
561 continue;
562 ix = -1 - ix;
563 object_ix[ix] = i + 1;
564 }
565 }
567 static unsigned name_hash(const char *name)
568 {
569 unsigned char c;
570 unsigned hash = 0;
572 /*
573 * This effectively just creates a sortable number from the
574 * last sixteen non-whitespace characters. Last characters
575 * count "most", so things that end in ".c" sort together.
576 */
577 while ((c = *name++) != 0) {
578 if (isspace(c))
579 continue;
580 hash = (hash >> 2) + (c << 24);
581 }
582 return hash;
583 }
585 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
586 {
587 unsigned int idx = nr_objects;
588 struct object_entry *entry;
589 struct packed_git *p;
590 unsigned int found_offset = 0;
591 struct packed_git *found_pack = NULL;
592 int ix, status = 0;
594 if (!exclude) {
595 for (p = packed_git; p; p = p->next) {
596 struct pack_entry e;
597 if (find_pack_entry_one(sha1, &e, p)) {
598 if (incremental)
599 return 0;
600 if (local && !p->pack_local)
601 return 0;
602 if (!found_pack) {
603 found_offset = e.offset;
604 found_pack = e.p;
605 }
606 }
607 }
608 }
609 if ((entry = locate_object_entry(sha1)) != NULL)
610 goto already_added;
612 if (idx >= nr_alloc) {
613 unsigned int needed = (idx + 1024) * 3 / 2;
614 objects = xrealloc(objects, needed * sizeof(*entry));
615 nr_alloc = needed;
616 }
617 entry = objects + idx;
618 nr_objects = idx + 1;
619 memset(entry, 0, sizeof(*entry));
620 hashcpy(entry->sha1, sha1);
621 entry->hash = hash;
623 if (object_ix_hashsz * 3 <= nr_objects * 4)
624 rehash_objects();
625 else {
626 ix = locate_object_entry_hash(entry->sha1);
627 if (0 <= ix)
628 die("internal error in object hashing.");
629 object_ix[-1 - ix] = idx + 1;
630 }
631 status = 1;
633 already_added:
634 if (progress_update) {
635 fprintf(stderr, "Counting objects...%d\r", nr_objects);
636 progress_update = 0;
637 }
638 if (exclude)
639 entry->preferred_base = 1;
640 else {
641 if (found_pack) {
642 entry->in_pack = found_pack;
643 entry->in_pack_offset = found_offset;
644 }
645 }
646 return status;
647 }
649 struct pbase_tree_cache {
650 unsigned char sha1[20];
651 int ref;
652 int temporary;
653 void *tree_data;
654 unsigned long tree_size;
655 };
657 static struct pbase_tree_cache *(pbase_tree_cache[256]);
658 static int pbase_tree_cache_ix(const unsigned char *sha1)
659 {
660 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
661 }
662 static int pbase_tree_cache_ix_incr(int ix)
663 {
664 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
665 }
667 static struct pbase_tree {
668 struct pbase_tree *next;
669 /* This is a phony "cache" entry; we are not
670 * going to evict it nor find it through _get()
671 * mechanism -- this is for the toplevel node that
672 * would almost always change with any commit.
673 */
674 struct pbase_tree_cache pcache;
675 } *pbase_tree;
677 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
678 {
679 struct pbase_tree_cache *ent, *nent;
680 void *data;
681 unsigned long size;
682 char type[20];
683 int neigh;
684 int my_ix = pbase_tree_cache_ix(sha1);
685 int available_ix = -1;
687 /* pbase-tree-cache acts as a limited hashtable.
688 * your object will be found at your index or within a few
689 * slots after that slot if it is cached.
690 */
691 for (neigh = 0; neigh < 8; neigh++) {
692 ent = pbase_tree_cache[my_ix];
693 if (ent && !hashcmp(ent->sha1, sha1)) {
694 ent->ref++;
695 return ent;
696 }
697 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
698 ((0 <= available_ix) &&
699 (!ent && pbase_tree_cache[available_ix])))
700 available_ix = my_ix;
701 if (!ent)
702 break;
703 my_ix = pbase_tree_cache_ix_incr(my_ix);
704 }
706 /* Did not find one. Either we got a bogus request or
707 * we need to read and perhaps cache.
708 */
709 data = read_sha1_file(sha1, type, &size);
710 if (!data)
711 return NULL;
712 if (strcmp(type, tree_type)) {
713 free(data);
714 return NULL;
715 }
717 /* We need to either cache or return a throwaway copy */
719 if (available_ix < 0)
720 ent = NULL;
721 else {
722 ent = pbase_tree_cache[available_ix];
723 my_ix = available_ix;
724 }
726 if (!ent) {
727 nent = xmalloc(sizeof(*nent));
728 nent->temporary = (available_ix < 0);
729 }
730 else {
731 /* evict and reuse */
732 free(ent->tree_data);
733 nent = ent;
734 }
735 hashcpy(nent->sha1, sha1);
736 nent->tree_data = data;
737 nent->tree_size = size;
738 nent->ref = 1;
739 if (!nent->temporary)
740 pbase_tree_cache[my_ix] = nent;
741 return nent;
742 }
744 static void pbase_tree_put(struct pbase_tree_cache *cache)
745 {
746 if (!cache->temporary) {
747 cache->ref--;
748 return;
749 }
750 free(cache->tree_data);
751 free(cache);
752 }
754 static int name_cmp_len(const char *name)
755 {
756 int i;
757 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
758 ;
759 return i;
760 }
762 static void add_pbase_object(struct tree_desc *tree,
763 const char *name,
764 int cmplen,
765 const char *fullname)
766 {
767 struct name_entry entry;
769 while (tree_entry(tree,&entry)) {
770 unsigned long size;
771 char type[20];
773 if (entry.pathlen != cmplen ||
774 memcmp(entry.path, name, cmplen) ||
775 !has_sha1_file(entry.sha1) ||
776 sha1_object_info(entry.sha1, type, &size))
777 continue;
778 if (name[cmplen] != '/') {
779 unsigned hash = name_hash(fullname);
780 add_object_entry(entry.sha1, hash, 1);
781 return;
782 }
783 if (!strcmp(type, tree_type)) {
784 struct tree_desc sub;
785 struct pbase_tree_cache *tree;
786 const char *down = name+cmplen+1;
787 int downlen = name_cmp_len(down);
789 tree = pbase_tree_get(entry.sha1);
790 if (!tree)
791 return;
792 sub.buf = tree->tree_data;
793 sub.size = tree->tree_size;
795 add_pbase_object(&sub, down, downlen, fullname);
796 pbase_tree_put(tree);
797 }
798 }
799 }
801 static unsigned *done_pbase_paths;
802 static int done_pbase_paths_num;
803 static int done_pbase_paths_alloc;
804 static int done_pbase_path_pos(unsigned hash)
805 {
806 int lo = 0;
807 int hi = done_pbase_paths_num;
808 while (lo < hi) {
809 int mi = (hi + lo) / 2;
810 if (done_pbase_paths[mi] == hash)
811 return mi;
812 if (done_pbase_paths[mi] < hash)
813 hi = mi;
814 else
815 lo = mi + 1;
816 }
817 return -lo-1;
818 }
820 static int check_pbase_path(unsigned hash)
821 {
822 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
823 if (0 <= pos)
824 return 1;
825 pos = -pos - 1;
826 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
827 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
828 done_pbase_paths = xrealloc(done_pbase_paths,
829 done_pbase_paths_alloc *
830 sizeof(unsigned));
831 }
832 done_pbase_paths_num++;
833 if (pos < done_pbase_paths_num)
834 memmove(done_pbase_paths + pos + 1,
835 done_pbase_paths + pos,
836 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
837 done_pbase_paths[pos] = hash;
838 return 0;
839 }
841 static void add_preferred_base_object(char *name, unsigned hash)
842 {
843 struct pbase_tree *it;
844 int cmplen = name_cmp_len(name);
846 if (check_pbase_path(hash))
847 return;
849 for (it = pbase_tree; it; it = it->next) {
850 if (cmplen == 0) {
851 hash = name_hash("");
852 add_object_entry(it->pcache.sha1, hash, 1);
853 }
854 else {
855 struct tree_desc tree;
856 tree.buf = it->pcache.tree_data;
857 tree.size = it->pcache.tree_size;
858 add_pbase_object(&tree, name, cmplen, name);
859 }
860 }
861 }
863 static void add_preferred_base(unsigned char *sha1)
864 {
865 struct pbase_tree *it;
866 void *data;
867 unsigned long size;
868 unsigned char tree_sha1[20];
870 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
871 if (!data)
872 return;
874 for (it = pbase_tree; it; it = it->next) {
875 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
876 free(data);
877 return;
878 }
879 }
881 it = xcalloc(1, sizeof(*it));
882 it->next = pbase_tree;
883 pbase_tree = it;
885 hashcpy(it->pcache.sha1, tree_sha1);
886 it->pcache.tree_data = data;
887 it->pcache.tree_size = size;
888 }
890 static void check_object(struct object_entry *entry)
891 {
892 char type[20];
894 if (entry->in_pack && !entry->preferred_base) {
895 unsigned char base[20];
896 unsigned long size;
897 struct object_entry *base_entry;
899 /* We want in_pack_type even if we do not reuse delta.
900 * There is no point not reusing non-delta representations.
901 */
902 check_reuse_pack_delta(entry->in_pack,
903 entry->in_pack_offset,
904 base, &size,
905 &entry->in_pack_type);
907 /* Check if it is delta, and the base is also an object
908 * we are going to pack. If so we will reuse the existing
909 * delta.
910 */
911 if (!no_reuse_delta &&
912 entry->in_pack_type == OBJ_DELTA &&
913 (base_entry = locate_object_entry(base)) &&
914 (!base_entry->preferred_base)) {
916 /* Depth value does not matter - find_deltas()
917 * will never consider reused delta as the
918 * base object to deltify other objects
919 * against, in order to avoid circular deltas.
920 */
922 /* uncompressed size of the delta data */
923 entry->size = entry->delta_size = size;
924 entry->delta = base_entry;
925 entry->type = OBJ_DELTA;
927 entry->delta_sibling = base_entry->delta_child;
928 base_entry->delta_child = entry;
930 return;
931 }
932 /* Otherwise we would do the usual */
933 }
935 if (sha1_object_info(entry->sha1, type, &entry->size))
936 die("unable to get type of object %s",
937 sha1_to_hex(entry->sha1));
939 if (!strcmp(type, commit_type)) {
940 entry->type = OBJ_COMMIT;
941 } else if (!strcmp(type, tree_type)) {
942 entry->type = OBJ_TREE;
943 } else if (!strcmp(type, blob_type)) {
944 entry->type = OBJ_BLOB;
945 } else if (!strcmp(type, tag_type)) {
946 entry->type = OBJ_TAG;
947 } else
948 die("unable to pack object %s of type %s",
949 sha1_to_hex(entry->sha1), type);
950 }
952 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
953 {
954 struct object_entry *child = me->delta_child;
955 unsigned int m = n;
956 while (child) {
957 unsigned int c = check_delta_limit(child, n + 1);
958 if (m < c)
959 m = c;
960 child = child->delta_sibling;
961 }
962 return m;
963 }
965 static void get_object_details(void)
966 {
967 int i;
968 struct object_entry *entry;
970 prepare_pack_ix();
971 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
972 check_object(entry);
974 if (nr_objects == nr_result) {
975 /*
976 * Depth of objects that depend on the entry -- this
977 * is subtracted from depth-max to break too deep
978 * delta chain because of delta data reusing.
979 * However, we loosen this restriction when we know we
980 * are creating a thin pack -- it will have to be
981 * expanded on the other end anyway, so do not
982 * artificially cut the delta chain and let it go as
983 * deep as it wants.
984 */
985 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
986 if (!entry->delta && entry->delta_child)
987 entry->delta_limit =
988 check_delta_limit(entry, 1);
989 }
990 }
992 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
994 static entry_sort_t current_sort;
996 static int sort_comparator(const void *_a, const void *_b)
997 {
998 struct object_entry *a = *(struct object_entry **)_a;
999 struct object_entry *b = *(struct object_entry **)_b;
1000 return current_sort(a,b);
1001 }
1003 static struct object_entry **create_sorted_list(entry_sort_t sort)
1004 {
1005 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1006 int i;
1008 for (i = 0; i < nr_objects; i++)
1009 list[i] = objects + i;
1010 current_sort = sort;
1011 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1012 return list;
1013 }
1015 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1016 {
1017 return hashcmp(a->sha1, b->sha1);
1018 }
1020 static struct object_entry **create_final_object_list(void)
1021 {
1022 struct object_entry **list;
1023 int i, j;
1025 for (i = nr_result = 0; i < nr_objects; i++)
1026 if (!objects[i].preferred_base)
1027 nr_result++;
1028 list = xmalloc(nr_result * sizeof(struct object_entry *));
1029 for (i = j = 0; i < nr_objects; i++) {
1030 if (!objects[i].preferred_base)
1031 list[j++] = objects + i;
1032 }
1033 current_sort = sha1_sort;
1034 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1035 return list;
1036 }
1038 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1039 {
1040 if (a->type < b->type)
1041 return -1;
1042 if (a->type > b->type)
1043 return 1;
1044 if (a->hash < b->hash)
1045 return -1;
1046 if (a->hash > b->hash)
1047 return 1;
1048 if (a->preferred_base < b->preferred_base)
1049 return -1;
1050 if (a->preferred_base > b->preferred_base)
1051 return 1;
1052 if (a->size < b->size)
1053 return -1;
1054 if (a->size > b->size)
1055 return 1;
1056 return a < b ? -1 : (a > b);
1057 }
1059 struct unpacked {
1060 struct object_entry *entry;
1061 void *data;
1062 struct delta_index *index;
1063 };
1065 /*
1066 * We search for deltas _backwards_ in a list sorted by type and
1067 * by size, so that we see progressively smaller and smaller files.
1068 * That's because we prefer deltas to be from the bigger file
1069 * to the smaller - deletes are potentially cheaper, but perhaps
1070 * more importantly, the bigger file is likely the more recent
1071 * one.
1072 */
1073 static int try_delta(struct unpacked *trg, struct unpacked *src,
1074 unsigned max_depth)
1075 {
1076 struct object_entry *trg_entry = trg->entry;
1077 struct object_entry *src_entry = src->entry;
1078 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1079 char type[10];
1080 void *delta_buf;
1082 /* Don't bother doing diffs between different types */
1083 if (trg_entry->type != src_entry->type)
1084 return -1;
1086 /* We do not compute delta to *create* objects we are not
1087 * going to pack.
1088 */
1089 if (trg_entry->preferred_base)
1090 return -1;
1092 /*
1093 * We do not bother to try a delta that we discarded
1094 * on an earlier try, but only when reusing delta data.
1095 */
1096 if (!no_reuse_delta && trg_entry->in_pack &&
1097 trg_entry->in_pack == src_entry->in_pack)
1098 return 0;
1100 /*
1101 * If the current object is at pack edge, take the depth the
1102 * objects that depend on the current object into account --
1103 * otherwise they would become too deep.
1104 */
1105 if (trg_entry->delta_child) {
1106 if (max_depth <= trg_entry->delta_limit)
1107 return 0;
1108 max_depth -= trg_entry->delta_limit;
1109 }
1110 if (src_entry->depth >= max_depth)
1111 return 0;
1113 /* Now some size filtering heuristics. */
1114 trg_size = trg_entry->size;
1115 max_size = trg_size/2 - 20;
1116 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1117 if (max_size == 0)
1118 return 0;
1119 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1120 max_size = trg_entry->delta_size-1;
1121 src_size = src_entry->size;
1122 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1123 if (sizediff >= max_size)
1124 return 0;
1126 /* Load data if not already done */
1127 if (!trg->data) {
1128 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1129 if (sz != trg_size)
1130 die("object %s inconsistent object length (%lu vs %lu)",
1131 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1132 }
1133 if (!src->data) {
1134 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1135 if (sz != src_size)
1136 die("object %s inconsistent object length (%lu vs %lu)",
1137 sha1_to_hex(src_entry->sha1), sz, src_size);
1138 }
1139 if (!src->index) {
1140 src->index = create_delta_index(src->data, src_size);
1141 if (!src->index)
1142 die("out of memory");
1143 }
1145 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1146 if (!delta_buf)
1147 return 0;
1149 trg_entry->delta = src_entry;
1150 trg_entry->delta_size = delta_size;
1151 trg_entry->depth = src_entry->depth + 1;
1152 free(delta_buf);
1153 return 1;
1154 }
1156 static void progress_interval(int signum)
1157 {
1158 progress_update = 1;
1159 }
1161 static void find_deltas(struct object_entry **list, int window, int depth)
1162 {
1163 int i, idx;
1164 unsigned int array_size = window * sizeof(struct unpacked);
1165 struct unpacked *array = xmalloc(array_size);
1166 unsigned processed = 0;
1167 unsigned last_percent = 999;
1169 memset(array, 0, array_size);
1170 i = nr_objects;
1171 idx = 0;
1172 if (progress)
1173 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1175 while (--i >= 0) {
1176 struct object_entry *entry = list[i];
1177 struct unpacked *n = array + idx;
1178 int j;
1180 if (!entry->preferred_base)
1181 processed++;
1183 if (progress) {
1184 unsigned percent = processed * 100 / nr_result;
1185 if (percent != last_percent || progress_update) {
1186 fprintf(stderr, "%4u%% (%u/%u) done\r",
1187 percent, processed, nr_result);
1188 progress_update = 0;
1189 last_percent = percent;
1190 }
1191 }
1193 if (entry->delta)
1194 /* This happens if we decided to reuse existing
1195 * delta from a pack. "!no_reuse_delta &&" is implied.
1196 */
1197 continue;
1199 if (entry->size < 50)
1200 continue;
1201 free_delta_index(n->index);
1202 n->index = NULL;
1203 free(n->data);
1204 n->data = NULL;
1205 n->entry = entry;
1207 j = window;
1208 while (--j > 0) {
1209 unsigned int other_idx = idx + j;
1210 struct unpacked *m;
1211 if (other_idx >= window)
1212 other_idx -= window;
1213 m = array + other_idx;
1214 if (!m->entry)
1215 break;
1216 if (try_delta(n, m, depth) < 0)
1217 break;
1218 }
1219 /* if we made n a delta, and if n is already at max
1220 * depth, leaving it in the window is pointless. we
1221 * should evict it first.
1222 */
1223 if (entry->delta && depth <= entry->depth)
1224 continue;
1226 idx++;
1227 if (idx >= window)
1228 idx = 0;
1229 }
1231 if (progress)
1232 fputc('\n', stderr);
1234 for (i = 0; i < window; ++i) {
1235 free_delta_index(array[i].index);
1236 free(array[i].data);
1237 }
1238 free(array);
1239 }
1241 static void prepare_pack(int window, int depth)
1242 {
1243 get_object_details();
1244 sorted_by_type = create_sorted_list(type_size_sort);
1245 if (window && depth)
1246 find_deltas(sorted_by_type, window+1, depth);
1247 }
1249 static int reuse_cached_pack(unsigned char *sha1)
1250 {
1251 static const char cache[] = "pack-cache/pack-%s.%s";
1252 char *cached_pack, *cached_idx;
1253 int ifd, ofd, ifd_ix = -1;
1255 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1256 ifd = open(cached_pack, O_RDONLY);
1257 if (ifd < 0)
1258 return 0;
1260 if (!pack_to_stdout) {
1261 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1262 ifd_ix = open(cached_idx, O_RDONLY);
1263 if (ifd_ix < 0) {
1264 close(ifd);
1265 return 0;
1266 }
1267 }
1269 if (progress)
1270 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1271 sha1_to_hex(sha1));
1273 if (pack_to_stdout) {
1274 if (copy_fd(ifd, 1))
1275 exit(1);
1276 close(ifd);
1277 }
1278 else {
1279 char name[PATH_MAX];
1280 snprintf(name, sizeof(name),
1281 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1282 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1283 if (ofd < 0)
1284 die("unable to open %s (%s)", name, strerror(errno));
1285 if (copy_fd(ifd, ofd))
1286 exit(1);
1287 close(ifd);
1289 snprintf(name, sizeof(name),
1290 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1291 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1292 if (ofd < 0)
1293 die("unable to open %s (%s)", name, strerror(errno));
1294 if (copy_fd(ifd_ix, ofd))
1295 exit(1);
1296 close(ifd_ix);
1297 puts(sha1_to_hex(sha1));
1298 }
1300 return 1;
1301 }
1303 static void setup_progress_signal(void)
1304 {
1305 struct sigaction sa;
1306 struct itimerval v;
1308 memset(&sa, 0, sizeof(sa));
1309 sa.sa_handler = progress_interval;
1310 sigemptyset(&sa.sa_mask);
1311 sa.sa_flags = SA_RESTART;
1312 sigaction(SIGALRM, &sa, NULL);
1314 v.it_interval.tv_sec = 1;
1315 v.it_interval.tv_usec = 0;
1316 v.it_value = v.it_interval;
1317 setitimer(ITIMER_REAL, &v, NULL);
1318 }
1320 static int git_pack_config(const char *k, const char *v)
1321 {
1322 if(!strcmp(k, "pack.window")) {
1323 window = git_config_int(k, v);
1324 return 0;
1325 }
1326 return git_default_config(k, v);
1327 }
1329 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1330 {
1331 SHA_CTX ctx;
1332 char line[40 + 1 + PATH_MAX + 2];
1333 int depth = 10;
1334 struct object_entry **list;
1335 int num_preferred_base = 0;
1336 int i;
1338 git_config(git_pack_config);
1340 progress = isatty(2);
1341 for (i = 1; i < argc; i++) {
1342 const char *arg = argv[i];
1344 if (*arg == '-') {
1345 if (!strcmp("--non-empty", arg)) {
1346 non_empty = 1;
1347 continue;
1348 }
1349 if (!strcmp("--local", arg)) {
1350 local = 1;
1351 continue;
1352 }
1353 if (!strcmp("--progress", arg)) {
1354 progress = 1;
1355 continue;
1356 }
1357 if (!strcmp("--incremental", arg)) {
1358 incremental = 1;
1359 continue;
1360 }
1361 if (!strncmp("--window=", arg, 9)) {
1362 char *end;
1363 window = strtoul(arg+9, &end, 0);
1364 if (!arg[9] || *end)
1365 usage(pack_usage);
1366 continue;
1367 }
1368 if (!strncmp("--depth=", arg, 8)) {
1369 char *end;
1370 depth = strtoul(arg+8, &end, 0);
1371 if (!arg[8] || *end)
1372 usage(pack_usage);
1373 continue;
1374 }
1375 if (!strcmp("--progress", arg)) {
1376 progress = 1;
1377 continue;
1378 }
1379 if (!strcmp("-q", arg)) {
1380 progress = 0;
1381 continue;
1382 }
1383 if (!strcmp("--no-reuse-delta", arg)) {
1384 no_reuse_delta = 1;
1385 continue;
1386 }
1387 if (!strcmp("--stdout", arg)) {
1388 pack_to_stdout = 1;
1389 continue;
1390 }
1391 usage(pack_usage);
1392 }
1393 if (base_name)
1394 usage(pack_usage);
1395 base_name = arg;
1396 }
1398 if (pack_to_stdout != !base_name)
1399 usage(pack_usage);
1401 prepare_packed_git();
1403 if (progress) {
1404 fprintf(stderr, "Generating pack...\n");
1405 setup_progress_signal();
1406 }
1408 for (;;) {
1409 unsigned char sha1[20];
1410 unsigned hash;
1412 if (!fgets(line, sizeof(line), stdin)) {
1413 if (feof(stdin))
1414 break;
1415 if (!ferror(stdin))
1416 die("fgets returned NULL, not EOF, not error!");
1417 if (errno != EINTR)
1418 die("fgets: %s", strerror(errno));
1419 clearerr(stdin);
1420 continue;
1421 }
1423 if (line[0] == '-') {
1424 if (get_sha1_hex(line+1, sha1))
1425 die("expected edge sha1, got garbage:\n %s",
1426 line+1);
1427 if (num_preferred_base++ < window)
1428 add_preferred_base(sha1);
1429 continue;
1430 }
1431 if (get_sha1_hex(line, sha1))
1432 die("expected sha1, got garbage:\n %s", line);
1433 hash = name_hash(line+41);
1434 add_preferred_base_object(line+41, hash);
1435 add_object_entry(sha1, hash, 0);
1436 }
1437 if (progress)
1438 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1439 sorted_by_sha = create_final_object_list();
1440 if (non_empty && !nr_result)
1441 return 0;
1443 SHA1_Init(&ctx);
1444 list = sorted_by_sha;
1445 for (i = 0; i < nr_result; i++) {
1446 struct object_entry *entry = *list++;
1447 SHA1_Update(&ctx, entry->sha1, 20);
1448 }
1449 SHA1_Final(object_list_sha1, &ctx);
1450 if (progress && (nr_objects != nr_result))
1451 fprintf(stderr, "Result has %d objects.\n", nr_result);
1453 if (reuse_cached_pack(object_list_sha1))
1454 ;
1455 else {
1456 if (nr_result)
1457 prepare_pack(window, depth);
1458 if (progress && pack_to_stdout) {
1459 /* the other end usually displays progress itself */
1460 struct itimerval v = {{0,},};
1461 setitimer(ITIMER_REAL, &v, NULL);
1462 signal(SIGALRM, SIG_IGN );
1463 progress_update = 0;
1464 }
1465 write_pack_file();
1466 if (!pack_to_stdout) {
1467 write_index_file();
1468 puts(sha1_to_hex(object_list_sha1));
1469 }
1470 }
1471 if (progress)
1472 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1473 nr_result, written, written_delta, reused, reused_delta);
1474 return 0;
1475 }