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 revalidate_one(struct object_entry *entry,
247 void *data, char *type, unsigned long size)
248 {
249 int err;
250 if ((!data) ||
251 ((entry->type != OBJ_DELTA) &&
252 ( (size != entry->size) ||
253 strcmp(type_names[entry->type], type))))
254 err = -1;
255 else
256 err = check_sha1_signature(entry->sha1, data, size, type);
257 free(data);
258 return err;
259 }
261 /*
262 * we are going to reuse the existing pack entry data. make
263 * sure it is not corrupt.
264 */
265 static int revalidate_pack_entry(struct object_entry *entry)
266 {
267 void *data;
268 char type[20];
269 unsigned long size;
270 struct pack_entry e;
272 if (pack_to_stdout)
273 return 0;
275 e.p = entry->in_pack;
276 e.offset = entry->in_pack_offset;
278 /* the caller has already called use_packed_git() for us */
279 data = unpack_entry_gently(&e, type, &size);
280 return revalidate_one(entry, data, type, size);
281 }
283 static int revalidate_loose_object(struct object_entry *entry,
284 unsigned char *map,
285 unsigned long mapsize)
286 {
287 /* we already know this is a loose object with new type header. */
288 void *data;
289 char type[20];
290 unsigned long size;
292 if (pack_to_stdout)
293 return 0;
295 data = unpack_sha1_file(map, mapsize, type, &size);
296 return revalidate_one(entry, data, type, size);
297 }
299 static unsigned long write_object(struct sha1file *f,
300 struct object_entry *entry)
301 {
302 unsigned long size;
303 char type[10];
304 void *buf;
305 unsigned char header[10];
306 unsigned hdrlen, datalen;
307 enum object_type obj_type;
308 int to_reuse = 0;
310 if (entry->preferred_base)
311 return 0;
313 obj_type = entry->type;
314 if (! entry->in_pack)
315 to_reuse = 0; /* can't reuse what we don't have */
316 else if (obj_type == OBJ_DELTA)
317 to_reuse = 1; /* check_object() decided it for us */
318 else if (obj_type != entry->in_pack_type)
319 to_reuse = 0; /* pack has delta which is unusable */
320 else if (entry->delta)
321 to_reuse = 0; /* we want to pack afresh */
322 else
323 to_reuse = 1; /* we have it in-pack undeltified,
324 * and we do not need to deltify it.
325 */
327 if (!entry->in_pack && !entry->delta) {
328 unsigned char *map;
329 unsigned long mapsize;
330 map = map_sha1_file(entry->sha1, &mapsize);
331 if (map && !legacy_loose_object(map)) {
332 /* We can copy straight into the pack file */
333 if (revalidate_loose_object(entry, map, mapsize))
334 die("corrupt loose object %s",
335 sha1_to_hex(entry->sha1));
336 sha1write(f, map, mapsize);
337 munmap(map, mapsize);
338 written++;
339 reused++;
340 return mapsize;
341 }
342 if (map)
343 munmap(map, mapsize);
344 }
346 if (!to_reuse) {
347 buf = read_sha1_file(entry->sha1, type, &size);
348 if (!buf)
349 die("unable to read %s", sha1_to_hex(entry->sha1));
350 if (size != entry->size)
351 die("object %s size inconsistency (%lu vs %lu)",
352 sha1_to_hex(entry->sha1), size, entry->size);
353 if (entry->delta) {
354 buf = delta_against(buf, size, entry);
355 size = entry->delta_size;
356 obj_type = OBJ_DELTA;
357 }
358 /*
359 * The object header is a byte of 'type' followed by zero or
360 * more bytes of length. For deltas, the 20 bytes of delta
361 * sha1 follows that.
362 */
363 hdrlen = encode_header(obj_type, size, header);
364 sha1write(f, header, hdrlen);
366 if (entry->delta) {
367 sha1write(f, entry->delta, 20);
368 hdrlen += 20;
369 }
370 datalen = sha1write_compressed(f, buf, size);
371 free(buf);
372 }
373 else {
374 struct packed_git *p = entry->in_pack;
375 use_packed_git(p);
377 datalen = find_packed_object_size(p, entry->in_pack_offset);
378 buf = (char *) p->pack_base + entry->in_pack_offset;
380 if (revalidate_pack_entry(entry))
381 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
382 sha1write(f, buf, datalen);
383 unuse_packed_git(p);
384 hdrlen = 0; /* not really */
385 if (obj_type == OBJ_DELTA)
386 reused_delta++;
387 reused++;
388 }
389 if (obj_type == OBJ_DELTA)
390 written_delta++;
391 written++;
392 return hdrlen + datalen;
393 }
395 static unsigned long write_one(struct sha1file *f,
396 struct object_entry *e,
397 unsigned long offset)
398 {
399 if (e->offset)
400 /* offset starts from header size and cannot be zero
401 * if it is written already.
402 */
403 return offset;
404 e->offset = offset;
405 offset += write_object(f, e);
406 /* if we are deltified, write out its base object. */
407 if (e->delta)
408 offset = write_one(f, e->delta, offset);
409 return offset;
410 }
412 static void write_pack_file(void)
413 {
414 int i;
415 struct sha1file *f;
416 unsigned long offset;
417 struct pack_header hdr;
418 unsigned last_percent = 999;
419 int do_progress = 0;
421 if (!base_name)
422 f = sha1fd(1, "<stdout>");
423 else {
424 f = sha1create("%s-%s.%s", base_name,
425 sha1_to_hex(object_list_sha1), "pack");
426 do_progress = progress;
427 }
428 if (do_progress)
429 fprintf(stderr, "Writing %d objects.\n", nr_result);
431 hdr.hdr_signature = htonl(PACK_SIGNATURE);
432 hdr.hdr_version = htonl(PACK_VERSION);
433 hdr.hdr_entries = htonl(nr_result);
434 sha1write(f, &hdr, sizeof(hdr));
435 offset = sizeof(hdr);
436 if (!nr_result)
437 goto done;
438 for (i = 0; i < nr_objects; i++) {
439 offset = write_one(f, objects + i, offset);
440 if (do_progress) {
441 unsigned percent = written * 100 / nr_result;
442 if (progress_update || percent != last_percent) {
443 fprintf(stderr, "%4u%% (%u/%u) done\r",
444 percent, written, nr_result);
445 progress_update = 0;
446 last_percent = percent;
447 }
448 }
449 }
450 if (do_progress)
451 fputc('\n', stderr);
452 done:
453 sha1close(f, pack_file_sha1, 1);
454 }
456 static void write_index_file(void)
457 {
458 int i;
459 struct sha1file *f = sha1create("%s-%s.%s", base_name,
460 sha1_to_hex(object_list_sha1), "idx");
461 struct object_entry **list = sorted_by_sha;
462 struct object_entry **last = list + nr_result;
463 unsigned int array[256];
465 /*
466 * Write the first-level table (the list is sorted,
467 * but we use a 256-entry lookup to be able to avoid
468 * having to do eight extra binary search iterations).
469 */
470 for (i = 0; i < 256; i++) {
471 struct object_entry **next = list;
472 while (next < last) {
473 struct object_entry *entry = *next;
474 if (entry->sha1[0] != i)
475 break;
476 next++;
477 }
478 array[i] = htonl(next - sorted_by_sha);
479 list = next;
480 }
481 sha1write(f, array, 256 * sizeof(int));
483 /*
484 * Write the actual SHA1 entries..
485 */
486 list = sorted_by_sha;
487 for (i = 0; i < nr_result; i++) {
488 struct object_entry *entry = *list++;
489 unsigned int offset = htonl(entry->offset);
490 sha1write(f, &offset, 4);
491 sha1write(f, entry->sha1, 20);
492 }
493 sha1write(f, pack_file_sha1, 20);
494 sha1close(f, NULL, 1);
495 }
497 static int locate_object_entry_hash(const unsigned char *sha1)
498 {
499 int i;
500 unsigned int ui;
501 memcpy(&ui, sha1, sizeof(unsigned int));
502 i = ui % object_ix_hashsz;
503 while (0 < object_ix[i]) {
504 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
505 return i;
506 if (++i == object_ix_hashsz)
507 i = 0;
508 }
509 return -1 - i;
510 }
512 static struct object_entry *locate_object_entry(const unsigned char *sha1)
513 {
514 int i;
516 if (!object_ix_hashsz)
517 return NULL;
519 i = locate_object_entry_hash(sha1);
520 if (0 <= i)
521 return &objects[object_ix[i]-1];
522 return NULL;
523 }
525 static void rehash_objects(void)
526 {
527 int i;
528 struct object_entry *oe;
530 object_ix_hashsz = nr_objects * 3;
531 if (object_ix_hashsz < 1024)
532 object_ix_hashsz = 1024;
533 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
534 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
535 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
536 int ix = locate_object_entry_hash(oe->sha1);
537 if (0 <= ix)
538 continue;
539 ix = -1 - ix;
540 object_ix[ix] = i + 1;
541 }
542 }
544 static unsigned name_hash(const char *name)
545 {
546 unsigned char c;
547 unsigned hash = 0;
549 /*
550 * This effectively just creates a sortable number from the
551 * last sixteen non-whitespace characters. Last characters
552 * count "most", so things that end in ".c" sort together.
553 */
554 while ((c = *name++) != 0) {
555 if (isspace(c))
556 continue;
557 hash = (hash >> 2) + (c << 24);
558 }
559 return hash;
560 }
562 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
563 {
564 unsigned int idx = nr_objects;
565 struct object_entry *entry;
566 struct packed_git *p;
567 unsigned int found_offset = 0;
568 struct packed_git *found_pack = NULL;
569 int ix, status = 0;
571 if (!exclude) {
572 for (p = packed_git; p; p = p->next) {
573 struct pack_entry e;
574 if (find_pack_entry_one(sha1, &e, p)) {
575 if (incremental)
576 return 0;
577 if (local && !p->pack_local)
578 return 0;
579 if (!found_pack) {
580 found_offset = e.offset;
581 found_pack = e.p;
582 }
583 }
584 }
585 }
586 if ((entry = locate_object_entry(sha1)) != NULL)
587 goto already_added;
589 if (idx >= nr_alloc) {
590 unsigned int needed = (idx + 1024) * 3 / 2;
591 objects = xrealloc(objects, needed * sizeof(*entry));
592 nr_alloc = needed;
593 }
594 entry = objects + idx;
595 nr_objects = idx + 1;
596 memset(entry, 0, sizeof(*entry));
597 hashcpy(entry->sha1, sha1);
598 entry->hash = hash;
600 if (object_ix_hashsz * 3 <= nr_objects * 4)
601 rehash_objects();
602 else {
603 ix = locate_object_entry_hash(entry->sha1);
604 if (0 <= ix)
605 die("internal error in object hashing.");
606 object_ix[-1 - ix] = idx + 1;
607 }
608 status = 1;
610 already_added:
611 if (progress_update) {
612 fprintf(stderr, "Counting objects...%d\r", nr_objects);
613 progress_update = 0;
614 }
615 if (exclude)
616 entry->preferred_base = 1;
617 else {
618 if (found_pack) {
619 entry->in_pack = found_pack;
620 entry->in_pack_offset = found_offset;
621 }
622 }
623 return status;
624 }
626 struct pbase_tree_cache {
627 unsigned char sha1[20];
628 int ref;
629 int temporary;
630 void *tree_data;
631 unsigned long tree_size;
632 };
634 static struct pbase_tree_cache *(pbase_tree_cache[256]);
635 static int pbase_tree_cache_ix(const unsigned char *sha1)
636 {
637 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
638 }
639 static int pbase_tree_cache_ix_incr(int ix)
640 {
641 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
642 }
644 static struct pbase_tree {
645 struct pbase_tree *next;
646 /* This is a phony "cache" entry; we are not
647 * going to evict it nor find it through _get()
648 * mechanism -- this is for the toplevel node that
649 * would almost always change with any commit.
650 */
651 struct pbase_tree_cache pcache;
652 } *pbase_tree;
654 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
655 {
656 struct pbase_tree_cache *ent, *nent;
657 void *data;
658 unsigned long size;
659 char type[20];
660 int neigh;
661 int my_ix = pbase_tree_cache_ix(sha1);
662 int available_ix = -1;
664 /* pbase-tree-cache acts as a limited hashtable.
665 * your object will be found at your index or within a few
666 * slots after that slot if it is cached.
667 */
668 for (neigh = 0; neigh < 8; neigh++) {
669 ent = pbase_tree_cache[my_ix];
670 if (ent && !hashcmp(ent->sha1, sha1)) {
671 ent->ref++;
672 return ent;
673 }
674 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
675 ((0 <= available_ix) &&
676 (!ent && pbase_tree_cache[available_ix])))
677 available_ix = my_ix;
678 if (!ent)
679 break;
680 my_ix = pbase_tree_cache_ix_incr(my_ix);
681 }
683 /* Did not find one. Either we got a bogus request or
684 * we need to read and perhaps cache.
685 */
686 data = read_sha1_file(sha1, type, &size);
687 if (!data)
688 return NULL;
689 if (strcmp(type, tree_type)) {
690 free(data);
691 return NULL;
692 }
694 /* We need to either cache or return a throwaway copy */
696 if (available_ix < 0)
697 ent = NULL;
698 else {
699 ent = pbase_tree_cache[available_ix];
700 my_ix = available_ix;
701 }
703 if (!ent) {
704 nent = xmalloc(sizeof(*nent));
705 nent->temporary = (available_ix < 0);
706 }
707 else {
708 /* evict and reuse */
709 free(ent->tree_data);
710 nent = ent;
711 }
712 hashcpy(nent->sha1, sha1);
713 nent->tree_data = data;
714 nent->tree_size = size;
715 nent->ref = 1;
716 if (!nent->temporary)
717 pbase_tree_cache[my_ix] = nent;
718 return nent;
719 }
721 static void pbase_tree_put(struct pbase_tree_cache *cache)
722 {
723 if (!cache->temporary) {
724 cache->ref--;
725 return;
726 }
727 free(cache->tree_data);
728 free(cache);
729 }
731 static int name_cmp_len(const char *name)
732 {
733 int i;
734 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
735 ;
736 return i;
737 }
739 static void add_pbase_object(struct tree_desc *tree,
740 const char *name,
741 int cmplen,
742 const char *fullname)
743 {
744 struct name_entry entry;
746 while (tree_entry(tree,&entry)) {
747 unsigned long size;
748 char type[20];
750 if (entry.pathlen != cmplen ||
751 memcmp(entry.path, name, cmplen) ||
752 !has_sha1_file(entry.sha1) ||
753 sha1_object_info(entry.sha1, type, &size))
754 continue;
755 if (name[cmplen] != '/') {
756 unsigned hash = name_hash(fullname);
757 add_object_entry(entry.sha1, hash, 1);
758 return;
759 }
760 if (!strcmp(type, tree_type)) {
761 struct tree_desc sub;
762 struct pbase_tree_cache *tree;
763 const char *down = name+cmplen+1;
764 int downlen = name_cmp_len(down);
766 tree = pbase_tree_get(entry.sha1);
767 if (!tree)
768 return;
769 sub.buf = tree->tree_data;
770 sub.size = tree->tree_size;
772 add_pbase_object(&sub, down, downlen, fullname);
773 pbase_tree_put(tree);
774 }
775 }
776 }
778 static unsigned *done_pbase_paths;
779 static int done_pbase_paths_num;
780 static int done_pbase_paths_alloc;
781 static int done_pbase_path_pos(unsigned hash)
782 {
783 int lo = 0;
784 int hi = done_pbase_paths_num;
785 while (lo < hi) {
786 int mi = (hi + lo) / 2;
787 if (done_pbase_paths[mi] == hash)
788 return mi;
789 if (done_pbase_paths[mi] < hash)
790 hi = mi;
791 else
792 lo = mi + 1;
793 }
794 return -lo-1;
795 }
797 static int check_pbase_path(unsigned hash)
798 {
799 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
800 if (0 <= pos)
801 return 1;
802 pos = -pos - 1;
803 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
804 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
805 done_pbase_paths = xrealloc(done_pbase_paths,
806 done_pbase_paths_alloc *
807 sizeof(unsigned));
808 }
809 done_pbase_paths_num++;
810 if (pos < done_pbase_paths_num)
811 memmove(done_pbase_paths + pos + 1,
812 done_pbase_paths + pos,
813 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
814 done_pbase_paths[pos] = hash;
815 return 0;
816 }
818 static void add_preferred_base_object(char *name, unsigned hash)
819 {
820 struct pbase_tree *it;
821 int cmplen = name_cmp_len(name);
823 if (check_pbase_path(hash))
824 return;
826 for (it = pbase_tree; it; it = it->next) {
827 if (cmplen == 0) {
828 hash = name_hash("");
829 add_object_entry(it->pcache.sha1, hash, 1);
830 }
831 else {
832 struct tree_desc tree;
833 tree.buf = it->pcache.tree_data;
834 tree.size = it->pcache.tree_size;
835 add_pbase_object(&tree, name, cmplen, name);
836 }
837 }
838 }
840 static void add_preferred_base(unsigned char *sha1)
841 {
842 struct pbase_tree *it;
843 void *data;
844 unsigned long size;
845 unsigned char tree_sha1[20];
847 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
848 if (!data)
849 return;
851 for (it = pbase_tree; it; it = it->next) {
852 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
853 free(data);
854 return;
855 }
856 }
858 it = xcalloc(1, sizeof(*it));
859 it->next = pbase_tree;
860 pbase_tree = it;
862 hashcpy(it->pcache.sha1, tree_sha1);
863 it->pcache.tree_data = data;
864 it->pcache.tree_size = size;
865 }
867 static void check_object(struct object_entry *entry)
868 {
869 char type[20];
871 if (entry->in_pack && !entry->preferred_base) {
872 unsigned char base[20];
873 unsigned long size;
874 struct object_entry *base_entry;
876 /* We want in_pack_type even if we do not reuse delta.
877 * There is no point not reusing non-delta representations.
878 */
879 check_reuse_pack_delta(entry->in_pack,
880 entry->in_pack_offset,
881 base, &size,
882 &entry->in_pack_type);
884 /* Check if it is delta, and the base is also an object
885 * we are going to pack. If so we will reuse the existing
886 * delta.
887 */
888 if (!no_reuse_delta &&
889 entry->in_pack_type == OBJ_DELTA &&
890 (base_entry = locate_object_entry(base)) &&
891 (!base_entry->preferred_base)) {
893 /* Depth value does not matter - find_deltas()
894 * will never consider reused delta as the
895 * base object to deltify other objects
896 * against, in order to avoid circular deltas.
897 */
899 /* uncompressed size of the delta data */
900 entry->size = entry->delta_size = size;
901 entry->delta = base_entry;
902 entry->type = OBJ_DELTA;
904 entry->delta_sibling = base_entry->delta_child;
905 base_entry->delta_child = entry;
907 return;
908 }
909 /* Otherwise we would do the usual */
910 }
912 if (sha1_object_info(entry->sha1, type, &entry->size))
913 die("unable to get type of object %s",
914 sha1_to_hex(entry->sha1));
916 if (!strcmp(type, commit_type)) {
917 entry->type = OBJ_COMMIT;
918 } else if (!strcmp(type, tree_type)) {
919 entry->type = OBJ_TREE;
920 } else if (!strcmp(type, blob_type)) {
921 entry->type = OBJ_BLOB;
922 } else if (!strcmp(type, tag_type)) {
923 entry->type = OBJ_TAG;
924 } else
925 die("unable to pack object %s of type %s",
926 sha1_to_hex(entry->sha1), type);
927 }
929 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
930 {
931 struct object_entry *child = me->delta_child;
932 unsigned int m = n;
933 while (child) {
934 unsigned int c = check_delta_limit(child, n + 1);
935 if (m < c)
936 m = c;
937 child = child->delta_sibling;
938 }
939 return m;
940 }
942 static void get_object_details(void)
943 {
944 int i;
945 struct object_entry *entry;
947 prepare_pack_ix();
948 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
949 check_object(entry);
951 if (nr_objects == nr_result) {
952 /*
953 * Depth of objects that depend on the entry -- this
954 * is subtracted from depth-max to break too deep
955 * delta chain because of delta data reusing.
956 * However, we loosen this restriction when we know we
957 * are creating a thin pack -- it will have to be
958 * expanded on the other end anyway, so do not
959 * artificially cut the delta chain and let it go as
960 * deep as it wants.
961 */
962 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
963 if (!entry->delta && entry->delta_child)
964 entry->delta_limit =
965 check_delta_limit(entry, 1);
966 }
967 }
969 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
971 static entry_sort_t current_sort;
973 static int sort_comparator(const void *_a, const void *_b)
974 {
975 struct object_entry *a = *(struct object_entry **)_a;
976 struct object_entry *b = *(struct object_entry **)_b;
977 return current_sort(a,b);
978 }
980 static struct object_entry **create_sorted_list(entry_sort_t sort)
981 {
982 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
983 int i;
985 for (i = 0; i < nr_objects; i++)
986 list[i] = objects + i;
987 current_sort = sort;
988 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
989 return list;
990 }
992 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
993 {
994 return hashcmp(a->sha1, b->sha1);
995 }
997 static struct object_entry **create_final_object_list(void)
998 {
999 struct object_entry **list;
1000 int i, j;
1002 for (i = nr_result = 0; i < nr_objects; i++)
1003 if (!objects[i].preferred_base)
1004 nr_result++;
1005 list = xmalloc(nr_result * sizeof(struct object_entry *));
1006 for (i = j = 0; i < nr_objects; i++) {
1007 if (!objects[i].preferred_base)
1008 list[j++] = objects + i;
1009 }
1010 current_sort = sha1_sort;
1011 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1012 return list;
1013 }
1015 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1016 {
1017 if (a->type < b->type)
1018 return -1;
1019 if (a->type > b->type)
1020 return 1;
1021 if (a->hash < b->hash)
1022 return -1;
1023 if (a->hash > b->hash)
1024 return 1;
1025 if (a->preferred_base < b->preferred_base)
1026 return -1;
1027 if (a->preferred_base > b->preferred_base)
1028 return 1;
1029 if (a->size < b->size)
1030 return -1;
1031 if (a->size > b->size)
1032 return 1;
1033 return a < b ? -1 : (a > b);
1034 }
1036 struct unpacked {
1037 struct object_entry *entry;
1038 void *data;
1039 struct delta_index *index;
1040 };
1042 /*
1043 * We search for deltas _backwards_ in a list sorted by type and
1044 * by size, so that we see progressively smaller and smaller files.
1045 * That's because we prefer deltas to be from the bigger file
1046 * to the smaller - deletes are potentially cheaper, but perhaps
1047 * more importantly, the bigger file is likely the more recent
1048 * one.
1049 */
1050 static int try_delta(struct unpacked *trg, struct unpacked *src,
1051 unsigned max_depth)
1052 {
1053 struct object_entry *trg_entry = trg->entry;
1054 struct object_entry *src_entry = src->entry;
1055 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1056 char type[10];
1057 void *delta_buf;
1059 /* Don't bother doing diffs between different types */
1060 if (trg_entry->type != src_entry->type)
1061 return -1;
1063 /* We do not compute delta to *create* objects we are not
1064 * going to pack.
1065 */
1066 if (trg_entry->preferred_base)
1067 return -1;
1069 /*
1070 * We do not bother to try a delta that we discarded
1071 * on an earlier try, but only when reusing delta data.
1072 */
1073 if (!no_reuse_delta && trg_entry->in_pack &&
1074 trg_entry->in_pack == src_entry->in_pack)
1075 return 0;
1077 /*
1078 * If the current object is at pack edge, take the depth the
1079 * objects that depend on the current object into account --
1080 * otherwise they would become too deep.
1081 */
1082 if (trg_entry->delta_child) {
1083 if (max_depth <= trg_entry->delta_limit)
1084 return 0;
1085 max_depth -= trg_entry->delta_limit;
1086 }
1087 if (src_entry->depth >= max_depth)
1088 return 0;
1090 /* Now some size filtering heuristics. */
1091 trg_size = trg_entry->size;
1092 max_size = trg_size/2 - 20;
1093 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1094 if (max_size == 0)
1095 return 0;
1096 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1097 max_size = trg_entry->delta_size-1;
1098 src_size = src_entry->size;
1099 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1100 if (sizediff >= max_size)
1101 return 0;
1103 /* Load data if not already done */
1104 if (!trg->data) {
1105 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1106 if (sz != trg_size)
1107 die("object %s inconsistent object length (%lu vs %lu)",
1108 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1109 }
1110 if (!src->data) {
1111 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1112 if (sz != src_size)
1113 die("object %s inconsistent object length (%lu vs %lu)",
1114 sha1_to_hex(src_entry->sha1), sz, src_size);
1115 }
1116 if (!src->index) {
1117 src->index = create_delta_index(src->data, src_size);
1118 if (!src->index)
1119 die("out of memory");
1120 }
1122 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1123 if (!delta_buf)
1124 return 0;
1126 trg_entry->delta = src_entry;
1127 trg_entry->delta_size = delta_size;
1128 trg_entry->depth = src_entry->depth + 1;
1129 free(delta_buf);
1130 return 1;
1131 }
1133 static void progress_interval(int signum)
1134 {
1135 progress_update = 1;
1136 }
1138 static void find_deltas(struct object_entry **list, int window, int depth)
1139 {
1140 int i, idx;
1141 unsigned int array_size = window * sizeof(struct unpacked);
1142 struct unpacked *array = xmalloc(array_size);
1143 unsigned processed = 0;
1144 unsigned last_percent = 999;
1146 memset(array, 0, array_size);
1147 i = nr_objects;
1148 idx = 0;
1149 if (progress)
1150 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1152 while (--i >= 0) {
1153 struct object_entry *entry = list[i];
1154 struct unpacked *n = array + idx;
1155 int j;
1157 if (!entry->preferred_base)
1158 processed++;
1160 if (progress) {
1161 unsigned percent = processed * 100 / nr_result;
1162 if (percent != last_percent || progress_update) {
1163 fprintf(stderr, "%4u%% (%u/%u) done\r",
1164 percent, processed, nr_result);
1165 progress_update = 0;
1166 last_percent = percent;
1167 }
1168 }
1170 if (entry->delta)
1171 /* This happens if we decided to reuse existing
1172 * delta from a pack. "!no_reuse_delta &&" is implied.
1173 */
1174 continue;
1176 if (entry->size < 50)
1177 continue;
1178 free_delta_index(n->index);
1179 n->index = NULL;
1180 free(n->data);
1181 n->data = NULL;
1182 n->entry = entry;
1184 j = window;
1185 while (--j > 0) {
1186 unsigned int other_idx = idx + j;
1187 struct unpacked *m;
1188 if (other_idx >= window)
1189 other_idx -= window;
1190 m = array + other_idx;
1191 if (!m->entry)
1192 break;
1193 if (try_delta(n, m, depth) < 0)
1194 break;
1195 }
1196 /* if we made n a delta, and if n is already at max
1197 * depth, leaving it in the window is pointless. we
1198 * should evict it first.
1199 */
1200 if (entry->delta && depth <= entry->depth)
1201 continue;
1203 idx++;
1204 if (idx >= window)
1205 idx = 0;
1206 }
1208 if (progress)
1209 fputc('\n', stderr);
1211 for (i = 0; i < window; ++i) {
1212 free_delta_index(array[i].index);
1213 free(array[i].data);
1214 }
1215 free(array);
1216 }
1218 static void prepare_pack(int window, int depth)
1219 {
1220 get_object_details();
1221 sorted_by_type = create_sorted_list(type_size_sort);
1222 if (window && depth)
1223 find_deltas(sorted_by_type, window+1, depth);
1224 }
1226 static int reuse_cached_pack(unsigned char *sha1)
1227 {
1228 static const char cache[] = "pack-cache/pack-%s.%s";
1229 char *cached_pack, *cached_idx;
1230 int ifd, ofd, ifd_ix = -1;
1232 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1233 ifd = open(cached_pack, O_RDONLY);
1234 if (ifd < 0)
1235 return 0;
1237 if (!pack_to_stdout) {
1238 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1239 ifd_ix = open(cached_idx, O_RDONLY);
1240 if (ifd_ix < 0) {
1241 close(ifd);
1242 return 0;
1243 }
1244 }
1246 if (progress)
1247 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1248 sha1_to_hex(sha1));
1250 if (pack_to_stdout) {
1251 if (copy_fd(ifd, 1))
1252 exit(1);
1253 close(ifd);
1254 }
1255 else {
1256 char name[PATH_MAX];
1257 snprintf(name, sizeof(name),
1258 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1259 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1260 if (ofd < 0)
1261 die("unable to open %s (%s)", name, strerror(errno));
1262 if (copy_fd(ifd, ofd))
1263 exit(1);
1264 close(ifd);
1266 snprintf(name, sizeof(name),
1267 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1268 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1269 if (ofd < 0)
1270 die("unable to open %s (%s)", name, strerror(errno));
1271 if (copy_fd(ifd_ix, ofd))
1272 exit(1);
1273 close(ifd_ix);
1274 puts(sha1_to_hex(sha1));
1275 }
1277 return 1;
1278 }
1280 static void setup_progress_signal(void)
1281 {
1282 struct sigaction sa;
1283 struct itimerval v;
1285 memset(&sa, 0, sizeof(sa));
1286 sa.sa_handler = progress_interval;
1287 sigemptyset(&sa.sa_mask);
1288 sa.sa_flags = SA_RESTART;
1289 sigaction(SIGALRM, &sa, NULL);
1291 v.it_interval.tv_sec = 1;
1292 v.it_interval.tv_usec = 0;
1293 v.it_value = v.it_interval;
1294 setitimer(ITIMER_REAL, &v, NULL);
1295 }
1297 static int git_pack_config(const char *k, const char *v)
1298 {
1299 if(!strcmp(k, "pack.window")) {
1300 window = git_config_int(k, v);
1301 return 0;
1302 }
1303 return git_default_config(k, v);
1304 }
1306 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1307 {
1308 SHA_CTX ctx;
1309 char line[40 + 1 + PATH_MAX + 2];
1310 int depth = 10;
1311 struct object_entry **list;
1312 int num_preferred_base = 0;
1313 int i;
1315 git_config(git_pack_config);
1317 progress = isatty(2);
1318 for (i = 1; i < argc; i++) {
1319 const char *arg = argv[i];
1321 if (*arg == '-') {
1322 if (!strcmp("--non-empty", arg)) {
1323 non_empty = 1;
1324 continue;
1325 }
1326 if (!strcmp("--local", arg)) {
1327 local = 1;
1328 continue;
1329 }
1330 if (!strcmp("--progress", arg)) {
1331 progress = 1;
1332 continue;
1333 }
1334 if (!strcmp("--incremental", arg)) {
1335 incremental = 1;
1336 continue;
1337 }
1338 if (!strncmp("--window=", arg, 9)) {
1339 char *end;
1340 window = strtoul(arg+9, &end, 0);
1341 if (!arg[9] || *end)
1342 usage(pack_usage);
1343 continue;
1344 }
1345 if (!strncmp("--depth=", arg, 8)) {
1346 char *end;
1347 depth = strtoul(arg+8, &end, 0);
1348 if (!arg[8] || *end)
1349 usage(pack_usage);
1350 continue;
1351 }
1352 if (!strcmp("--progress", arg)) {
1353 progress = 1;
1354 continue;
1355 }
1356 if (!strcmp("-q", arg)) {
1357 progress = 0;
1358 continue;
1359 }
1360 if (!strcmp("--no-reuse-delta", arg)) {
1361 no_reuse_delta = 1;
1362 continue;
1363 }
1364 if (!strcmp("--stdout", arg)) {
1365 pack_to_stdout = 1;
1366 continue;
1367 }
1368 usage(pack_usage);
1369 }
1370 if (base_name)
1371 usage(pack_usage);
1372 base_name = arg;
1373 }
1375 if (pack_to_stdout != !base_name)
1376 usage(pack_usage);
1378 prepare_packed_git();
1380 if (progress) {
1381 fprintf(stderr, "Generating pack...\n");
1382 setup_progress_signal();
1383 }
1385 for (;;) {
1386 unsigned char sha1[20];
1387 unsigned hash;
1389 if (!fgets(line, sizeof(line), stdin)) {
1390 if (feof(stdin))
1391 break;
1392 if (!ferror(stdin))
1393 die("fgets returned NULL, not EOF, not error!");
1394 if (errno != EINTR)
1395 die("fgets: %s", strerror(errno));
1396 clearerr(stdin);
1397 continue;
1398 }
1400 if (line[0] == '-') {
1401 if (get_sha1_hex(line+1, sha1))
1402 die("expected edge sha1, got garbage:\n %s",
1403 line+1);
1404 if (num_preferred_base++ < window)
1405 add_preferred_base(sha1);
1406 continue;
1407 }
1408 if (get_sha1_hex(line, sha1))
1409 die("expected sha1, got garbage:\n %s", line);
1410 hash = name_hash(line+41);
1411 add_preferred_base_object(line+41, hash);
1412 add_object_entry(sha1, hash, 0);
1413 }
1414 if (progress)
1415 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1416 sorted_by_sha = create_final_object_list();
1417 if (non_empty && !nr_result)
1418 return 0;
1420 SHA1_Init(&ctx);
1421 list = sorted_by_sha;
1422 for (i = 0; i < nr_result; i++) {
1423 struct object_entry *entry = *list++;
1424 SHA1_Update(&ctx, entry->sha1, 20);
1425 }
1426 SHA1_Final(object_list_sha1, &ctx);
1427 if (progress && (nr_objects != nr_result))
1428 fprintf(stderr, "Result has %d objects.\n", nr_result);
1430 if (reuse_cached_pack(object_list_sha1))
1431 ;
1432 else {
1433 if (nr_result)
1434 prepare_pack(window, depth);
1435 if (progress && pack_to_stdout) {
1436 /* the other end usually displays progress itself */
1437 struct itimerval v = {{0,},};
1438 setitimer(ITIMER_REAL, &v, NULL);
1439 signal(SIGALRM, SIG_IGN );
1440 progress_update = 0;
1441 }
1442 write_pack_file();
1443 if (!pack_to_stdout) {
1444 write_index_file();
1445 puts(sha1_to_hex(object_list_sha1));
1446 }
1447 }
1448 if (progress)
1449 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1450 nr_result, written, written_delta, reused, reused_delta);
1451 return 0;
1452 }