![[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 = 0;
57 static int no_reuse_delta = 0;
58 static int local = 0;
59 static int incremental = 0;
60 static struct object_entry **sorted_by_sha, **sorted_by_type;
61 static struct object_entry *objects = NULL;
62 static int nr_objects = 0, nr_alloc = 0, nr_result = 0;
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 = 0;
67 static int window = 10;
69 /*
70 * The object names in objects array are hashed with this hashtable,
71 * to help looking up the entry by object name. Binary search from
72 * sorted_by_sha is also possible but this was easier to code and faster.
73 * This hashtable is built after all the objects are seen.
74 */
75 static int *object_ix = NULL;
76 static int object_ix_hashsz = 0;
78 /*
79 * Pack index for existing packs give us easy access to the offsets into
80 * corresponding pack file where each object's data starts, but the entries
81 * do not store the size of the compressed representation (uncompressed
82 * size is easily available by examining the pack entry header). We build
83 * a hashtable of existing packs (pack_revindex), and keep reverse index
84 * here -- pack index file is sorted by object name mapping to offset; this
85 * pack_revindex[].revindex array is an ordered list of offsets, so if you
86 * know the offset of an object, next offset is where its packed
87 * representation ends.
88 */
89 struct pack_revindex {
90 struct packed_git *p;
91 unsigned long *revindex;
92 } *pack_revindex = NULL;
93 static int pack_revindex_hashsz = 0;
95 /*
96 * stats
97 */
98 static int written = 0;
99 static int written_delta = 0;
100 static int reused = 0;
101 static int reused_delta = 0;
103 static int pack_revindex_ix(struct packed_git *p)
104 {
105 unsigned long ui = (unsigned long)p;
106 int i;
108 ui = ui ^ (ui >> 16); /* defeat structure alignment */
109 i = (int)(ui % pack_revindex_hashsz);
110 while (pack_revindex[i].p) {
111 if (pack_revindex[i].p == p)
112 return i;
113 if (++i == pack_revindex_hashsz)
114 i = 0;
115 }
116 return -1 - i;
117 }
119 static void prepare_pack_ix(void)
120 {
121 int num;
122 struct packed_git *p;
123 for (num = 0, p = packed_git; p; p = p->next)
124 num++;
125 if (!num)
126 return;
127 pack_revindex_hashsz = num * 11;
128 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
129 for (p = packed_git; p; p = p->next) {
130 num = pack_revindex_ix(p);
131 num = - 1 - num;
132 pack_revindex[num].p = p;
133 }
134 /* revindex elements are lazily initialized */
135 }
137 static int cmp_offset(const void *a_, const void *b_)
138 {
139 unsigned long a = *(unsigned long *) a_;
140 unsigned long b = *(unsigned long *) b_;
141 if (a < b)
142 return -1;
143 else if (a == b)
144 return 0;
145 else
146 return 1;
147 }
149 /*
150 * Ordered list of offsets of objects in the pack.
151 */
152 static void prepare_pack_revindex(struct pack_revindex *rix)
153 {
154 struct packed_git *p = rix->p;
155 int num_ent = num_packed_objects(p);
156 int i;
157 void *index = p->index_base + 256;
159 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
160 for (i = 0; i < num_ent; i++) {
161 unsigned int hl = *((unsigned int *)((char *) index + 24*i));
162 rix->revindex[i] = ntohl(hl);
163 }
164 /* This knows the pack format -- the 20-byte trailer
165 * follows immediately after the last object data.
166 */
167 rix->revindex[num_ent] = p->pack_size - 20;
168 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
169 }
171 static unsigned long find_packed_object_size(struct packed_git *p,
172 unsigned long ofs)
173 {
174 int num;
175 int lo, hi;
176 struct pack_revindex *rix;
177 unsigned long *revindex;
178 num = pack_revindex_ix(p);
179 if (num < 0)
180 die("internal error: pack revindex uninitialized");
181 rix = &pack_revindex[num];
182 if (!rix->revindex)
183 prepare_pack_revindex(rix);
184 revindex = rix->revindex;
185 lo = 0;
186 hi = num_packed_objects(p) + 1;
187 do {
188 int mi = (lo + hi) / 2;
189 if (revindex[mi] == ofs) {
190 return revindex[mi+1] - ofs;
191 }
192 else if (ofs < revindex[mi])
193 hi = mi;
194 else
195 lo = mi + 1;
196 } while (lo < hi);
197 die("internal error: pack revindex corrupt");
198 }
200 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
201 {
202 unsigned long othersize, delta_size;
203 char type[10];
204 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
205 void *delta_buf;
207 if (!otherbuf)
208 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
209 delta_buf = diff_delta(otherbuf, othersize,
210 buf, size, &delta_size, 0);
211 if (!delta_buf || delta_size != entry->delta_size)
212 die("delta size changed");
213 free(buf);
214 free(otherbuf);
215 return delta_buf;
216 }
218 /*
219 * The per-object header is a pretty dense thing, which is
220 * - first byte: low four bits are "size", then three bits of "type",
221 * and the high bit is "size continues".
222 * - each byte afterwards: low seven bits are size continuation,
223 * with the high bit being "size continues"
224 */
225 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
226 {
227 int n = 1;
228 unsigned char c;
230 if (type < OBJ_COMMIT || type > OBJ_DELTA)
231 die("bad type %d", type);
233 c = (type << 4) | (size & 15);
234 size >>= 4;
235 while (size) {
236 *hdr++ = c | 0x80;
237 c = size & 0x7f;
238 size >>= 7;
239 n++;
240 }
241 *hdr = c;
242 return n;
243 }
245 static unsigned long write_object(struct sha1file *f,
246 struct object_entry *entry)
247 {
248 unsigned long size;
249 char type[10];
250 void *buf;
251 unsigned char header[10];
252 unsigned hdrlen, datalen;
253 enum object_type obj_type;
254 int to_reuse = 0;
256 if (entry->preferred_base)
257 return 0;
259 obj_type = entry->type;
260 if (! entry->in_pack)
261 to_reuse = 0; /* can't reuse what we don't have */
262 else if (obj_type == OBJ_DELTA)
263 to_reuse = 1; /* check_object() decided it for us */
264 else if (obj_type != entry->in_pack_type)
265 to_reuse = 0; /* pack has delta which is unusable */
266 else if (entry->delta)
267 to_reuse = 0; /* we want to pack afresh */
268 else
269 to_reuse = 1; /* we have it in-pack undeltified,
270 * and we do not need to deltify it.
271 */
273 if (! to_reuse) {
274 buf = read_sha1_file(entry->sha1, type, &size);
275 if (!buf)
276 die("unable to read %s", sha1_to_hex(entry->sha1));
277 if (size != entry->size)
278 die("object %s size inconsistency (%lu vs %lu)",
279 sha1_to_hex(entry->sha1), size, entry->size);
280 if (entry->delta) {
281 buf = delta_against(buf, size, entry);
282 size = entry->delta_size;
283 obj_type = OBJ_DELTA;
284 }
285 /*
286 * The object header is a byte of 'type' followed by zero or
287 * more bytes of length. For deltas, the 20 bytes of delta
288 * sha1 follows that.
289 */
290 hdrlen = encode_header(obj_type, size, header);
291 sha1write(f, header, hdrlen);
293 if (entry->delta) {
294 sha1write(f, entry->delta, 20);
295 hdrlen += 20;
296 }
297 datalen = sha1write_compressed(f, buf, size);
298 free(buf);
299 }
300 else {
301 struct packed_git *p = entry->in_pack;
302 use_packed_git(p);
304 datalen = find_packed_object_size(p, entry->in_pack_offset);
305 buf = (char *) p->pack_base + entry->in_pack_offset;
306 sha1write(f, buf, datalen);
307 unuse_packed_git(p);
308 hdrlen = 0; /* not really */
309 if (obj_type == OBJ_DELTA)
310 reused_delta++;
311 reused++;
312 }
313 if (obj_type == OBJ_DELTA)
314 written_delta++;
315 written++;
316 return hdrlen + datalen;
317 }
319 static unsigned long write_one(struct sha1file *f,
320 struct object_entry *e,
321 unsigned long offset)
322 {
323 if (e->offset)
324 /* offset starts from header size and cannot be zero
325 * if it is written already.
326 */
327 return offset;
328 e->offset = offset;
329 offset += write_object(f, e);
330 /* if we are deltified, write out its base object. */
331 if (e->delta)
332 offset = write_one(f, e->delta, offset);
333 return offset;
334 }
336 static void write_pack_file(void)
337 {
338 int i;
339 struct sha1file *f;
340 unsigned long offset;
341 struct pack_header hdr;
342 unsigned last_percent = 999;
343 int do_progress = 0;
345 if (!base_name)
346 f = sha1fd(1, "<stdout>");
347 else {
348 f = sha1create("%s-%s.%s", base_name,
349 sha1_to_hex(object_list_sha1), "pack");
350 do_progress = progress;
351 }
352 if (do_progress)
353 fprintf(stderr, "Writing %d objects.\n", nr_result);
355 hdr.hdr_signature = htonl(PACK_SIGNATURE);
356 hdr.hdr_version = htonl(PACK_VERSION);
357 hdr.hdr_entries = htonl(nr_result);
358 sha1write(f, &hdr, sizeof(hdr));
359 offset = sizeof(hdr);
360 if (!nr_result)
361 goto done;
362 for (i = 0; i < nr_objects; i++) {
363 offset = write_one(f, objects + i, offset);
364 if (do_progress) {
365 unsigned percent = written * 100 / nr_result;
366 if (progress_update || percent != last_percent) {
367 fprintf(stderr, "%4u%% (%u/%u) done\r",
368 percent, written, nr_result);
369 progress_update = 0;
370 last_percent = percent;
371 }
372 }
373 }
374 if (do_progress)
375 fputc('\n', stderr);
376 done:
377 sha1close(f, pack_file_sha1, 1);
378 }
380 static void write_index_file(void)
381 {
382 int i;
383 struct sha1file *f = sha1create("%s-%s.%s", base_name,
384 sha1_to_hex(object_list_sha1), "idx");
385 struct object_entry **list = sorted_by_sha;
386 struct object_entry **last = list + nr_result;
387 unsigned int array[256];
389 /*
390 * Write the first-level table (the list is sorted,
391 * but we use a 256-entry lookup to be able to avoid
392 * having to do eight extra binary search iterations).
393 */
394 for (i = 0; i < 256; i++) {
395 struct object_entry **next = list;
396 while (next < last) {
397 struct object_entry *entry = *next;
398 if (entry->sha1[0] != i)
399 break;
400 next++;
401 }
402 array[i] = htonl(next - sorted_by_sha);
403 list = next;
404 }
405 sha1write(f, array, 256 * sizeof(int));
407 /*
408 * Write the actual SHA1 entries..
409 */
410 list = sorted_by_sha;
411 for (i = 0; i < nr_result; i++) {
412 struct object_entry *entry = *list++;
413 unsigned int offset = htonl(entry->offset);
414 sha1write(f, &offset, 4);
415 sha1write(f, entry->sha1, 20);
416 }
417 sha1write(f, pack_file_sha1, 20);
418 sha1close(f, NULL, 1);
419 }
421 static int locate_object_entry_hash(const unsigned char *sha1)
422 {
423 int i;
424 unsigned int ui;
425 memcpy(&ui, sha1, sizeof(unsigned int));
426 i = ui % object_ix_hashsz;
427 while (0 < object_ix[i]) {
428 if (!memcmp(sha1, objects[object_ix[i]-1].sha1, 20))
429 return i;
430 if (++i == object_ix_hashsz)
431 i = 0;
432 }
433 return -1 - i;
434 }
436 static struct object_entry *locate_object_entry(const unsigned char *sha1)
437 {
438 int i;
440 if (!object_ix_hashsz)
441 return NULL;
443 i = locate_object_entry_hash(sha1);
444 if (0 <= i)
445 return &objects[object_ix[i]-1];
446 return NULL;
447 }
449 static void rehash_objects(void)
450 {
451 int i;
452 struct object_entry *oe;
454 object_ix_hashsz = nr_objects * 3;
455 if (object_ix_hashsz < 1024)
456 object_ix_hashsz = 1024;
457 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
458 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
459 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
460 int ix = locate_object_entry_hash(oe->sha1);
461 if (0 <= ix)
462 continue;
463 ix = -1 - ix;
464 object_ix[ix] = i + 1;
465 }
466 }
468 static unsigned name_hash(const char *name)
469 {
470 unsigned char c;
471 unsigned hash = 0;
473 /*
474 * This effectively just creates a sortable number from the
475 * last sixteen non-whitespace characters. Last characters
476 * count "most", so things that end in ".c" sort together.
477 */
478 while ((c = *name++) != 0) {
479 if (isspace(c))
480 continue;
481 hash = (hash >> 2) + (c << 24);
482 }
483 return hash;
484 }
486 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
487 {
488 unsigned int idx = nr_objects;
489 struct object_entry *entry;
490 struct packed_git *p;
491 unsigned int found_offset = 0;
492 struct packed_git *found_pack = NULL;
493 int ix, status = 0;
495 if (!exclude) {
496 for (p = packed_git; p; p = p->next) {
497 struct pack_entry e;
498 if (find_pack_entry_one(sha1, &e, p)) {
499 if (incremental)
500 return 0;
501 if (local && !p->pack_local)
502 return 0;
503 if (!found_pack) {
504 found_offset = e.offset;
505 found_pack = e.p;
506 }
507 }
508 }
509 }
510 if ((entry = locate_object_entry(sha1)) != NULL)
511 goto already_added;
513 if (idx >= nr_alloc) {
514 unsigned int needed = (idx + 1024) * 3 / 2;
515 objects = xrealloc(objects, needed * sizeof(*entry));
516 nr_alloc = needed;
517 }
518 entry = objects + idx;
519 nr_objects = idx + 1;
520 memset(entry, 0, sizeof(*entry));
521 memcpy(entry->sha1, sha1, 20);
522 entry->hash = hash;
524 if (object_ix_hashsz * 3 <= nr_objects * 4)
525 rehash_objects();
526 else {
527 ix = locate_object_entry_hash(entry->sha1);
528 if (0 <= ix)
529 die("internal error in object hashing.");
530 object_ix[-1 - ix] = idx + 1;
531 }
532 status = 1;
534 already_added:
535 if (progress_update) {
536 fprintf(stderr, "Counting objects...%d\r", nr_objects);
537 progress_update = 0;
538 }
539 if (exclude)
540 entry->preferred_base = 1;
541 else {
542 if (found_pack) {
543 entry->in_pack = found_pack;
544 entry->in_pack_offset = found_offset;
545 }
546 }
547 return status;
548 }
550 struct pbase_tree_cache {
551 unsigned char sha1[20];
552 int ref;
553 int temporary;
554 void *tree_data;
555 unsigned long tree_size;
556 };
558 static struct pbase_tree_cache *(pbase_tree_cache[256]);
559 static int pbase_tree_cache_ix(const unsigned char *sha1)
560 {
561 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
562 }
563 static int pbase_tree_cache_ix_incr(int ix)
564 {
565 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
566 }
568 static struct pbase_tree {
569 struct pbase_tree *next;
570 /* This is a phony "cache" entry; we are not
571 * going to evict it nor find it through _get()
572 * mechanism -- this is for the toplevel node that
573 * would almost always change with any commit.
574 */
575 struct pbase_tree_cache pcache;
576 } *pbase_tree;
578 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
579 {
580 struct pbase_tree_cache *ent, *nent;
581 void *data;
582 unsigned long size;
583 char type[20];
584 int neigh;
585 int my_ix = pbase_tree_cache_ix(sha1);
586 int available_ix = -1;
588 /* pbase-tree-cache acts as a limited hashtable.
589 * your object will be found at your index or within a few
590 * slots after that slot if it is cached.
591 */
592 for (neigh = 0; neigh < 8; neigh++) {
593 ent = pbase_tree_cache[my_ix];
594 if (ent && !memcmp(ent->sha1, sha1, 20)) {
595 ent->ref++;
596 return ent;
597 }
598 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
599 ((0 <= available_ix) &&
600 (!ent && pbase_tree_cache[available_ix])))
601 available_ix = my_ix;
602 if (!ent)
603 break;
604 my_ix = pbase_tree_cache_ix_incr(my_ix);
605 }
607 /* Did not find one. Either we got a bogus request or
608 * we need to read and perhaps cache.
609 */
610 data = read_sha1_file(sha1, type, &size);
611 if (!data)
612 return NULL;
613 if (strcmp(type, tree_type)) {
614 free(data);
615 return NULL;
616 }
618 /* We need to either cache or return a throwaway copy */
620 if (available_ix < 0)
621 ent = NULL;
622 else {
623 ent = pbase_tree_cache[available_ix];
624 my_ix = available_ix;
625 }
627 if (!ent) {
628 nent = xmalloc(sizeof(*nent));
629 nent->temporary = (available_ix < 0);
630 }
631 else {
632 /* evict and reuse */
633 free(ent->tree_data);
634 nent = ent;
635 }
636 memcpy(nent->sha1, sha1, 20);
637 nent->tree_data = data;
638 nent->tree_size = size;
639 nent->ref = 1;
640 if (!nent->temporary)
641 pbase_tree_cache[my_ix] = nent;
642 return nent;
643 }
645 static void pbase_tree_put(struct pbase_tree_cache *cache)
646 {
647 if (!cache->temporary) {
648 cache->ref--;
649 return;
650 }
651 free(cache->tree_data);
652 free(cache);
653 }
655 static int name_cmp_len(const char *name)
656 {
657 int i;
658 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
659 ;
660 return i;
661 }
663 static void add_pbase_object(struct tree_desc *tree,
664 const char *name,
665 int cmplen,
666 const char *fullname)
667 {
668 struct name_entry entry;
670 while (tree_entry(tree,&entry)) {
671 unsigned long size;
672 char type[20];
674 if (entry.pathlen != cmplen ||
675 memcmp(entry.path, name, cmplen) ||
676 !has_sha1_file(entry.sha1) ||
677 sha1_object_info(entry.sha1, type, &size))
678 continue;
679 if (name[cmplen] != '/') {
680 unsigned hash = name_hash(fullname);
681 add_object_entry(entry.sha1, hash, 1);
682 return;
683 }
684 if (!strcmp(type, tree_type)) {
685 struct tree_desc sub;
686 struct pbase_tree_cache *tree;
687 const char *down = name+cmplen+1;
688 int downlen = name_cmp_len(down);
690 tree = pbase_tree_get(entry.sha1);
691 if (!tree)
692 return;
693 sub.buf = tree->tree_data;
694 sub.size = tree->tree_size;
696 add_pbase_object(&sub, down, downlen, fullname);
697 pbase_tree_put(tree);
698 }
699 }
700 }
702 static unsigned *done_pbase_paths;
703 static int done_pbase_paths_num;
704 static int done_pbase_paths_alloc;
705 static int done_pbase_path_pos(unsigned hash)
706 {
707 int lo = 0;
708 int hi = done_pbase_paths_num;
709 while (lo < hi) {
710 int mi = (hi + lo) / 2;
711 if (done_pbase_paths[mi] == hash)
712 return mi;
713 if (done_pbase_paths[mi] < hash)
714 hi = mi;
715 else
716 lo = mi + 1;
717 }
718 return -lo-1;
719 }
721 static int check_pbase_path(unsigned hash)
722 {
723 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
724 if (0 <= pos)
725 return 1;
726 pos = -pos - 1;
727 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
728 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
729 done_pbase_paths = xrealloc(done_pbase_paths,
730 done_pbase_paths_alloc *
731 sizeof(unsigned));
732 }
733 done_pbase_paths_num++;
734 if (pos < done_pbase_paths_num)
735 memmove(done_pbase_paths + pos + 1,
736 done_pbase_paths + pos,
737 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
738 done_pbase_paths[pos] = hash;
739 return 0;
740 }
742 static void add_preferred_base_object(char *name, unsigned hash)
743 {
744 struct pbase_tree *it;
745 int cmplen = name_cmp_len(name);
747 if (check_pbase_path(hash))
748 return;
750 for (it = pbase_tree; it; it = it->next) {
751 if (cmplen == 0) {
752 hash = name_hash("");
753 add_object_entry(it->pcache.sha1, hash, 1);
754 }
755 else {
756 struct tree_desc tree;
757 tree.buf = it->pcache.tree_data;
758 tree.size = it->pcache.tree_size;
759 add_pbase_object(&tree, name, cmplen, name);
760 }
761 }
762 }
764 static void add_preferred_base(unsigned char *sha1)
765 {
766 struct pbase_tree *it;
767 void *data;
768 unsigned long size;
769 unsigned char tree_sha1[20];
771 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
772 if (!data)
773 return;
775 for (it = pbase_tree; it; it = it->next) {
776 if (!memcmp(it->pcache.sha1, tree_sha1, 20)) {
777 free(data);
778 return;
779 }
780 }
782 it = xcalloc(1, sizeof(*it));
783 it->next = pbase_tree;
784 pbase_tree = it;
786 memcpy(it->pcache.sha1, tree_sha1, 20);
787 it->pcache.tree_data = data;
788 it->pcache.tree_size = size;
789 }
791 static void check_object(struct object_entry *entry)
792 {
793 char type[20];
795 if (entry->in_pack && !entry->preferred_base) {
796 unsigned char base[20];
797 unsigned long size;
798 struct object_entry *base_entry;
800 /* We want in_pack_type even if we do not reuse delta.
801 * There is no point not reusing non-delta representations.
802 */
803 check_reuse_pack_delta(entry->in_pack,
804 entry->in_pack_offset,
805 base, &size,
806 &entry->in_pack_type);
808 /* Check if it is delta, and the base is also an object
809 * we are going to pack. If so we will reuse the existing
810 * delta.
811 */
812 if (!no_reuse_delta &&
813 entry->in_pack_type == OBJ_DELTA &&
814 (base_entry = locate_object_entry(base)) &&
815 (!base_entry->preferred_base)) {
817 /* Depth value does not matter - find_deltas()
818 * will never consider reused delta as the
819 * base object to deltify other objects
820 * against, in order to avoid circular deltas.
821 */
823 /* uncompressed size of the delta data */
824 entry->size = entry->delta_size = size;
825 entry->delta = base_entry;
826 entry->type = OBJ_DELTA;
828 entry->delta_sibling = base_entry->delta_child;
829 base_entry->delta_child = entry;
831 return;
832 }
833 /* Otherwise we would do the usual */
834 }
836 if (sha1_object_info(entry->sha1, type, &entry->size))
837 die("unable to get type of object %s",
838 sha1_to_hex(entry->sha1));
840 if (!strcmp(type, commit_type)) {
841 entry->type = OBJ_COMMIT;
842 } else if (!strcmp(type, tree_type)) {
843 entry->type = OBJ_TREE;
844 } else if (!strcmp(type, blob_type)) {
845 entry->type = OBJ_BLOB;
846 } else if (!strcmp(type, tag_type)) {
847 entry->type = OBJ_TAG;
848 } else
849 die("unable to pack object %s of type %s",
850 sha1_to_hex(entry->sha1), type);
851 }
853 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
854 {
855 struct object_entry *child = me->delta_child;
856 unsigned int m = n;
857 while (child) {
858 unsigned int c = check_delta_limit(child, n + 1);
859 if (m < c)
860 m = c;
861 child = child->delta_sibling;
862 }
863 return m;
864 }
866 static void get_object_details(void)
867 {
868 int i;
869 struct object_entry *entry;
871 prepare_pack_ix();
872 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
873 check_object(entry);
875 if (nr_objects == nr_result) {
876 /*
877 * Depth of objects that depend on the entry -- this
878 * is subtracted from depth-max to break too deep
879 * delta chain because of delta data reusing.
880 * However, we loosen this restriction when we know we
881 * are creating a thin pack -- it will have to be
882 * expanded on the other end anyway, so do not
883 * artificially cut the delta chain and let it go as
884 * deep as it wants.
885 */
886 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
887 if (!entry->delta && entry->delta_child)
888 entry->delta_limit =
889 check_delta_limit(entry, 1);
890 }
891 }
893 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
895 static entry_sort_t current_sort;
897 static int sort_comparator(const void *_a, const void *_b)
898 {
899 struct object_entry *a = *(struct object_entry **)_a;
900 struct object_entry *b = *(struct object_entry **)_b;
901 return current_sort(a,b);
902 }
904 static struct object_entry **create_sorted_list(entry_sort_t sort)
905 {
906 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
907 int i;
909 for (i = 0; i < nr_objects; i++)
910 list[i] = objects + i;
911 current_sort = sort;
912 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
913 return list;
914 }
916 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
917 {
918 return memcmp(a->sha1, b->sha1, 20);
919 }
921 static struct object_entry **create_final_object_list(void)
922 {
923 struct object_entry **list;
924 int i, j;
926 for (i = nr_result = 0; i < nr_objects; i++)
927 if (!objects[i].preferred_base)
928 nr_result++;
929 list = xmalloc(nr_result * sizeof(struct object_entry *));
930 for (i = j = 0; i < nr_objects; i++) {
931 if (!objects[i].preferred_base)
932 list[j++] = objects + i;
933 }
934 current_sort = sha1_sort;
935 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
936 return list;
937 }
939 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
940 {
941 if (a->type < b->type)
942 return -1;
943 if (a->type > b->type)
944 return 1;
945 if (a->hash < b->hash)
946 return -1;
947 if (a->hash > b->hash)
948 return 1;
949 if (a->preferred_base < b->preferred_base)
950 return -1;
951 if (a->preferred_base > b->preferred_base)
952 return 1;
953 if (a->size < b->size)
954 return -1;
955 if (a->size > b->size)
956 return 1;
957 return a < b ? -1 : (a > b);
958 }
960 struct unpacked {
961 struct object_entry *entry;
962 void *data;
963 struct delta_index *index;
964 };
966 /*
967 * We search for deltas _backwards_ in a list sorted by type and
968 * by size, so that we see progressively smaller and smaller files.
969 * That's because we prefer deltas to be from the bigger file
970 * to the smaller - deletes are potentially cheaper, but perhaps
971 * more importantly, the bigger file is likely the more recent
972 * one.
973 */
974 static int try_delta(struct unpacked *trg, struct unpacked *src,
975 unsigned max_depth)
976 {
977 struct object_entry *trg_entry = trg->entry;
978 struct object_entry *src_entry = src->entry;
979 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
980 char type[10];
981 void *delta_buf;
983 /* Don't bother doing diffs between different types */
984 if (trg_entry->type != src_entry->type)
985 return -1;
987 /* We do not compute delta to *create* objects we are not
988 * going to pack.
989 */
990 if (trg_entry->preferred_base)
991 return -1;
993 /*
994 * We do not bother to try a delta that we discarded
995 * on an earlier try, but only when reusing delta data.
996 */
997 if (!no_reuse_delta && trg_entry->in_pack &&
998 trg_entry->in_pack == src_entry->in_pack)
999 return 0;
1001 /*
1002 * If the current object is at pack edge, take the depth the
1003 * objects that depend on the current object into account --
1004 * otherwise they would become too deep.
1005 */
1006 if (trg_entry->delta_child) {
1007 if (max_depth <= trg_entry->delta_limit)
1008 return 0;
1009 max_depth -= trg_entry->delta_limit;
1010 }
1011 if (src_entry->depth >= max_depth)
1012 return 0;
1014 /* Now some size filtering heuristics. */
1015 trg_size = trg_entry->size;
1016 max_size = trg_size/2 - 20;
1017 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1018 if (max_size == 0)
1019 return 0;
1020 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1021 max_size = trg_entry->delta_size-1;
1022 src_size = src_entry->size;
1023 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1024 if (sizediff >= max_size)
1025 return 0;
1027 /* Load data if not already done */
1028 if (!trg->data) {
1029 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1030 if (sz != trg_size)
1031 die("object %s inconsistent object length (%lu vs %lu)",
1032 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1033 }
1034 if (!src->data) {
1035 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1036 if (sz != src_size)
1037 die("object %s inconsistent object length (%lu vs %lu)",
1038 sha1_to_hex(src_entry->sha1), sz, src_size);
1039 }
1040 if (!src->index) {
1041 src->index = create_delta_index(src->data, src_size);
1042 if (!src->index)
1043 die("out of memory");
1044 }
1046 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1047 if (!delta_buf)
1048 return 0;
1050 trg_entry->delta = src_entry;
1051 trg_entry->delta_size = delta_size;
1052 trg_entry->depth = src_entry->depth + 1;
1053 free(delta_buf);
1054 return 1;
1055 }
1057 static void progress_interval(int signum)
1058 {
1059 progress_update = 1;
1060 }
1062 static void find_deltas(struct object_entry **list, int window, int depth)
1063 {
1064 int i, idx;
1065 unsigned int array_size = window * sizeof(struct unpacked);
1066 struct unpacked *array = xmalloc(array_size);
1067 unsigned processed = 0;
1068 unsigned last_percent = 999;
1070 memset(array, 0, array_size);
1071 i = nr_objects;
1072 idx = 0;
1073 if (progress)
1074 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1076 while (--i >= 0) {
1077 struct object_entry *entry = list[i];
1078 struct unpacked *n = array + idx;
1079 int j;
1081 if (!entry->preferred_base)
1082 processed++;
1084 if (progress) {
1085 unsigned percent = processed * 100 / nr_result;
1086 if (percent != last_percent || progress_update) {
1087 fprintf(stderr, "%4u%% (%u/%u) done\r",
1088 percent, processed, nr_result);
1089 progress_update = 0;
1090 last_percent = percent;
1091 }
1092 }
1094 if (entry->delta)
1095 /* This happens if we decided to reuse existing
1096 * delta from a pack. "!no_reuse_delta &&" is implied.
1097 */
1098 continue;
1100 if (entry->size < 50)
1101 continue;
1102 free_delta_index(n->index);
1103 n->index = NULL;
1104 free(n->data);
1105 n->data = NULL;
1106 n->entry = entry;
1108 j = window;
1109 while (--j > 0) {
1110 unsigned int other_idx = idx + j;
1111 struct unpacked *m;
1112 if (other_idx >= window)
1113 other_idx -= window;
1114 m = array + other_idx;
1115 if (!m->entry)
1116 break;
1117 if (try_delta(n, m, depth) < 0)
1118 break;
1119 }
1120 /* if we made n a delta, and if n is already at max
1121 * depth, leaving it in the window is pointless. we
1122 * should evict it first.
1123 */
1124 if (entry->delta && depth <= entry->depth)
1125 continue;
1127 idx++;
1128 if (idx >= window)
1129 idx = 0;
1130 }
1132 if (progress)
1133 fputc('\n', stderr);
1135 for (i = 0; i < window; ++i) {
1136 free_delta_index(array[i].index);
1137 free(array[i].data);
1138 }
1139 free(array);
1140 }
1142 static void prepare_pack(int window, int depth)
1143 {
1144 get_object_details();
1145 sorted_by_type = create_sorted_list(type_size_sort);
1146 if (window && depth)
1147 find_deltas(sorted_by_type, window+1, depth);
1148 }
1150 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
1151 {
1152 static const char cache[] = "pack-cache/pack-%s.%s";
1153 char *cached_pack, *cached_idx;
1154 int ifd, ofd, ifd_ix = -1;
1156 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1157 ifd = open(cached_pack, O_RDONLY);
1158 if (ifd < 0)
1159 return 0;
1161 if (!pack_to_stdout) {
1162 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1163 ifd_ix = open(cached_idx, O_RDONLY);
1164 if (ifd_ix < 0) {
1165 close(ifd);
1166 return 0;
1167 }
1168 }
1170 if (progress)
1171 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1172 sha1_to_hex(sha1));
1174 if (pack_to_stdout) {
1175 if (copy_fd(ifd, 1))
1176 exit(1);
1177 close(ifd);
1178 }
1179 else {
1180 char name[PATH_MAX];
1181 snprintf(name, sizeof(name),
1182 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1183 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1184 if (ofd < 0)
1185 die("unable to open %s (%s)", name, strerror(errno));
1186 if (copy_fd(ifd, ofd))
1187 exit(1);
1188 close(ifd);
1190 snprintf(name, sizeof(name),
1191 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1192 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1193 if (ofd < 0)
1194 die("unable to open %s (%s)", name, strerror(errno));
1195 if (copy_fd(ifd_ix, ofd))
1196 exit(1);
1197 close(ifd_ix);
1198 puts(sha1_to_hex(sha1));
1199 }
1201 return 1;
1202 }
1204 static void setup_progress_signal(void)
1205 {
1206 struct sigaction sa;
1207 struct itimerval v;
1209 memset(&sa, 0, sizeof(sa));
1210 sa.sa_handler = progress_interval;
1211 sigemptyset(&sa.sa_mask);
1212 sa.sa_flags = SA_RESTART;
1213 sigaction(SIGALRM, &sa, NULL);
1215 v.it_interval.tv_sec = 1;
1216 v.it_interval.tv_usec = 0;
1217 v.it_value = v.it_interval;
1218 setitimer(ITIMER_REAL, &v, NULL);
1219 }
1221 static int git_pack_config(const char *k, const char *v)
1222 {
1223 if(!strcmp(k, "pack.window")) {
1224 window = git_config_int(k, v);
1225 return 0;
1226 }
1227 return git_default_config(k, v);
1228 }
1230 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1231 {
1232 SHA_CTX ctx;
1233 char line[40 + 1 + PATH_MAX + 2];
1234 int depth = 10, pack_to_stdout = 0;
1235 struct object_entry **list;
1236 int num_preferred_base = 0;
1237 int i;
1239 git_config(git_pack_config);
1241 progress = isatty(2);
1242 for (i = 1; i < argc; i++) {
1243 const char *arg = argv[i];
1245 if (*arg == '-') {
1246 if (!strcmp("--non-empty", arg)) {
1247 non_empty = 1;
1248 continue;
1249 }
1250 if (!strcmp("--local", arg)) {
1251 local = 1;
1252 continue;
1253 }
1254 if (!strcmp("--progress", arg)) {
1255 progress = 1;
1256 continue;
1257 }
1258 if (!strcmp("--incremental", arg)) {
1259 incremental = 1;
1260 continue;
1261 }
1262 if (!strncmp("--window=", arg, 9)) {
1263 char *end;
1264 window = strtoul(arg+9, &end, 0);
1265 if (!arg[9] || *end)
1266 usage(pack_usage);
1267 continue;
1268 }
1269 if (!strncmp("--depth=", arg, 8)) {
1270 char *end;
1271 depth = strtoul(arg+8, &end, 0);
1272 if (!arg[8] || *end)
1273 usage(pack_usage);
1274 continue;
1275 }
1276 if (!strcmp("--progress", arg)) {
1277 progress = 1;
1278 continue;
1279 }
1280 if (!strcmp("-q", arg)) {
1281 progress = 0;
1282 continue;
1283 }
1284 if (!strcmp("--no-reuse-delta", arg)) {
1285 no_reuse_delta = 1;
1286 continue;
1287 }
1288 if (!strcmp("--stdout", arg)) {
1289 pack_to_stdout = 1;
1290 continue;
1291 }
1292 usage(pack_usage);
1293 }
1294 if (base_name)
1295 usage(pack_usage);
1296 base_name = arg;
1297 }
1299 if (pack_to_stdout != !base_name)
1300 usage(pack_usage);
1302 prepare_packed_git();
1304 if (progress) {
1305 fprintf(stderr, "Generating pack...\n");
1306 setup_progress_signal();
1307 }
1309 for (;;) {
1310 unsigned char sha1[20];
1311 unsigned hash;
1313 if (!fgets(line, sizeof(line), stdin)) {
1314 if (feof(stdin))
1315 break;
1316 if (!ferror(stdin))
1317 die("fgets returned NULL, not EOF, not error!");
1318 if (errno != EINTR)
1319 die("fgets: %s", strerror(errno));
1320 clearerr(stdin);
1321 continue;
1322 }
1324 if (line[0] == '-') {
1325 if (get_sha1_hex(line+1, sha1))
1326 die("expected edge sha1, got garbage:\n %s",
1327 line+1);
1328 if (num_preferred_base++ < window)
1329 add_preferred_base(sha1);
1330 continue;
1331 }
1332 if (get_sha1_hex(line, sha1))
1333 die("expected sha1, got garbage:\n %s", line);
1334 hash = name_hash(line+41);
1335 add_preferred_base_object(line+41, hash);
1336 add_object_entry(sha1, hash, 0);
1337 }
1338 if (progress)
1339 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1340 sorted_by_sha = create_final_object_list();
1341 if (non_empty && !nr_result)
1342 return 0;
1344 SHA1_Init(&ctx);
1345 list = sorted_by_sha;
1346 for (i = 0; i < nr_result; i++) {
1347 struct object_entry *entry = *list++;
1348 SHA1_Update(&ctx, entry->sha1, 20);
1349 }
1350 SHA1_Final(object_list_sha1, &ctx);
1351 if (progress && (nr_objects != nr_result))
1352 fprintf(stderr, "Result has %d objects.\n", nr_result);
1354 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1355 ;
1356 else {
1357 if (nr_result)
1358 prepare_pack(window, depth);
1359 if (progress && pack_to_stdout) {
1360 /* the other end usually displays progress itself */
1361 struct itimerval v = {{0,},};
1362 setitimer(ITIMER_REAL, &v, NULL);
1363 signal(SIGALRM, SIG_IGN );
1364 progress_update = 0;
1365 }
1366 write_pack_file();
1367 if (!pack_to_stdout) {
1368 write_index_file();
1369 puts(sha1_to_hex(object_list_sha1));
1370 }
1371 }
1372 if (progress)
1373 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1374 nr_result, written, written_delta, reused, reused_delta);
1375 return 0;
1376 }