Merge branch 'master' into next
[git/git.git] / pack-objects.c
1 #include "cache.h"
2 #include "object.h"
3 #include "delta.h"
4 #include "pack.h"
5 #include "csum-file.h"
6 #include "diff.h"
7 #include <sys/time.h>
8 #include <signal.h>
9
10 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";
11
12 struct object_entry {
13 unsigned char sha1[20];
14 unsigned long size; /* uncompressed size */
15 unsigned long offset; /* offset into the final pack file;
16 * nonzero if already written.
17 */
18 unsigned int depth; /* delta depth */
19 unsigned int delta_limit; /* base adjustment for in-pack delta */
20 unsigned int hash; /* name hint hash */
21 enum object_type type;
22 enum object_type in_pack_type; /* could be delta */
23 unsigned long delta_size; /* delta data size (uncompressed) */
24 struct object_entry *delta; /* delta base object */
25 struct packed_git *in_pack; /* already in pack */
26 unsigned int in_pack_offset;
27 struct object_entry *delta_child; /* delitified objects who bases me */
28 struct object_entry *delta_sibling; /* other deltified objects who
29 * uses the same base as me
30 */
31 int preferred_base; /* we do not pack this, but is encouraged to
32 * be used as the base objectto delta huge
33 * objects against.
34 */
35 };
36
37 /*
38 * Objects we are going to pack are colected in objects array (dynamically
39 * expanded). nr_objects & nr_alloc controls this array. They are stored
40 * in the order we see -- typically rev-list --objects order that gives us
41 * nice "minimum seek" order.
42 *
43 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
44 * elements in the objects array. The former is used to build the pack
45 * index (lists object names in the ascending order to help offset lookup),
46 * and the latter is used to group similar things together by try_delta()
47 * heuristics.
48 */
49
50 static unsigned char object_list_sha1[20];
51 static int non_empty = 0;
52 static int no_reuse_delta = 0;
53 static int local = 0;
54 static int incremental = 0;
55 static struct object_entry **sorted_by_sha, **sorted_by_type;
56 static struct object_entry *objects = NULL;
57 static int nr_objects = 0, nr_alloc = 0, nr_result = 0;
58 static const char *base_name;
59 static unsigned char pack_file_sha1[20];
60 static int progress = 1;
61 static volatile int progress_update = 0;
62
63 /*
64 * The object names in objects array are hashed with this hashtable,
65 * to help looking up the entry by object name. Binary search from
66 * sorted_by_sha is also possible but this was easier to code and faster.
67 * This hashtable is built after all the objects are seen.
68 */
69 static int *object_ix = NULL;
70 static int object_ix_hashsz = 0;
71
72 /*
73 * Pack index for existing packs give us easy access to the offsets into
74 * corresponding pack file where each object's data starts, but the entries
75 * do not store the size of the compressed representation (uncompressed
76 * size is easily available by examining the pack entry header). We build
77 * a hashtable of existing packs (pack_revindex), and keep reverse index
78 * here -- pack index file is sorted by object name mapping to offset; this
79 * pack_revindex[].revindex array is an ordered list of offsets, so if you
80 * know the offset of an object, next offset is where its packed
81 * representation ends.
82 */
83 struct pack_revindex {
84 struct packed_git *p;
85 unsigned long *revindex;
86 } *pack_revindex = NULL;
87 static int pack_revindex_hashsz = 0;
88
89 /*
90 * stats
91 */
92 static int written = 0;
93 static int written_delta = 0;
94 static int reused = 0;
95 static int reused_delta = 0;
96
97 static int pack_revindex_ix(struct packed_git *p)
98 {
99 unsigned long ui = (unsigned long)p;
100 int i;
101
102 ui = ui ^ (ui >> 16); /* defeat structure alignment */
103 i = (int)(ui % pack_revindex_hashsz);
104 while (pack_revindex[i].p) {
105 if (pack_revindex[i].p == p)
106 return i;
107 if (++i == pack_revindex_hashsz)
108 i = 0;
109 }
110 return -1 - i;
111 }
112
113 static void prepare_pack_ix(void)
114 {
115 int num;
116 struct packed_git *p;
117 for (num = 0, p = packed_git; p; p = p->next)
118 num++;
119 if (!num)
120 return;
121 pack_revindex_hashsz = num * 11;
122 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
123 for (p = packed_git; p; p = p->next) {
124 num = pack_revindex_ix(p);
125 num = - 1 - num;
126 pack_revindex[num].p = p;
127 }
128 /* revindex elements are lazily initialized */
129 }
130
131 static int cmp_offset(const void *a_, const void *b_)
132 {
133 unsigned long a = *(unsigned long *) a_;
134 unsigned long b = *(unsigned long *) b_;
135 if (a < b)
136 return -1;
137 else if (a == b)
138 return 0;
139 else
140 return 1;
141 }
142
143 /*
144 * Ordered list of offsets of objects in the pack.
145 */
146 static void prepare_pack_revindex(struct pack_revindex *rix)
147 {
148 struct packed_git *p = rix->p;
149 int num_ent = num_packed_objects(p);
150 int i;
151 void *index = p->index_base + 256;
152
153 rix->revindex = xmalloc(sizeof(unsigned long) * (num_ent + 1));
154 for (i = 0; i < num_ent; i++) {
155 long hl = *((long *)(index + 24 * i));
156 rix->revindex[i] = ntohl(hl);
157 }
158 /* This knows the pack format -- the 20-byte trailer
159 * follows immediately after the last object data.
160 */
161 rix->revindex[num_ent] = p->pack_size - 20;
162 qsort(rix->revindex, num_ent, sizeof(unsigned long), cmp_offset);
163 }
164
165 static unsigned long find_packed_object_size(struct packed_git *p,
166 unsigned long ofs)
167 {
168 int num;
169 int lo, hi;
170 struct pack_revindex *rix;
171 unsigned long *revindex;
172 num = pack_revindex_ix(p);
173 if (num < 0)
174 die("internal error: pack revindex uninitialized");
175 rix = &pack_revindex[num];
176 if (!rix->revindex)
177 prepare_pack_revindex(rix);
178 revindex = rix->revindex;
179 lo = 0;
180 hi = num_packed_objects(p) + 1;
181 do {
182 int mi = (lo + hi) / 2;
183 if (revindex[mi] == ofs) {
184 return revindex[mi+1] - ofs;
185 }
186 else if (ofs < revindex[mi])
187 hi = mi;
188 else
189 lo = mi + 1;
190 } while (lo < hi);
191 die("internal error: pack revindex corrupt");
192 }
193
194 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
195 {
196 unsigned long othersize, delta_size;
197 char type[10];
198 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
199 void *delta_buf;
200
201 if (!otherbuf)
202 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
203 delta_buf = diff_delta(otherbuf, othersize,
204 buf, size, &delta_size, 0);
205 if (!delta_buf || delta_size != entry->delta_size)
206 die("delta size changed");
207 free(buf);
208 free(otherbuf);
209 return delta_buf;
210 }
211
212 /*
213 * The per-object header is a pretty dense thing, which is
214 * - first byte: low four bits are "size", then three bits of "type",
215 * and the high bit is "size continues".
216 * - each byte afterwards: low seven bits are size continuation,
217 * with the high bit being "size continues"
218 */
219 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
220 {
221 int n = 1;
222 unsigned char c;
223
224 if (type < OBJ_COMMIT || type > OBJ_DELTA)
225 die("bad type %d", type);
226
227 c = (type << 4) | (size & 15);
228 size >>= 4;
229 while (size) {
230 *hdr++ = c | 0x80;
231 c = size & 0x7f;
232 size >>= 7;
233 n++;
234 }
235 *hdr = c;
236 return n;
237 }
238
239 static unsigned long write_object(struct sha1file *f,
240 struct object_entry *entry)
241 {
242 unsigned long size;
243 char type[10];
244 void *buf;
245 unsigned char header[10];
246 unsigned hdrlen, datalen;
247 enum object_type obj_type;
248 int to_reuse = 0;
249
250 if (entry->preferred_base)
251 return 0;
252
253 obj_type = entry->type;
254 if (! entry->in_pack)
255 to_reuse = 0; /* can't reuse what we don't have */
256 else if (obj_type == OBJ_DELTA)
257 to_reuse = 1; /* check_object() decided it for us */
258 else if (obj_type != entry->in_pack_type)
259 to_reuse = 0; /* pack has delta which is unusable */
260 else if (entry->delta)
261 to_reuse = 0; /* we want to pack afresh */
262 else
263 to_reuse = 1; /* we have it in-pack undeltified,
264 * and we do not need to deltify it.
265 */
266
267 if (! to_reuse) {
268 buf = read_sha1_file(entry->sha1, type, &size);
269 if (!buf)
270 die("unable to read %s", sha1_to_hex(entry->sha1));
271 if (size != entry->size)
272 die("object %s size inconsistency (%lu vs %lu)",
273 sha1_to_hex(entry->sha1), size, entry->size);
274 if (entry->delta) {
275 buf = delta_against(buf, size, entry);
276 size = entry->delta_size;
277 obj_type = OBJ_DELTA;
278 }
279 /*
280 * The object header is a byte of 'type' followed by zero or
281 * more bytes of length. For deltas, the 20 bytes of delta
282 * sha1 follows that.
283 */
284 hdrlen = encode_header(obj_type, size, header);
285 sha1write(f, header, hdrlen);
286
287 if (entry->delta) {
288 sha1write(f, entry->delta, 20);
289 hdrlen += 20;
290 }
291 datalen = sha1write_compressed(f, buf, size);
292 free(buf);
293 }
294 else {
295 struct packed_git *p = entry->in_pack;
296 use_packed_git(p);
297
298 datalen = find_packed_object_size(p, entry->in_pack_offset);
299 buf = p->pack_base + entry->in_pack_offset;
300 sha1write(f, buf, datalen);
301 unuse_packed_git(p);
302 hdrlen = 0; /* not really */
303 if (obj_type == OBJ_DELTA)
304 reused_delta++;
305 reused++;
306 }
307 if (obj_type == OBJ_DELTA)
308 written_delta++;
309 written++;
310 return hdrlen + datalen;
311 }
312
313 static unsigned long write_one(struct sha1file *f,
314 struct object_entry *e,
315 unsigned long offset)
316 {
317 if (e->offset)
318 /* offset starts from header size and cannot be zero
319 * if it is written already.
320 */
321 return offset;
322 e->offset = offset;
323 offset += write_object(f, e);
324 /* if we are deltified, write out its base object. */
325 if (e->delta)
326 offset = write_one(f, e->delta, offset);
327 return offset;
328 }
329
330 static void write_pack_file(void)
331 {
332 int i;
333 struct sha1file *f;
334 unsigned long offset;
335 struct pack_header hdr;
336 unsigned last_percent = 999;
337 int do_progress = 0;
338
339 if (!base_name)
340 f = sha1fd(1, "<stdout>");
341 else {
342 f = sha1create("%s-%s.%s", base_name,
343 sha1_to_hex(object_list_sha1), "pack");
344 do_progress = progress;
345 }
346 if (do_progress)
347 fprintf(stderr, "Writing %d objects.\n", nr_result);
348
349 hdr.hdr_signature = htonl(PACK_SIGNATURE);
350 hdr.hdr_version = htonl(PACK_VERSION);
351 hdr.hdr_entries = htonl(nr_result);
352 sha1write(f, &hdr, sizeof(hdr));
353 offset = sizeof(hdr);
354 if (!nr_result)
355 goto done;
356 for (i = 0; i < nr_objects; i++) {
357 offset = write_one(f, objects + i, offset);
358 if (do_progress) {
359 unsigned percent = written * 100 / nr_result;
360 if (progress_update || percent != last_percent) {
361 fprintf(stderr, "%4u%% (%u/%u) done\r",
362 percent, written, nr_result);
363 progress_update = 0;
364 last_percent = percent;
365 }
366 }
367 }
368 if (do_progress)
369 fputc('\n', stderr);
370 done:
371 sha1close(f, pack_file_sha1, 1);
372 }
373
374 static void write_index_file(void)
375 {
376 int i;
377 struct sha1file *f = sha1create("%s-%s.%s", base_name,
378 sha1_to_hex(object_list_sha1), "idx");
379 struct object_entry **list = sorted_by_sha;
380 struct object_entry **last = list + nr_result;
381 unsigned int array[256];
382
383 /*
384 * Write the first-level table (the list is sorted,
385 * but we use a 256-entry lookup to be able to avoid
386 * having to do eight extra binary search iterations).
387 */
388 for (i = 0; i < 256; i++) {
389 struct object_entry **next = list;
390 while (next < last) {
391 struct object_entry *entry = *next;
392 if (entry->sha1[0] != i)
393 break;
394 next++;
395 }
396 array[i] = htonl(next - sorted_by_sha);
397 list = next;
398 }
399 sha1write(f, array, 256 * sizeof(int));
400
401 /*
402 * Write the actual SHA1 entries..
403 */
404 list = sorted_by_sha;
405 for (i = 0; i < nr_result; i++) {
406 struct object_entry *entry = *list++;
407 unsigned int offset = htonl(entry->offset);
408 sha1write(f, &offset, 4);
409 sha1write(f, entry->sha1, 20);
410 }
411 sha1write(f, pack_file_sha1, 20);
412 sha1close(f, NULL, 1);
413 }
414
415 static int locate_object_entry_hash(const unsigned char *sha1)
416 {
417 int i;
418 unsigned int ui;
419 memcpy(&ui, sha1, sizeof(unsigned int));
420 i = ui % object_ix_hashsz;
421 while (0 < object_ix[i]) {
422 if (!memcmp(sha1, objects[object_ix[i]-1].sha1, 20))
423 return i;
424 if (++i == object_ix_hashsz)
425 i = 0;
426 }
427 return -1 - i;
428 }
429
430 static struct object_entry *locate_object_entry(const unsigned char *sha1)
431 {
432 int i;
433
434 if (!object_ix_hashsz)
435 return NULL;
436
437 i = locate_object_entry_hash(sha1);
438 if (0 <= i)
439 return &objects[object_ix[i]-1];
440 return NULL;
441 }
442
443 static void rehash_objects(void)
444 {
445 int i;
446 struct object_entry *oe;
447
448 object_ix_hashsz = nr_objects * 3;
449 if (object_ix_hashsz < 1024)
450 object_ix_hashsz = 1024;
451 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
452 object_ix = memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
453 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
454 int ix = locate_object_entry_hash(oe->sha1);
455 if (0 <= ix)
456 continue;
457 ix = -1 - ix;
458 object_ix[ix] = i + 1;
459 }
460 }
461
462 struct name_path {
463 struct name_path *up;
464 const char *elem;
465 int len;
466 };
467
468 #define DIRBITS 12
469
470 static unsigned name_hash(struct name_path *path, const char *name)
471 {
472 struct name_path *p = path;
473 const char *n = name + strlen(name);
474 unsigned hash = 0, name_hash = 0, name_done = 0;
475
476 if (n != name && n[-1] == '\n')
477 n--;
478 while (name <= --n) {
479 unsigned char c = *n;
480 if (c == '/' && !name_done) {
481 name_hash = hash;
482 name_done = 1;
483 hash = 0;
484 }
485 hash = hash * 11 + c;
486 }
487 if (!name_done) {
488 name_hash = hash;
489 hash = 0;
490 }
491 for (p = path; p; p = p->up) {
492 hash = hash * 11 + '/';
493 n = p->elem + p->len;
494 while (p->elem <= --n) {
495 unsigned char c = *n;
496 hash = hash * 11 + c;
497 }
498 }
499 /*
500 * Make sure "Makefile" and "t/Makefile" are hashed separately
501 * but close enough.
502 */
503 hash = (name_hash<<DIRBITS) | (hash & ((1U<<DIRBITS )-1));
504
505 if (0) { /* debug */
506 n = name + strlen(name);
507 if (n != name && n[-1] == '\n')
508 n--;
509 while (name <= --n)
510 fputc(*n, stderr);
511 for (p = path; p; p = p->up) {
512 fputc('/', stderr);
513 n = p->elem + p->len;
514 while (p->elem <= --n)
515 fputc(*n, stderr);
516 }
517 fprintf(stderr, "\t%08x\n", hash);
518 }
519 return hash;
520 }
521
522 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
523 {
524 unsigned int idx = nr_objects;
525 struct object_entry *entry;
526 struct packed_git *p;
527 unsigned int found_offset = 0;
528 struct packed_git *found_pack = NULL;
529 int ix, status = 0;
530
531 if (!exclude) {
532 for (p = packed_git; p; p = p->next) {
533 struct pack_entry e;
534 if (find_pack_entry_one(sha1, &e, p)) {
535 if (incremental)
536 return 0;
537 if (local && !p->pack_local)
538 return 0;
539 if (!found_pack) {
540 found_offset = e.offset;
541 found_pack = e.p;
542 }
543 }
544 }
545 }
546 if ((entry = locate_object_entry(sha1)) != NULL)
547 goto already_added;
548
549 if (idx >= nr_alloc) {
550 unsigned int needed = (idx + 1024) * 3 / 2;
551 objects = xrealloc(objects, needed * sizeof(*entry));
552 nr_alloc = needed;
553 }
554 entry = objects + idx;
555 nr_objects = idx + 1;
556 memset(entry, 0, sizeof(*entry));
557 memcpy(entry->sha1, sha1, 20);
558 entry->hash = hash;
559
560 if (object_ix_hashsz * 3 <= nr_objects * 4)
561 rehash_objects();
562 else {
563 ix = locate_object_entry_hash(entry->sha1);
564 if (0 <= ix)
565 die("internal error in object hashing.");
566 object_ix[-1 - ix] = idx + 1;
567 }
568 status = 1;
569
570 already_added:
571 if (progress_update) {
572 fprintf(stderr, "Counting objects...%d\r", nr_objects);
573 progress_update = 0;
574 }
575 if (exclude)
576 entry->preferred_base = 1;
577 else {
578 if (found_pack) {
579 entry->in_pack = found_pack;
580 entry->in_pack_offset = found_offset;
581 }
582 }
583 return status;
584 }
585
586 static void add_pbase_tree(struct tree_desc *tree, struct name_path *up)
587 {
588 while (tree->size) {
589 const unsigned char *sha1;
590 const char *name;
591 unsigned mode, hash;
592 unsigned long size;
593 char type[20];
594
595 sha1 = tree_entry_extract(tree, &name, &mode);
596 update_tree_entry(tree);
597 if (!has_sha1_file(sha1))
598 continue;
599 if (sha1_object_info(sha1, type, &size))
600 continue;
601
602 hash = name_hash(up, name);
603 if (!add_object_entry(sha1, hash, 1))
604 continue;
605
606 if (!strcmp(type, "tree")) {
607 struct tree_desc sub;
608 void *elem;
609 struct name_path me;
610
611 elem = read_sha1_file(sha1, type, &sub.size);
612 sub.buf = elem;
613 if (sub.buf) {
614 me.up = up;
615 me.elem = name;
616 me.len = strlen(name);
617 add_pbase_tree(&sub, &me);
618 free(elem);
619 }
620 }
621 }
622 }
623
624 static void add_preferred_base(unsigned char *sha1)
625 {
626 struct tree_desc tree;
627 void *elem;
628
629 elem = read_object_with_reference(sha1, "tree", &tree.size, NULL);
630 tree.buf = elem;
631 if (!tree.buf)
632 return;
633 if (add_object_entry(sha1, name_hash(NULL, ""), 1))
634 add_pbase_tree(&tree, NULL);
635 free(elem);
636 }
637
638 static void check_object(struct object_entry *entry)
639 {
640 char type[20];
641
642 if (entry->in_pack && !entry->preferred_base) {
643 unsigned char base[20];
644 unsigned long size;
645 struct object_entry *base_entry;
646
647 /* We want in_pack_type even if we do not reuse delta.
648 * There is no point not reusing non-delta representations.
649 */
650 check_reuse_pack_delta(entry->in_pack,
651 entry->in_pack_offset,
652 base, &size,
653 &entry->in_pack_type);
654
655 /* Check if it is delta, and the base is also an object
656 * we are going to pack. If so we will reuse the existing
657 * delta.
658 */
659 if (!no_reuse_delta &&
660 entry->in_pack_type == OBJ_DELTA &&
661 (base_entry = locate_object_entry(base)) &&
662 (!base_entry->preferred_base)) {
663
664 /* Depth value does not matter - find_deltas()
665 * will never consider reused delta as the
666 * base object to deltify other objects
667 * against, in order to avoid circular deltas.
668 */
669
670 /* uncompressed size of the delta data */
671 entry->size = entry->delta_size = size;
672 entry->delta = base_entry;
673 entry->type = OBJ_DELTA;
674
675 entry->delta_sibling = base_entry->delta_child;
676 base_entry->delta_child = entry;
677
678 return;
679 }
680 /* Otherwise we would do the usual */
681 }
682
683 if (sha1_object_info(entry->sha1, type, &entry->size))
684 die("unable to get type of object %s",
685 sha1_to_hex(entry->sha1));
686
687 if (!strcmp(type, "commit")) {
688 entry->type = OBJ_COMMIT;
689 } else if (!strcmp(type, "tree")) {
690 entry->type = OBJ_TREE;
691 } else if (!strcmp(type, "blob")) {
692 entry->type = OBJ_BLOB;
693 } else if (!strcmp(type, "tag")) {
694 entry->type = OBJ_TAG;
695 } else
696 die("unable to pack object %s of type %s",
697 sha1_to_hex(entry->sha1), type);
698 }
699
700 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
701 {
702 struct object_entry *child = me->delta_child;
703 unsigned int m = n;
704 while (child) {
705 unsigned int c = check_delta_limit(child, n + 1);
706 if (m < c)
707 m = c;
708 child = child->delta_sibling;
709 }
710 return m;
711 }
712
713 static void get_object_details(void)
714 {
715 int i;
716 struct object_entry *entry;
717
718 prepare_pack_ix();
719 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
720 check_object(entry);
721
722 if (nr_objects == nr_result) {
723 /*
724 * Depth of objects that depend on the entry -- this
725 * is subtracted from depth-max to break too deep
726 * delta chain because of delta data reusing.
727 * However, we loosen this restriction when we know we
728 * are creating a thin pack -- it will have to be
729 * expanded on the other end anyway, so do not
730 * artificially cut the delta chain and let it go as
731 * deep as it wants.
732 */
733 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
734 if (!entry->delta && entry->delta_child)
735 entry->delta_limit =
736 check_delta_limit(entry, 1);
737 }
738 }
739
740 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
741
742 static entry_sort_t current_sort;
743
744 static int sort_comparator(const void *_a, const void *_b)
745 {
746 struct object_entry *a = *(struct object_entry **)_a;
747 struct object_entry *b = *(struct object_entry **)_b;
748 return current_sort(a,b);
749 }
750
751 static struct object_entry **create_sorted_list(entry_sort_t sort)
752 {
753 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
754 int i;
755
756 for (i = 0; i < nr_objects; i++)
757 list[i] = objects + i;
758 current_sort = sort;
759 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
760 return list;
761 }
762
763 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
764 {
765 return memcmp(a->sha1, b->sha1, 20);
766 }
767
768 static struct object_entry **create_final_object_list(void)
769 {
770 struct object_entry **list;
771 int i, j;
772
773 for (i = nr_result = 0; i < nr_objects; i++)
774 if (!objects[i].preferred_base)
775 nr_result++;
776 list = xmalloc(nr_result * sizeof(struct object_entry *));
777 for (i = j = 0; i < nr_objects; i++) {
778 if (!objects[i].preferred_base)
779 list[j++] = objects + i;
780 }
781 current_sort = sha1_sort;
782 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
783 return list;
784 }
785
786 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
787 {
788 if (a->type < b->type)
789 return -1;
790 if (a->type > b->type)
791 return 1;
792 if (a->hash < b->hash)
793 return -1;
794 if (a->hash > b->hash)
795 return 1;
796 if (a->preferred_base < b->preferred_base)
797 return -1;
798 if (a->preferred_base > b->preferred_base)
799 return 1;
800 if (a->size < b->size)
801 return -1;
802 if (a->size > b->size)
803 return 1;
804 return a < b ? -1 : (a > b);
805 }
806
807 struct unpacked {
808 struct object_entry *entry;
809 void *data;
810 };
811
812 /*
813 * We search for deltas _backwards_ in a list sorted by type and
814 * by size, so that we see progressively smaller and smaller files.
815 * That's because we prefer deltas to be from the bigger file
816 * to the smaller - deletes are potentially cheaper, but perhaps
817 * more importantly, the bigger file is likely the more recent
818 * one.
819 */
820 static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
821 {
822 struct object_entry *cur_entry = cur->entry;
823 struct object_entry *old_entry = old->entry;
824 unsigned long size, oldsize, delta_size, sizediff;
825 long max_size;
826 void *delta_buf;
827
828 /* Don't bother doing diffs between different types */
829 if (cur_entry->type != old_entry->type)
830 return -1;
831
832 /* We do not compute delta to *create* objects we are not
833 * going to pack.
834 */
835 if (cur_entry->preferred_base)
836 return -1;
837
838 /* If the current object is at pack edge, take the depth the
839 * objects that depend on the current object into account --
840 * otherwise they would become too deep.
841 */
842 if (cur_entry->delta_child) {
843 if (max_depth <= cur_entry->delta_limit)
844 return 0;
845 max_depth -= cur_entry->delta_limit;
846 }
847
848 size = cur_entry->size;
849 oldsize = old_entry->size;
850 sizediff = oldsize > size ? oldsize - size : size - oldsize;
851
852 if (size < 50)
853 return -1;
854 if (old_entry->depth >= max_depth)
855 return 0;
856
857 /*
858 * NOTE!
859 *
860 * We always delta from the bigger to the smaller, since that's
861 * more space-efficient (deletes don't have to say _what_ they
862 * delete).
863 */
864 max_size = size / 2 - 20;
865 if (cur_entry->delta)
866 max_size = cur_entry->delta_size-1;
867 if (sizediff >= max_size)
868 return -1;
869 delta_buf = diff_delta(old->data, oldsize,
870 cur->data, size, &delta_size, max_size);
871 if (!delta_buf)
872 return 0;
873 cur_entry->delta = old_entry;
874 cur_entry->delta_size = delta_size;
875 cur_entry->depth = old_entry->depth + 1;
876 free(delta_buf);
877 return 0;
878 }
879
880 static void progress_interval(int signum)
881 {
882 signal(SIGALRM, progress_interval);
883 progress_update = 1;
884 }
885
886 static void find_deltas(struct object_entry **list, int window, int depth)
887 {
888 int i, idx;
889 unsigned int array_size = window * sizeof(struct unpacked);
890 struct unpacked *array = xmalloc(array_size);
891 unsigned processed = 0;
892 unsigned last_percent = 999;
893
894 memset(array, 0, array_size);
895 i = nr_objects;
896 idx = 0;
897 if (progress)
898 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
899
900 while (--i >= 0) {
901 struct object_entry *entry = list[i];
902 struct unpacked *n = array + idx;
903 unsigned long size;
904 char type[10];
905 int j;
906
907 if (!entry->preferred_base)
908 processed++;
909
910 if (progress) {
911 unsigned percent = processed * 100 / nr_result;
912 if (percent != last_percent || progress_update) {
913 fprintf(stderr, "%4u%% (%u/%u) done\r",
914 percent, processed, nr_result);
915 progress_update = 0;
916 last_percent = percent;
917 }
918 }
919
920 if (entry->delta)
921 /* This happens if we decided to reuse existing
922 * delta from a pack. "!no_reuse_delta &&" is implied.
923 */
924 continue;
925
926 free(n->data);
927 n->entry = entry;
928 n->data = read_sha1_file(entry->sha1, type, &size);
929 if (size != entry->size)
930 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
931
932 j = window;
933 while (--j > 0) {
934 unsigned int other_idx = idx + j;
935 struct unpacked *m;
936 if (other_idx >= window)
937 other_idx -= window;
938 m = array + other_idx;
939 if (!m->entry)
940 break;
941 if (try_delta(n, m, depth) < 0)
942 break;
943 }
944 #if 0
945 /* if we made n a delta, and if n is already at max
946 * depth, leaving it in the window is pointless. we
947 * should evict it first.
948 * ... in theory only; somehow this makes things worse.
949 */
950 if (entry->delta && depth <= entry->depth)
951 continue;
952 #endif
953 idx++;
954 if (idx >= window)
955 idx = 0;
956 }
957
958 if (progress)
959 fputc('\n', stderr);
960
961 for (i = 0; i < window; ++i)
962 free(array[i].data);
963 free(array);
964 }
965
966 static void prepare_pack(int window, int depth)
967 {
968 get_object_details();
969 sorted_by_type = create_sorted_list(type_size_sort);
970 if (window && depth)
971 find_deltas(sorted_by_type, window+1, depth);
972 }
973
974 static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
975 {
976 static const char cache[] = "pack-cache/pack-%s.%s";
977 char *cached_pack, *cached_idx;
978 int ifd, ofd, ifd_ix = -1;
979
980 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
981 ifd = open(cached_pack, O_RDONLY);
982 if (ifd < 0)
983 return 0;
984
985 if (!pack_to_stdout) {
986 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
987 ifd_ix = open(cached_idx, O_RDONLY);
988 if (ifd_ix < 0) {
989 close(ifd);
990 return 0;
991 }
992 }
993
994 if (progress)
995 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
996 sha1_to_hex(sha1));
997
998 if (pack_to_stdout) {
999 if (copy_fd(ifd, 1))
1000 exit(1);
1001 close(ifd);
1002 }
1003 else {
1004 char name[PATH_MAX];
1005 snprintf(name, sizeof(name),
1006 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1007 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1008 if (ofd < 0)
1009 die("unable to open %s (%s)", name, strerror(errno));
1010 if (copy_fd(ifd, ofd))
1011 exit(1);
1012 close(ifd);
1013
1014 snprintf(name, sizeof(name),
1015 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1016 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1017 if (ofd < 0)
1018 die("unable to open %s (%s)", name, strerror(errno));
1019 if (copy_fd(ifd_ix, ofd))
1020 exit(1);
1021 close(ifd_ix);
1022 puts(sha1_to_hex(sha1));
1023 }
1024
1025 return 1;
1026 }
1027
1028 int main(int argc, char **argv)
1029 {
1030 SHA_CTX ctx;
1031 char line[PATH_MAX + 20];
1032 int window = 10, depth = 10, pack_to_stdout = 0;
1033 struct object_entry **list;
1034 int i;
1035
1036 setup_git_directory();
1037
1038 for (i = 1; i < argc; i++) {
1039 const char *arg = argv[i];
1040
1041 if (*arg == '-') {
1042 if (!strcmp("--non-empty", arg)) {
1043 non_empty = 1;
1044 continue;
1045 }
1046 if (!strcmp("--local", arg)) {
1047 local = 1;
1048 continue;
1049 }
1050 if (!strcmp("--incremental", arg)) {
1051 incremental = 1;
1052 continue;
1053 }
1054 if (!strncmp("--window=", arg, 9)) {
1055 char *end;
1056 window = strtoul(arg+9, &end, 0);
1057 if (!arg[9] || *end)
1058 usage(pack_usage);
1059 continue;
1060 }
1061 if (!strncmp("--depth=", arg, 8)) {
1062 char *end;
1063 depth = strtoul(arg+8, &end, 0);
1064 if (!arg[8] || *end)
1065 usage(pack_usage);
1066 continue;
1067 }
1068 if (!strcmp("-q", arg)) {
1069 progress = 0;
1070 continue;
1071 }
1072 if (!strcmp("--no-reuse-delta", arg)) {
1073 no_reuse_delta = 1;
1074 continue;
1075 }
1076 if (!strcmp("--stdout", arg)) {
1077 pack_to_stdout = 1;
1078 continue;
1079 }
1080 usage(pack_usage);
1081 }
1082 if (base_name)
1083 usage(pack_usage);
1084 base_name = arg;
1085 }
1086
1087 if (pack_to_stdout != !base_name)
1088 usage(pack_usage);
1089
1090 prepare_packed_git();
1091
1092 if (progress) {
1093 struct itimerval v;
1094 v.it_interval.tv_sec = 1;
1095 v.it_interval.tv_usec = 0;
1096 v.it_value = v.it_interval;
1097 signal(SIGALRM, progress_interval);
1098 setitimer(ITIMER_REAL, &v, NULL);
1099 fprintf(stderr, "Generating pack...\n");
1100 }
1101
1102 while (fgets(line, sizeof(line), stdin) != NULL) {
1103 unsigned char sha1[20];
1104
1105 if (line[0] == '-') {
1106 if (get_sha1_hex(line+1, sha1))
1107 die("expected edge sha1, got garbage:\n %s",
1108 line+1);
1109 add_preferred_base(sha1);
1110 continue;
1111 }
1112 if (get_sha1_hex(line, sha1))
1113 die("expected sha1, got garbage:\n %s", line);
1114 add_object_entry(sha1, name_hash(NULL, line+41), 0);
1115 }
1116 if (progress)
1117 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1118 sorted_by_sha = create_final_object_list();
1119 if (non_empty && !nr_result)
1120 return 0;
1121
1122 SHA1_Init(&ctx);
1123 list = sorted_by_sha;
1124 for (i = 0; i < nr_result; i++) {
1125 struct object_entry *entry = *list++;
1126 SHA1_Update(&ctx, entry->sha1, 20);
1127 }
1128 SHA1_Final(object_list_sha1, &ctx);
1129 if (progress && (nr_objects != nr_result))
1130 fprintf(stderr, "Result has %d objects.\n", nr_result);
1131
1132 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1133 ;
1134 else {
1135 if (nr_result)
1136 prepare_pack(window, depth);
1137 if (progress && pack_to_stdout) {
1138 /* the other end usually displays progress itself */
1139 struct itimerval v = {{0,},};
1140 setitimer(ITIMER_REAL, &v, NULL);
1141 signal(SIGALRM, SIG_IGN );
1142 progress_update = 0;
1143 }
1144 write_pack_file();
1145 if (!pack_to_stdout) {
1146 write_index_file();
1147 puts(sha1_to_hex(object_list_sha1));
1148 }
1149 }
1150 if (progress)
1151 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1152 nr_result, written, written_delta, reused, reused_delta);
1153 return 0;
1154 }