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