2320bcf3101558d603e0b8c988c9b5f4abcfc05c
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
448static unsigned name_hash(struct name_path *path, const char *name)
449{
450 struct name_path *p = path;
451 const char *n = name + strlen(name);
452 unsigned hash = 0;
453
454 if (n != name && n[-1] == '\n')
455 n--;
456 while (name <= --n) {
457 unsigned char c = *n;
458 hash = hash * 11 + c;
459 }
460 for (p = path; p; p = p->up) {
461 hash = hash * 11 + '/';
462 n = p->elem + p->len;
463 while (p->elem <= --n) {
464 unsigned char c = *n;
465 hash = hash * 11 + c;
466 }
467 }
468 return hash;
469}
470
471static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
472{
473 unsigned int idx = nr_objects;
474 struct object_entry *entry;
475 struct packed_git *p;
476 unsigned int found_offset = 0;
477 struct packed_git *found_pack = NULL;
478 int ix, status = 0;
479
480 if (!exclude) {
481 for (p = packed_git; p; p = p->next) {
482 struct pack_entry e;
483 if (find_pack_entry_one(sha1, &e, p)) {
484 if (incremental)
485 return 0;
486 if (local && !p->pack_local)
487 return 0;
488 if (!found_pack) {
489 found_offset = e.offset;
490 found_pack = e.p;
491 }
492 }
493 }
494 }
495 if ((entry = locate_object_entry(sha1)) != NULL)
496 goto already_added;
497
498 if (idx >= nr_alloc) {
499 unsigned int needed = (idx + 1024) * 3 / 2;
500 objects = xrealloc(objects, needed * sizeof(*entry));
501 nr_alloc = needed;
502 }
503 entry = objects + idx;
504 nr_objects = idx + 1;
505 memset(entry, 0, sizeof(*entry));
506 memcpy(entry->sha1, sha1, 20);
507 entry->hash = hash;
508
509 if (object_ix_hashsz * 3 <= nr_objects * 4)
510 rehash_objects();
511 else {
512 ix = locate_object_entry_hash(entry->sha1);
513 if (0 <= ix)
514 die("internal error in object hashing.");
515 object_ix[-1 - ix] = idx + 1;
516 }
517 status = 1;
518
519 already_added:
520 if (exclude)
521 entry->preferred_base = 1;
522 else {
523 if (found_pack) {
524 entry->in_pack = found_pack;
525 entry->in_pack_offset = found_offset;
526 }
527 }
528 return status;
529}
530
531static void add_pbase_tree(struct tree_desc *tree, struct name_path *up)
532{
533 while (tree->size) {
534 const unsigned char *sha1;
535 const char *name;
536 unsigned mode, hash;
537 unsigned long size;
538 char type[20];
539
540 sha1 = tree_entry_extract(tree, &name, &mode);
541 update_tree_entry(tree);
542 if (!has_sha1_file(sha1))
543 continue;
544 if (sha1_object_info(sha1, type, &size))
545 continue;
546
547 hash = name_hash(up, name);
548 if (!add_object_entry(sha1, hash, 1))
549 continue;
550
551 if (!strcmp(type, "tree")) {
552 struct tree_desc sub;
553 void *elem;
554 struct name_path me;
555
556 elem = read_sha1_file(sha1, type, &sub.size);
557 sub.buf = elem;
558 if (sub.buf) {
559 me.up = up;
560 me.elem = name;
561 me.len = strlen(name);
562 add_pbase_tree(&sub, &me);
563 free(elem);
564 }
565 }
566 }
567}
568
569static void add_preferred_base(unsigned char *sha1)
570{
571 struct tree_desc tree;
572 void *elem;
573
574 elem = read_object_with_reference(sha1, "tree", &tree.size, NULL);
575 tree.buf = elem;
576 if (!tree.buf)
577 return;
578 if (add_object_entry(sha1, name_hash(NULL, ""), 1))
579 add_pbase_tree(&tree, NULL);
580 free(elem);
581}
582
583static void check_object(struct object_entry *entry)
584{
585 char type[20];
586
587 if (entry->in_pack && !entry->preferred_base) {
588 unsigned char base[20];
589 unsigned long size;
590 struct object_entry *base_entry;
591
592 /* We want in_pack_type even if we do not reuse delta.
593 * There is no point not reusing non-delta representations.
594 */
595 check_reuse_pack_delta(entry->in_pack,
596 entry->in_pack_offset,
597 base, &size,
598 &entry->in_pack_type);
599
600 /* Check if it is delta, and the base is also an object
601 * we are going to pack. If so we will reuse the existing
602 * delta.
603 */
604 if (!no_reuse_delta &&
605 entry->in_pack_type == OBJ_DELTA &&
606 (base_entry = locate_object_entry(base)) &&
607 (!base_entry->preferred_base)) {
608
609 /* Depth value does not matter - find_deltas()
610 * will never consider reused delta as the
611 * base object to deltify other objects
612 * against, in order to avoid circular deltas.
613 */
614
615 /* uncompressed size of the delta data */
616 entry->size = entry->delta_size = size;
617 entry->delta = base_entry;
618 entry->type = OBJ_DELTA;
619
620 entry->delta_sibling = base_entry->delta_child;
621 base_entry->delta_child = entry;
622
623 return;
624 }
625 /* Otherwise we would do the usual */
626 }
627
628 if (sha1_object_info(entry->sha1, type, &entry->size))
629 die("unable to get type of object %s",
630 sha1_to_hex(entry->sha1));
631
632 if (!strcmp(type, "commit")) {
633 entry->type = OBJ_COMMIT;
634 } else if (!strcmp(type, "tree")) {
635 entry->type = OBJ_TREE;
636 } else if (!strcmp(type, "blob")) {
637 entry->type = OBJ_BLOB;
638 } else if (!strcmp(type, "tag")) {
639 entry->type = OBJ_TAG;
640 } else
641 die("unable to pack object %s of type %s",
642 sha1_to_hex(entry->sha1), type);
643}
644
645static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
646{
647 struct object_entry *child = me->delta_child;
648 unsigned int m = n;
649 while (child) {
650 unsigned int c = check_delta_limit(child, n + 1);
651 if (m < c)
652 m = c;
653 child = child->delta_sibling;
654 }
655 return m;
656}
657
658static void get_object_details(void)
659{
660 int i;
661 struct object_entry *entry;
662
663 prepare_pack_ix();
664 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
665 check_object(entry);
666
667 if (nr_objects == nr_result) {
668 /*
669 * Depth of objects that depend on the entry -- this
670 * is subtracted from depth-max to break too deep
671 * delta chain because of delta data reusing.
672 * However, we loosen this restriction when we know we
673 * are creating a thin pack -- it will have to be
674 * expanded on the other end anyway, so do not
675 * artificially cut the delta chain and let it go as
676 * deep as it wants.
677 */
678 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
679 if (!entry->delta && entry->delta_child)
680 entry->delta_limit =
681 check_delta_limit(entry, 1);
682 }
683}
684
685typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
686
687static entry_sort_t current_sort;
688
689static int sort_comparator(const void *_a, const void *_b)
690{
691 struct object_entry *a = *(struct object_entry **)_a;
692 struct object_entry *b = *(struct object_entry **)_b;
693 return current_sort(a,b);
694}
695
696static struct object_entry **create_sorted_list(entry_sort_t sort)
697{
698 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
699 int i;
700
701 for (i = 0; i < nr_objects; i++)
702 list[i] = objects + i;
703 current_sort = sort;
704 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
705 return list;
706}
707
708static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
709{
710 return memcmp(a->sha1, b->sha1, 20);
711}
712
713static struct object_entry **create_final_object_list()
714{
715 struct object_entry **list;
716 int i, j;
717
718 for (i = nr_result = 0; i < nr_objects; i++)
719 if (!objects[i].preferred_base)
720 nr_result++;
721 list = xmalloc(nr_result * sizeof(struct object_entry *));
722 for (i = j = 0; i < nr_objects; i++) {
723 if (!objects[i].preferred_base)
724 list[j++] = objects + i;
725 }
726 current_sort = sha1_sort;
727 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
728 return list;
729}
730
731static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
732{
733 if (a->type < b->type)
734 return -1;
735 if (a->type > b->type)
736 return 1;
737 if (a->hash < b->hash)
738 return -1;
739 if (a->hash > b->hash)
740 return 1;
741 if (a->preferred_base < b->preferred_base)
742 return -1;
743 if (a->preferred_base > b->preferred_base)
744 return 1;
745 if (a->size < b->size)
746 return -1;
747 if (a->size > b->size)
748 return 1;
749 return a < b ? -1 : (a > b);
750}
751
752struct unpacked {
753 struct object_entry *entry;
754 void *data;
755};
756
757/*
758 * We search for deltas _backwards_ in a list sorted by type and
759 * by size, so that we see progressively smaller and smaller files.
760 * That's because we prefer deltas to be from the bigger file
761 * to the smaller - deletes are potentially cheaper, but perhaps
762 * more importantly, the bigger file is likely the more recent
763 * one.
764 */
765static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
766{
767 struct object_entry *cur_entry = cur->entry;
768 struct object_entry *old_entry = old->entry;
769 int old_preferred = (old_entry->preferred_base ||
770 old_entry->based_on_preferred);
771 unsigned long size, oldsize, delta_size, sizediff;
772 long max_size;
773 void *delta_buf;
774
775 /* Don't bother doing diffs between different types */
776 if (cur_entry->type != old_entry->type)
777 return -1;
778
779 /* We do not compute delta to *create* objects we are not
780 * going to pack.
781 */
782 if (cur_entry->preferred_base)
783 return -1;
784
785 /* If the current object is at pack edge, take the depth the
786 * objects that depend on the current object into account --
787 * otherwise they would become too deep.
788 */
789 if (cur_entry->delta_child) {
790 if (max_depth <= cur_entry->delta_limit)
791 return 0;
792 max_depth -= cur_entry->delta_limit;
793 }
794
795 size = cur_entry->size;
796 if (size < 50)
797 return -1;
798 oldsize = old_entry->size;
799 sizediff = oldsize > size ? oldsize - size : size - oldsize;
800 if (sizediff > size / 8)
801 return -1;
802 if (old_entry->depth >= max_depth)
803 return 0;
804
805 /*
806 * NOTE!
807 *
808 * We always delta from the bigger to the smaller, since that's
809 * more space-efficient (deletes don't have to say _what_ they
810 * delete).
811 */
812 max_size = size / 2 - 20;
813 if (cur_entry->delta) {
814 if (cur_entry->based_on_preferred) {
815 if (old_preferred)
816 max_size = cur_entry->delta_size-1;
817 else
818 /* trying with non-preferred one when we
819 * already have a delta based on preferred
820 * one is pointless.
821 */
822 return -1;
823 }
824 else if (!old_preferred)
825 max_size = cur_entry->delta_size-1;
826 else
827 /* otherwise... even if delta with a
828 * preferred one produces a bigger result than
829 * what we currently have, which is based on a
830 * non-preferred one, it is OK.
831 */
832 ;
833 }
834 if (sizediff >= max_size)
835 return -1;
836 delta_buf = diff_delta(old->data, oldsize,
837 cur->data, size, &delta_size, max_size);
838 if (!delta_buf)
839 return 0;
840 cur_entry->delta = old_entry;
841 cur_entry->delta_size = delta_size;
842 cur_entry->depth = old_entry->depth + 1;
843 cur_entry->based_on_preferred = old_preferred;
844 free(delta_buf);
845 return 0;
846}
847
848static void find_deltas(struct object_entry **list, int window, int depth)
849{
850 int i, idx;
851 unsigned int array_size = window * sizeof(struct unpacked);
852 struct unpacked *array = xmalloc(array_size);
853 int eye_candy;
854
855 memset(array, 0, array_size);
856 i = nr_objects;
857 idx = 0;
858 eye_candy = i - (nr_objects / 20);
859
860 while (--i >= 0) {
861 struct object_entry *entry = list[i];
862 struct unpacked *n = array + idx;
863 unsigned long size;
864 char type[10];
865 int j;
866
867 if (progress && i <= eye_candy) {
868 eye_candy -= nr_objects / 20;
869 fputc('.', stderr);
870 }
871
872 if (entry->delta)
873 /* This happens if we decided to reuse existing
874 * delta from a pack. "!no_reuse_delta &&" is implied.
875 */
876 continue;
877
878 free(n->data);
879 n->entry = entry;
880 n->data = read_sha1_file(entry->sha1, type, &size);
881 if (size != entry->size)
882 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
883
884 j = window;
885 while (--j > 0) {
886 unsigned int other_idx = idx + j;
887 struct unpacked *m;
888 if (other_idx >= window)
889 other_idx -= window;
890 m = array + other_idx;
891 if (!m->entry)
892 break;
893 if (try_delta(n, m, depth) < 0)
894 break;
895 }
896 idx++;
897 if (idx >= window)
898 idx = 0;
899 }
900
901 for (i = 0; i < window; ++i)
902 free(array[i].data);
903 free(array);
904}
905
906static void prepare_pack(int window, int depth)
907{
908 if (progress)
909 fprintf(stderr, "Packing %d objects", nr_result);
910 get_object_details();
911 if (progress)
912 fputc('.', stderr);
913
914 sorted_by_type = create_sorted_list(type_size_sort);
915 if (window && depth)
916 find_deltas(sorted_by_type, window+1, depth);
917 if (progress)
918 fputc('\n', stderr);
919 write_pack_file();
920}
921
922static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
923{
924 static const char cache[] = "pack-cache/pack-%s.%s";
925 char *cached_pack, *cached_idx;
926 int ifd, ofd, ifd_ix = -1;
927
928 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
929 ifd = open(cached_pack, O_RDONLY);
930 if (ifd < 0)
931 return 0;
932
933 if (!pack_to_stdout) {
934 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
935 ifd_ix = open(cached_idx, O_RDONLY);
936 if (ifd_ix < 0) {
937 close(ifd);
938 return 0;
939 }
940 }
941
942 if (progress)
943 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
944 sha1_to_hex(sha1));
945
946 if (pack_to_stdout) {
947 if (copy_fd(ifd, 1))
948 exit(1);
949 close(ifd);
950 }
951 else {
952 char name[PATH_MAX];
953 snprintf(name, sizeof(name),
954 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
955 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
956 if (ofd < 0)
957 die("unable to open %s (%s)", name, strerror(errno));
958 if (copy_fd(ifd, ofd))
959 exit(1);
960 close(ifd);
961
962 snprintf(name, sizeof(name),
963 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
964 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
965 if (ofd < 0)
966 die("unable to open %s (%s)", name, strerror(errno));
967 if (copy_fd(ifd_ix, ofd))
968 exit(1);
969 close(ifd_ix);
970 puts(sha1_to_hex(sha1));
971 }
972
973 return 1;
974}
975
976int main(int argc, char **argv)
977{
978 SHA_CTX ctx;
979 char line[PATH_MAX + 20];
980 int window = 10, depth = 10, pack_to_stdout = 0;
981 struct object_entry **list;
982 int i;
983 struct timeval prev_tv;
984 int eye_candy = 0;
985 int eye_candy_incr = 500;
986
987
988 setup_git_directory();
989
990 for (i = 1; i < argc; i++) {
991 const char *arg = argv[i];
992
993 if (*arg == '-') {
994 if (!strcmp("--non-empty", arg)) {
995 non_empty = 1;
996 continue;
997 }
998 if (!strcmp("--local", arg)) {
999 local = 1;
1000 continue;
1001 }
1002 if (!strcmp("--incremental", arg)) {
1003 incremental = 1;
1004 continue;
1005 }
1006 if (!strncmp("--window=", arg, 9)) {
1007 char *end;
1008 window = strtoul(arg+9, &end, 0);
1009 if (!arg[9] || *end)
1010 usage(pack_usage);
1011 continue;
1012 }
1013 if (!strncmp("--depth=", arg, 8)) {
1014 char *end;
1015 depth = strtoul(arg+8, &end, 0);
1016 if (!arg[8] || *end)
1017 usage(pack_usage);
1018 continue;
1019 }
1020 if (!strcmp("-q", arg)) {
1021 progress = 0;
1022 continue;
1023 }
1024 if (!strcmp("--no-reuse-delta", arg)) {
1025 no_reuse_delta = 1;
1026 continue;
1027 }
1028 if (!strcmp("--stdout", arg)) {
1029 pack_to_stdout = 1;
1030 continue;
1031 }
1032 usage(pack_usage);
1033 }
1034 if (base_name)
1035 usage(pack_usage);
1036 base_name = arg;
1037 }
1038
1039 if (pack_to_stdout != !base_name)
1040 usage(pack_usage);
1041
1042 prepare_packed_git();
1043 if (progress) {
1044 fprintf(stderr, "Generating pack...\n");
1045 gettimeofday(&prev_tv, NULL);
1046 }
1047 while (fgets(line, sizeof(line), stdin) != NULL) {
1048 unsigned char sha1[20];
1049
1050 if (progress && (eye_candy <= nr_objects)) {
1051 fprintf(stderr, "Counting objects...%d\r", nr_objects);
1052 if (eye_candy && (50 <= eye_candy_incr)) {
1053 struct timeval tv;
1054 int time_diff;
1055 gettimeofday(&tv, NULL);
1056 time_diff = (tv.tv_sec - prev_tv.tv_sec);
1057 time_diff <<= 10;
1058 time_diff += (tv.tv_usec - prev_tv.tv_usec);
1059 if ((1 << 9) < time_diff)
1060 eye_candy_incr += 50;
1061 else if (50 < eye_candy_incr)
1062 eye_candy_incr -= 50;
1063 }
1064 eye_candy += eye_candy_incr;
1065 }
1066 if (line[0] == '-') {
1067 if (get_sha1_hex(line+1, sha1))
1068 die("expected edge sha1, got garbage:\n %s",
1069 line+1);
1070 add_preferred_base(sha1);
1071 continue;
1072 }
1073 if (get_sha1_hex(line, sha1))
1074 die("expected sha1, got garbage:\n %s", line);
1075 add_object_entry(sha1, name_hash(NULL, line+41), 0);
1076 }
1077 if (progress)
1078 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1079 if (non_empty && !nr_objects)
1080 return 0;
1081
1082 sorted_by_sha = create_final_object_list();
1083 SHA1_Init(&ctx);
1084 list = sorted_by_sha;
1085 for (i = 0; i < nr_result; i++) {
1086 struct object_entry *entry = *list++;
1087 SHA1_Update(&ctx, entry->sha1, 20);
1088 }
1089 SHA1_Final(object_list_sha1, &ctx);
1090 if (progress && (nr_objects != nr_result))
1091 fprintf(stderr, "Result has %d objects.\n", nr_result);
1092
1093 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1094 ;
1095 else {
1096 prepare_pack(window, depth);
1097 if (!pack_to_stdout) {
1098 write_index_file();
1099 puts(sha1_to_hex(object_list_sha1));
1100 }
1101 }
1102 if (progress)
1103 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1104 nr_result, written, written_delta, reused, reused_delta);
1105 return 0;
1106}