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
442static int add_object_entry(const unsigned char *sha1, const char *name, int exclude)
443{
444 unsigned int hash = 0;
445 unsigned int idx = nr_objects;
446 struct object_entry *entry;
447 struct packed_git *p;
448 unsigned int found_offset = 0;
449 struct packed_git *found_pack = NULL;
450 int ix;
451
452 if (!exclude) {
453 for (p = packed_git; p; p = p->next) {
454 struct pack_entry e;
455 if (find_pack_entry_one(sha1, &e, p)) {
456 if (incremental)
457 return 0;
458 if (local && !p->pack_local)
459 return 0;
460 if (!found_pack) {
461 found_offset = e.offset;
462 found_pack = e.p;
463 }
464 }
465 }
466 }
467 if ((entry = locate_object_entry(sha1)) != NULL)
468 goto already_added;
469
470 while (*name) {
471 unsigned char c = *name++;
472 if (isspace(c))
473 continue;
474 hash = hash * 11 + c;
475 }
476
477 if (idx >= nr_alloc) {
478 unsigned int needed = (idx + 1024) * 3 / 2;
479 objects = xrealloc(objects, needed * sizeof(*entry));
480 nr_alloc = needed;
481 }
482 entry = objects + idx;
483 nr_objects = idx + 1;
484 memset(entry, 0, sizeof(*entry));
485 memcpy(entry->sha1, sha1, 20);
486 entry->hash = hash;
487
488 if (object_ix_hashsz * 3 <= nr_objects * 4)
489 rehash_objects();
490 else {
491 ix = locate_object_entry_hash(entry->sha1);
492 if (0 <= ix)
493 die("internal error in object hashing.");
494 object_ix[-1 - ix] = idx + 1;
495 }
496
497 already_added:
498 if (exclude)
499 entry->preferred_base = 1;
500 else {
501 if (found_pack) {
502 entry->in_pack = found_pack;
503 entry->in_pack_offset = found_offset;
504 }
505 }
506 return 1;
507}
508
509static void add_pbase_tree(struct tree_desc *tree)
510{
511 while (tree->size) {
512 const unsigned char *sha1;
513 const char *name;
514 unsigned mode;
515 unsigned long size;
516 char type[20];
517
518 sha1 = tree_entry_extract(tree, &name, &mode);
519 update_tree_entry(tree);
520 if (!has_sha1_file(sha1))
521 continue;
522 if (sha1_object_info(sha1, type, &size))
523 continue;
524 add_object_entry(sha1, name, 1);
525 if (!strcmp(type, "tree")) {
526 struct tree_desc sub;
527 void *elem;
528 elem = read_sha1_file(sha1, type, &sub.size);
529 sub.buf = elem;
530 if (sub.buf) {
531 add_pbase_tree(&sub);
532 free(elem);
533 }
534 }
535 }
536}
537
538static void add_preferred_base(unsigned char *sha1)
539{
540 struct tree_desc tree;
541 void *elem;
542 elem = read_object_with_reference(sha1, "tree", &tree.size, NULL);
543 tree.buf = elem;
544 if (!tree.buf)
545 return;
546 add_object_entry(sha1, "", 1);
547 add_pbase_tree(&tree);
548 free(elem);
549}
550
551static void check_object(struct object_entry *entry)
552{
553 char type[20];
554
555 if (entry->in_pack && !entry->preferred_base) {
556 unsigned char base[20];
557 unsigned long size;
558 struct object_entry *base_entry;
559
560 /* We want in_pack_type even if we do not reuse delta.
561 * There is no point not reusing non-delta representations.
562 */
563 check_reuse_pack_delta(entry->in_pack,
564 entry->in_pack_offset,
565 base, &size,
566 &entry->in_pack_type);
567
568 /* Check if it is delta, and the base is also an object
569 * we are going to pack. If so we will reuse the existing
570 * delta.
571 */
572 if (!no_reuse_delta &&
573 entry->in_pack_type == OBJ_DELTA &&
574 (base_entry = locate_object_entry(base)) &&
575 (!base_entry->preferred_base)) {
576
577 /* Depth value does not matter - find_deltas()
578 * will never consider reused delta as the
579 * base object to deltify other objects
580 * against, in order to avoid circular deltas.
581 */
582
583 /* uncompressed size of the delta data */
584 entry->size = entry->delta_size = size;
585 entry->delta = base_entry;
586 entry->type = OBJ_DELTA;
587
588 entry->delta_sibling = base_entry->delta_child;
589 base_entry->delta_child = entry;
590
591 return;
592 }
593 /* Otherwise we would do the usual */
594 }
595
596 if (sha1_object_info(entry->sha1, type, &entry->size))
597 die("unable to get type of object %s",
598 sha1_to_hex(entry->sha1));
599
600 if (!strcmp(type, "commit")) {
601 entry->type = OBJ_COMMIT;
602 } else if (!strcmp(type, "tree")) {
603 entry->type = OBJ_TREE;
604 } else if (!strcmp(type, "blob")) {
605 entry->type = OBJ_BLOB;
606 } else if (!strcmp(type, "tag")) {
607 entry->type = OBJ_TAG;
608 } else
609 die("unable to pack object %s of type %s",
610 sha1_to_hex(entry->sha1), type);
611}
612
613static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
614{
615 struct object_entry *child = me->delta_child;
616 unsigned int m = n;
617 while (child) {
618 unsigned int c = check_delta_limit(child, n + 1);
619 if (m < c)
620 m = c;
621 child = child->delta_sibling;
622 }
623 return m;
624}
625
626static void get_object_details(void)
627{
628 int i;
629 struct object_entry *entry;
630
631 prepare_pack_ix();
632 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
633 check_object(entry);
634 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
635 if (!entry->delta && entry->delta_child)
636 entry->delta_limit =
637 check_delta_limit(entry, 1);
638}
639
640typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
641
642static entry_sort_t current_sort;
643
644static int sort_comparator(const void *_a, const void *_b)
645{
646 struct object_entry *a = *(struct object_entry **)_a;
647 struct object_entry *b = *(struct object_entry **)_b;
648 return current_sort(a,b);
649}
650
651static struct object_entry **create_sorted_list(entry_sort_t sort)
652{
653 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
654 int i;
655
656 for (i = 0; i < nr_objects; i++)
657 list[i] = objects + i;
658 current_sort = sort;
659 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
660 return list;
661}
662
663static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
664{
665 return memcmp(a->sha1, b->sha1, 20);
666}
667
668static struct object_entry **create_final_object_list()
669{
670 struct object_entry **list;
671 int i, j;
672
673 for (i = nr_result = 0; i < nr_objects; i++)
674 if (!objects[i].preferred_base)
675 nr_result++;
676 list = xmalloc(nr_result * sizeof(struct object_entry *));
677 for (i = j = 0; i < nr_objects; i++) {
678 if (!objects[i].preferred_base)
679 list[j++] = objects + i;
680 }
681 current_sort = sha1_sort;
682 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
683 return list;
684}
685
686static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
687{
688 if (a->type < b->type)
689 return -1;
690 if (a->type > b->type)
691 return 1;
692 if (a->hash < b->hash)
693 return -1;
694 if (a->hash > b->hash)
695 return 1;
696 if (a->preferred_base < b->preferred_base)
697 return -1;
698 if (a->preferred_base > b->preferred_base)
699 return 1;
700 if (a->size < b->size)
701 return -1;
702 if (a->size > b->size)
703 return 1;
704 return a < b ? -1 : (a > b);
705}
706
707struct unpacked {
708 struct object_entry *entry;
709 void *data;
710};
711
712/*
713 * We search for deltas _backwards_ in a list sorted by type and
714 * by size, so that we see progressively smaller and smaller files.
715 * That's because we prefer deltas to be from the bigger file
716 * to the smaller - deletes are potentially cheaper, but perhaps
717 * more importantly, the bigger file is likely the more recent
718 * one.
719 */
720static int try_delta(struct unpacked *cur, struct unpacked *old, unsigned max_depth)
721{
722 struct object_entry *cur_entry = cur->entry;
723 struct object_entry *old_entry = old->entry;
724 int old_preferred = (old_entry->preferred_base ||
725 old_entry->based_on_preferred);
726 unsigned long size, oldsize, delta_size, sizediff;
727 long max_size;
728 void *delta_buf;
729
730 /* Don't bother doing diffs between different types */
731 if (cur_entry->type != old_entry->type)
732 return -1;
733
734 /* We do not compute delta to *create* objects we are not
735 * going to pack.
736 */
737 if (cur_entry->preferred_base)
738 return -1;
739
740 /* If the current object is at pack edge, take the depth the
741 * objects that depend on the current object into account --
742 * otherwise they would become too deep.
743 */
744 if (cur_entry->delta_child) {
745 if (max_depth <= cur_entry->delta_limit)
746 return 0;
747 max_depth -= cur_entry->delta_limit;
748 }
749
750 size = cur_entry->size;
751 oldsize = old_entry->size;
752 sizediff = oldsize > size ? oldsize - size : size - oldsize;
753
754 if (size < 50)
755 return -1;
756 if (old_entry->depth >= max_depth)
757 return 0;
758
759 /*
760 * NOTE!
761 *
762 * We always delta from the bigger to the smaller, since that's
763 * more space-efficient (deletes don't have to say _what_ they
764 * delete).
765 */
766 max_size = size / 2 - 20;
767 if (cur_entry->delta) {
768 if (cur_entry->based_on_preferred) {
769 if (old_preferred)
770 max_size = cur_entry->delta_size-1;
771 else
772 /* trying with non-preferred one when we
773 * already have a delta based on preferred
774 * one is pointless.
775 */
776 return 0;
777 }
778 else if (!old_preferred)
779 max_size = cur_entry->delta_size-1;
780 else
781 /* otherwise... even if delta with a
782 * preferred one produces a bigger result than
783 * what we currently have, which is based on a
784 * non-preferred one, it is OK.
785 */
786 ;
787 }
788 if (sizediff >= max_size)
789 return -1;
790 delta_buf = diff_delta(old->data, oldsize,
791 cur->data, size, &delta_size, max_size);
792 if (!delta_buf)
793 return 0;
794 cur_entry->delta = old_entry;
795 cur_entry->delta_size = delta_size;
796 cur_entry->depth = old_entry->depth + 1;
797 cur_entry->based_on_preferred = old_preferred;
798 free(delta_buf);
799 return 0;
800}
801
802static void find_deltas(struct object_entry **list, int window, int depth)
803{
804 int i, idx;
805 unsigned int array_size = window * sizeof(struct unpacked);
806 struct unpacked *array = xmalloc(array_size);
807 int eye_candy;
808
809 memset(array, 0, array_size);
810 i = nr_objects;
811 idx = 0;
812 eye_candy = i - (nr_objects / 20);
813
814 while (--i >= 0) {
815 struct object_entry *entry = list[i];
816 struct unpacked *n = array + idx;
817 unsigned long size;
818 char type[10];
819 int j;
820
821 if (progress && i <= eye_candy) {
822 eye_candy -= nr_objects / 20;
823 fputc('.', stderr);
824 }
825
826 if (entry->delta)
827 /* This happens if we decided to reuse existing
828 * delta from a pack. "!no_reuse_delta &&" is implied.
829 */
830 continue;
831
832 free(n->data);
833 n->entry = entry;
834 n->data = read_sha1_file(entry->sha1, type, &size);
835 if (size != entry->size)
836 die("object %s inconsistent object length (%lu vs %lu)", sha1_to_hex(entry->sha1), size, entry->size);
837
838 j = window;
839 while (--j > 0) {
840 unsigned int other_idx = idx + j;
841 struct unpacked *m;
842 if (other_idx >= window)
843 other_idx -= window;
844 m = array + other_idx;
845 if (!m->entry)
846 break;
847 if (try_delta(n, m, depth) < 0)
848 break;
849 }
850 idx++;
851 if (idx >= window)
852 idx = 0;
853 }
854
855 for (i = 0; i < window; ++i)
856 free(array[i].data);
857 free(array);
858}
859
860static void prepare_pack(int window, int depth)
861{
862 if (progress)
863 fprintf(stderr, "Packing %d objects", nr_result);
864 get_object_details();
865 if (progress)
866 fputc('.', stderr);
867
868 sorted_by_type = create_sorted_list(type_size_sort);
869 if (window && depth)
870 find_deltas(sorted_by_type, window+1, depth);
871 if (progress)
872 fputc('\n', stderr);
873 write_pack_file();
874}
875
876static int reuse_cached_pack(unsigned char *sha1, int pack_to_stdout)
877{
878 static const char cache[] = "pack-cache/pack-%s.%s";
879 char *cached_pack, *cached_idx;
880 int ifd, ofd, ifd_ix = -1;
881
882 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
883 ifd = open(cached_pack, O_RDONLY);
884 if (ifd < 0)
885 return 0;
886
887 if (!pack_to_stdout) {
888 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
889 ifd_ix = open(cached_idx, O_RDONLY);
890 if (ifd_ix < 0) {
891 close(ifd);
892 return 0;
893 }
894 }
895
896 if (progress)
897 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
898 sha1_to_hex(sha1));
899
900 if (pack_to_stdout) {
901 if (copy_fd(ifd, 1))
902 exit(1);
903 close(ifd);
904 }
905 else {
906 char name[PATH_MAX];
907 snprintf(name, sizeof(name),
908 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
909 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
910 if (ofd < 0)
911 die("unable to open %s (%s)", name, strerror(errno));
912 if (copy_fd(ifd, ofd))
913 exit(1);
914 close(ifd);
915
916 snprintf(name, sizeof(name),
917 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
918 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
919 if (ofd < 0)
920 die("unable to open %s (%s)", name, strerror(errno));
921 if (copy_fd(ifd_ix, ofd))
922 exit(1);
923 close(ifd_ix);
924 puts(sha1_to_hex(sha1));
925 }
926
927 return 1;
928}
929
930int main(int argc, char **argv)
931{
932 SHA_CTX ctx;
933 char line[PATH_MAX + 20];
934 int window = 10, depth = 10, pack_to_stdout = 0;
935 struct object_entry **list;
936 int i;
937 struct timeval prev_tv;
938 int eye_candy = 0;
939 int eye_candy_incr = 500;
940
941
942 setup_git_directory();
943
944 for (i = 1; i < argc; i++) {
945 const char *arg = argv[i];
946
947 if (*arg == '-') {
948 if (!strcmp("--non-empty", arg)) {
949 non_empty = 1;
950 continue;
951 }
952 if (!strcmp("--local", arg)) {
953 local = 1;
954 continue;
955 }
956 if (!strcmp("--incremental", arg)) {
957 incremental = 1;
958 continue;
959 }
960 if (!strncmp("--window=", arg, 9)) {
961 char *end;
962 window = strtoul(arg+9, &end, 0);
963 if (!arg[9] || *end)
964 usage(pack_usage);
965 continue;
966 }
967 if (!strncmp("--depth=", arg, 8)) {
968 char *end;
969 depth = strtoul(arg+8, &end, 0);
970 if (!arg[8] || *end)
971 usage(pack_usage);
972 continue;
973 }
974 if (!strcmp("-q", arg)) {
975 progress = 0;
976 continue;
977 }
978 if (!strcmp("--no-reuse-delta", arg)) {
979 no_reuse_delta = 1;
980 continue;
981 }
982 if (!strcmp("--stdout", arg)) {
983 pack_to_stdout = 1;
984 continue;
985 }
986 usage(pack_usage);
987 }
988 if (base_name)
989 usage(pack_usage);
990 base_name = arg;
991 }
992
993 if (pack_to_stdout != !base_name)
994 usage(pack_usage);
995
996 prepare_packed_git();
997 if (progress) {
998 fprintf(stderr, "Generating pack...\n");
999 gettimeofday(&prev_tv, NULL);
1000 }
1001 while (fgets(line, sizeof(line), stdin) != NULL) {
1002 unsigned char sha1[20];
1003
1004 if (progress && (eye_candy <= nr_objects)) {
1005 fprintf(stderr, "Counting objects...%d\r", nr_objects);
1006 if (eye_candy && (50 <= eye_candy_incr)) {
1007 struct timeval tv;
1008 int time_diff;
1009 gettimeofday(&tv, NULL);
1010 time_diff = (tv.tv_sec - prev_tv.tv_sec);
1011 time_diff <<= 10;
1012 time_diff += (tv.tv_usec - prev_tv.tv_usec);
1013 if ((1 << 9) < time_diff)
1014 eye_candy_incr += 50;
1015 else if (50 < eye_candy_incr)
1016 eye_candy_incr -= 50;
1017 }
1018 eye_candy += eye_candy_incr;
1019 }
1020 if (line[0] == '-') {
1021 if (get_sha1_hex(line+1, sha1))
1022 die("expected edge sha1, got garbage:\n %s",
1023 line+1);
1024 add_preferred_base(sha1);
1025 continue;
1026 }
1027 if (get_sha1_hex(line, sha1))
1028 die("expected sha1, got garbage:\n %s", line);
1029 add_object_entry(sha1, line+40, 0);
1030 }
1031 if (progress)
1032 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1033 if (non_empty && !nr_objects)
1034 return 0;
1035
1036 sorted_by_sha = create_final_object_list();
1037 SHA1_Init(&ctx);
1038 list = sorted_by_sha;
1039 for (i = 0; i < nr_result; i++) {
1040 struct object_entry *entry = *list++;
1041 SHA1_Update(&ctx, entry->sha1, 20);
1042 }
1043 SHA1_Final(object_list_sha1, &ctx);
1044 if (progress && (nr_objects != nr_result))
1045 fprintf(stderr, "Result has %d objects.\n", nr_result);
1046
1047 if (reuse_cached_pack(object_list_sha1, pack_to_stdout))
1048 ;
1049 else {
1050 prepare_pack(window, depth);
1051 if (!pack_to_stdout) {
1052 write_index_file();
1053 puts(sha1_to_hex(object_list_sha1));
1054 }
1055 }
1056 if (progress)
1057 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1058 nr_result, written, written_delta, reused, reused_delta);
1059 return 0;
1060}