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