GIT 1.4.4.3
[git/git.git] / builtin-pack-objects.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "object.h"
4 #include "blob.h"
5 #include "commit.h"
6 #include "tag.h"
7 #include "tree.h"
8 #include "delta.h"
9 #include "pack.h"
10 #include "csum-file.h"
11 #include "tree-walk.h"
12 #include "diff.h"
13 #include "revision.h"
14 #include "list-objects.h"
15 #include <sys/time.h>
16 #include <signal.h>
17
18 static const char pack_usage[] = "\
19 git-pack-objects [{ -q | --progress | --all-progress }] \n\
20 [--local] [--incremental] [--window=N] [--depth=N] \n\
21 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
22 [--revs [--unpacked | --all]*] [--stdout | base-name] \n\
23 [<ref-list | <object-list]";
24
25 struct object_entry {
26 unsigned char sha1[20];
27 unsigned long size; /* uncompressed size */
28 unsigned long offset; /* offset into the final pack file;
29 * nonzero if already written.
30 */
31 unsigned int depth; /* delta depth */
32 unsigned int delta_limit; /* base adjustment for in-pack delta */
33 unsigned int hash; /* name hint hash */
34 enum object_type type;
35 enum object_type in_pack_type; /* could be delta */
36 unsigned long delta_size; /* delta data size (uncompressed) */
37 #define in_pack_header_size delta_size /* only when reusing pack data */
38 struct object_entry *delta; /* delta base object */
39 struct packed_git *in_pack; /* already in pack */
40 unsigned int in_pack_offset;
41 struct object_entry *delta_child; /* deltified objects who bases me */
42 struct object_entry *delta_sibling; /* other deltified objects who
43 * uses the same base as me
44 */
45 int preferred_base; /* we do not pack this, but is encouraged to
46 * be used as the base objectto delta huge
47 * objects against.
48 */
49 };
50
51 /*
52 * Objects we are going to pack are collected in objects array (dynamically
53 * expanded). nr_objects & nr_alloc controls this array. They are stored
54 * in the order we see -- typically rev-list --objects order that gives us
55 * nice "minimum seek" order.
56 *
57 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
58 * elements in the objects array. The former is used to build the pack
59 * index (lists object names in the ascending order to help offset lookup),
60 * and the latter is used to group similar things together by try_delta()
61 * heuristics.
62 */
63
64 static unsigned char object_list_sha1[20];
65 static int non_empty;
66 static int no_reuse_delta;
67 static int local;
68 static int incremental;
69 static int allow_ofs_delta;
70
71 static struct object_entry **sorted_by_sha, **sorted_by_type;
72 static struct object_entry *objects;
73 static int nr_objects, nr_alloc, nr_result;
74 static const char *base_name;
75 static unsigned char pack_file_sha1[20];
76 static int progress = 1;
77 static volatile sig_atomic_t progress_update;
78 static int window = 10;
79 static int pack_to_stdout;
80 static int num_preferred_base;
81
82 /*
83 * The object names in objects array are hashed with this hashtable,
84 * to help looking up the entry by object name. Binary search from
85 * sorted_by_sha is also possible but this was easier to code and faster.
86 * This hashtable is built after all the objects are seen.
87 */
88 static int *object_ix;
89 static int object_ix_hashsz;
90
91 /*
92 * Pack index for existing packs give us easy access to the offsets into
93 * corresponding pack file where each object's data starts, but the entries
94 * do not store the size of the compressed representation (uncompressed
95 * size is easily available by examining the pack entry header). It is
96 * also rather expensive to find the sha1 for an object given its offset.
97 *
98 * We build a hashtable of existing packs (pack_revindex), and keep reverse
99 * index here -- pack index file is sorted by object name mapping to offset;
100 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
101 * ordered by offset, so if you know the offset of an object, next offset
102 * is where its packed representation ends and the index_nr can be used to
103 * get the object sha1 from the main index.
104 */
105 struct revindex_entry {
106 unsigned int offset;
107 unsigned int nr;
108 };
109 struct pack_revindex {
110 struct packed_git *p;
111 struct revindex_entry *revindex;
112 };
113 static struct pack_revindex *pack_revindex;
114 static int pack_revindex_hashsz;
115
116 /*
117 * stats
118 */
119 static int written;
120 static int written_delta;
121 static int reused;
122 static int reused_delta;
123
124 static int pack_revindex_ix(struct packed_git *p)
125 {
126 unsigned long ui = (unsigned long)p;
127 int i;
128
129 ui = ui ^ (ui >> 16); /* defeat structure alignment */
130 i = (int)(ui % pack_revindex_hashsz);
131 while (pack_revindex[i].p) {
132 if (pack_revindex[i].p == p)
133 return i;
134 if (++i == pack_revindex_hashsz)
135 i = 0;
136 }
137 return -1 - i;
138 }
139
140 static void prepare_pack_ix(void)
141 {
142 int num;
143 struct packed_git *p;
144 for (num = 0, p = packed_git; p; p = p->next)
145 num++;
146 if (!num)
147 return;
148 pack_revindex_hashsz = num * 11;
149 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
150 for (p = packed_git; p; p = p->next) {
151 num = pack_revindex_ix(p);
152 num = - 1 - num;
153 pack_revindex[num].p = p;
154 }
155 /* revindex elements are lazily initialized */
156 }
157
158 static int cmp_offset(const void *a_, const void *b_)
159 {
160 const struct revindex_entry *a = a_;
161 const struct revindex_entry *b = b_;
162 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
163 }
164
165 /*
166 * Ordered list of offsets of objects in the pack.
167 */
168 static void prepare_pack_revindex(struct pack_revindex *rix)
169 {
170 struct packed_git *p = rix->p;
171 int num_ent = num_packed_objects(p);
172 int i;
173 void *index = p->index_base + 256;
174
175 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
176 for (i = 0; i < num_ent; i++) {
177 unsigned int hl = *((unsigned int *)((char *) index + 24*i));
178 rix->revindex[i].offset = ntohl(hl);
179 rix->revindex[i].nr = i;
180 }
181 /* This knows the pack format -- the 20-byte trailer
182 * follows immediately after the last object data.
183 */
184 rix->revindex[num_ent].offset = p->pack_size - 20;
185 rix->revindex[num_ent].nr = -1;
186 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
187 }
188
189 static struct revindex_entry * find_packed_object(struct packed_git *p,
190 unsigned int ofs)
191 {
192 int num;
193 int lo, hi;
194 struct pack_revindex *rix;
195 struct revindex_entry *revindex;
196 num = pack_revindex_ix(p);
197 if (num < 0)
198 die("internal error: pack revindex uninitialized");
199 rix = &pack_revindex[num];
200 if (!rix->revindex)
201 prepare_pack_revindex(rix);
202 revindex = rix->revindex;
203 lo = 0;
204 hi = num_packed_objects(p) + 1;
205 do {
206 int mi = (lo + hi) / 2;
207 if (revindex[mi].offset == ofs) {
208 return revindex + mi;
209 }
210 else if (ofs < revindex[mi].offset)
211 hi = mi;
212 else
213 lo = mi + 1;
214 } while (lo < hi);
215 die("internal error: pack revindex corrupt");
216 }
217
218 static unsigned long find_packed_object_size(struct packed_git *p,
219 unsigned long ofs)
220 {
221 struct revindex_entry *entry = find_packed_object(p, ofs);
222 return entry[1].offset - ofs;
223 }
224
225 static unsigned char *find_packed_object_name(struct packed_git *p,
226 unsigned long ofs)
227 {
228 struct revindex_entry *entry = find_packed_object(p, ofs);
229 return (unsigned char *)(p->index_base + 256) + 24 * entry->nr + 4;
230 }
231
232 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
233 {
234 unsigned long othersize, delta_size;
235 char type[10];
236 void *otherbuf = read_sha1_file(entry->delta->sha1, type, &othersize);
237 void *delta_buf;
238
239 if (!otherbuf)
240 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
241 delta_buf = diff_delta(otherbuf, othersize,
242 buf, size, &delta_size, 0);
243 if (!delta_buf || delta_size != entry->delta_size)
244 die("delta size changed");
245 free(buf);
246 free(otherbuf);
247 return delta_buf;
248 }
249
250 /*
251 * The per-object header is a pretty dense thing, which is
252 * - first byte: low four bits are "size", then three bits of "type",
253 * and the high bit is "size continues".
254 * - each byte afterwards: low seven bits are size continuation,
255 * with the high bit being "size continues"
256 */
257 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
258 {
259 int n = 1;
260 unsigned char c;
261
262 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
263 die("bad type %d", type);
264
265 c = (type << 4) | (size & 15);
266 size >>= 4;
267 while (size) {
268 *hdr++ = c | 0x80;
269 c = size & 0x7f;
270 size >>= 7;
271 n++;
272 }
273 *hdr = c;
274 return n;
275 }
276
277 /*
278 * we are going to reuse the existing object data as is. make
279 * sure it is not corrupt.
280 */
281 static int check_inflate(unsigned char *data, unsigned long len, unsigned long expect)
282 {
283 z_stream stream;
284 unsigned char fakebuf[4096];
285 int st;
286
287 memset(&stream, 0, sizeof(stream));
288 stream.next_in = data;
289 stream.avail_in = len;
290 stream.next_out = fakebuf;
291 stream.avail_out = sizeof(fakebuf);
292 inflateInit(&stream);
293
294 while (1) {
295 st = inflate(&stream, Z_FINISH);
296 if (st == Z_STREAM_END || st == Z_OK) {
297 st = (stream.total_out == expect &&
298 stream.total_in == len) ? 0 : -1;
299 break;
300 }
301 if (st != Z_BUF_ERROR) {
302 st = -1;
303 break;
304 }
305 stream.next_out = fakebuf;
306 stream.avail_out = sizeof(fakebuf);
307 }
308 inflateEnd(&stream);
309 return st;
310 }
311
312 static int revalidate_loose_object(struct object_entry *entry,
313 unsigned char *map,
314 unsigned long mapsize)
315 {
316 /* we already know this is a loose object with new type header. */
317 enum object_type type;
318 unsigned long size, used;
319
320 if (pack_to_stdout)
321 return 0;
322
323 used = unpack_object_header_gently(map, mapsize, &type, &size);
324 if (!used)
325 return -1;
326 map += used;
327 mapsize -= used;
328 return check_inflate(map, mapsize, size);
329 }
330
331 static unsigned long write_object(struct sha1file *f,
332 struct object_entry *entry)
333 {
334 unsigned long size;
335 char type[10];
336 void *buf;
337 unsigned char header[10];
338 unsigned hdrlen, datalen;
339 enum object_type obj_type;
340 int to_reuse = 0;
341
342 obj_type = entry->type;
343 if (! entry->in_pack)
344 to_reuse = 0; /* can't reuse what we don't have */
345 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
346 to_reuse = 1; /* check_object() decided it for us */
347 else if (obj_type != entry->in_pack_type)
348 to_reuse = 0; /* pack has delta which is unusable */
349 else if (entry->delta)
350 to_reuse = 0; /* we want to pack afresh */
351 else
352 to_reuse = 1; /* we have it in-pack undeltified,
353 * and we do not need to deltify it.
354 */
355
356 if (!entry->in_pack && !entry->delta) {
357 unsigned char *map;
358 unsigned long mapsize;
359 map = map_sha1_file(entry->sha1, &mapsize);
360 if (map && !legacy_loose_object(map)) {
361 /* We can copy straight into the pack file */
362 if (revalidate_loose_object(entry, map, mapsize))
363 die("corrupt loose object %s",
364 sha1_to_hex(entry->sha1));
365 sha1write(f, map, mapsize);
366 munmap(map, mapsize);
367 written++;
368 reused++;
369 return mapsize;
370 }
371 if (map)
372 munmap(map, mapsize);
373 }
374
375 if (!to_reuse) {
376 buf = read_sha1_file(entry->sha1, type, &size);
377 if (!buf)
378 die("unable to read %s", sha1_to_hex(entry->sha1));
379 if (size != entry->size)
380 die("object %s size inconsistency (%lu vs %lu)",
381 sha1_to_hex(entry->sha1), size, entry->size);
382 if (entry->delta) {
383 buf = delta_against(buf, size, entry);
384 size = entry->delta_size;
385 obj_type = (allow_ofs_delta && entry->delta->offset) ?
386 OBJ_OFS_DELTA : OBJ_REF_DELTA;
387 }
388 /*
389 * The object header is a byte of 'type' followed by zero or
390 * more bytes of length.
391 */
392 hdrlen = encode_header(obj_type, size, header);
393 sha1write(f, header, hdrlen);
394
395 if (obj_type == OBJ_OFS_DELTA) {
396 /*
397 * Deltas with relative base contain an additional
398 * encoding of the relative offset for the delta
399 * base from this object's position in the pack.
400 */
401 unsigned long ofs = entry->offset - entry->delta->offset;
402 unsigned pos = sizeof(header) - 1;
403 header[pos] = ofs & 127;
404 while (ofs >>= 7)
405 header[--pos] = 128 | (--ofs & 127);
406 sha1write(f, header + pos, sizeof(header) - pos);
407 hdrlen += sizeof(header) - pos;
408 } else if (obj_type == OBJ_REF_DELTA) {
409 /*
410 * Deltas with a base reference contain
411 * an additional 20 bytes for the base sha1.
412 */
413 sha1write(f, entry->delta->sha1, 20);
414 hdrlen += 20;
415 }
416 datalen = sha1write_compressed(f, buf, size);
417 free(buf);
418 }
419 else {
420 struct packed_git *p = entry->in_pack;
421
422 if (entry->delta) {
423 obj_type = (allow_ofs_delta && entry->delta->offset) ?
424 OBJ_OFS_DELTA : OBJ_REF_DELTA;
425 reused_delta++;
426 }
427 hdrlen = encode_header(obj_type, entry->size, header);
428 sha1write(f, header, hdrlen);
429 if (obj_type == OBJ_OFS_DELTA) {
430 unsigned long ofs = entry->offset - entry->delta->offset;
431 unsigned pos = sizeof(header) - 1;
432 header[pos] = ofs & 127;
433 while (ofs >>= 7)
434 header[--pos] = 128 | (--ofs & 127);
435 sha1write(f, header + pos, sizeof(header) - pos);
436 hdrlen += sizeof(header) - pos;
437 } else if (obj_type == OBJ_REF_DELTA) {
438 sha1write(f, entry->delta->sha1, 20);
439 hdrlen += 20;
440 }
441
442 use_packed_git(p);
443 buf = (char *) p->pack_base
444 + entry->in_pack_offset
445 + entry->in_pack_header_size;
446 datalen = find_packed_object_size(p, entry->in_pack_offset)
447 - entry->in_pack_header_size;
448 if (!pack_to_stdout && check_inflate(buf, datalen, entry->size))
449 die("corrupt delta in pack %s", sha1_to_hex(entry->sha1));
450 sha1write(f, buf, datalen);
451 unuse_packed_git(p);
452 reused++;
453 }
454 if (entry->delta)
455 written_delta++;
456 written++;
457 return hdrlen + datalen;
458 }
459
460 static unsigned long write_one(struct sha1file *f,
461 struct object_entry *e,
462 unsigned long offset)
463 {
464 if (e->offset || e->preferred_base)
465 /* offset starts from header size and cannot be zero
466 * if it is written already.
467 */
468 return offset;
469 /* if we are deltified, write out its base object first. */
470 if (e->delta)
471 offset = write_one(f, e->delta, offset);
472 e->offset = offset;
473 return offset + write_object(f, e);
474 }
475
476 static void write_pack_file(void)
477 {
478 int i;
479 struct sha1file *f;
480 unsigned long offset;
481 struct pack_header hdr;
482 unsigned last_percent = 999;
483 int do_progress = progress;
484
485 if (!base_name) {
486 f = sha1fd(1, "<stdout>");
487 do_progress >>= 1;
488 }
489 else
490 f = sha1create("%s-%s.%s", base_name,
491 sha1_to_hex(object_list_sha1), "pack");
492 if (do_progress)
493 fprintf(stderr, "Writing %d objects.\n", nr_result);
494
495 hdr.hdr_signature = htonl(PACK_SIGNATURE);
496 hdr.hdr_version = htonl(PACK_VERSION);
497 hdr.hdr_entries = htonl(nr_result);
498 sha1write(f, &hdr, sizeof(hdr));
499 offset = sizeof(hdr);
500 if (!nr_result)
501 goto done;
502 for (i = 0; i < nr_objects; i++) {
503 offset = write_one(f, objects + i, offset);
504 if (do_progress) {
505 unsigned percent = written * 100 / nr_result;
506 if (progress_update || percent != last_percent) {
507 fprintf(stderr, "%4u%% (%u/%u) done\r",
508 percent, written, nr_result);
509 progress_update = 0;
510 last_percent = percent;
511 }
512 }
513 }
514 if (do_progress)
515 fputc('\n', stderr);
516 done:
517 sha1close(f, pack_file_sha1, 1);
518 }
519
520 static void write_index_file(void)
521 {
522 int i;
523 struct sha1file *f = sha1create("%s-%s.%s", base_name,
524 sha1_to_hex(object_list_sha1), "idx");
525 struct object_entry **list = sorted_by_sha;
526 struct object_entry **last = list + nr_result;
527 unsigned int array[256];
528
529 /*
530 * Write the first-level table (the list is sorted,
531 * but we use a 256-entry lookup to be able to avoid
532 * having to do eight extra binary search iterations).
533 */
534 for (i = 0; i < 256; i++) {
535 struct object_entry **next = list;
536 while (next < last) {
537 struct object_entry *entry = *next;
538 if (entry->sha1[0] != i)
539 break;
540 next++;
541 }
542 array[i] = htonl(next - sorted_by_sha);
543 list = next;
544 }
545 sha1write(f, array, 256 * sizeof(int));
546
547 /*
548 * Write the actual SHA1 entries..
549 */
550 list = sorted_by_sha;
551 for (i = 0; i < nr_result; i++) {
552 struct object_entry *entry = *list++;
553 unsigned int offset = htonl(entry->offset);
554 sha1write(f, &offset, 4);
555 sha1write(f, entry->sha1, 20);
556 }
557 sha1write(f, pack_file_sha1, 20);
558 sha1close(f, NULL, 1);
559 }
560
561 static int locate_object_entry_hash(const unsigned char *sha1)
562 {
563 int i;
564 unsigned int ui;
565 memcpy(&ui, sha1, sizeof(unsigned int));
566 i = ui % object_ix_hashsz;
567 while (0 < object_ix[i]) {
568 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
569 return i;
570 if (++i == object_ix_hashsz)
571 i = 0;
572 }
573 return -1 - i;
574 }
575
576 static struct object_entry *locate_object_entry(const unsigned char *sha1)
577 {
578 int i;
579
580 if (!object_ix_hashsz)
581 return NULL;
582
583 i = locate_object_entry_hash(sha1);
584 if (0 <= i)
585 return &objects[object_ix[i]-1];
586 return NULL;
587 }
588
589 static void rehash_objects(void)
590 {
591 int i;
592 struct object_entry *oe;
593
594 object_ix_hashsz = nr_objects * 3;
595 if (object_ix_hashsz < 1024)
596 object_ix_hashsz = 1024;
597 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
598 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
599 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
600 int ix = locate_object_entry_hash(oe->sha1);
601 if (0 <= ix)
602 continue;
603 ix = -1 - ix;
604 object_ix[ix] = i + 1;
605 }
606 }
607
608 static unsigned name_hash(const char *name)
609 {
610 unsigned char c;
611 unsigned hash = 0;
612
613 /*
614 * This effectively just creates a sortable number from the
615 * last sixteen non-whitespace characters. Last characters
616 * count "most", so things that end in ".c" sort together.
617 */
618 while ((c = *name++) != 0) {
619 if (isspace(c))
620 continue;
621 hash = (hash >> 2) + (c << 24);
622 }
623 return hash;
624 }
625
626 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
627 {
628 unsigned int idx = nr_objects;
629 struct object_entry *entry;
630 struct packed_git *p;
631 unsigned int found_offset = 0;
632 struct packed_git *found_pack = NULL;
633 int ix, status = 0;
634
635 if (!exclude) {
636 for (p = packed_git; p; p = p->next) {
637 unsigned long offset = find_pack_entry_one(sha1, p);
638 if (offset) {
639 if (incremental)
640 return 0;
641 if (local && !p->pack_local)
642 return 0;
643 if (!found_pack) {
644 found_offset = offset;
645 found_pack = p;
646 }
647 }
648 }
649 }
650 if ((entry = locate_object_entry(sha1)) != NULL)
651 goto already_added;
652
653 if (idx >= nr_alloc) {
654 unsigned int needed = (idx + 1024) * 3 / 2;
655 objects = xrealloc(objects, needed * sizeof(*entry));
656 nr_alloc = needed;
657 }
658 entry = objects + idx;
659 nr_objects = idx + 1;
660 memset(entry, 0, sizeof(*entry));
661 hashcpy(entry->sha1, sha1);
662 entry->hash = hash;
663
664 if (object_ix_hashsz * 3 <= nr_objects * 4)
665 rehash_objects();
666 else {
667 ix = locate_object_entry_hash(entry->sha1);
668 if (0 <= ix)
669 die("internal error in object hashing.");
670 object_ix[-1 - ix] = idx + 1;
671 }
672 status = 1;
673
674 already_added:
675 if (progress_update) {
676 fprintf(stderr, "Counting objects...%d\r", nr_objects);
677 progress_update = 0;
678 }
679 if (exclude)
680 entry->preferred_base = 1;
681 else {
682 if (found_pack) {
683 entry->in_pack = found_pack;
684 entry->in_pack_offset = found_offset;
685 }
686 }
687 return status;
688 }
689
690 struct pbase_tree_cache {
691 unsigned char sha1[20];
692 int ref;
693 int temporary;
694 void *tree_data;
695 unsigned long tree_size;
696 };
697
698 static struct pbase_tree_cache *(pbase_tree_cache[256]);
699 static int pbase_tree_cache_ix(const unsigned char *sha1)
700 {
701 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
702 }
703 static int pbase_tree_cache_ix_incr(int ix)
704 {
705 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
706 }
707
708 static struct pbase_tree {
709 struct pbase_tree *next;
710 /* This is a phony "cache" entry; we are not
711 * going to evict it nor find it through _get()
712 * mechanism -- this is for the toplevel node that
713 * would almost always change with any commit.
714 */
715 struct pbase_tree_cache pcache;
716 } *pbase_tree;
717
718 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
719 {
720 struct pbase_tree_cache *ent, *nent;
721 void *data;
722 unsigned long size;
723 char type[20];
724 int neigh;
725 int my_ix = pbase_tree_cache_ix(sha1);
726 int available_ix = -1;
727
728 /* pbase-tree-cache acts as a limited hashtable.
729 * your object will be found at your index or within a few
730 * slots after that slot if it is cached.
731 */
732 for (neigh = 0; neigh < 8; neigh++) {
733 ent = pbase_tree_cache[my_ix];
734 if (ent && !hashcmp(ent->sha1, sha1)) {
735 ent->ref++;
736 return ent;
737 }
738 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
739 ((0 <= available_ix) &&
740 (!ent && pbase_tree_cache[available_ix])))
741 available_ix = my_ix;
742 if (!ent)
743 break;
744 my_ix = pbase_tree_cache_ix_incr(my_ix);
745 }
746
747 /* Did not find one. Either we got a bogus request or
748 * we need to read and perhaps cache.
749 */
750 data = read_sha1_file(sha1, type, &size);
751 if (!data)
752 return NULL;
753 if (strcmp(type, tree_type)) {
754 free(data);
755 return NULL;
756 }
757
758 /* We need to either cache or return a throwaway copy */
759
760 if (available_ix < 0)
761 ent = NULL;
762 else {
763 ent = pbase_tree_cache[available_ix];
764 my_ix = available_ix;
765 }
766
767 if (!ent) {
768 nent = xmalloc(sizeof(*nent));
769 nent->temporary = (available_ix < 0);
770 }
771 else {
772 /* evict and reuse */
773 free(ent->tree_data);
774 nent = ent;
775 }
776 hashcpy(nent->sha1, sha1);
777 nent->tree_data = data;
778 nent->tree_size = size;
779 nent->ref = 1;
780 if (!nent->temporary)
781 pbase_tree_cache[my_ix] = nent;
782 return nent;
783 }
784
785 static void pbase_tree_put(struct pbase_tree_cache *cache)
786 {
787 if (!cache->temporary) {
788 cache->ref--;
789 return;
790 }
791 free(cache->tree_data);
792 free(cache);
793 }
794
795 static int name_cmp_len(const char *name)
796 {
797 int i;
798 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
799 ;
800 return i;
801 }
802
803 static void add_pbase_object(struct tree_desc *tree,
804 const char *name,
805 int cmplen,
806 const char *fullname)
807 {
808 struct name_entry entry;
809
810 while (tree_entry(tree,&entry)) {
811 unsigned long size;
812 char type[20];
813
814 if (entry.pathlen != cmplen ||
815 memcmp(entry.path, name, cmplen) ||
816 !has_sha1_file(entry.sha1) ||
817 sha1_object_info(entry.sha1, type, &size))
818 continue;
819 if (name[cmplen] != '/') {
820 unsigned hash = name_hash(fullname);
821 add_object_entry(entry.sha1, hash, 1);
822 return;
823 }
824 if (!strcmp(type, tree_type)) {
825 struct tree_desc sub;
826 struct pbase_tree_cache *tree;
827 const char *down = name+cmplen+1;
828 int downlen = name_cmp_len(down);
829
830 tree = pbase_tree_get(entry.sha1);
831 if (!tree)
832 return;
833 sub.buf = tree->tree_data;
834 sub.size = tree->tree_size;
835
836 add_pbase_object(&sub, down, downlen, fullname);
837 pbase_tree_put(tree);
838 }
839 }
840 }
841
842 static unsigned *done_pbase_paths;
843 static int done_pbase_paths_num;
844 static int done_pbase_paths_alloc;
845 static int done_pbase_path_pos(unsigned hash)
846 {
847 int lo = 0;
848 int hi = done_pbase_paths_num;
849 while (lo < hi) {
850 int mi = (hi + lo) / 2;
851 if (done_pbase_paths[mi] == hash)
852 return mi;
853 if (done_pbase_paths[mi] < hash)
854 hi = mi;
855 else
856 lo = mi + 1;
857 }
858 return -lo-1;
859 }
860
861 static int check_pbase_path(unsigned hash)
862 {
863 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
864 if (0 <= pos)
865 return 1;
866 pos = -pos - 1;
867 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
868 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
869 done_pbase_paths = xrealloc(done_pbase_paths,
870 done_pbase_paths_alloc *
871 sizeof(unsigned));
872 }
873 done_pbase_paths_num++;
874 if (pos < done_pbase_paths_num)
875 memmove(done_pbase_paths + pos + 1,
876 done_pbase_paths + pos,
877 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
878 done_pbase_paths[pos] = hash;
879 return 0;
880 }
881
882 static void add_preferred_base_object(const char *name, unsigned hash)
883 {
884 struct pbase_tree *it;
885 int cmplen = name_cmp_len(name);
886
887 if (check_pbase_path(hash))
888 return;
889
890 for (it = pbase_tree; it; it = it->next) {
891 if (cmplen == 0) {
892 hash = name_hash("");
893 add_object_entry(it->pcache.sha1, hash, 1);
894 }
895 else {
896 struct tree_desc tree;
897 tree.buf = it->pcache.tree_data;
898 tree.size = it->pcache.tree_size;
899 add_pbase_object(&tree, name, cmplen, name);
900 }
901 }
902 }
903
904 static void add_preferred_base(unsigned char *sha1)
905 {
906 struct pbase_tree *it;
907 void *data;
908 unsigned long size;
909 unsigned char tree_sha1[20];
910
911 if (window <= num_preferred_base++)
912 return;
913
914 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
915 if (!data)
916 return;
917
918 for (it = pbase_tree; it; it = it->next) {
919 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
920 free(data);
921 return;
922 }
923 }
924
925 it = xcalloc(1, sizeof(*it));
926 it->next = pbase_tree;
927 pbase_tree = it;
928
929 hashcpy(it->pcache.sha1, tree_sha1);
930 it->pcache.tree_data = data;
931 it->pcache.tree_size = size;
932 }
933
934 static void check_object(struct object_entry *entry)
935 {
936 char type[20];
937
938 if (entry->in_pack && !entry->preferred_base) {
939 struct packed_git *p = entry->in_pack;
940 unsigned long left = p->pack_size - entry->in_pack_offset;
941 unsigned long size, used;
942 unsigned char *buf;
943 struct object_entry *base_entry = NULL;
944
945 use_packed_git(p);
946 buf = p->pack_base;
947 buf += entry->in_pack_offset;
948
949 /* We want in_pack_type even if we do not reuse delta.
950 * There is no point not reusing non-delta representations.
951 */
952 used = unpack_object_header_gently(buf, left,
953 &entry->in_pack_type, &size);
954 if (!used || left - used <= 20)
955 die("corrupt pack for %s", sha1_to_hex(entry->sha1));
956
957 /* Check if it is delta, and the base is also an object
958 * we are going to pack. If so we will reuse the existing
959 * delta.
960 */
961 if (!no_reuse_delta) {
962 unsigned char c, *base_name;
963 unsigned long ofs;
964 /* there is at least 20 bytes left in the pack */
965 switch (entry->in_pack_type) {
966 case OBJ_REF_DELTA:
967 base_name = buf + used;
968 used += 20;
969 break;
970 case OBJ_OFS_DELTA:
971 c = buf[used++];
972 ofs = c & 127;
973 while (c & 128) {
974 ofs += 1;
975 if (!ofs || ofs & ~(~0UL >> 7))
976 die("delta base offset overflow in pack for %s",
977 sha1_to_hex(entry->sha1));
978 c = buf[used++];
979 ofs = (ofs << 7) + (c & 127);
980 }
981 if (ofs >= entry->in_pack_offset)
982 die("delta base offset out of bound for %s",
983 sha1_to_hex(entry->sha1));
984 ofs = entry->in_pack_offset - ofs;
985 base_name = find_packed_object_name(p, ofs);
986 break;
987 default:
988 base_name = NULL;
989 }
990 if (base_name)
991 base_entry = locate_object_entry(base_name);
992 }
993 unuse_packed_git(p);
994 entry->in_pack_header_size = used;
995
996 if (base_entry) {
997
998 /* Depth value does not matter - find_deltas()
999 * will never consider reused delta as the
1000 * base object to deltify other objects
1001 * against, in order to avoid circular deltas.
1002 */
1003
1004 /* uncompressed size of the delta data */
1005 entry->size = size;
1006 entry->delta = base_entry;
1007 entry->type = entry->in_pack_type;
1008
1009 entry->delta_sibling = base_entry->delta_child;
1010 base_entry->delta_child = entry;
1011
1012 return;
1013 }
1014 /* Otherwise we would do the usual */
1015 }
1016
1017 if (sha1_object_info(entry->sha1, type, &entry->size))
1018 die("unable to get type of object %s",
1019 sha1_to_hex(entry->sha1));
1020
1021 if (!strcmp(type, commit_type)) {
1022 entry->type = OBJ_COMMIT;
1023 } else if (!strcmp(type, tree_type)) {
1024 entry->type = OBJ_TREE;
1025 } else if (!strcmp(type, blob_type)) {
1026 entry->type = OBJ_BLOB;
1027 } else if (!strcmp(type, tag_type)) {
1028 entry->type = OBJ_TAG;
1029 } else
1030 die("unable to pack object %s of type %s",
1031 sha1_to_hex(entry->sha1), type);
1032 }
1033
1034 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1035 {
1036 struct object_entry *child = me->delta_child;
1037 unsigned int m = n;
1038 while (child) {
1039 unsigned int c = check_delta_limit(child, n + 1);
1040 if (m < c)
1041 m = c;
1042 child = child->delta_sibling;
1043 }
1044 return m;
1045 }
1046
1047 static void get_object_details(void)
1048 {
1049 int i;
1050 struct object_entry *entry;
1051
1052 prepare_pack_ix();
1053 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1054 check_object(entry);
1055
1056 if (nr_objects == nr_result) {
1057 /*
1058 * Depth of objects that depend on the entry -- this
1059 * is subtracted from depth-max to break too deep
1060 * delta chain because of delta data reusing.
1061 * However, we loosen this restriction when we know we
1062 * are creating a thin pack -- it will have to be
1063 * expanded on the other end anyway, so do not
1064 * artificially cut the delta chain and let it go as
1065 * deep as it wants.
1066 */
1067 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1068 if (!entry->delta && entry->delta_child)
1069 entry->delta_limit =
1070 check_delta_limit(entry, 1);
1071 }
1072 }
1073
1074 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1075
1076 static entry_sort_t current_sort;
1077
1078 static int sort_comparator(const void *_a, const void *_b)
1079 {
1080 struct object_entry *a = *(struct object_entry **)_a;
1081 struct object_entry *b = *(struct object_entry **)_b;
1082 return current_sort(a,b);
1083 }
1084
1085 static struct object_entry **create_sorted_list(entry_sort_t sort)
1086 {
1087 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1088 int i;
1089
1090 for (i = 0; i < nr_objects; i++)
1091 list[i] = objects + i;
1092 current_sort = sort;
1093 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1094 return list;
1095 }
1096
1097 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1098 {
1099 return hashcmp(a->sha1, b->sha1);
1100 }
1101
1102 static struct object_entry **create_final_object_list(void)
1103 {
1104 struct object_entry **list;
1105 int i, j;
1106
1107 for (i = nr_result = 0; i < nr_objects; i++)
1108 if (!objects[i].preferred_base)
1109 nr_result++;
1110 list = xmalloc(nr_result * sizeof(struct object_entry *));
1111 for (i = j = 0; i < nr_objects; i++) {
1112 if (!objects[i].preferred_base)
1113 list[j++] = objects + i;
1114 }
1115 current_sort = sha1_sort;
1116 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1117 return list;
1118 }
1119
1120 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1121 {
1122 if (a->type < b->type)
1123 return -1;
1124 if (a->type > b->type)
1125 return 1;
1126 if (a->hash < b->hash)
1127 return -1;
1128 if (a->hash > b->hash)
1129 return 1;
1130 if (a->preferred_base < b->preferred_base)
1131 return -1;
1132 if (a->preferred_base > b->preferred_base)
1133 return 1;
1134 if (a->size < b->size)
1135 return -1;
1136 if (a->size > b->size)
1137 return 1;
1138 return a < b ? -1 : (a > b);
1139 }
1140
1141 struct unpacked {
1142 struct object_entry *entry;
1143 void *data;
1144 struct delta_index *index;
1145 };
1146
1147 /*
1148 * We search for deltas _backwards_ in a list sorted by type and
1149 * by size, so that we see progressively smaller and smaller files.
1150 * That's because we prefer deltas to be from the bigger file
1151 * to the smaller - deletes are potentially cheaper, but perhaps
1152 * more importantly, the bigger file is likely the more recent
1153 * one.
1154 */
1155 static int try_delta(struct unpacked *trg, struct unpacked *src,
1156 unsigned max_depth)
1157 {
1158 struct object_entry *trg_entry = trg->entry;
1159 struct object_entry *src_entry = src->entry;
1160 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1161 char type[10];
1162 void *delta_buf;
1163
1164 /* Don't bother doing diffs between different types */
1165 if (trg_entry->type != src_entry->type)
1166 return -1;
1167
1168 /* We do not compute delta to *create* objects we are not
1169 * going to pack.
1170 */
1171 if (trg_entry->preferred_base)
1172 return -1;
1173
1174 /*
1175 * We do not bother to try a delta that we discarded
1176 * on an earlier try, but only when reusing delta data.
1177 */
1178 if (!no_reuse_delta && trg_entry->in_pack &&
1179 trg_entry->in_pack == src_entry->in_pack)
1180 return 0;
1181
1182 /*
1183 * If the current object is at pack edge, take the depth the
1184 * objects that depend on the current object into account --
1185 * otherwise they would become too deep.
1186 */
1187 if (trg_entry->delta_child) {
1188 if (max_depth <= trg_entry->delta_limit)
1189 return 0;
1190 max_depth -= trg_entry->delta_limit;
1191 }
1192 if (src_entry->depth >= max_depth)
1193 return 0;
1194
1195 /* Now some size filtering heuristics. */
1196 trg_size = trg_entry->size;
1197 max_size = trg_size/2 - 20;
1198 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1199 if (max_size == 0)
1200 return 0;
1201 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1202 max_size = trg_entry->delta_size-1;
1203 src_size = src_entry->size;
1204 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1205 if (sizediff >= max_size)
1206 return 0;
1207
1208 /* Load data if not already done */
1209 if (!trg->data) {
1210 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1211 if (sz != trg_size)
1212 die("object %s inconsistent object length (%lu vs %lu)",
1213 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1214 }
1215 if (!src->data) {
1216 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1217 if (sz != src_size)
1218 die("object %s inconsistent object length (%lu vs %lu)",
1219 sha1_to_hex(src_entry->sha1), sz, src_size);
1220 }
1221 if (!src->index) {
1222 src->index = create_delta_index(src->data, src_size);
1223 if (!src->index)
1224 die("out of memory");
1225 }
1226
1227 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1228 if (!delta_buf)
1229 return 0;
1230
1231 trg_entry->delta = src_entry;
1232 trg_entry->delta_size = delta_size;
1233 trg_entry->depth = src_entry->depth + 1;
1234 free(delta_buf);
1235 return 1;
1236 }
1237
1238 static void progress_interval(int signum)
1239 {
1240 progress_update = 1;
1241 }
1242
1243 static void find_deltas(struct object_entry **list, int window, int depth)
1244 {
1245 int i, idx;
1246 unsigned int array_size = window * sizeof(struct unpacked);
1247 struct unpacked *array = xmalloc(array_size);
1248 unsigned processed = 0;
1249 unsigned last_percent = 999;
1250
1251 memset(array, 0, array_size);
1252 i = nr_objects;
1253 idx = 0;
1254 if (progress)
1255 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1256
1257 while (--i >= 0) {
1258 struct object_entry *entry = list[i];
1259 struct unpacked *n = array + idx;
1260 int j;
1261
1262 if (!entry->preferred_base)
1263 processed++;
1264
1265 if (progress) {
1266 unsigned percent = processed * 100 / nr_result;
1267 if (percent != last_percent || progress_update) {
1268 fprintf(stderr, "%4u%% (%u/%u) done\r",
1269 percent, processed, nr_result);
1270 progress_update = 0;
1271 last_percent = percent;
1272 }
1273 }
1274
1275 if (entry->delta)
1276 /* This happens if we decided to reuse existing
1277 * delta from a pack. "!no_reuse_delta &&" is implied.
1278 */
1279 continue;
1280
1281 if (entry->size < 50)
1282 continue;
1283 free_delta_index(n->index);
1284 n->index = NULL;
1285 free(n->data);
1286 n->data = NULL;
1287 n->entry = entry;
1288
1289 j = window;
1290 while (--j > 0) {
1291 unsigned int other_idx = idx + j;
1292 struct unpacked *m;
1293 if (other_idx >= window)
1294 other_idx -= window;
1295 m = array + other_idx;
1296 if (!m->entry)
1297 break;
1298 if (try_delta(n, m, depth) < 0)
1299 break;
1300 }
1301 /* if we made n a delta, and if n is already at max
1302 * depth, leaving it in the window is pointless. we
1303 * should evict it first.
1304 */
1305 if (entry->delta && depth <= entry->depth)
1306 continue;
1307
1308 idx++;
1309 if (idx >= window)
1310 idx = 0;
1311 }
1312
1313 if (progress)
1314 fputc('\n', stderr);
1315
1316 for (i = 0; i < window; ++i) {
1317 free_delta_index(array[i].index);
1318 free(array[i].data);
1319 }
1320 free(array);
1321 }
1322
1323 static void prepare_pack(int window, int depth)
1324 {
1325 get_object_details();
1326 sorted_by_type = create_sorted_list(type_size_sort);
1327 if (window && depth)
1328 find_deltas(sorted_by_type, window+1, depth);
1329 }
1330
1331 static int reuse_cached_pack(unsigned char *sha1)
1332 {
1333 static const char cache[] = "pack-cache/pack-%s.%s";
1334 char *cached_pack, *cached_idx;
1335 int ifd, ofd, ifd_ix = -1;
1336
1337 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1338 ifd = open(cached_pack, O_RDONLY);
1339 if (ifd < 0)
1340 return 0;
1341
1342 if (!pack_to_stdout) {
1343 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1344 ifd_ix = open(cached_idx, O_RDONLY);
1345 if (ifd_ix < 0) {
1346 close(ifd);
1347 return 0;
1348 }
1349 }
1350
1351 if (progress)
1352 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1353 sha1_to_hex(sha1));
1354
1355 if (pack_to_stdout) {
1356 if (copy_fd(ifd, 1))
1357 exit(1);
1358 close(ifd);
1359 }
1360 else {
1361 char name[PATH_MAX];
1362 snprintf(name, sizeof(name),
1363 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1364 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1365 if (ofd < 0)
1366 die("unable to open %s (%s)", name, strerror(errno));
1367 if (copy_fd(ifd, ofd))
1368 exit(1);
1369 close(ifd);
1370
1371 snprintf(name, sizeof(name),
1372 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1373 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1374 if (ofd < 0)
1375 die("unable to open %s (%s)", name, strerror(errno));
1376 if (copy_fd(ifd_ix, ofd))
1377 exit(1);
1378 close(ifd_ix);
1379 puts(sha1_to_hex(sha1));
1380 }
1381
1382 return 1;
1383 }
1384
1385 static void setup_progress_signal(void)
1386 {
1387 struct sigaction sa;
1388 struct itimerval v;
1389
1390 memset(&sa, 0, sizeof(sa));
1391 sa.sa_handler = progress_interval;
1392 sigemptyset(&sa.sa_mask);
1393 sa.sa_flags = SA_RESTART;
1394 sigaction(SIGALRM, &sa, NULL);
1395
1396 v.it_interval.tv_sec = 1;
1397 v.it_interval.tv_usec = 0;
1398 v.it_value = v.it_interval;
1399 setitimer(ITIMER_REAL, &v, NULL);
1400 }
1401
1402 static int git_pack_config(const char *k, const char *v)
1403 {
1404 if(!strcmp(k, "pack.window")) {
1405 window = git_config_int(k, v);
1406 return 0;
1407 }
1408 return git_default_config(k, v);
1409 }
1410
1411 static void read_object_list_from_stdin(void)
1412 {
1413 char line[40 + 1 + PATH_MAX + 2];
1414 unsigned char sha1[20];
1415 unsigned hash;
1416
1417 for (;;) {
1418 if (!fgets(line, sizeof(line), stdin)) {
1419 if (feof(stdin))
1420 break;
1421 if (!ferror(stdin))
1422 die("fgets returned NULL, not EOF, not error!");
1423 if (errno != EINTR)
1424 die("fgets: %s", strerror(errno));
1425 clearerr(stdin);
1426 continue;
1427 }
1428 if (line[0] == '-') {
1429 if (get_sha1_hex(line+1, sha1))
1430 die("expected edge sha1, got garbage:\n %s",
1431 line);
1432 add_preferred_base(sha1);
1433 continue;
1434 }
1435 if (get_sha1_hex(line, sha1))
1436 die("expected sha1, got garbage:\n %s", line);
1437
1438 hash = name_hash(line+41);
1439 add_preferred_base_object(line+41, hash);
1440 add_object_entry(sha1, hash, 0);
1441 }
1442 }
1443
1444 static void show_commit(struct commit *commit)
1445 {
1446 unsigned hash = name_hash("");
1447 add_preferred_base_object("", hash);
1448 add_object_entry(commit->object.sha1, hash, 0);
1449 }
1450
1451 static void show_object(struct object_array_entry *p)
1452 {
1453 unsigned hash = name_hash(p->name);
1454 add_preferred_base_object(p->name, hash);
1455 add_object_entry(p->item->sha1, hash, 0);
1456 }
1457
1458 static void show_edge(struct commit *commit)
1459 {
1460 add_preferred_base(commit->object.sha1);
1461 }
1462
1463 static void get_object_list(int ac, const char **av)
1464 {
1465 struct rev_info revs;
1466 char line[1000];
1467 int flags = 0;
1468
1469 init_revisions(&revs, NULL);
1470 save_commit_buffer = 0;
1471 track_object_refs = 0;
1472 setup_revisions(ac, av, &revs, NULL);
1473
1474 while (fgets(line, sizeof(line), stdin) != NULL) {
1475 int len = strlen(line);
1476 if (line[len - 1] == '\n')
1477 line[--len] = 0;
1478 if (!len)
1479 break;
1480 if (*line == '-') {
1481 if (!strcmp(line, "--not")) {
1482 flags ^= UNINTERESTING;
1483 continue;
1484 }
1485 die("not a rev '%s'", line);
1486 }
1487 if (handle_revision_arg(line, &revs, flags, 1))
1488 die("bad revision '%s'", line);
1489 }
1490
1491 prepare_revision_walk(&revs);
1492 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1493 traverse_commit_list(&revs, show_commit, show_object);
1494 }
1495
1496 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1497 {
1498 SHA_CTX ctx;
1499 int depth = 10;
1500 struct object_entry **list;
1501 int use_internal_rev_list = 0;
1502 int thin = 0;
1503 int i;
1504 const char *rp_av[64];
1505 int rp_ac;
1506
1507 rp_av[0] = "pack-objects";
1508 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1509 rp_ac = 2;
1510
1511 git_config(git_pack_config);
1512
1513 progress = isatty(2);
1514 for (i = 1; i < argc; i++) {
1515 const char *arg = argv[i];
1516
1517 if (*arg != '-')
1518 break;
1519
1520 if (!strcmp("--non-empty", arg)) {
1521 non_empty = 1;
1522 continue;
1523 }
1524 if (!strcmp("--local", arg)) {
1525 local = 1;
1526 continue;
1527 }
1528 if (!strcmp("--incremental", arg)) {
1529 incremental = 1;
1530 continue;
1531 }
1532 if (!strncmp("--window=", arg, 9)) {
1533 char *end;
1534 window = strtoul(arg+9, &end, 0);
1535 if (!arg[9] || *end)
1536 usage(pack_usage);
1537 continue;
1538 }
1539 if (!strncmp("--depth=", arg, 8)) {
1540 char *end;
1541 depth = strtoul(arg+8, &end, 0);
1542 if (!arg[8] || *end)
1543 usage(pack_usage);
1544 continue;
1545 }
1546 if (!strcmp("--progress", arg)) {
1547 progress = 1;
1548 continue;
1549 }
1550 if (!strcmp("--all-progress", arg)) {
1551 progress = 2;
1552 continue;
1553 }
1554 if (!strcmp("-q", arg)) {
1555 progress = 0;
1556 continue;
1557 }
1558 if (!strcmp("--no-reuse-delta", arg)) {
1559 no_reuse_delta = 1;
1560 continue;
1561 }
1562 if (!strcmp("--delta-base-offset", arg)) {
1563 allow_ofs_delta = 1;
1564 continue;
1565 }
1566 if (!strcmp("--stdout", arg)) {
1567 pack_to_stdout = 1;
1568 continue;
1569 }
1570 if (!strcmp("--revs", arg)) {
1571 use_internal_rev_list = 1;
1572 continue;
1573 }
1574 if (!strcmp("--unpacked", arg) ||
1575 !strncmp("--unpacked=", arg, 11) ||
1576 !strcmp("--all", arg)) {
1577 use_internal_rev_list = 1;
1578 if (ARRAY_SIZE(rp_av) - 1 <= rp_ac)
1579 die("too many internal rev-list options");
1580 rp_av[rp_ac++] = arg;
1581 continue;
1582 }
1583 if (!strcmp("--thin", arg)) {
1584 use_internal_rev_list = 1;
1585 thin = 1;
1586 rp_av[1] = "--objects-edge";
1587 continue;
1588 }
1589 usage(pack_usage);
1590 }
1591
1592 /* Traditionally "pack-objects [options] base extra" failed;
1593 * we would however want to take refs parameter that would
1594 * have been given to upstream rev-list ourselves, which means
1595 * we somehow want to say what the base name is. So the
1596 * syntax would be:
1597 *
1598 * pack-objects [options] base <refs...>
1599 *
1600 * in other words, we would treat the first non-option as the
1601 * base_name and send everything else to the internal revision
1602 * walker.
1603 */
1604
1605 if (!pack_to_stdout)
1606 base_name = argv[i++];
1607
1608 if (pack_to_stdout != !base_name)
1609 usage(pack_usage);
1610
1611 if (!pack_to_stdout && thin)
1612 die("--thin cannot be used to build an indexable pack.");
1613
1614 prepare_packed_git();
1615
1616 if (progress) {
1617 fprintf(stderr, "Generating pack...\n");
1618 setup_progress_signal();
1619 }
1620
1621 if (!use_internal_rev_list)
1622 read_object_list_from_stdin();
1623 else {
1624 rp_av[rp_ac] = NULL;
1625 get_object_list(rp_ac, rp_av);
1626 }
1627
1628 if (progress)
1629 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1630 sorted_by_sha = create_final_object_list();
1631 if (non_empty && !nr_result)
1632 return 0;
1633
1634 SHA1_Init(&ctx);
1635 list = sorted_by_sha;
1636 for (i = 0; i < nr_result; i++) {
1637 struct object_entry *entry = *list++;
1638 SHA1_Update(&ctx, entry->sha1, 20);
1639 }
1640 SHA1_Final(object_list_sha1, &ctx);
1641 if (progress && (nr_objects != nr_result))
1642 fprintf(stderr, "Result has %d objects.\n", nr_result);
1643
1644 if (reuse_cached_pack(object_list_sha1))
1645 ;
1646 else {
1647 if (nr_result)
1648 prepare_pack(window, depth);
1649 if (progress == 1 && pack_to_stdout) {
1650 /* the other end usually displays progress itself */
1651 struct itimerval v = {{0,},};
1652 setitimer(ITIMER_REAL, &v, NULL);
1653 signal(SIGALRM, SIG_IGN );
1654 progress_update = 0;
1655 }
1656 write_pack_file();
1657 if (!pack_to_stdout) {
1658 write_index_file();
1659 puts(sha1_to_hex(object_list_sha1));
1660 }
1661 }
1662 if (progress)
1663 fprintf(stderr, "Total %d, written %d (delta %d), reused %d (delta %d)\n",
1664 nr_result, written, written_delta, reused, reused_delta);
1665 return 0;
1666 }