![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
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>
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]";
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 };
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 */
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;
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;
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;
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;
116 /*
117 * stats
118 */
119 static int written;
120 static int written_delta;
121 static int reused;
122 static int reused_delta;
124 static int pack_revindex_ix(struct packed_git *p)
125 {
126 unsigned long ui = (unsigned long)p;
127 int i;
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 }
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 }
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 }
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;
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 }
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 }
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 }
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 }
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;
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 }
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;
262 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
263 die("bad type %d", type);
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 }
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;
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);
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 }
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;
320 if (pack_to_stdout)
321 return 0;
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 }
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;
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 */
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 }
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);
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;
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 }
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 }
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 }
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;
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);
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 if (written != nr_result)
518 die("wrote %d objects while expecting %d", written, nr_result);
519 sha1close(f, pack_file_sha1, 1);
520 }
522 static void write_index_file(void)
523 {
524 int i;
525 struct sha1file *f = sha1create("%s-%s.%s", base_name,
526 sha1_to_hex(object_list_sha1), "idx");
527 struct object_entry **list = sorted_by_sha;
528 struct object_entry **last = list + nr_result;
529 unsigned int array[256];
531 /*
532 * Write the first-level table (the list is sorted,
533 * but we use a 256-entry lookup to be able to avoid
534 * having to do eight extra binary search iterations).
535 */
536 for (i = 0; i < 256; i++) {
537 struct object_entry **next = list;
538 while (next < last) {
539 struct object_entry *entry = *next;
540 if (entry->sha1[0] != i)
541 break;
542 next++;
543 }
544 array[i] = htonl(next - sorted_by_sha);
545 list = next;
546 }
547 sha1write(f, array, 256 * sizeof(int));
549 /*
550 * Write the actual SHA1 entries..
551 */
552 list = sorted_by_sha;
553 for (i = 0; i < nr_result; i++) {
554 struct object_entry *entry = *list++;
555 unsigned int offset = htonl(entry->offset);
556 sha1write(f, &offset, 4);
557 sha1write(f, entry->sha1, 20);
558 }
559 sha1write(f, pack_file_sha1, 20);
560 sha1close(f, NULL, 1);
561 }
563 static int locate_object_entry_hash(const unsigned char *sha1)
564 {
565 int i;
566 unsigned int ui;
567 memcpy(&ui, sha1, sizeof(unsigned int));
568 i = ui % object_ix_hashsz;
569 while (0 < object_ix[i]) {
570 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
571 return i;
572 if (++i == object_ix_hashsz)
573 i = 0;
574 }
575 return -1 - i;
576 }
578 static struct object_entry *locate_object_entry(const unsigned char *sha1)
579 {
580 int i;
582 if (!object_ix_hashsz)
583 return NULL;
585 i = locate_object_entry_hash(sha1);
586 if (0 <= i)
587 return &objects[object_ix[i]-1];
588 return NULL;
589 }
591 static void rehash_objects(void)
592 {
593 int i;
594 struct object_entry *oe;
596 object_ix_hashsz = nr_objects * 3;
597 if (object_ix_hashsz < 1024)
598 object_ix_hashsz = 1024;
599 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
600 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
601 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
602 int ix = locate_object_entry_hash(oe->sha1);
603 if (0 <= ix)
604 continue;
605 ix = -1 - ix;
606 object_ix[ix] = i + 1;
607 }
608 }
610 static unsigned name_hash(const char *name)
611 {
612 unsigned char c;
613 unsigned hash = 0;
615 /*
616 * This effectively just creates a sortable number from the
617 * last sixteen non-whitespace characters. Last characters
618 * count "most", so things that end in ".c" sort together.
619 */
620 while ((c = *name++) != 0) {
621 if (isspace(c))
622 continue;
623 hash = (hash >> 2) + (c << 24);
624 }
625 return hash;
626 }
628 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
629 {
630 unsigned int idx = nr_objects;
631 struct object_entry *entry;
632 struct packed_git *p;
633 unsigned int found_offset = 0;
634 struct packed_git *found_pack = NULL;
635 int ix, status = 0;
637 if (!exclude) {
638 for (p = packed_git; p; p = p->next) {
639 unsigned long offset = find_pack_entry_one(sha1, p);
640 if (offset) {
641 if (incremental)
642 return 0;
643 if (local && !p->pack_local)
644 return 0;
645 if (!found_pack) {
646 found_offset = offset;
647 found_pack = p;
648 }
649 }
650 }
651 }
652 if ((entry = locate_object_entry(sha1)) != NULL)
653 goto already_added;
655 if (idx >= nr_alloc) {
656 unsigned int needed = (idx + 1024) * 3 / 2;
657 objects = xrealloc(objects, needed * sizeof(*entry));
658 nr_alloc = needed;
659 }
660 entry = objects + idx;
661 nr_objects = idx + 1;
662 memset(entry, 0, sizeof(*entry));
663 hashcpy(entry->sha1, sha1);
664 entry->hash = hash;
666 if (object_ix_hashsz * 3 <= nr_objects * 4)
667 rehash_objects();
668 else {
669 ix = locate_object_entry_hash(entry->sha1);
670 if (0 <= ix)
671 die("internal error in object hashing.");
672 object_ix[-1 - ix] = idx + 1;
673 }
674 status = 1;
676 already_added:
677 if (progress_update) {
678 fprintf(stderr, "Counting objects...%d\r", nr_objects);
679 progress_update = 0;
680 }
681 if (exclude)
682 entry->preferred_base = 1;
683 else {
684 if (found_pack) {
685 entry->in_pack = found_pack;
686 entry->in_pack_offset = found_offset;
687 }
688 }
689 return status;
690 }
692 struct pbase_tree_cache {
693 unsigned char sha1[20];
694 int ref;
695 int temporary;
696 void *tree_data;
697 unsigned long tree_size;
698 };
700 static struct pbase_tree_cache *(pbase_tree_cache[256]);
701 static int pbase_tree_cache_ix(const unsigned char *sha1)
702 {
703 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
704 }
705 static int pbase_tree_cache_ix_incr(int ix)
706 {
707 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
708 }
710 static struct pbase_tree {
711 struct pbase_tree *next;
712 /* This is a phony "cache" entry; we are not
713 * going to evict it nor find it through _get()
714 * mechanism -- this is for the toplevel node that
715 * would almost always change with any commit.
716 */
717 struct pbase_tree_cache pcache;
718 } *pbase_tree;
720 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
721 {
722 struct pbase_tree_cache *ent, *nent;
723 void *data;
724 unsigned long size;
725 char type[20];
726 int neigh;
727 int my_ix = pbase_tree_cache_ix(sha1);
728 int available_ix = -1;
730 /* pbase-tree-cache acts as a limited hashtable.
731 * your object will be found at your index or within a few
732 * slots after that slot if it is cached.
733 */
734 for (neigh = 0; neigh < 8; neigh++) {
735 ent = pbase_tree_cache[my_ix];
736 if (ent && !hashcmp(ent->sha1, sha1)) {
737 ent->ref++;
738 return ent;
739 }
740 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
741 ((0 <= available_ix) &&
742 (!ent && pbase_tree_cache[available_ix])))
743 available_ix = my_ix;
744 if (!ent)
745 break;
746 my_ix = pbase_tree_cache_ix_incr(my_ix);
747 }
749 /* Did not find one. Either we got a bogus request or
750 * we need to read and perhaps cache.
751 */
752 data = read_sha1_file(sha1, type, &size);
753 if (!data)
754 return NULL;
755 if (strcmp(type, tree_type)) {
756 free(data);
757 return NULL;
758 }
760 /* We need to either cache or return a throwaway copy */
762 if (available_ix < 0)
763 ent = NULL;
764 else {
765 ent = pbase_tree_cache[available_ix];
766 my_ix = available_ix;
767 }
769 if (!ent) {
770 nent = xmalloc(sizeof(*nent));
771 nent->temporary = (available_ix < 0);
772 }
773 else {
774 /* evict and reuse */
775 free(ent->tree_data);
776 nent = ent;
777 }
778 hashcpy(nent->sha1, sha1);
779 nent->tree_data = data;
780 nent->tree_size = size;
781 nent->ref = 1;
782 if (!nent->temporary)
783 pbase_tree_cache[my_ix] = nent;
784 return nent;
785 }
787 static void pbase_tree_put(struct pbase_tree_cache *cache)
788 {
789 if (!cache->temporary) {
790 cache->ref--;
791 return;
792 }
793 free(cache->tree_data);
794 free(cache);
795 }
797 static int name_cmp_len(const char *name)
798 {
799 int i;
800 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
801 ;
802 return i;
803 }
805 static void add_pbase_object(struct tree_desc *tree,
806 const char *name,
807 int cmplen,
808 const char *fullname)
809 {
810 struct name_entry entry;
812 while (tree_entry(tree,&entry)) {
813 unsigned long size;
814 char type[20];
816 if (entry.pathlen != cmplen ||
817 memcmp(entry.path, name, cmplen) ||
818 !has_sha1_file(entry.sha1) ||
819 sha1_object_info(entry.sha1, type, &size))
820 continue;
821 if (name[cmplen] != '/') {
822 unsigned hash = name_hash(fullname);
823 add_object_entry(entry.sha1, hash, 1);
824 return;
825 }
826 if (!strcmp(type, tree_type)) {
827 struct tree_desc sub;
828 struct pbase_tree_cache *tree;
829 const char *down = name+cmplen+1;
830 int downlen = name_cmp_len(down);
832 tree = pbase_tree_get(entry.sha1);
833 if (!tree)
834 return;
835 sub.buf = tree->tree_data;
836 sub.size = tree->tree_size;
838 add_pbase_object(&sub, down, downlen, fullname);
839 pbase_tree_put(tree);
840 }
841 }
842 }
844 static unsigned *done_pbase_paths;
845 static int done_pbase_paths_num;
846 static int done_pbase_paths_alloc;
847 static int done_pbase_path_pos(unsigned hash)
848 {
849 int lo = 0;
850 int hi = done_pbase_paths_num;
851 while (lo < hi) {
852 int mi = (hi + lo) / 2;
853 if (done_pbase_paths[mi] == hash)
854 return mi;
855 if (done_pbase_paths[mi] < hash)
856 hi = mi;
857 else
858 lo = mi + 1;
859 }
860 return -lo-1;
861 }
863 static int check_pbase_path(unsigned hash)
864 {
865 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
866 if (0 <= pos)
867 return 1;
868 pos = -pos - 1;
869 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
870 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
871 done_pbase_paths = xrealloc(done_pbase_paths,
872 done_pbase_paths_alloc *
873 sizeof(unsigned));
874 }
875 done_pbase_paths_num++;
876 if (pos < done_pbase_paths_num)
877 memmove(done_pbase_paths + pos + 1,
878 done_pbase_paths + pos,
879 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
880 done_pbase_paths[pos] = hash;
881 return 0;
882 }
884 static void add_preferred_base_object(const char *name, unsigned hash)
885 {
886 struct pbase_tree *it;
887 int cmplen = name_cmp_len(name);
889 if (check_pbase_path(hash))
890 return;
892 for (it = pbase_tree; it; it = it->next) {
893 if (cmplen == 0) {
894 hash = name_hash("");
895 add_object_entry(it->pcache.sha1, hash, 1);
896 }
897 else {
898 struct tree_desc tree;
899 tree.buf = it->pcache.tree_data;
900 tree.size = it->pcache.tree_size;
901 add_pbase_object(&tree, name, cmplen, name);
902 }
903 }
904 }
906 static void add_preferred_base(unsigned char *sha1)
907 {
908 struct pbase_tree *it;
909 void *data;
910 unsigned long size;
911 unsigned char tree_sha1[20];
913 if (window <= num_preferred_base++)
914 return;
916 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
917 if (!data)
918 return;
920 for (it = pbase_tree; it; it = it->next) {
921 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
922 free(data);
923 return;
924 }
925 }
927 it = xcalloc(1, sizeof(*it));
928 it->next = pbase_tree;
929 pbase_tree = it;
931 hashcpy(it->pcache.sha1, tree_sha1);
932 it->pcache.tree_data = data;
933 it->pcache.tree_size = size;
934 }
936 static void check_object(struct object_entry *entry)
937 {
938 char type[20];
940 if (entry->in_pack && !entry->preferred_base) {
941 struct packed_git *p = entry->in_pack;
942 unsigned long left = p->pack_size - entry->in_pack_offset;
943 unsigned long size, used;
944 unsigned char *buf;
945 struct object_entry *base_entry = NULL;
947 use_packed_git(p);
948 buf = p->pack_base;
949 buf += entry->in_pack_offset;
951 /* We want in_pack_type even if we do not reuse delta.
952 * There is no point not reusing non-delta representations.
953 */
954 used = unpack_object_header_gently(buf, left,
955 &entry->in_pack_type, &size);
956 if (!used || left - used <= 20)
957 die("corrupt pack for %s", sha1_to_hex(entry->sha1));
959 /* Check if it is delta, and the base is also an object
960 * we are going to pack. If so we will reuse the existing
961 * delta.
962 */
963 if (!no_reuse_delta) {
964 unsigned char c, *base_name;
965 unsigned long ofs;
966 /* there is at least 20 bytes left in the pack */
967 switch (entry->in_pack_type) {
968 case OBJ_REF_DELTA:
969 base_name = buf + used;
970 used += 20;
971 break;
972 case OBJ_OFS_DELTA:
973 c = buf[used++];
974 ofs = c & 127;
975 while (c & 128) {
976 ofs += 1;
977 if (!ofs || ofs & ~(~0UL >> 7))
978 die("delta base offset overflow in pack for %s",
979 sha1_to_hex(entry->sha1));
980 c = buf[used++];
981 ofs = (ofs << 7) + (c & 127);
982 }
983 if (ofs >= entry->in_pack_offset)
984 die("delta base offset out of bound for %s",
985 sha1_to_hex(entry->sha1));
986 ofs = entry->in_pack_offset - ofs;
987 base_name = find_packed_object_name(p, ofs);
988 break;
989 default:
990 base_name = NULL;
991 }
992 if (base_name)
993 base_entry = locate_object_entry(base_name);
994 }
995 unuse_packed_git(p);
996 entry->in_pack_header_size = used;
998 if (base_entry) {
1000 /* Depth value does not matter - find_deltas()
1001 * will never consider reused delta as the
1002 * base object to deltify other objects
1003 * against, in order to avoid circular deltas.
1004 */
1006 /* uncompressed size of the delta data */
1007 entry->size = size;
1008 entry->delta = base_entry;
1009 entry->type = entry->in_pack_type;
1011 entry->delta_sibling = base_entry->delta_child;
1012 base_entry->delta_child = entry;
1014 return;
1015 }
1016 /* Otherwise we would do the usual */
1017 }
1019 if (sha1_object_info(entry->sha1, type, &entry->size))
1020 die("unable to get type of object %s",
1021 sha1_to_hex(entry->sha1));
1023 if (!strcmp(type, commit_type)) {
1024 entry->type = OBJ_COMMIT;
1025 } else if (!strcmp(type, tree_type)) {
1026 entry->type = OBJ_TREE;
1027 } else if (!strcmp(type, blob_type)) {
1028 entry->type = OBJ_BLOB;
1029 } else if (!strcmp(type, tag_type)) {
1030 entry->type = OBJ_TAG;
1031 } else
1032 die("unable to pack object %s of type %s",
1033 sha1_to_hex(entry->sha1), type);
1034 }
1036 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1037 {
1038 struct object_entry *child = me->delta_child;
1039 unsigned int m = n;
1040 while (child) {
1041 unsigned int c = check_delta_limit(child, n + 1);
1042 if (m < c)
1043 m = c;
1044 child = child->delta_sibling;
1045 }
1046 return m;
1047 }
1049 static void get_object_details(void)
1050 {
1051 int i;
1052 struct object_entry *entry;
1054 prepare_pack_ix();
1055 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1056 check_object(entry);
1058 if (nr_objects == nr_result) {
1059 /*
1060 * Depth of objects that depend on the entry -- this
1061 * is subtracted from depth-max to break too deep
1062 * delta chain because of delta data reusing.
1063 * However, we loosen this restriction when we know we
1064 * are creating a thin pack -- it will have to be
1065 * expanded on the other end anyway, so do not
1066 * artificially cut the delta chain and let it go as
1067 * deep as it wants.
1068 */
1069 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1070 if (!entry->delta && entry->delta_child)
1071 entry->delta_limit =
1072 check_delta_limit(entry, 1);
1073 }
1074 }
1076 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1078 static entry_sort_t current_sort;
1080 static int sort_comparator(const void *_a, const void *_b)
1081 {
1082 struct object_entry *a = *(struct object_entry **)_a;
1083 struct object_entry *b = *(struct object_entry **)_b;
1084 return current_sort(a,b);
1085 }
1087 static struct object_entry **create_sorted_list(entry_sort_t sort)
1088 {
1089 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1090 int i;
1092 for (i = 0; i < nr_objects; i++)
1093 list[i] = objects + i;
1094 current_sort = sort;
1095 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1096 return list;
1097 }
1099 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1100 {
1101 return hashcmp(a->sha1, b->sha1);
1102 }
1104 static struct object_entry **create_final_object_list(void)
1105 {
1106 struct object_entry **list;
1107 int i, j;
1109 for (i = nr_result = 0; i < nr_objects; i++)
1110 if (!objects[i].preferred_base)
1111 nr_result++;
1112 list = xmalloc(nr_result * sizeof(struct object_entry *));
1113 for (i = j = 0; i < nr_objects; i++) {
1114 if (!objects[i].preferred_base)
1115 list[j++] = objects + i;
1116 }
1117 current_sort = sha1_sort;
1118 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1119 return list;
1120 }
1122 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1123 {
1124 if (a->type < b->type)
1125 return -1;
1126 if (a->type > b->type)
1127 return 1;
1128 if (a->hash < b->hash)
1129 return -1;
1130 if (a->hash > b->hash)
1131 return 1;
1132 if (a->preferred_base < b->preferred_base)
1133 return -1;
1134 if (a->preferred_base > b->preferred_base)
1135 return 1;
1136 if (a->size < b->size)
1137 return -1;
1138 if (a->size > b->size)
1139 return 1;
1140 return a < b ? -1 : (a > b);
1141 }
1143 struct unpacked {
1144 struct object_entry *entry;
1145 void *data;
1146 struct delta_index *index;
1147 };
1149 /*
1150 * We search for deltas _backwards_ in a list sorted by type and
1151 * by size, so that we see progressively smaller and smaller files.
1152 * That's because we prefer deltas to be from the bigger file
1153 * to the smaller - deletes are potentially cheaper, but perhaps
1154 * more importantly, the bigger file is likely the more recent
1155 * one.
1156 */
1157 static int try_delta(struct unpacked *trg, struct unpacked *src,
1158 unsigned max_depth)
1159 {
1160 struct object_entry *trg_entry = trg->entry;
1161 struct object_entry *src_entry = src->entry;
1162 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1163 char type[10];
1164 void *delta_buf;
1166 /* Don't bother doing diffs between different types */
1167 if (trg_entry->type != src_entry->type)
1168 return -1;
1170 /* We do not compute delta to *create* objects we are not
1171 * going to pack.
1172 */
1173 if (trg_entry->preferred_base)
1174 return -1;
1176 /*
1177 * We do not bother to try a delta that we discarded
1178 * on an earlier try, but only when reusing delta data.
1179 */
1180 if (!no_reuse_delta && trg_entry->in_pack &&
1181 trg_entry->in_pack == src_entry->in_pack &&
1182 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1183 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1184 return 0;
1186 /*
1187 * If the current object is at pack edge, take the depth the
1188 * objects that depend on the current object into account --
1189 * otherwise they would become too deep.
1190 */
1191 if (trg_entry->delta_child) {
1192 if (max_depth <= trg_entry->delta_limit)
1193 return 0;
1194 max_depth -= trg_entry->delta_limit;
1195 }
1196 if (src_entry->depth >= max_depth)
1197 return 0;
1199 /* Now some size filtering heuristics. */
1200 trg_size = trg_entry->size;
1201 max_size = trg_size/2 - 20;
1202 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1203 if (max_size == 0)
1204 return 0;
1205 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1206 max_size = trg_entry->delta_size-1;
1207 src_size = src_entry->size;
1208 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1209 if (sizediff >= max_size)
1210 return 0;
1212 /* Load data if not already done */
1213 if (!trg->data) {
1214 trg->data = read_sha1_file(trg_entry->sha1, type, &sz);
1215 if (sz != trg_size)
1216 die("object %s inconsistent object length (%lu vs %lu)",
1217 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1218 }
1219 if (!src->data) {
1220 src->data = read_sha1_file(src_entry->sha1, type, &sz);
1221 if (sz != src_size)
1222 die("object %s inconsistent object length (%lu vs %lu)",
1223 sha1_to_hex(src_entry->sha1), sz, src_size);
1224 }
1225 if (!src->index) {
1226 src->index = create_delta_index(src->data, src_size);
1227 if (!src->index)
1228 die("out of memory");
1229 }
1231 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1232 if (!delta_buf)
1233 return 0;
1235 trg_entry->delta = src_entry;
1236 trg_entry->delta_size = delta_size;
1237 trg_entry->depth = src_entry->depth + 1;
1238 free(delta_buf);
1239 return 1;
1240 }
1242 static void progress_interval(int signum)
1243 {
1244 progress_update = 1;
1245 }
1247 static void find_deltas(struct object_entry **list, int window, int depth)
1248 {
1249 int i, idx;
1250 unsigned int array_size = window * sizeof(struct unpacked);
1251 struct unpacked *array = xmalloc(array_size);
1252 unsigned processed = 0;
1253 unsigned last_percent = 999;
1255 memset(array, 0, array_size);
1256 i = nr_objects;
1257 idx = 0;
1258 if (progress)
1259 fprintf(stderr, "Deltifying %d objects.\n", nr_result);
1261 while (--i >= 0) {
1262 struct object_entry *entry = list[i];
1263 struct unpacked *n = array + idx;
1264 int j;
1266 if (!entry->preferred_base)
1267 processed++;
1269 if (progress) {
1270 unsigned percent = processed * 100 / nr_result;
1271 if (percent != last_percent || progress_update) {
1272 fprintf(stderr, "%4u%% (%u/%u) done\r",
1273 percent, processed, nr_result);
1274 progress_update = 0;
1275 last_percent = percent;
1276 }
1277 }
1279 if (entry->delta)
1280 /* This happens if we decided to reuse existing
1281 * delta from a pack. "!no_reuse_delta &&" is implied.
1282 */
1283 continue;
1285 if (entry->size < 50)
1286 continue;
1287 free_delta_index(n->index);
1288 n->index = NULL;
1289 free(n->data);
1290 n->data = NULL;
1291 n->entry = entry;
1293 j = window;
1294 while (--j > 0) {
1295 unsigned int other_idx = idx + j;
1296 struct unpacked *m;
1297 if (other_idx >= window)
1298 other_idx -= window;
1299 m = array + other_idx;
1300 if (!m->entry)
1301 break;
1302 if (try_delta(n, m, depth) < 0)
1303 break;
1304 }
1305 /* if we made n a delta, and if n is already at max
1306 * depth, leaving it in the window is pointless. we
1307 * should evict it first.
1308 */
1309 if (entry->delta && depth <= entry->depth)
1310 continue;
1312 idx++;
1313 if (idx >= window)
1314 idx = 0;
1315 }
1317 if (progress)
1318 fputc('\n', stderr);
1320 for (i = 0; i < window; ++i) {
1321 free_delta_index(array[i].index);
1322 free(array[i].data);
1323 }
1324 free(array);
1325 }
1327 static void prepare_pack(int window, int depth)
1328 {
1329 get_object_details();
1330 sorted_by_type = create_sorted_list(type_size_sort);
1331 if (window && depth)
1332 find_deltas(sorted_by_type, window+1, depth);
1333 }
1335 static int reuse_cached_pack(unsigned char *sha1)
1336 {
1337 static const char cache[] = "pack-cache/pack-%s.%s";
1338 char *cached_pack, *cached_idx;
1339 int ifd, ofd, ifd_ix = -1;
1341 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1342 ifd = open(cached_pack, O_RDONLY);
1343 if (ifd < 0)
1344 return 0;
1346 if (!pack_to_stdout) {
1347 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1348 ifd_ix = open(cached_idx, O_RDONLY);
1349 if (ifd_ix < 0) {
1350 close(ifd);
1351 return 0;
1352 }
1353 }
1355 if (progress)
1356 fprintf(stderr, "Reusing %d objects pack %s\n", nr_objects,
1357 sha1_to_hex(sha1));
1359 if (pack_to_stdout) {
1360 if (copy_fd(ifd, 1))
1361 exit(1);
1362 close(ifd);
1363 }
1364 else {
1365 char name[PATH_MAX];
1366 snprintf(name, sizeof(name),
1367 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1368 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1369 if (ofd < 0)
1370 die("unable to open %s (%s)", name, strerror(errno));
1371 if (copy_fd(ifd, ofd))
1372 exit(1);
1373 close(ifd);
1375 snprintf(name, sizeof(name),
1376 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1377 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1378 if (ofd < 0)
1379 die("unable to open %s (%s)", name, strerror(errno));
1380 if (copy_fd(ifd_ix, ofd))
1381 exit(1);
1382 close(ifd_ix);
1383 puts(sha1_to_hex(sha1));
1384 }
1386 return 1;
1387 }
1389 static void setup_progress_signal(void)
1390 {
1391 struct sigaction sa;
1392 struct itimerval v;
1394 memset(&sa, 0, sizeof(sa));
1395 sa.sa_handler = progress_interval;
1396 sigemptyset(&sa.sa_mask);
1397 sa.sa_flags = SA_RESTART;
1398 sigaction(SIGALRM, &sa, NULL);
1400 v.it_interval.tv_sec = 1;
1401 v.it_interval.tv_usec = 0;
1402 v.it_value = v.it_interval;
1403 setitimer(ITIMER_REAL, &v, NULL);
1404 }
1406 static int git_pack_config(const char *k, const char *v)
1407 {
1408 if(!strcmp(k, "pack.window")) {
1409 window = git_config_int(k, v);
1410 return 0;
1411 }
1412 return git_default_config(k, v);
1413 }
1415 static void read_object_list_from_stdin(void)
1416 {
1417 char line[40 + 1 + PATH_MAX + 2];
1418 unsigned char sha1[20];
1419 unsigned hash;
1421 for (;;) {
1422 if (!fgets(line, sizeof(line), stdin)) {
1423 if (feof(stdin))
1424 break;
1425 if (!ferror(stdin))
1426 die("fgets returned NULL, not EOF, not error!");
1427 if (errno != EINTR)
1428 die("fgets: %s", strerror(errno));
1429 clearerr(stdin);
1430 continue;
1431 }
1432 if (line[0] == '-') {
1433 if (get_sha1_hex(line+1, sha1))
1434 die("expected edge sha1, got garbage:\n %s",
1435 line);
1436 add_preferred_base(sha1);
1437 continue;
1438 }
1439 if (get_sha1_hex(line, sha1))
1440 die("expected sha1, got garbage:\n %s", line);
1442 hash = name_hash(line+41);
1443 add_preferred_base_object(line+41, hash);
1444 add_object_entry(sha1, hash, 0);
1445 }
1446 }
1448 static void show_commit(struct commit *commit)
1449 {
1450 unsigned hash = name_hash("");
1451 add_preferred_base_object("", hash);
1452 add_object_entry(commit->object.sha1, hash, 0);
1453 }
1455 static void show_object(struct object_array_entry *p)
1456 {
1457 unsigned hash = name_hash(p->name);
1458 add_preferred_base_object(p->name, hash);
1459 add_object_entry(p->item->sha1, hash, 0);
1460 }
1462 static void show_edge(struct commit *commit)
1463 {
1464 add_preferred_base(commit->object.sha1);
1465 }
1467 static void get_object_list(int ac, const char **av)
1468 {
1469 struct rev_info revs;
1470 char line[1000];
1471 int flags = 0;
1473 init_revisions(&revs, NULL);
1474 save_commit_buffer = 0;
1475 track_object_refs = 0;
1476 setup_revisions(ac, av, &revs, NULL);
1478 while (fgets(line, sizeof(line), stdin) != NULL) {
1479 int len = strlen(line);
1480 if (line[len - 1] == '\n')
1481 line[--len] = 0;
1482 if (!len)
1483 break;
1484 if (*line == '-') {
1485 if (!strcmp(line, "--not")) {
1486 flags ^= UNINTERESTING;
1487 continue;
1488 }
1489 die("not a rev '%s'", line);
1490 }
1491 if (handle_revision_arg(line, &revs, flags, 1))
1492 die("bad revision '%s'", line);
1493 }
1495 prepare_revision_walk(&revs);
1496 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1497 traverse_commit_list(&revs, show_commit, show_object);
1498 }
1500 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1501 {
1502 SHA_CTX ctx;
1503 int depth = 10;
1504 struct object_entry **list;
1505 int use_internal_rev_list = 0;
1506 int thin = 0;
1507 int i;
1508 const char *rp_av[64];
1509 int rp_ac;
1511 rp_av[0] = "pack-objects";
1512 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1513 rp_ac = 2;
1515 git_config(git_pack_config);
1517 progress = isatty(2);
1518 for (i = 1; i < argc; i++) {
1519 const char *arg = argv[i];
1521 if (*arg != '-')
1522 break;
1524 if (!strcmp("--non-empty", arg)) {
1525 non_empty = 1;
1526 continue;
1527 }
1528 if (!strcmp("--local", arg)) {
1529 local = 1;
1530 continue;
1531 }
1532 if (!strcmp("--incremental", arg)) {
1533 incremental = 1;
1534 continue;
1535 }
1536 if (!strncmp("--window=", arg, 9)) {
1537 char *end;
1538 window = strtoul(arg+9, &end, 0);
1539 if (!arg[9] || *end)
1540 usage(pack_usage);
1541 continue;
1542 }
1543 if (!strncmp("--depth=", arg, 8)) {
1544 char *end;
1545 depth = strtoul(arg+8, &end, 0);
1546 if (!arg[8] || *end)
1547 usage(pack_usage);
1548 continue;
1549 }
1550 if (!strcmp("--progress", arg)) {
1551 progress = 1;
1552 continue;
1553 }
1554 if (!strcmp("--all-progress", arg)) {
1555 progress = 2;
1556 continue;
1557 }
1558 if (!strcmp("-q", arg)) {
1559 progress = 0;
1560 continue;
1561 }
1562 if (!strcmp("--no-reuse-delta", arg)) {
1563 no_reuse_delta = 1;
1564 continue;
1565 }
1566 if (!strcmp("--delta-base-offset", arg)) {
1567 allow_ofs_delta = 1;
1568 continue;
1569 }
1570 if (!strcmp("--stdout", arg)) {
1571 pack_to_stdout = 1;
1572 continue;
1573 }
1574 if (!strcmp("--revs", arg)) {
1575 use_internal_rev_list = 1;
1576 continue;
1577 }
1578 if (!strcmp("--unpacked", arg) ||
1579 !strncmp("--unpacked=", arg, 11) ||
1580 !strcmp("--all", arg)) {
1581 use_internal_rev_list = 1;
1582 if (ARRAY_SIZE(rp_av) - 1 <= rp_ac)
1583 die("too many internal rev-list options");
1584 rp_av[rp_ac++] = arg;
1585 continue;
1586 }
1587 if (!strcmp("--thin", arg)) {
1588 use_internal_rev_list = 1;
1589 thin = 1;
1590 rp_av[1] = "--objects-edge";
1591 continue;
1592 }
1593 usage(pack_usage);
1594 }
1596 /* Traditionally "pack-objects [options] base extra" failed;
1597 * we would however want to take refs parameter that would
1598 * have been given to upstream rev-list ourselves, which means
1599 * we somehow want to say what the base name is. So the
1600 * syntax would be:
1601 *
1602 * pack-objects [options] base <refs...>
1603 *
1604 * in other words, we would treat the first non-option as the
1605 * base_name and send everything else to the internal revision
1606 * walker.
1607 */
1609 if (!pack_to_stdout)
1610 base_name = argv[i++];
1612 if (pack_to_stdout != !base_name)
1613 usage(pack_usage);
1615 if (!pack_to_stdout && thin)
1616 die("--thin cannot be used to build an indexable pack.");
1618 prepare_packed_git();
1620 if (progress) {
1621 fprintf(stderr, "Generating pack...\n");
1622 setup_progress_signal();
1623 }
1625 if (!use_internal_rev_list)
1626 read_object_list_from_stdin();
1627 else {
1628 rp_av[rp_ac] = NULL;
1629 get_object_list(rp_ac, rp_av);
1630 }
1632 if (progress)
1633 fprintf(stderr, "Done counting %d objects.\n", nr_objects);
1634 sorted_by_sha = create_final_object_list();
1635 if (non_empty && !nr_result)
1636 return 0;
1638 SHA1_Init(&ctx);
1639 list = sorted_by_sha;
1640 for (i = 0; i < nr_result; i++) {
1641 struct object_entry *entry = *list++;
1642 SHA1_Update(&ctx, entry->sha1, 20);
1643 }
1644 SHA1_Final(object_list_sha1, &ctx);
1645 if (progress && (nr_objects != nr_result))
1646 fprintf(stderr, "Result has %d objects.\n", nr_result);
1648 if (reuse_cached_pack(object_list_sha1))
1649 ;
1650 else {
1651 if (nr_result)
1652 prepare_pack(window, depth);
1653 if (progress == 1 && pack_to_stdout) {
1654 /* the other end usually displays progress itself */
1655 struct itimerval v = {{0,},};
1656 setitimer(ITIMER_REAL, &v, NULL);
1657 signal(SIGALRM, SIG_IGN );
1658 progress_update = 0;
1659 }
1660 write_pack_file();
1661 if (!pack_to_stdout) {
1662 write_index_file();
1663 puts(sha1_to_hex(object_list_sha1));
1664 }
1665 }
1666 if (progress)
1667 fprintf(stderr, "Total %d (delta %d), reused %d (delta %d)\n",
1668 written, written_delta, reused, reused_delta);
1669 return 0;
1670 }