pack-objects: optimize preferred base handling a bit
[git/jnareb-git.git] / builtin-pack-objects.c
blob62a011e2e92a57ebbdb5a99e796013bedfa6cd8d
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"
16 static const char pack_usage[] = "\
17 git-pack-objects [{ -q | --progress | --all-progress }] \n\
18 [--local] [--incremental] [--window=N] [--depth=N] \n\
19 [--no-reuse-delta] [--delta-base-offset] [--non-empty] \n\
20 [--revs [--unpacked | --all]*] [--reflog] [--stdout | base-name] \n\
21 [<ref-list | <object-list]";
23 struct object_entry {
24 unsigned char sha1[20];
25 unsigned long size; /* uncompressed size */
26 off_t offset; /* offset into the final pack file;
27 * nonzero if already written.
29 unsigned int depth; /* delta depth */
30 unsigned int delta_limit; /* base adjustment for in-pack delta */
31 unsigned int hash; /* name hint hash */
32 enum object_type type;
33 enum object_type in_pack_type; /* could be delta */
34 unsigned long delta_size; /* delta data size (uncompressed) */
35 #define in_pack_header_size delta_size /* only when reusing pack data */
36 struct object_entry *delta; /* delta base object */
37 struct packed_git *in_pack; /* already in pack */
38 off_t in_pack_offset;
39 struct object_entry *delta_child; /* deltified objects who bases me */
40 struct object_entry *delta_sibling; /* other deltified objects who
41 * uses the same base as me
43 int preferred_base; /* we do not pack this, but is encouraged to
44 * be used as the base objectto delta huge
45 * objects against.
47 uint32_t crc32; /* crc of raw pack data for this object */
51 * Objects we are going to pack are collected in objects array (dynamically
52 * expanded). nr_objects & nr_alloc controls this array. They are stored
53 * in the order we see -- typically rev-list --objects order that gives us
54 * nice "minimum seek" order.
56 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
57 * elements in the objects array. The former is used to build the pack
58 * index (lists object names in the ascending order to help offset lookup),
59 * and the latter is used to group similar things together by try_delta()
60 * heuristics.
63 static unsigned char object_list_sha1[20];
64 static int non_empty;
65 static int no_reuse_delta;
66 static int local;
67 static int incremental;
68 static int allow_ofs_delta;
70 static struct object_entry **sorted_by_sha, **sorted_by_type;
71 static struct object_entry *objects;
72 static uint32_t nr_objects, nr_alloc, nr_result;
73 static const char *base_name;
74 static unsigned char pack_file_sha1[20];
75 static int progress = 1;
76 static volatile sig_atomic_t progress_update;
77 static int window = 10;
78 static int pack_to_stdout;
79 static int num_preferred_base;
82 * The object names in objects array are hashed with this hashtable,
83 * to help looking up the entry by object name. Binary search from
84 * sorted_by_sha is also possible but this was easier to code and faster.
85 * This hashtable is built after all the objects are seen.
87 static int *object_ix;
88 static int object_ix_hashsz;
91 * Pack index for existing packs give us easy access to the offsets into
92 * corresponding pack file where each object's data starts, but the entries
93 * do not store the size of the compressed representation (uncompressed
94 * size is easily available by examining the pack entry header). It is
95 * also rather expensive to find the sha1 for an object given its offset.
97 * We build a hashtable of existing packs (pack_revindex), and keep reverse
98 * index here -- pack index file is sorted by object name mapping to offset;
99 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
100 * ordered by offset, so if you know the offset of an object, next offset
101 * is where its packed representation ends and the index_nr can be used to
102 * get the object sha1 from the main index.
104 struct revindex_entry {
105 off_t offset;
106 unsigned int nr;
108 struct pack_revindex {
109 struct packed_git *p;
110 struct revindex_entry *revindex;
112 static struct pack_revindex *pack_revindex;
113 static int pack_revindex_hashsz;
116 * stats
118 static uint32_t written, written_delta;
119 static uint32_t reused, reused_delta;
121 static int pack_revindex_ix(struct packed_git *p)
123 unsigned long ui = (unsigned long)p;
124 int i;
126 ui = ui ^ (ui >> 16); /* defeat structure alignment */
127 i = (int)(ui % pack_revindex_hashsz);
128 while (pack_revindex[i].p) {
129 if (pack_revindex[i].p == p)
130 return i;
131 if (++i == pack_revindex_hashsz)
132 i = 0;
134 return -1 - i;
137 static void prepare_pack_ix(void)
139 int num;
140 struct packed_git *p;
141 for (num = 0, p = packed_git; p; p = p->next)
142 num++;
143 if (!num)
144 return;
145 pack_revindex_hashsz = num * 11;
146 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
147 for (p = packed_git; p; p = p->next) {
148 num = pack_revindex_ix(p);
149 num = - 1 - num;
150 pack_revindex[num].p = p;
152 /* revindex elements are lazily initialized */
155 static int cmp_offset(const void *a_, const void *b_)
157 const struct revindex_entry *a = a_;
158 const struct revindex_entry *b = b_;
159 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
163 * Ordered list of offsets of objects in the pack.
165 static void prepare_pack_revindex(struct pack_revindex *rix)
167 struct packed_git *p = rix->p;
168 int num_ent = p->num_objects;
169 int i;
170 const char *index = p->index_data;
172 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
173 index += 4 * 256;
175 if (p->index_version > 1) {
176 const uint32_t *off_32 =
177 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
178 const uint32_t *off_64 = off_32 + p->num_objects;
179 for (i = 0; i < num_ent; i++) {
180 uint32_t off = ntohl(*off_32++);
181 if (!(off & 0x80000000)) {
182 rix->revindex[i].offset = off;
183 } else {
184 rix->revindex[i].offset =
185 ((uint64_t)ntohl(*off_64++)) << 32;
186 rix->revindex[i].offset |=
187 ntohl(*off_64++);
189 rix->revindex[i].nr = i;
191 } else {
192 for (i = 0; i < num_ent; i++) {
193 uint32_t hl = *((uint32_t *)(index + 24 * i));
194 rix->revindex[i].offset = ntohl(hl);
195 rix->revindex[i].nr = i;
199 /* This knows the pack format -- the 20-byte trailer
200 * follows immediately after the last object data.
202 rix->revindex[num_ent].offset = p->pack_size - 20;
203 rix->revindex[num_ent].nr = -1;
204 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
207 static struct revindex_entry * find_packed_object(struct packed_git *p,
208 off_t ofs)
210 int num;
211 int lo, hi;
212 struct pack_revindex *rix;
213 struct revindex_entry *revindex;
214 num = pack_revindex_ix(p);
215 if (num < 0)
216 die("internal error: pack revindex uninitialized");
217 rix = &pack_revindex[num];
218 if (!rix->revindex)
219 prepare_pack_revindex(rix);
220 revindex = rix->revindex;
221 lo = 0;
222 hi = p->num_objects + 1;
223 do {
224 int mi = (lo + hi) / 2;
225 if (revindex[mi].offset == ofs) {
226 return revindex + mi;
228 else if (ofs < revindex[mi].offset)
229 hi = mi;
230 else
231 lo = mi + 1;
232 } while (lo < hi);
233 die("internal error: pack revindex corrupt");
236 static const unsigned char *find_packed_object_name(struct packed_git *p,
237 off_t ofs)
239 struct revindex_entry *entry = find_packed_object(p, ofs);
240 return nth_packed_object_sha1(p, entry->nr);
243 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
245 unsigned long othersize, delta_size;
246 enum object_type type;
247 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
248 void *delta_buf;
250 if (!otherbuf)
251 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
252 delta_buf = diff_delta(otherbuf, othersize,
253 buf, size, &delta_size, 0);
254 if (!delta_buf || delta_size != entry->delta_size)
255 die("delta size changed");
256 free(buf);
257 free(otherbuf);
258 return delta_buf;
262 * The per-object header is a pretty dense thing, which is
263 * - first byte: low four bits are "size", then three bits of "type",
264 * and the high bit is "size continues".
265 * - each byte afterwards: low seven bits are size continuation,
266 * with the high bit being "size continues"
268 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
270 int n = 1;
271 unsigned char c;
273 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
274 die("bad type %d", type);
276 c = (type << 4) | (size & 15);
277 size >>= 4;
278 while (size) {
279 *hdr++ = c | 0x80;
280 c = size & 0x7f;
281 size >>= 7;
282 n++;
284 *hdr = c;
285 return n;
289 * we are going to reuse the existing object data as is. make
290 * sure it is not corrupt.
292 static int check_pack_inflate(struct packed_git *p,
293 struct pack_window **w_curs,
294 off_t offset,
295 off_t len,
296 unsigned long expect)
298 z_stream stream;
299 unsigned char fakebuf[4096], *in;
300 int st;
302 memset(&stream, 0, sizeof(stream));
303 inflateInit(&stream);
304 do {
305 in = use_pack(p, w_curs, offset, &stream.avail_in);
306 stream.next_in = in;
307 stream.next_out = fakebuf;
308 stream.avail_out = sizeof(fakebuf);
309 st = inflate(&stream, Z_FINISH);
310 offset += stream.next_in - in;
311 } while (st == Z_OK || st == Z_BUF_ERROR);
312 inflateEnd(&stream);
313 return (st == Z_STREAM_END &&
314 stream.total_out == expect &&
315 stream.total_in == len) ? 0 : -1;
318 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
319 off_t offset, off_t len, unsigned int nr)
321 const uint32_t *index_crc;
322 uint32_t data_crc = crc32(0, Z_NULL, 0);
324 do {
325 unsigned int avail;
326 void *data = use_pack(p, w_curs, offset, &avail);
327 if (avail > len)
328 avail = len;
329 data_crc = crc32(data_crc, data, avail);
330 offset += avail;
331 len -= avail;
332 } while (len);
334 index_crc = p->index_data;
335 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
337 return data_crc != ntohl(*index_crc);
340 static void copy_pack_data(struct sha1file *f,
341 struct packed_git *p,
342 struct pack_window **w_curs,
343 off_t offset,
344 off_t len)
346 unsigned char *in;
347 unsigned int avail;
349 while (len) {
350 in = use_pack(p, w_curs, offset, &avail);
351 if (avail > len)
352 avail = (unsigned int)len;
353 sha1write(f, in, avail);
354 offset += avail;
355 len -= avail;
359 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
361 z_stream stream;
362 unsigned char fakebuf[4096];
363 int st;
365 memset(&stream, 0, sizeof(stream));
366 stream.next_in = data;
367 stream.avail_in = len;
368 stream.next_out = fakebuf;
369 stream.avail_out = sizeof(fakebuf);
370 inflateInit(&stream);
372 while (1) {
373 st = inflate(&stream, Z_FINISH);
374 if (st == Z_STREAM_END || st == Z_OK) {
375 st = (stream.total_out == expect &&
376 stream.total_in == len) ? 0 : -1;
377 break;
379 if (st != Z_BUF_ERROR) {
380 st = -1;
381 break;
383 stream.next_out = fakebuf;
384 stream.avail_out = sizeof(fakebuf);
386 inflateEnd(&stream);
387 return st;
390 static int revalidate_loose_object(struct object_entry *entry,
391 unsigned char *map,
392 unsigned long mapsize)
394 /* we already know this is a loose object with new type header. */
395 enum object_type type;
396 unsigned long size, used;
398 if (pack_to_stdout)
399 return 0;
401 used = unpack_object_header_gently(map, mapsize, &type, &size);
402 if (!used)
403 return -1;
404 map += used;
405 mapsize -= used;
406 return check_loose_inflate(map, mapsize, size);
409 static unsigned long write_object(struct sha1file *f,
410 struct object_entry *entry)
412 unsigned long size;
413 enum object_type type;
414 void *buf;
415 unsigned char header[10];
416 unsigned hdrlen;
417 off_t datalen;
418 enum object_type obj_type;
419 int to_reuse = 0;
421 if (!pack_to_stdout)
422 crc32_begin(f);
424 obj_type = entry->type;
425 if (! entry->in_pack)
426 to_reuse = 0; /* can't reuse what we don't have */
427 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
428 to_reuse = 1; /* check_object() decided it for us */
429 else if (obj_type != entry->in_pack_type)
430 to_reuse = 0; /* pack has delta which is unusable */
431 else if (entry->delta)
432 to_reuse = 0; /* we want to pack afresh */
433 else
434 to_reuse = 1; /* we have it in-pack undeltified,
435 * and we do not need to deltify it.
438 if (!entry->in_pack && !entry->delta) {
439 unsigned char *map;
440 unsigned long mapsize;
441 map = map_sha1_file(entry->sha1, &mapsize);
442 if (map && !legacy_loose_object(map)) {
443 /* We can copy straight into the pack file */
444 if (revalidate_loose_object(entry, map, mapsize))
445 die("corrupt loose object %s",
446 sha1_to_hex(entry->sha1));
447 sha1write(f, map, mapsize);
448 munmap(map, mapsize);
449 written++;
450 reused++;
451 return mapsize;
453 if (map)
454 munmap(map, mapsize);
457 if (!to_reuse) {
458 buf = read_sha1_file(entry->sha1, &type, &size);
459 if (!buf)
460 die("unable to read %s", sha1_to_hex(entry->sha1));
461 if (size != entry->size)
462 die("object %s size inconsistency (%lu vs %lu)",
463 sha1_to_hex(entry->sha1), size, entry->size);
464 if (entry->delta) {
465 buf = delta_against(buf, size, entry);
466 size = entry->delta_size;
467 obj_type = (allow_ofs_delta && entry->delta->offset) ?
468 OBJ_OFS_DELTA : OBJ_REF_DELTA;
471 * The object header is a byte of 'type' followed by zero or
472 * more bytes of length.
474 hdrlen = encode_header(obj_type, size, header);
475 sha1write(f, header, hdrlen);
477 if (obj_type == OBJ_OFS_DELTA) {
479 * Deltas with relative base contain an additional
480 * encoding of the relative offset for the delta
481 * base from this object's position in the pack.
483 off_t ofs = entry->offset - entry->delta->offset;
484 unsigned pos = sizeof(header) - 1;
485 header[pos] = ofs & 127;
486 while (ofs >>= 7)
487 header[--pos] = 128 | (--ofs & 127);
488 sha1write(f, header + pos, sizeof(header) - pos);
489 hdrlen += sizeof(header) - pos;
490 } else if (obj_type == OBJ_REF_DELTA) {
492 * Deltas with a base reference contain
493 * an additional 20 bytes for the base sha1.
495 sha1write(f, entry->delta->sha1, 20);
496 hdrlen += 20;
498 datalen = sha1write_compressed(f, buf, size);
499 free(buf);
501 else {
502 struct packed_git *p = entry->in_pack;
503 struct pack_window *w_curs = NULL;
504 struct revindex_entry *revidx;
505 off_t offset;
507 if (entry->delta) {
508 obj_type = (allow_ofs_delta && entry->delta->offset) ?
509 OBJ_OFS_DELTA : OBJ_REF_DELTA;
510 reused_delta++;
512 hdrlen = encode_header(obj_type, entry->size, header);
513 sha1write(f, header, hdrlen);
514 if (obj_type == OBJ_OFS_DELTA) {
515 off_t ofs = entry->offset - entry->delta->offset;
516 unsigned pos = sizeof(header) - 1;
517 header[pos] = ofs & 127;
518 while (ofs >>= 7)
519 header[--pos] = 128 | (--ofs & 127);
520 sha1write(f, header + pos, sizeof(header) - pos);
521 hdrlen += sizeof(header) - pos;
522 } else if (obj_type == OBJ_REF_DELTA) {
523 sha1write(f, entry->delta->sha1, 20);
524 hdrlen += 20;
527 offset = entry->in_pack_offset;
528 revidx = find_packed_object(p, offset);
529 datalen = revidx[1].offset - offset;
530 if (!pack_to_stdout && p->index_version > 1 &&
531 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
532 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
533 offset += entry->in_pack_header_size;
534 datalen -= entry->in_pack_header_size;
535 if (!pack_to_stdout && p->index_version == 1 &&
536 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
537 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
538 copy_pack_data(f, p, &w_curs, offset, datalen);
539 unuse_pack(&w_curs);
540 reused++;
542 if (entry->delta)
543 written_delta++;
544 written++;
545 if (!pack_to_stdout)
546 entry->crc32 = crc32_end(f);
547 return hdrlen + datalen;
550 static off_t write_one(struct sha1file *f,
551 struct object_entry *e,
552 off_t offset)
554 unsigned long size;
556 /* offset is non zero if object is written already. */
557 if (e->offset || e->preferred_base)
558 return offset;
560 /* if we are deltified, write out base object first. */
561 if (e->delta)
562 offset = write_one(f, e->delta, offset);
564 e->offset = offset;
565 size = write_object(f, e);
567 /* make sure off_t is sufficiently large not to wrap */
568 if (offset > offset + size)
569 die("pack too large for current definition of off_t");
570 return offset + size;
573 static off_t write_pack_file(void)
575 uint32_t i;
576 struct sha1file *f;
577 off_t offset, last_obj_offset = 0;
578 struct pack_header hdr;
579 unsigned last_percent = 999;
580 int do_progress = progress;
582 if (!base_name) {
583 f = sha1fd(1, "<stdout>");
584 do_progress >>= 1;
586 else
587 f = sha1create("%s-%s.%s", base_name,
588 sha1_to_hex(object_list_sha1), "pack");
589 if (do_progress)
590 fprintf(stderr, "Writing %u objects.\n", nr_result);
592 hdr.hdr_signature = htonl(PACK_SIGNATURE);
593 hdr.hdr_version = htonl(PACK_VERSION);
594 hdr.hdr_entries = htonl(nr_result);
595 sha1write(f, &hdr, sizeof(hdr));
596 offset = sizeof(hdr);
597 if (!nr_result)
598 goto done;
599 for (i = 0; i < nr_objects; i++) {
600 last_obj_offset = offset;
601 offset = write_one(f, objects + i, offset);
602 if (do_progress) {
603 unsigned percent = written * 100 / nr_result;
604 if (progress_update || percent != last_percent) {
605 fprintf(stderr, "%4u%% (%u/%u) done\r",
606 percent, written, nr_result);
607 progress_update = 0;
608 last_percent = percent;
612 if (do_progress)
613 fputc('\n', stderr);
614 done:
615 if (written != nr_result)
616 die("wrote %u objects while expecting %u", written, nr_result);
617 sha1close(f, pack_file_sha1, 1);
619 return last_obj_offset;
622 static uint32_t index_default_version = 1;
623 static uint32_t index_off32_limit = 0x7fffffff;
625 static void write_index_file(off_t last_obj_offset)
627 uint32_t i;
628 struct sha1file *f = sha1create("%s-%s.%s", base_name,
629 sha1_to_hex(object_list_sha1), "idx");
630 struct object_entry **list = sorted_by_sha;
631 struct object_entry **last = list + nr_result;
632 uint32_t array[256];
633 uint32_t index_version;
635 /* if last object's offset is >= 2^31 we should use index V2 */
636 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
638 /* index versions 2 and above need a header */
639 if (index_version >= 2) {
640 struct pack_idx_header hdr;
641 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
642 hdr.idx_version = htonl(index_version);
643 sha1write(f, &hdr, sizeof(hdr));
647 * Write the first-level table (the list is sorted,
648 * but we use a 256-entry lookup to be able to avoid
649 * having to do eight extra binary search iterations).
651 for (i = 0; i < 256; i++) {
652 struct object_entry **next = list;
653 while (next < last) {
654 struct object_entry *entry = *next;
655 if (entry->sha1[0] != i)
656 break;
657 next++;
659 array[i] = htonl(next - sorted_by_sha);
660 list = next;
662 sha1write(f, array, 256 * 4);
665 * Write the actual SHA1 entries..
667 list = sorted_by_sha;
668 for (i = 0; i < nr_result; i++) {
669 struct object_entry *entry = *list++;
670 if (index_version < 2) {
671 uint32_t offset = htonl(entry->offset);
672 sha1write(f, &offset, 4);
674 sha1write(f, entry->sha1, 20);
677 if (index_version >= 2) {
678 unsigned int nr_large_offset = 0;
680 /* write the crc32 table */
681 list = sorted_by_sha;
682 for (i = 0; i < nr_objects; i++) {
683 struct object_entry *entry = *list++;
684 uint32_t crc32_val = htonl(entry->crc32);
685 sha1write(f, &crc32_val, 4);
688 /* write the 32-bit offset table */
689 list = sorted_by_sha;
690 for (i = 0; i < nr_objects; i++) {
691 struct object_entry *entry = *list++;
692 uint32_t offset = (entry->offset <= index_off32_limit) ?
693 entry->offset : (0x80000000 | nr_large_offset++);
694 offset = htonl(offset);
695 sha1write(f, &offset, 4);
698 /* write the large offset table */
699 list = sorted_by_sha;
700 while (nr_large_offset) {
701 struct object_entry *entry = *list++;
702 uint64_t offset = entry->offset;
703 if (offset > index_off32_limit) {
704 uint32_t split[2];
705 split[0] = htonl(offset >> 32);
706 split[1] = htonl(offset & 0xffffffff);
707 sha1write(f, split, 8);
708 nr_large_offset--;
713 sha1write(f, pack_file_sha1, 20);
714 sha1close(f, NULL, 1);
717 static int locate_object_entry_hash(const unsigned char *sha1)
719 int i;
720 unsigned int ui;
721 memcpy(&ui, sha1, sizeof(unsigned int));
722 i = ui % object_ix_hashsz;
723 while (0 < object_ix[i]) {
724 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
725 return i;
726 if (++i == object_ix_hashsz)
727 i = 0;
729 return -1 - i;
732 static struct object_entry *locate_object_entry(const unsigned char *sha1)
734 int i;
736 if (!object_ix_hashsz)
737 return NULL;
739 i = locate_object_entry_hash(sha1);
740 if (0 <= i)
741 return &objects[object_ix[i]-1];
742 return NULL;
745 static void rehash_objects(void)
747 uint32_t i;
748 struct object_entry *oe;
750 object_ix_hashsz = nr_objects * 3;
751 if (object_ix_hashsz < 1024)
752 object_ix_hashsz = 1024;
753 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
754 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
755 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
756 int ix = locate_object_entry_hash(oe->sha1);
757 if (0 <= ix)
758 continue;
759 ix = -1 - ix;
760 object_ix[ix] = i + 1;
764 static unsigned name_hash(const char *name)
766 unsigned char c;
767 unsigned hash = 0;
770 * This effectively just creates a sortable number from the
771 * last sixteen non-whitespace characters. Last characters
772 * count "most", so things that end in ".c" sort together.
774 while ((c = *name++) != 0) {
775 if (isspace(c))
776 continue;
777 hash = (hash >> 2) + (c << 24);
779 return hash;
782 static int add_object_entry(const unsigned char *sha1, unsigned hash, int exclude)
784 struct object_entry *entry;
785 struct packed_git *p, *found_pack = NULL;
786 off_t found_offset = 0;
787 int ix;
789 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
790 if (ix >= 0) {
791 if (exclude) {
792 entry = objects + object_ix[ix] - 1;
793 entry->preferred_base = 1;
795 return 0;
798 if (!exclude) {
799 for (p = packed_git; p; p = p->next) {
800 off_t offset = find_pack_entry_one(sha1, p);
801 if (offset) {
802 if (incremental)
803 return 0;
804 if (local && !p->pack_local)
805 return 0;
806 if (!found_pack) {
807 found_offset = offset;
808 found_pack = p;
814 if (nr_objects >= nr_alloc) {
815 nr_alloc = (nr_alloc + 1024) * 3 / 2;
816 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
819 entry = objects + nr_objects++;
820 memset(entry, 0, sizeof(*entry));
821 hashcpy(entry->sha1, sha1);
822 entry->hash = hash;
823 if (exclude)
824 entry->preferred_base = 1;
825 if (found_pack) {
826 entry->in_pack = found_pack;
827 entry->in_pack_offset = found_offset;
830 if (object_ix_hashsz * 3 <= nr_objects * 4)
831 rehash_objects();
832 else
833 object_ix[-1 - ix] = nr_objects;
835 if (progress_update) {
836 fprintf(stderr, "Counting objects...%u\r", nr_objects);
837 progress_update = 0;
840 return 1;
843 struct pbase_tree_cache {
844 unsigned char sha1[20];
845 int ref;
846 int temporary;
847 void *tree_data;
848 unsigned long tree_size;
851 static struct pbase_tree_cache *(pbase_tree_cache[256]);
852 static int pbase_tree_cache_ix(const unsigned char *sha1)
854 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
856 static int pbase_tree_cache_ix_incr(int ix)
858 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
861 static struct pbase_tree {
862 struct pbase_tree *next;
863 /* This is a phony "cache" entry; we are not
864 * going to evict it nor find it through _get()
865 * mechanism -- this is for the toplevel node that
866 * would almost always change with any commit.
868 struct pbase_tree_cache pcache;
869 } *pbase_tree;
871 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
873 struct pbase_tree_cache *ent, *nent;
874 void *data;
875 unsigned long size;
876 enum object_type type;
877 int neigh;
878 int my_ix = pbase_tree_cache_ix(sha1);
879 int available_ix = -1;
881 /* pbase-tree-cache acts as a limited hashtable.
882 * your object will be found at your index or within a few
883 * slots after that slot if it is cached.
885 for (neigh = 0; neigh < 8; neigh++) {
886 ent = pbase_tree_cache[my_ix];
887 if (ent && !hashcmp(ent->sha1, sha1)) {
888 ent->ref++;
889 return ent;
891 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
892 ((0 <= available_ix) &&
893 (!ent && pbase_tree_cache[available_ix])))
894 available_ix = my_ix;
895 if (!ent)
896 break;
897 my_ix = pbase_tree_cache_ix_incr(my_ix);
900 /* Did not find one. Either we got a bogus request or
901 * we need to read and perhaps cache.
903 data = read_sha1_file(sha1, &type, &size);
904 if (!data)
905 return NULL;
906 if (type != OBJ_TREE) {
907 free(data);
908 return NULL;
911 /* We need to either cache or return a throwaway copy */
913 if (available_ix < 0)
914 ent = NULL;
915 else {
916 ent = pbase_tree_cache[available_ix];
917 my_ix = available_ix;
920 if (!ent) {
921 nent = xmalloc(sizeof(*nent));
922 nent->temporary = (available_ix < 0);
924 else {
925 /* evict and reuse */
926 free(ent->tree_data);
927 nent = ent;
929 hashcpy(nent->sha1, sha1);
930 nent->tree_data = data;
931 nent->tree_size = size;
932 nent->ref = 1;
933 if (!nent->temporary)
934 pbase_tree_cache[my_ix] = nent;
935 return nent;
938 static void pbase_tree_put(struct pbase_tree_cache *cache)
940 if (!cache->temporary) {
941 cache->ref--;
942 return;
944 free(cache->tree_data);
945 free(cache);
948 static int name_cmp_len(const char *name)
950 int i;
951 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
953 return i;
956 static void add_pbase_object(struct tree_desc *tree,
957 const char *name,
958 int cmplen,
959 const char *fullname)
961 struct name_entry entry;
962 int cmp;
964 while (tree_entry(tree,&entry)) {
965 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
966 memcmp(name, entry.path, cmplen);
967 if (cmp > 0)
968 continue;
969 if (cmp < 0)
970 return;
971 if (name[cmplen] != '/') {
972 unsigned hash = name_hash(fullname);
973 add_object_entry(entry.sha1, hash, 1);
974 return;
976 if (S_ISDIR(entry.mode)) {
977 struct tree_desc sub;
978 struct pbase_tree_cache *tree;
979 const char *down = name+cmplen+1;
980 int downlen = name_cmp_len(down);
982 tree = pbase_tree_get(entry.sha1);
983 if (!tree)
984 return;
985 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
987 add_pbase_object(&sub, down, downlen, fullname);
988 pbase_tree_put(tree);
993 static unsigned *done_pbase_paths;
994 static int done_pbase_paths_num;
995 static int done_pbase_paths_alloc;
996 static int done_pbase_path_pos(unsigned hash)
998 int lo = 0;
999 int hi = done_pbase_paths_num;
1000 while (lo < hi) {
1001 int mi = (hi + lo) / 2;
1002 if (done_pbase_paths[mi] == hash)
1003 return mi;
1004 if (done_pbase_paths[mi] < hash)
1005 hi = mi;
1006 else
1007 lo = mi + 1;
1009 return -lo-1;
1012 static int check_pbase_path(unsigned hash)
1014 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1015 if (0 <= pos)
1016 return 1;
1017 pos = -pos - 1;
1018 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1019 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1020 done_pbase_paths = xrealloc(done_pbase_paths,
1021 done_pbase_paths_alloc *
1022 sizeof(unsigned));
1024 done_pbase_paths_num++;
1025 if (pos < done_pbase_paths_num)
1026 memmove(done_pbase_paths + pos + 1,
1027 done_pbase_paths + pos,
1028 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1029 done_pbase_paths[pos] = hash;
1030 return 0;
1033 static void add_preferred_base_object(const char *name, unsigned hash)
1035 struct pbase_tree *it;
1036 int cmplen;
1038 if (!num_preferred_base || check_pbase_path(hash))
1039 return;
1041 cmplen = name_cmp_len(name);
1042 for (it = pbase_tree; it; it = it->next) {
1043 if (cmplen == 0) {
1044 add_object_entry(it->pcache.sha1, 0, 1);
1046 else {
1047 struct tree_desc tree;
1048 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1049 add_pbase_object(&tree, name, cmplen, name);
1054 static void add_preferred_base(unsigned char *sha1)
1056 struct pbase_tree *it;
1057 void *data;
1058 unsigned long size;
1059 unsigned char tree_sha1[20];
1061 if (window <= num_preferred_base++)
1062 return;
1064 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1065 if (!data)
1066 return;
1068 for (it = pbase_tree; it; it = it->next) {
1069 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1070 free(data);
1071 return;
1075 it = xcalloc(1, sizeof(*it));
1076 it->next = pbase_tree;
1077 pbase_tree = it;
1079 hashcpy(it->pcache.sha1, tree_sha1);
1080 it->pcache.tree_data = data;
1081 it->pcache.tree_size = size;
1084 static void check_object(struct object_entry *entry)
1086 if (entry->in_pack && !entry->preferred_base) {
1087 struct packed_git *p = entry->in_pack;
1088 struct pack_window *w_curs = NULL;
1089 unsigned long size, used;
1090 unsigned int avail;
1091 unsigned char *buf;
1092 struct object_entry *base_entry = NULL;
1094 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1096 /* We want in_pack_type even if we do not reuse delta.
1097 * There is no point not reusing non-delta representations.
1099 used = unpack_object_header_gently(buf, avail,
1100 &entry->in_pack_type, &size);
1102 /* Check if it is delta, and the base is also an object
1103 * we are going to pack. If so we will reuse the existing
1104 * delta.
1106 if (!no_reuse_delta) {
1107 unsigned char c;
1108 const unsigned char *base_name;
1109 off_t ofs;
1110 unsigned long used_0;
1111 /* there is at least 20 bytes left in the pack */
1112 switch (entry->in_pack_type) {
1113 case OBJ_REF_DELTA:
1114 base_name = use_pack(p, &w_curs,
1115 entry->in_pack_offset + used, NULL);
1116 used += 20;
1117 break;
1118 case OBJ_OFS_DELTA:
1119 buf = use_pack(p, &w_curs,
1120 entry->in_pack_offset + used, NULL);
1121 used_0 = 0;
1122 c = buf[used_0++];
1123 ofs = c & 127;
1124 while (c & 128) {
1125 ofs += 1;
1126 if (!ofs || MSB(ofs, 7))
1127 die("delta base offset overflow in pack for %s",
1128 sha1_to_hex(entry->sha1));
1129 c = buf[used_0++];
1130 ofs = (ofs << 7) + (c & 127);
1132 if (ofs >= entry->in_pack_offset)
1133 die("delta base offset out of bound for %s",
1134 sha1_to_hex(entry->sha1));
1135 ofs = entry->in_pack_offset - ofs;
1136 base_name = find_packed_object_name(p, ofs);
1137 used += used_0;
1138 break;
1139 default:
1140 base_name = NULL;
1142 if (base_name)
1143 base_entry = locate_object_entry(base_name);
1145 unuse_pack(&w_curs);
1146 entry->in_pack_header_size = used;
1148 if (base_entry) {
1150 /* Depth value does not matter - find_deltas()
1151 * will never consider reused delta as the
1152 * base object to deltify other objects
1153 * against, in order to avoid circular deltas.
1156 /* uncompressed size of the delta data */
1157 entry->size = size;
1158 entry->delta = base_entry;
1159 entry->type = entry->in_pack_type;
1161 entry->delta_sibling = base_entry->delta_child;
1162 base_entry->delta_child = entry;
1164 return;
1166 /* Otherwise we would do the usual */
1169 entry->type = sha1_object_info(entry->sha1, &entry->size);
1170 if (entry->type < 0)
1171 die("unable to get type of object %s",
1172 sha1_to_hex(entry->sha1));
1175 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1177 struct object_entry *child = me->delta_child;
1178 unsigned int m = n;
1179 while (child) {
1180 unsigned int c = check_delta_limit(child, n + 1);
1181 if (m < c)
1182 m = c;
1183 child = child->delta_sibling;
1185 return m;
1188 static void get_object_details(void)
1190 uint32_t i;
1191 struct object_entry *entry;
1193 prepare_pack_ix();
1194 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1195 check_object(entry);
1197 if (nr_objects == nr_result) {
1199 * Depth of objects that depend on the entry -- this
1200 * is subtracted from depth-max to break too deep
1201 * delta chain because of delta data reusing.
1202 * However, we loosen this restriction when we know we
1203 * are creating a thin pack -- it will have to be
1204 * expanded on the other end anyway, so do not
1205 * artificially cut the delta chain and let it go as
1206 * deep as it wants.
1208 for (i = 0, entry = objects; i < nr_objects; i++, entry++)
1209 if (!entry->delta && entry->delta_child)
1210 entry->delta_limit =
1211 check_delta_limit(entry, 1);
1215 typedef int (*entry_sort_t)(const struct object_entry *, const struct object_entry *);
1217 static entry_sort_t current_sort;
1219 static int sort_comparator(const void *_a, const void *_b)
1221 struct object_entry *a = *(struct object_entry **)_a;
1222 struct object_entry *b = *(struct object_entry **)_b;
1223 return current_sort(a,b);
1226 static struct object_entry **create_sorted_list(entry_sort_t sort)
1228 struct object_entry **list = xmalloc(nr_objects * sizeof(struct object_entry *));
1229 uint32_t i;
1231 for (i = 0; i < nr_objects; i++)
1232 list[i] = objects + i;
1233 current_sort = sort;
1234 qsort(list, nr_objects, sizeof(struct object_entry *), sort_comparator);
1235 return list;
1238 static int sha1_sort(const struct object_entry *a, const struct object_entry *b)
1240 return hashcmp(a->sha1, b->sha1);
1243 static struct object_entry **create_final_object_list(void)
1245 struct object_entry **list;
1246 uint32_t i, j;
1248 for (i = nr_result = 0; i < nr_objects; i++)
1249 if (!objects[i].preferred_base)
1250 nr_result++;
1251 list = xmalloc(nr_result * sizeof(struct object_entry *));
1252 for (i = j = 0; i < nr_objects; i++) {
1253 if (!objects[i].preferred_base)
1254 list[j++] = objects + i;
1256 current_sort = sha1_sort;
1257 qsort(list, nr_result, sizeof(struct object_entry *), sort_comparator);
1258 return list;
1261 static int type_size_sort(const struct object_entry *a, const struct object_entry *b)
1263 if (a->type < b->type)
1264 return -1;
1265 if (a->type > b->type)
1266 return 1;
1267 if (a->hash < b->hash)
1268 return -1;
1269 if (a->hash > b->hash)
1270 return 1;
1271 if (a->preferred_base < b->preferred_base)
1272 return -1;
1273 if (a->preferred_base > b->preferred_base)
1274 return 1;
1275 if (a->size < b->size)
1276 return -1;
1277 if (a->size > b->size)
1278 return 1;
1279 return a < b ? -1 : (a > b);
1282 struct unpacked {
1283 struct object_entry *entry;
1284 void *data;
1285 struct delta_index *index;
1289 * We search for deltas _backwards_ in a list sorted by type and
1290 * by size, so that we see progressively smaller and smaller files.
1291 * That's because we prefer deltas to be from the bigger file
1292 * to the smaller - deletes are potentially cheaper, but perhaps
1293 * more importantly, the bigger file is likely the more recent
1294 * one.
1296 static int try_delta(struct unpacked *trg, struct unpacked *src,
1297 unsigned max_depth)
1299 struct object_entry *trg_entry = trg->entry;
1300 struct object_entry *src_entry = src->entry;
1301 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1302 enum object_type type;
1303 void *delta_buf;
1305 /* Don't bother doing diffs between different types */
1306 if (trg_entry->type != src_entry->type)
1307 return -1;
1309 /* We do not compute delta to *create* objects we are not
1310 * going to pack.
1312 if (trg_entry->preferred_base)
1313 return -1;
1316 * We do not bother to try a delta that we discarded
1317 * on an earlier try, but only when reusing delta data.
1319 if (!no_reuse_delta && trg_entry->in_pack &&
1320 trg_entry->in_pack == src_entry->in_pack &&
1321 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1322 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1323 return 0;
1326 * If the current object is at pack edge, take the depth the
1327 * objects that depend on the current object into account --
1328 * otherwise they would become too deep.
1330 if (trg_entry->delta_child) {
1331 if (max_depth <= trg_entry->delta_limit)
1332 return 0;
1333 max_depth -= trg_entry->delta_limit;
1335 if (src_entry->depth >= max_depth)
1336 return 0;
1338 /* Now some size filtering heuristics. */
1339 trg_size = trg_entry->size;
1340 max_size = trg_size/2 - 20;
1341 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1342 if (max_size == 0)
1343 return 0;
1344 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1345 max_size = trg_entry->delta_size-1;
1346 src_size = src_entry->size;
1347 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1348 if (sizediff >= max_size)
1349 return 0;
1351 /* Load data if not already done */
1352 if (!trg->data) {
1353 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1354 if (sz != trg_size)
1355 die("object %s inconsistent object length (%lu vs %lu)",
1356 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1358 if (!src->data) {
1359 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1360 if (sz != src_size)
1361 die("object %s inconsistent object length (%lu vs %lu)",
1362 sha1_to_hex(src_entry->sha1), sz, src_size);
1364 if (!src->index) {
1365 src->index = create_delta_index(src->data, src_size);
1366 if (!src->index)
1367 die("out of memory");
1370 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1371 if (!delta_buf)
1372 return 0;
1374 trg_entry->delta = src_entry;
1375 trg_entry->delta_size = delta_size;
1376 trg_entry->depth = src_entry->depth + 1;
1377 free(delta_buf);
1378 return 1;
1381 static void progress_interval(int signum)
1383 progress_update = 1;
1386 static void find_deltas(struct object_entry **list, int window, int depth)
1388 uint32_t i = nr_objects, idx = 0, processed = 0;
1389 unsigned int array_size = window * sizeof(struct unpacked);
1390 struct unpacked *array;
1391 unsigned last_percent = 999;
1393 if (!nr_objects)
1394 return;
1395 array = xmalloc(array_size);
1396 memset(array, 0, array_size);
1397 if (progress)
1398 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
1400 do {
1401 struct object_entry *entry = list[--i];
1402 struct unpacked *n = array + idx;
1403 int j;
1405 if (!entry->preferred_base)
1406 processed++;
1408 if (progress) {
1409 unsigned percent = processed * 100 / nr_result;
1410 if (percent != last_percent || progress_update) {
1411 fprintf(stderr, "%4u%% (%u/%u) done\r",
1412 percent, processed, nr_result);
1413 progress_update = 0;
1414 last_percent = percent;
1418 if (entry->delta)
1419 /* This happens if we decided to reuse existing
1420 * delta from a pack. "!no_reuse_delta &&" is implied.
1422 continue;
1424 if (entry->size < 50)
1425 continue;
1426 free_delta_index(n->index);
1427 n->index = NULL;
1428 free(n->data);
1429 n->data = NULL;
1430 n->entry = entry;
1432 j = window;
1433 while (--j > 0) {
1434 uint32_t other_idx = idx + j;
1435 struct unpacked *m;
1436 if (other_idx >= window)
1437 other_idx -= window;
1438 m = array + other_idx;
1439 if (!m->entry)
1440 break;
1441 if (try_delta(n, m, depth) < 0)
1442 break;
1444 /* if we made n a delta, and if n is already at max
1445 * depth, leaving it in the window is pointless. we
1446 * should evict it first.
1448 if (entry->delta && depth <= entry->depth)
1449 continue;
1451 idx++;
1452 if (idx >= window)
1453 idx = 0;
1454 } while (i > 0);
1456 if (progress)
1457 fputc('\n', stderr);
1459 for (i = 0; i < window; ++i) {
1460 free_delta_index(array[i].index);
1461 free(array[i].data);
1463 free(array);
1466 static void prepare_pack(int window, int depth)
1468 get_object_details();
1469 sorted_by_type = create_sorted_list(type_size_sort);
1470 if (window && depth)
1471 find_deltas(sorted_by_type, window+1, depth);
1474 static int reuse_cached_pack(unsigned char *sha1)
1476 static const char cache[] = "pack-cache/pack-%s.%s";
1477 char *cached_pack, *cached_idx;
1478 int ifd, ofd, ifd_ix = -1;
1480 cached_pack = git_path(cache, sha1_to_hex(sha1), "pack");
1481 ifd = open(cached_pack, O_RDONLY);
1482 if (ifd < 0)
1483 return 0;
1485 if (!pack_to_stdout) {
1486 cached_idx = git_path(cache, sha1_to_hex(sha1), "idx");
1487 ifd_ix = open(cached_idx, O_RDONLY);
1488 if (ifd_ix < 0) {
1489 close(ifd);
1490 return 0;
1494 if (progress)
1495 fprintf(stderr, "Reusing %u objects pack %s\n", nr_objects,
1496 sha1_to_hex(sha1));
1498 if (pack_to_stdout) {
1499 if (copy_fd(ifd, 1))
1500 exit(1);
1501 close(ifd);
1503 else {
1504 char name[PATH_MAX];
1505 snprintf(name, sizeof(name),
1506 "%s-%s.%s", base_name, sha1_to_hex(sha1), "pack");
1507 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1508 if (ofd < 0)
1509 die("unable to open %s (%s)", name, strerror(errno));
1510 if (copy_fd(ifd, ofd))
1511 exit(1);
1512 close(ifd);
1514 snprintf(name, sizeof(name),
1515 "%s-%s.%s", base_name, sha1_to_hex(sha1), "idx");
1516 ofd = open(name, O_CREAT | O_EXCL | O_WRONLY, 0666);
1517 if (ofd < 0)
1518 die("unable to open %s (%s)", name, strerror(errno));
1519 if (copy_fd(ifd_ix, ofd))
1520 exit(1);
1521 close(ifd_ix);
1522 puts(sha1_to_hex(sha1));
1525 return 1;
1528 static void setup_progress_signal(void)
1530 struct sigaction sa;
1531 struct itimerval v;
1533 memset(&sa, 0, sizeof(sa));
1534 sa.sa_handler = progress_interval;
1535 sigemptyset(&sa.sa_mask);
1536 sa.sa_flags = SA_RESTART;
1537 sigaction(SIGALRM, &sa, NULL);
1539 v.it_interval.tv_sec = 1;
1540 v.it_interval.tv_usec = 0;
1541 v.it_value = v.it_interval;
1542 setitimer(ITIMER_REAL, &v, NULL);
1545 static int git_pack_config(const char *k, const char *v)
1547 if(!strcmp(k, "pack.window")) {
1548 window = git_config_int(k, v);
1549 return 0;
1551 return git_default_config(k, v);
1554 static void read_object_list_from_stdin(void)
1556 char line[40 + 1 + PATH_MAX + 2];
1557 unsigned char sha1[20];
1558 unsigned hash;
1560 for (;;) {
1561 if (!fgets(line, sizeof(line), stdin)) {
1562 if (feof(stdin))
1563 break;
1564 if (!ferror(stdin))
1565 die("fgets returned NULL, not EOF, not error!");
1566 if (errno != EINTR)
1567 die("fgets: %s", strerror(errno));
1568 clearerr(stdin);
1569 continue;
1571 if (line[0] == '-') {
1572 if (get_sha1_hex(line+1, sha1))
1573 die("expected edge sha1, got garbage:\n %s",
1574 line);
1575 add_preferred_base(sha1);
1576 continue;
1578 if (get_sha1_hex(line, sha1))
1579 die("expected sha1, got garbage:\n %s", line);
1581 hash = name_hash(line+41);
1582 add_preferred_base_object(line+41, hash);
1583 add_object_entry(sha1, hash, 0);
1587 static void show_commit(struct commit *commit)
1589 add_object_entry(commit->object.sha1, 0, 0);
1592 static void show_object(struct object_array_entry *p)
1594 unsigned hash = name_hash(p->name);
1595 add_preferred_base_object(p->name, hash);
1596 add_object_entry(p->item->sha1, hash, 0);
1599 static void show_edge(struct commit *commit)
1601 add_preferred_base(commit->object.sha1);
1604 static void get_object_list(int ac, const char **av)
1606 struct rev_info revs;
1607 char line[1000];
1608 int flags = 0;
1610 init_revisions(&revs, NULL);
1611 save_commit_buffer = 0;
1612 track_object_refs = 0;
1613 setup_revisions(ac, av, &revs, NULL);
1615 while (fgets(line, sizeof(line), stdin) != NULL) {
1616 int len = strlen(line);
1617 if (line[len - 1] == '\n')
1618 line[--len] = 0;
1619 if (!len)
1620 break;
1621 if (*line == '-') {
1622 if (!strcmp(line, "--not")) {
1623 flags ^= UNINTERESTING;
1624 continue;
1626 die("not a rev '%s'", line);
1628 if (handle_revision_arg(line, &revs, flags, 1))
1629 die("bad revision '%s'", line);
1632 prepare_revision_walk(&revs);
1633 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1634 traverse_commit_list(&revs, show_commit, show_object);
1637 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1639 SHA_CTX ctx;
1640 int depth = 10;
1641 struct object_entry **list;
1642 int use_internal_rev_list = 0;
1643 int thin = 0;
1644 uint32_t i;
1645 const char **rp_av;
1646 int rp_ac_alloc = 64;
1647 int rp_ac;
1649 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1651 rp_av[0] = "pack-objects";
1652 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1653 rp_ac = 2;
1655 git_config(git_pack_config);
1657 progress = isatty(2);
1658 for (i = 1; i < argc; i++) {
1659 const char *arg = argv[i];
1661 if (*arg != '-')
1662 break;
1664 if (!strcmp("--non-empty", arg)) {
1665 non_empty = 1;
1666 continue;
1668 if (!strcmp("--local", arg)) {
1669 local = 1;
1670 continue;
1672 if (!strcmp("--incremental", arg)) {
1673 incremental = 1;
1674 continue;
1676 if (!prefixcmp(arg, "--window=")) {
1677 char *end;
1678 window = strtoul(arg+9, &end, 0);
1679 if (!arg[9] || *end)
1680 usage(pack_usage);
1681 continue;
1683 if (!prefixcmp(arg, "--depth=")) {
1684 char *end;
1685 depth = strtoul(arg+8, &end, 0);
1686 if (!arg[8] || *end)
1687 usage(pack_usage);
1688 continue;
1690 if (!strcmp("--progress", arg)) {
1691 progress = 1;
1692 continue;
1694 if (!strcmp("--all-progress", arg)) {
1695 progress = 2;
1696 continue;
1698 if (!strcmp("-q", arg)) {
1699 progress = 0;
1700 continue;
1702 if (!strcmp("--no-reuse-delta", arg)) {
1703 no_reuse_delta = 1;
1704 continue;
1706 if (!strcmp("--delta-base-offset", arg)) {
1707 allow_ofs_delta = 1;
1708 continue;
1710 if (!strcmp("--stdout", arg)) {
1711 pack_to_stdout = 1;
1712 continue;
1714 if (!strcmp("--revs", arg)) {
1715 use_internal_rev_list = 1;
1716 continue;
1718 if (!strcmp("--unpacked", arg) ||
1719 !prefixcmp(arg, "--unpacked=") ||
1720 !strcmp("--reflog", arg) ||
1721 !strcmp("--all", arg)) {
1722 use_internal_rev_list = 1;
1723 if (rp_ac >= rp_ac_alloc - 1) {
1724 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1725 rp_av = xrealloc(rp_av,
1726 rp_ac_alloc * sizeof(*rp_av));
1728 rp_av[rp_ac++] = arg;
1729 continue;
1731 if (!strcmp("--thin", arg)) {
1732 use_internal_rev_list = 1;
1733 thin = 1;
1734 rp_av[1] = "--objects-edge";
1735 continue;
1737 if (!prefixcmp(arg, "--index-version=")) {
1738 char *c;
1739 index_default_version = strtoul(arg + 16, &c, 10);
1740 if (index_default_version > 2)
1741 die("bad %s", arg);
1742 if (*c == ',')
1743 index_off32_limit = strtoul(c+1, &c, 0);
1744 if (*c || index_off32_limit & 0x80000000)
1745 die("bad %s", arg);
1746 continue;
1748 usage(pack_usage);
1751 /* Traditionally "pack-objects [options] base extra" failed;
1752 * we would however want to take refs parameter that would
1753 * have been given to upstream rev-list ourselves, which means
1754 * we somehow want to say what the base name is. So the
1755 * syntax would be:
1757 * pack-objects [options] base <refs...>
1759 * in other words, we would treat the first non-option as the
1760 * base_name and send everything else to the internal revision
1761 * walker.
1764 if (!pack_to_stdout)
1765 base_name = argv[i++];
1767 if (pack_to_stdout != !base_name)
1768 usage(pack_usage);
1770 if (!pack_to_stdout && thin)
1771 die("--thin cannot be used to build an indexable pack.");
1773 prepare_packed_git();
1775 if (progress) {
1776 fprintf(stderr, "Generating pack...\n");
1777 setup_progress_signal();
1780 if (!use_internal_rev_list)
1781 read_object_list_from_stdin();
1782 else {
1783 rp_av[rp_ac] = NULL;
1784 get_object_list(rp_ac, rp_av);
1787 if (progress)
1788 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1789 sorted_by_sha = create_final_object_list();
1790 if (non_empty && !nr_result)
1791 return 0;
1793 SHA1_Init(&ctx);
1794 list = sorted_by_sha;
1795 for (i = 0; i < nr_result; i++) {
1796 struct object_entry *entry = *list++;
1797 SHA1_Update(&ctx, entry->sha1, 20);
1799 SHA1_Final(object_list_sha1, &ctx);
1800 if (progress && (nr_objects != nr_result))
1801 fprintf(stderr, "Result has %u objects.\n", nr_result);
1803 if (reuse_cached_pack(object_list_sha1))
1805 else {
1806 off_t last_obj_offset;
1807 if (nr_result)
1808 prepare_pack(window, depth);
1809 if (progress == 1 && pack_to_stdout) {
1810 /* the other end usually displays progress itself */
1811 struct itimerval v = {{0,},};
1812 setitimer(ITIMER_REAL, &v, NULL);
1813 signal(SIGALRM, SIG_IGN );
1814 progress_update = 0;
1816 last_obj_offset = write_pack_file();
1817 if (!pack_to_stdout) {
1818 write_index_file(last_obj_offset);
1819 puts(sha1_to_hex(object_list_sha1));
1822 if (progress)
1823 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1824 written, written_delta, reused, reused_delta);
1825 return 0;