![[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"
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 uint32_t crc32; /* crc of raw pack data for this object */
26 off_t offset; /* offset into the final pack file */
27 unsigned long size; /* uncompressed size */
28 unsigned int hash; /* name hint hash */
29 unsigned int depth; /* delta depth */
30 struct packed_git *in_pack; /* already in pack */
31 off_t in_pack_offset;
32 struct object_entry *delta; /* delta base object */
33 struct object_entry *delta_child; /* deltified objects who bases me */
34 struct object_entry *delta_sibling; /* other deltified objects who
35 * uses the same base as me
36 */
37 unsigned long delta_size; /* delta data size (uncompressed) */
38 enum object_type type;
39 enum object_type in_pack_type; /* could be delta */
40 unsigned char in_pack_header_size;
41 unsigned char preferred_base; /* we do not pack this, but is available
42 * to be used as the base objectto delta
43 * objects against.
44 */
45 };
47 /*
48 * Objects we are going to pack are collected in objects array (dynamically
49 * expanded). nr_objects & nr_alloc controls this array. They are stored
50 * in the order we see -- typically rev-list --objects order that gives us
51 * nice "minimum seek" order.
52 *
53 * sorted-by-sha ans sorted-by-type are arrays of pointers that point at
54 * elements in the objects array. The former is used to build the pack
55 * index (lists object names in the ascending order to help offset lookup),
56 * and the latter is used to group similar things together by try_delta()
57 * heuristics.
58 */
60 static int non_empty;
61 static int no_reuse_delta;
62 static int local;
63 static int incremental;
64 static int allow_ofs_delta;
66 static struct object_entry *objects;
67 static uint32_t nr_objects, nr_alloc, nr_result;
68 static const char *pack_tmp_name, *idx_tmp_name;
69 static char tmpname[PATH_MAX];
70 static unsigned char pack_file_sha1[20];
71 static int progress = 1;
72 static volatile sig_atomic_t progress_update;
73 static int window = 10;
74 static int pack_to_stdout;
75 static int num_preferred_base;
77 /*
78 * The object names in objects array are hashed with this hashtable,
79 * to help looking up the entry by object name. Binary search from
80 * sorted_by_sha is also possible but this was easier to code and faster.
81 * This hashtable is built after all the objects are seen.
82 */
83 static int *object_ix;
84 static int object_ix_hashsz;
86 /*
87 * Pack index for existing packs give us easy access to the offsets into
88 * corresponding pack file where each object's data starts, but the entries
89 * do not store the size of the compressed representation (uncompressed
90 * size is easily available by examining the pack entry header). It is
91 * also rather expensive to find the sha1 for an object given its offset.
92 *
93 * We build a hashtable of existing packs (pack_revindex), and keep reverse
94 * index here -- pack index file is sorted by object name mapping to offset;
95 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
96 * ordered by offset, so if you know the offset of an object, next offset
97 * is where its packed representation ends and the index_nr can be used to
98 * get the object sha1 from the main index.
99 */
100 struct revindex_entry {
101 off_t offset;
102 unsigned int nr;
103 };
104 struct pack_revindex {
105 struct packed_git *p;
106 struct revindex_entry *revindex;
107 };
108 static struct pack_revindex *pack_revindex;
109 static int pack_revindex_hashsz;
111 /*
112 * stats
113 */
114 static uint32_t written, written_delta;
115 static uint32_t reused, reused_delta;
117 static int pack_revindex_ix(struct packed_git *p)
118 {
119 unsigned long ui = (unsigned long)p;
120 int i;
122 ui = ui ^ (ui >> 16); /* defeat structure alignment */
123 i = (int)(ui % pack_revindex_hashsz);
124 while (pack_revindex[i].p) {
125 if (pack_revindex[i].p == p)
126 return i;
127 if (++i == pack_revindex_hashsz)
128 i = 0;
129 }
130 return -1 - i;
131 }
133 static void prepare_pack_ix(void)
134 {
135 int num;
136 struct packed_git *p;
137 for (num = 0, p = packed_git; p; p = p->next)
138 num++;
139 if (!num)
140 return;
141 pack_revindex_hashsz = num * 11;
142 pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
143 for (p = packed_git; p; p = p->next) {
144 num = pack_revindex_ix(p);
145 num = - 1 - num;
146 pack_revindex[num].p = p;
147 }
148 /* revindex elements are lazily initialized */
149 }
151 static int cmp_offset(const void *a_, const void *b_)
152 {
153 const struct revindex_entry *a = a_;
154 const struct revindex_entry *b = b_;
155 return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
156 }
158 /*
159 * Ordered list of offsets of objects in the pack.
160 */
161 static void prepare_pack_revindex(struct pack_revindex *rix)
162 {
163 struct packed_git *p = rix->p;
164 int num_ent = p->num_objects;
165 int i;
166 const char *index = p->index_data;
168 rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
169 index += 4 * 256;
171 if (p->index_version > 1) {
172 const uint32_t *off_32 =
173 (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
174 const uint32_t *off_64 = off_32 + p->num_objects;
175 for (i = 0; i < num_ent; i++) {
176 uint32_t off = ntohl(*off_32++);
177 if (!(off & 0x80000000)) {
178 rix->revindex[i].offset = off;
179 } else {
180 rix->revindex[i].offset =
181 ((uint64_t)ntohl(*off_64++)) << 32;
182 rix->revindex[i].offset |=
183 ntohl(*off_64++);
184 }
185 rix->revindex[i].nr = i;
186 }
187 } else {
188 for (i = 0; i < num_ent; i++) {
189 uint32_t hl = *((uint32_t *)(index + 24 * i));
190 rix->revindex[i].offset = ntohl(hl);
191 rix->revindex[i].nr = i;
192 }
193 }
195 /* This knows the pack format -- the 20-byte trailer
196 * follows immediately after the last object data.
197 */
198 rix->revindex[num_ent].offset = p->pack_size - 20;
199 rix->revindex[num_ent].nr = -1;
200 qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
201 }
203 static struct revindex_entry * find_packed_object(struct packed_git *p,
204 off_t ofs)
205 {
206 int num;
207 int lo, hi;
208 struct pack_revindex *rix;
209 struct revindex_entry *revindex;
210 num = pack_revindex_ix(p);
211 if (num < 0)
212 die("internal error: pack revindex uninitialized");
213 rix = &pack_revindex[num];
214 if (!rix->revindex)
215 prepare_pack_revindex(rix);
216 revindex = rix->revindex;
217 lo = 0;
218 hi = p->num_objects + 1;
219 do {
220 int mi = (lo + hi) / 2;
221 if (revindex[mi].offset == ofs) {
222 return revindex + mi;
223 }
224 else if (ofs < revindex[mi].offset)
225 hi = mi;
226 else
227 lo = mi + 1;
228 } while (lo < hi);
229 die("internal error: pack revindex corrupt");
230 }
232 static const unsigned char *find_packed_object_name(struct packed_git *p,
233 off_t ofs)
234 {
235 struct revindex_entry *entry = find_packed_object(p, ofs);
236 return nth_packed_object_sha1(p, entry->nr);
237 }
239 static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
240 {
241 unsigned long othersize, delta_size;
242 enum object_type type;
243 void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
244 void *delta_buf;
246 if (!otherbuf)
247 die("unable to read %s", sha1_to_hex(entry->delta->sha1));
248 delta_buf = diff_delta(otherbuf, othersize,
249 buf, size, &delta_size, 0);
250 if (!delta_buf || delta_size != entry->delta_size)
251 die("delta size changed");
252 free(buf);
253 free(otherbuf);
254 return delta_buf;
255 }
257 /*
258 * The per-object header is a pretty dense thing, which is
259 * - first byte: low four bits are "size", then three bits of "type",
260 * and the high bit is "size continues".
261 * - each byte afterwards: low seven bits are size continuation,
262 * with the high bit being "size continues"
263 */
264 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
265 {
266 int n = 1;
267 unsigned char c;
269 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
270 die("bad type %d", type);
272 c = (type << 4) | (size & 15);
273 size >>= 4;
274 while (size) {
275 *hdr++ = c | 0x80;
276 c = size & 0x7f;
277 size >>= 7;
278 n++;
279 }
280 *hdr = c;
281 return n;
282 }
284 /*
285 * we are going to reuse the existing object data as is. make
286 * sure it is not corrupt.
287 */
288 static int check_pack_inflate(struct packed_git *p,
289 struct pack_window **w_curs,
290 off_t offset,
291 off_t len,
292 unsigned long expect)
293 {
294 z_stream stream;
295 unsigned char fakebuf[4096], *in;
296 int st;
298 memset(&stream, 0, sizeof(stream));
299 inflateInit(&stream);
300 do {
301 in = use_pack(p, w_curs, offset, &stream.avail_in);
302 stream.next_in = in;
303 stream.next_out = fakebuf;
304 stream.avail_out = sizeof(fakebuf);
305 st = inflate(&stream, Z_FINISH);
306 offset += stream.next_in - in;
307 } while (st == Z_OK || st == Z_BUF_ERROR);
308 inflateEnd(&stream);
309 return (st == Z_STREAM_END &&
310 stream.total_out == expect &&
311 stream.total_in == len) ? 0 : -1;
312 }
314 static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
315 off_t offset, off_t len, unsigned int nr)
316 {
317 const uint32_t *index_crc;
318 uint32_t data_crc = crc32(0, Z_NULL, 0);
320 do {
321 unsigned int avail;
322 void *data = use_pack(p, w_curs, offset, &avail);
323 if (avail > len)
324 avail = len;
325 data_crc = crc32(data_crc, data, avail);
326 offset += avail;
327 len -= avail;
328 } while (len);
330 index_crc = p->index_data;
331 index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
333 return data_crc != ntohl(*index_crc);
334 }
336 static void copy_pack_data(struct sha1file *f,
337 struct packed_git *p,
338 struct pack_window **w_curs,
339 off_t offset,
340 off_t len)
341 {
342 unsigned char *in;
343 unsigned int avail;
345 while (len) {
346 in = use_pack(p, w_curs, offset, &avail);
347 if (avail > len)
348 avail = (unsigned int)len;
349 sha1write(f, in, avail);
350 offset += avail;
351 len -= avail;
352 }
353 }
355 static int check_loose_inflate(unsigned char *data, unsigned long len, unsigned long expect)
356 {
357 z_stream stream;
358 unsigned char fakebuf[4096];
359 int st;
361 memset(&stream, 0, sizeof(stream));
362 stream.next_in = data;
363 stream.avail_in = len;
364 stream.next_out = fakebuf;
365 stream.avail_out = sizeof(fakebuf);
366 inflateInit(&stream);
368 while (1) {
369 st = inflate(&stream, Z_FINISH);
370 if (st == Z_STREAM_END || st == Z_OK) {
371 st = (stream.total_out == expect &&
372 stream.total_in == len) ? 0 : -1;
373 break;
374 }
375 if (st != Z_BUF_ERROR) {
376 st = -1;
377 break;
378 }
379 stream.next_out = fakebuf;
380 stream.avail_out = sizeof(fakebuf);
381 }
382 inflateEnd(&stream);
383 return st;
384 }
386 static int revalidate_loose_object(struct object_entry *entry,
387 unsigned char *map,
388 unsigned long mapsize)
389 {
390 /* we already know this is a loose object with new type header. */
391 enum object_type type;
392 unsigned long size, used;
394 if (pack_to_stdout)
395 return 0;
397 used = unpack_object_header_gently(map, mapsize, &type, &size);
398 if (!used)
399 return -1;
400 map += used;
401 mapsize -= used;
402 return check_loose_inflate(map, mapsize, size);
403 }
405 static unsigned long write_object(struct sha1file *f,
406 struct object_entry *entry)
407 {
408 unsigned long size;
409 enum object_type type;
410 void *buf;
411 unsigned char header[10];
412 unsigned hdrlen;
413 off_t datalen;
414 enum object_type obj_type;
415 int to_reuse = 0;
417 if (!pack_to_stdout)
418 crc32_begin(f);
420 obj_type = entry->type;
421 if (! entry->in_pack)
422 to_reuse = 0; /* can't reuse what we don't have */
423 else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
424 to_reuse = 1; /* check_object() decided it for us */
425 else if (obj_type != entry->in_pack_type)
426 to_reuse = 0; /* pack has delta which is unusable */
427 else if (entry->delta)
428 to_reuse = 0; /* we want to pack afresh */
429 else
430 to_reuse = 1; /* we have it in-pack undeltified,
431 * and we do not need to deltify it.
432 */
434 if (!entry->in_pack && !entry->delta) {
435 unsigned char *map;
436 unsigned long mapsize;
437 map = map_sha1_file(entry->sha1, &mapsize);
438 if (map && !legacy_loose_object(map)) {
439 /* We can copy straight into the pack file */
440 if (revalidate_loose_object(entry, map, mapsize))
441 die("corrupt loose object %s",
442 sha1_to_hex(entry->sha1));
443 sha1write(f, map, mapsize);
444 munmap(map, mapsize);
445 written++;
446 reused++;
447 return mapsize;
448 }
449 if (map)
450 munmap(map, mapsize);
451 }
453 if (!to_reuse) {
454 buf = read_sha1_file(entry->sha1, &type, &size);
455 if (!buf)
456 die("unable to read %s", sha1_to_hex(entry->sha1));
457 if (size != entry->size)
458 die("object %s size inconsistency (%lu vs %lu)",
459 sha1_to_hex(entry->sha1), size, entry->size);
460 if (entry->delta) {
461 buf = delta_against(buf, size, entry);
462 size = entry->delta_size;
463 obj_type = (allow_ofs_delta && entry->delta->offset) ?
464 OBJ_OFS_DELTA : OBJ_REF_DELTA;
465 }
466 /*
467 * The object header is a byte of 'type' followed by zero or
468 * more bytes of length.
469 */
470 hdrlen = encode_header(obj_type, size, header);
471 sha1write(f, header, hdrlen);
473 if (obj_type == OBJ_OFS_DELTA) {
474 /*
475 * Deltas with relative base contain an additional
476 * encoding of the relative offset for the delta
477 * base from this object's position in the pack.
478 */
479 off_t ofs = entry->offset - entry->delta->offset;
480 unsigned pos = sizeof(header) - 1;
481 header[pos] = ofs & 127;
482 while (ofs >>= 7)
483 header[--pos] = 128 | (--ofs & 127);
484 sha1write(f, header + pos, sizeof(header) - pos);
485 hdrlen += sizeof(header) - pos;
486 } else if (obj_type == OBJ_REF_DELTA) {
487 /*
488 * Deltas with a base reference contain
489 * an additional 20 bytes for the base sha1.
490 */
491 sha1write(f, entry->delta->sha1, 20);
492 hdrlen += 20;
493 }
494 datalen = sha1write_compressed(f, buf, size);
495 free(buf);
496 }
497 else {
498 struct packed_git *p = entry->in_pack;
499 struct pack_window *w_curs = NULL;
500 struct revindex_entry *revidx;
501 off_t offset;
503 if (entry->delta) {
504 obj_type = (allow_ofs_delta && entry->delta->offset) ?
505 OBJ_OFS_DELTA : OBJ_REF_DELTA;
506 reused_delta++;
507 }
508 hdrlen = encode_header(obj_type, entry->size, header);
509 sha1write(f, header, hdrlen);
510 if (obj_type == OBJ_OFS_DELTA) {
511 off_t ofs = entry->offset - entry->delta->offset;
512 unsigned pos = sizeof(header) - 1;
513 header[pos] = ofs & 127;
514 while (ofs >>= 7)
515 header[--pos] = 128 | (--ofs & 127);
516 sha1write(f, header + pos, sizeof(header) - pos);
517 hdrlen += sizeof(header) - pos;
518 } else if (obj_type == OBJ_REF_DELTA) {
519 sha1write(f, entry->delta->sha1, 20);
520 hdrlen += 20;
521 }
523 offset = entry->in_pack_offset;
524 revidx = find_packed_object(p, offset);
525 datalen = revidx[1].offset - offset;
526 if (!pack_to_stdout && p->index_version > 1 &&
527 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
528 die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
529 offset += entry->in_pack_header_size;
530 datalen -= entry->in_pack_header_size;
531 if (!pack_to_stdout && p->index_version == 1 &&
532 check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
533 die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
534 copy_pack_data(f, p, &w_curs, offset, datalen);
535 unuse_pack(&w_curs);
536 reused++;
537 }
538 if (entry->delta)
539 written_delta++;
540 written++;
541 if (!pack_to_stdout)
542 entry->crc32 = crc32_end(f);
543 return hdrlen + datalen;
544 }
546 static off_t write_one(struct sha1file *f,
547 struct object_entry *e,
548 off_t offset)
549 {
550 unsigned long size;
552 /* offset is non zero if object is written already. */
553 if (e->offset || e->preferred_base)
554 return offset;
556 /* if we are deltified, write out base object first. */
557 if (e->delta)
558 offset = write_one(f, e->delta, offset);
560 e->offset = offset;
561 size = write_object(f, e);
563 /* make sure off_t is sufficiently large not to wrap */
564 if (offset > offset + size)
565 die("pack too large for current definition of off_t");
566 return offset + size;
567 }
569 static off_t write_pack_file(void)
570 {
571 uint32_t i;
572 struct sha1file *f;
573 off_t offset, last_obj_offset = 0;
574 struct pack_header hdr;
575 unsigned last_percent = 999;
576 int do_progress = progress;
578 if (pack_to_stdout) {
579 f = sha1fd(1, "<stdout>");
580 do_progress >>= 1;
581 } else {
582 int fd;
583 snprintf(tmpname, sizeof(tmpname), "tmp_pack_XXXXXX");
584 fd = mkstemp(tmpname);
585 if (fd < 0)
586 die("unable to create %s: %s\n", tmpname, strerror(errno));
587 pack_tmp_name = xstrdup(tmpname);
588 f = sha1fd(fd, pack_tmp_name);
589 }
591 if (do_progress)
592 fprintf(stderr, "Writing %u objects.\n", nr_result);
594 hdr.hdr_signature = htonl(PACK_SIGNATURE);
595 hdr.hdr_version = htonl(PACK_VERSION);
596 hdr.hdr_entries = htonl(nr_result);
597 sha1write(f, &hdr, sizeof(hdr));
598 offset = sizeof(hdr);
599 if (!nr_result)
600 goto done;
601 for (i = 0; i < nr_objects; i++) {
602 last_obj_offset = offset;
603 offset = write_one(f, objects + i, offset);
604 if (do_progress) {
605 unsigned percent = written * 100 / nr_result;
606 if (progress_update || percent != last_percent) {
607 fprintf(stderr, "%4u%% (%u/%u) done\r",
608 percent, written, nr_result);
609 progress_update = 0;
610 last_percent = percent;
611 }
612 }
613 }
614 if (do_progress)
615 fputc('\n', stderr);
616 done:
617 if (written != nr_result)
618 die("wrote %u objects while expecting %u", written, nr_result);
619 sha1close(f, pack_file_sha1, 1);
621 return last_obj_offset;
622 }
624 static int sha1_sort(const void *_a, const void *_b)
625 {
626 const struct object_entry *a = *(struct object_entry **)_a;
627 const struct object_entry *b = *(struct object_entry **)_b;
628 return hashcmp(a->sha1, b->sha1);
629 }
631 static uint32_t index_default_version = 1;
632 static uint32_t index_off32_limit = 0x7fffffff;
634 static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
635 {
636 struct sha1file *f;
637 struct object_entry **sorted_by_sha, **list, **last;
638 uint32_t array[256];
639 uint32_t i, index_version;
640 SHA_CTX ctx;
641 int fd;
643 snprintf(tmpname, sizeof(tmpname), "tmp_idx_XXXXXX");
644 fd = mkstemp(tmpname);
645 if (fd < 0)
646 die("unable to create %s: %s\n", tmpname, strerror(errno));
647 idx_tmp_name = xstrdup(tmpname);
648 f = sha1fd(fd, idx_tmp_name);
650 if (nr_result) {
651 uint32_t j = 0;
652 sorted_by_sha =
653 xcalloc(nr_result, sizeof(struct object_entry *));
654 for (i = 0; i < nr_objects; i++)
655 if (!objects[i].preferred_base)
656 sorted_by_sha[j++] = objects + i;
657 if (j != nr_result)
658 die("listed %u objects while expecting %u", j, nr_result);
659 qsort(sorted_by_sha, nr_result, sizeof(*sorted_by_sha), sha1_sort);
660 list = sorted_by_sha;
661 last = sorted_by_sha + nr_result;
662 } else
663 sorted_by_sha = list = last = NULL;
665 /* if last object's offset is >= 2^31 we should use index V2 */
666 index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
668 /* index versions 2 and above need a header */
669 if (index_version >= 2) {
670 struct pack_idx_header hdr;
671 hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
672 hdr.idx_version = htonl(index_version);
673 sha1write(f, &hdr, sizeof(hdr));
674 }
676 /*
677 * Write the first-level table (the list is sorted,
678 * but we use a 256-entry lookup to be able to avoid
679 * having to do eight extra binary search iterations).
680 */
681 for (i = 0; i < 256; i++) {
682 struct object_entry **next = list;
683 while (next < last) {
684 struct object_entry *entry = *next;
685 if (entry->sha1[0] != i)
686 break;
687 next++;
688 }
689 array[i] = htonl(next - sorted_by_sha);
690 list = next;
691 }
692 sha1write(f, array, 256 * 4);
694 /* Compute the SHA1 hash of sorted object names. */
695 SHA1_Init(&ctx);
697 /* Write the actual SHA1 entries. */
698 list = sorted_by_sha;
699 for (i = 0; i < nr_result; i++) {
700 struct object_entry *entry = *list++;
701 if (index_version < 2) {
702 uint32_t offset = htonl(entry->offset);
703 sha1write(f, &offset, 4);
704 }
705 sha1write(f, entry->sha1, 20);
706 SHA1_Update(&ctx, entry->sha1, 20);
707 }
709 if (index_version >= 2) {
710 unsigned int nr_large_offset = 0;
712 /* write the crc32 table */
713 list = sorted_by_sha;
714 for (i = 0; i < nr_objects; i++) {
715 struct object_entry *entry = *list++;
716 uint32_t crc32_val = htonl(entry->crc32);
717 sha1write(f, &crc32_val, 4);
718 }
720 /* write the 32-bit offset table */
721 list = sorted_by_sha;
722 for (i = 0; i < nr_objects; i++) {
723 struct object_entry *entry = *list++;
724 uint32_t offset = (entry->offset <= index_off32_limit) ?
725 entry->offset : (0x80000000 | nr_large_offset++);
726 offset = htonl(offset);
727 sha1write(f, &offset, 4);
728 }
730 /* write the large offset table */
731 list = sorted_by_sha;
732 while (nr_large_offset) {
733 struct object_entry *entry = *list++;
734 uint64_t offset = entry->offset;
735 if (offset > index_off32_limit) {
736 uint32_t split[2];
737 split[0] = htonl(offset >> 32);
738 split[1] = htonl(offset & 0xffffffff);
739 sha1write(f, split, 8);
740 nr_large_offset--;
741 }
742 }
743 }
745 sha1write(f, pack_file_sha1, 20);
746 sha1close(f, NULL, 1);
747 free(sorted_by_sha);
748 SHA1_Final(sha1, &ctx);
749 }
751 static int locate_object_entry_hash(const unsigned char *sha1)
752 {
753 int i;
754 unsigned int ui;
755 memcpy(&ui, sha1, sizeof(unsigned int));
756 i = ui % object_ix_hashsz;
757 while (0 < object_ix[i]) {
758 if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
759 return i;
760 if (++i == object_ix_hashsz)
761 i = 0;
762 }
763 return -1 - i;
764 }
766 static struct object_entry *locate_object_entry(const unsigned char *sha1)
767 {
768 int i;
770 if (!object_ix_hashsz)
771 return NULL;
773 i = locate_object_entry_hash(sha1);
774 if (0 <= i)
775 return &objects[object_ix[i]-1];
776 return NULL;
777 }
779 static void rehash_objects(void)
780 {
781 uint32_t i;
782 struct object_entry *oe;
784 object_ix_hashsz = nr_objects * 3;
785 if (object_ix_hashsz < 1024)
786 object_ix_hashsz = 1024;
787 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
788 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
789 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
790 int ix = locate_object_entry_hash(oe->sha1);
791 if (0 <= ix)
792 continue;
793 ix = -1 - ix;
794 object_ix[ix] = i + 1;
795 }
796 }
798 static unsigned name_hash(const char *name)
799 {
800 unsigned char c;
801 unsigned hash = 0;
803 /*
804 * This effectively just creates a sortable number from the
805 * last sixteen non-whitespace characters. Last characters
806 * count "most", so things that end in ".c" sort together.
807 */
808 while ((c = *name++) != 0) {
809 if (isspace(c))
810 continue;
811 hash = (hash >> 2) + (c << 24);
812 }
813 return hash;
814 }
816 static int add_object_entry(const unsigned char *sha1, enum object_type type,
817 unsigned hash, int exclude)
818 {
819 struct object_entry *entry;
820 struct packed_git *p, *found_pack = NULL;
821 off_t found_offset = 0;
822 int ix;
824 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
825 if (ix >= 0) {
826 if (exclude) {
827 entry = objects + object_ix[ix] - 1;
828 if (!entry->preferred_base)
829 nr_result--;
830 entry->preferred_base = 1;
831 }
832 return 0;
833 }
835 for (p = packed_git; p; p = p->next) {
836 off_t offset = find_pack_entry_one(sha1, p);
837 if (offset) {
838 if (!found_pack) {
839 found_offset = offset;
840 found_pack = p;
841 }
842 if (exclude)
843 break;
844 if (incremental)
845 return 0;
846 if (local && !p->pack_local)
847 return 0;
848 }
849 }
851 if (nr_objects >= nr_alloc) {
852 nr_alloc = (nr_alloc + 1024) * 3 / 2;
853 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
854 }
856 entry = objects + nr_objects++;
857 memset(entry, 0, sizeof(*entry));
858 hashcpy(entry->sha1, sha1);
859 entry->hash = hash;
860 if (type)
861 entry->type = type;
862 if (exclude)
863 entry->preferred_base = 1;
864 else
865 nr_result++;
866 if (found_pack) {
867 entry->in_pack = found_pack;
868 entry->in_pack_offset = found_offset;
869 }
871 if (object_ix_hashsz * 3 <= nr_objects * 4)
872 rehash_objects();
873 else
874 object_ix[-1 - ix] = nr_objects;
876 if (progress_update) {
877 fprintf(stderr, "Counting objects...%u\r", nr_objects);
878 progress_update = 0;
879 }
881 return 1;
882 }
884 struct pbase_tree_cache {
885 unsigned char sha1[20];
886 int ref;
887 int temporary;
888 void *tree_data;
889 unsigned long tree_size;
890 };
892 static struct pbase_tree_cache *(pbase_tree_cache[256]);
893 static int pbase_tree_cache_ix(const unsigned char *sha1)
894 {
895 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
896 }
897 static int pbase_tree_cache_ix_incr(int ix)
898 {
899 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
900 }
902 static struct pbase_tree {
903 struct pbase_tree *next;
904 /* This is a phony "cache" entry; we are not
905 * going to evict it nor find it through _get()
906 * mechanism -- this is for the toplevel node that
907 * would almost always change with any commit.
908 */
909 struct pbase_tree_cache pcache;
910 } *pbase_tree;
912 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
913 {
914 struct pbase_tree_cache *ent, *nent;
915 void *data;
916 unsigned long size;
917 enum object_type type;
918 int neigh;
919 int my_ix = pbase_tree_cache_ix(sha1);
920 int available_ix = -1;
922 /* pbase-tree-cache acts as a limited hashtable.
923 * your object will be found at your index or within a few
924 * slots after that slot if it is cached.
925 */
926 for (neigh = 0; neigh < 8; neigh++) {
927 ent = pbase_tree_cache[my_ix];
928 if (ent && !hashcmp(ent->sha1, sha1)) {
929 ent->ref++;
930 return ent;
931 }
932 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
933 ((0 <= available_ix) &&
934 (!ent && pbase_tree_cache[available_ix])))
935 available_ix = my_ix;
936 if (!ent)
937 break;
938 my_ix = pbase_tree_cache_ix_incr(my_ix);
939 }
941 /* Did not find one. Either we got a bogus request or
942 * we need to read and perhaps cache.
943 */
944 data = read_sha1_file(sha1, &type, &size);
945 if (!data)
946 return NULL;
947 if (type != OBJ_TREE) {
948 free(data);
949 return NULL;
950 }
952 /* We need to either cache or return a throwaway copy */
954 if (available_ix < 0)
955 ent = NULL;
956 else {
957 ent = pbase_tree_cache[available_ix];
958 my_ix = available_ix;
959 }
961 if (!ent) {
962 nent = xmalloc(sizeof(*nent));
963 nent->temporary = (available_ix < 0);
964 }
965 else {
966 /* evict and reuse */
967 free(ent->tree_data);
968 nent = ent;
969 }
970 hashcpy(nent->sha1, sha1);
971 nent->tree_data = data;
972 nent->tree_size = size;
973 nent->ref = 1;
974 if (!nent->temporary)
975 pbase_tree_cache[my_ix] = nent;
976 return nent;
977 }
979 static void pbase_tree_put(struct pbase_tree_cache *cache)
980 {
981 if (!cache->temporary) {
982 cache->ref--;
983 return;
984 }
985 free(cache->tree_data);
986 free(cache);
987 }
989 static int name_cmp_len(const char *name)
990 {
991 int i;
992 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
993 ;
994 return i;
995 }
997 static void add_pbase_object(struct tree_desc *tree,
998 const char *name,
999 int cmplen,
1000 const char *fullname)
1001 {
1002 struct name_entry entry;
1003 int cmp;
1005 while (tree_entry(tree,&entry)) {
1006 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1007 memcmp(name, entry.path, cmplen);
1008 if (cmp > 0)
1009 continue;
1010 if (cmp < 0)
1011 return;
1012 if (name[cmplen] != '/') {
1013 unsigned hash = name_hash(fullname);
1014 add_object_entry(entry.sha1,
1015 S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1016 hash, 1);
1017 return;
1018 }
1019 if (S_ISDIR(entry.mode)) {
1020 struct tree_desc sub;
1021 struct pbase_tree_cache *tree;
1022 const char *down = name+cmplen+1;
1023 int downlen = name_cmp_len(down);
1025 tree = pbase_tree_get(entry.sha1);
1026 if (!tree)
1027 return;
1028 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1030 add_pbase_object(&sub, down, downlen, fullname);
1031 pbase_tree_put(tree);
1032 }
1033 }
1034 }
1036 static unsigned *done_pbase_paths;
1037 static int done_pbase_paths_num;
1038 static int done_pbase_paths_alloc;
1039 static int done_pbase_path_pos(unsigned hash)
1040 {
1041 int lo = 0;
1042 int hi = done_pbase_paths_num;
1043 while (lo < hi) {
1044 int mi = (hi + lo) / 2;
1045 if (done_pbase_paths[mi] == hash)
1046 return mi;
1047 if (done_pbase_paths[mi] < hash)
1048 hi = mi;
1049 else
1050 lo = mi + 1;
1051 }
1052 return -lo-1;
1053 }
1055 static int check_pbase_path(unsigned hash)
1056 {
1057 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1058 if (0 <= pos)
1059 return 1;
1060 pos = -pos - 1;
1061 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1062 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1063 done_pbase_paths = xrealloc(done_pbase_paths,
1064 done_pbase_paths_alloc *
1065 sizeof(unsigned));
1066 }
1067 done_pbase_paths_num++;
1068 if (pos < done_pbase_paths_num)
1069 memmove(done_pbase_paths + pos + 1,
1070 done_pbase_paths + pos,
1071 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1072 done_pbase_paths[pos] = hash;
1073 return 0;
1074 }
1076 static void add_preferred_base_object(const char *name, unsigned hash)
1077 {
1078 struct pbase_tree *it;
1079 int cmplen;
1081 if (!num_preferred_base || check_pbase_path(hash))
1082 return;
1084 cmplen = name_cmp_len(name);
1085 for (it = pbase_tree; it; it = it->next) {
1086 if (cmplen == 0) {
1087 add_object_entry(it->pcache.sha1, OBJ_TREE, 0, 1);
1088 }
1089 else {
1090 struct tree_desc tree;
1091 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1092 add_pbase_object(&tree, name, cmplen, name);
1093 }
1094 }
1095 }
1097 static void add_preferred_base(unsigned char *sha1)
1098 {
1099 struct pbase_tree *it;
1100 void *data;
1101 unsigned long size;
1102 unsigned char tree_sha1[20];
1104 if (window <= num_preferred_base++)
1105 return;
1107 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1108 if (!data)
1109 return;
1111 for (it = pbase_tree; it; it = it->next) {
1112 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1113 free(data);
1114 return;
1115 }
1116 }
1118 it = xcalloc(1, sizeof(*it));
1119 it->next = pbase_tree;
1120 pbase_tree = it;
1122 hashcpy(it->pcache.sha1, tree_sha1);
1123 it->pcache.tree_data = data;
1124 it->pcache.tree_size = size;
1125 }
1127 static void check_object(struct object_entry *entry)
1128 {
1129 if (entry->in_pack) {
1130 struct packed_git *p = entry->in_pack;
1131 struct pack_window *w_curs = NULL;
1132 const unsigned char *base_ref = NULL;
1133 struct object_entry *base_entry;
1134 unsigned long used, used_0;
1135 unsigned int avail;
1136 off_t ofs;
1137 unsigned char *buf, c;
1139 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1141 /*
1142 * We want in_pack_type even if we do not reuse delta.
1143 * There is no point not reusing non-delta representations.
1144 */
1145 used = unpack_object_header_gently(buf, avail,
1146 &entry->in_pack_type,
1147 &entry->size);
1149 /*
1150 * Determine if this is a delta and if so whether we can
1151 * reuse it or not. Otherwise let's find out as cheaply as
1152 * possible what the actual type and size for this object is.
1153 */
1154 switch (entry->in_pack_type) {
1155 default:
1156 /* Not a delta hence we've already got all we need. */
1157 entry->type = entry->in_pack_type;
1158 entry->in_pack_header_size = used;
1159 unuse_pack(&w_curs);
1160 return;
1161 case OBJ_REF_DELTA:
1162 if (!no_reuse_delta && !entry->preferred_base)
1163 base_ref = use_pack(p, &w_curs,
1164 entry->in_pack_offset + used, NULL);
1165 entry->in_pack_header_size = used + 20;
1166 break;
1167 case OBJ_OFS_DELTA:
1168 buf = use_pack(p, &w_curs,
1169 entry->in_pack_offset + used, NULL);
1170 used_0 = 0;
1171 c = buf[used_0++];
1172 ofs = c & 127;
1173 while (c & 128) {
1174 ofs += 1;
1175 if (!ofs || MSB(ofs, 7))
1176 die("delta base offset overflow in pack for %s",
1177 sha1_to_hex(entry->sha1));
1178 c = buf[used_0++];
1179 ofs = (ofs << 7) + (c & 127);
1180 }
1181 if (ofs >= entry->in_pack_offset)
1182 die("delta base offset out of bound for %s",
1183 sha1_to_hex(entry->sha1));
1184 ofs = entry->in_pack_offset - ofs;
1185 if (!no_reuse_delta && !entry->preferred_base)
1186 base_ref = find_packed_object_name(p, ofs);
1187 entry->in_pack_header_size = used + used_0;
1188 break;
1189 }
1191 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1192 /*
1193 * If base_ref was set above that means we wish to
1194 * reuse delta data, and we even found that base
1195 * in the list of objects we want to pack. Goodie!
1196 *
1197 * Depth value does not matter - find_deltas() will
1198 * never consider reused delta as the base object to
1199 * deltify other objects against, in order to avoid
1200 * circular deltas.
1201 */
1202 entry->type = entry->in_pack_type;
1203 entry->delta = base_entry;
1204 entry->delta_sibling = base_entry->delta_child;
1205 base_entry->delta_child = entry;
1206 unuse_pack(&w_curs);
1207 return;
1208 }
1210 if (entry->type) {
1211 /*
1212 * This must be a delta and we already know what the
1213 * final object type is. Let's extract the actual
1214 * object size from the delta header.
1215 */
1216 entry->size = get_size_from_delta(p, &w_curs,
1217 entry->in_pack_offset + entry->in_pack_header_size);
1218 unuse_pack(&w_curs);
1219 return;
1220 }
1222 /*
1223 * No choice but to fall back to the recursive delta walk
1224 * with sha1_object_info() to find about the object type
1225 * at this point...
1226 */
1227 unuse_pack(&w_curs);
1228 }
1230 entry->type = sha1_object_info(entry->sha1, &entry->size);
1231 if (entry->type < 0)
1232 die("unable to get type of object %s",
1233 sha1_to_hex(entry->sha1));
1234 }
1236 static int pack_offset_sort(const void *_a, const void *_b)
1237 {
1238 const struct object_entry *a = *(struct object_entry **)_a;
1239 const struct object_entry *b = *(struct object_entry **)_b;
1241 /* avoid filesystem trashing with loose objects */
1242 if (!a->in_pack && !b->in_pack)
1243 return hashcmp(a->sha1, b->sha1);
1245 if (a->in_pack < b->in_pack)
1246 return -1;
1247 if (a->in_pack > b->in_pack)
1248 return 1;
1249 return a->in_pack_offset < b->in_pack_offset ? -1 :
1250 (a->in_pack_offset > b->in_pack_offset);
1251 }
1253 static void get_object_details(void)
1254 {
1255 uint32_t i;
1256 struct object_entry **sorted_by_offset;
1258 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1259 for (i = 0; i < nr_objects; i++)
1260 sorted_by_offset[i] = objects + i;
1261 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1263 prepare_pack_ix();
1264 for (i = 0; i < nr_objects; i++)
1265 check_object(sorted_by_offset[i]);
1266 free(sorted_by_offset);
1267 }
1269 static int type_size_sort(const void *_a, const void *_b)
1270 {
1271 const struct object_entry *a = *(struct object_entry **)_a;
1272 const struct object_entry *b = *(struct object_entry **)_b;
1274 if (a->type < b->type)
1275 return -1;
1276 if (a->type > b->type)
1277 return 1;
1278 if (a->hash < b->hash)
1279 return -1;
1280 if (a->hash > b->hash)
1281 return 1;
1282 if (a->preferred_base < b->preferred_base)
1283 return -1;
1284 if (a->preferred_base > b->preferred_base)
1285 return 1;
1286 if (a->size < b->size)
1287 return -1;
1288 if (a->size > b->size)
1289 return 1;
1290 return a > b ? -1 : (a < b); /* newest last */
1291 }
1293 struct unpacked {
1294 struct object_entry *entry;
1295 void *data;
1296 struct delta_index *index;
1297 };
1299 /*
1300 * We search for deltas _backwards_ in a list sorted by type and
1301 * by size, so that we see progressively smaller and smaller files.
1302 * That's because we prefer deltas to be from the bigger file
1303 * to the smaller - deletes are potentially cheaper, but perhaps
1304 * more importantly, the bigger file is likely the more recent
1305 * one.
1306 */
1307 static int try_delta(struct unpacked *trg, struct unpacked *src,
1308 unsigned max_depth)
1309 {
1310 struct object_entry *trg_entry = trg->entry;
1311 struct object_entry *src_entry = src->entry;
1312 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1313 enum object_type type;
1314 void *delta_buf;
1316 /* Don't bother doing diffs between different types */
1317 if (trg_entry->type != src_entry->type)
1318 return -1;
1320 /* We do not compute delta to *create* objects we are not
1321 * going to pack.
1322 */
1323 if (trg_entry->preferred_base)
1324 return -1;
1326 /*
1327 * We do not bother to try a delta that we discarded
1328 * on an earlier try, but only when reusing delta data.
1329 */
1330 if (!no_reuse_delta && trg_entry->in_pack &&
1331 trg_entry->in_pack == src_entry->in_pack &&
1332 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1333 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1334 return 0;
1336 /* Let's not bust the allowed depth. */
1337 if (src_entry->depth >= max_depth)
1338 return 0;
1340 /* Now some size filtering heuristics. */
1341 trg_size = trg_entry->size;
1342 max_size = trg_size/2 - 20;
1343 max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1344 if (max_size == 0)
1345 return 0;
1346 if (trg_entry->delta && trg_entry->delta_size <= max_size)
1347 max_size = trg_entry->delta_size-1;
1348 src_size = src_entry->size;
1349 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1350 if (sizediff >= max_size)
1351 return 0;
1353 /* Load data if not already done */
1354 if (!trg->data) {
1355 trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1356 if (sz != trg_size)
1357 die("object %s inconsistent object length (%lu vs %lu)",
1358 sha1_to_hex(trg_entry->sha1), sz, trg_size);
1359 }
1360 if (!src->data) {
1361 src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1362 if (sz != src_size)
1363 die("object %s inconsistent object length (%lu vs %lu)",
1364 sha1_to_hex(src_entry->sha1), sz, src_size);
1365 }
1366 if (!src->index) {
1367 src->index = create_delta_index(src->data, src_size);
1368 if (!src->index)
1369 die("out of memory");
1370 }
1372 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1373 if (!delta_buf)
1374 return 0;
1376 trg_entry->delta = src_entry;
1377 trg_entry->delta_size = delta_size;
1378 trg_entry->depth = src_entry->depth + 1;
1379 free(delta_buf);
1380 return 1;
1381 }
1383 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1384 {
1385 struct object_entry *child = me->delta_child;
1386 unsigned int m = n;
1387 while (child) {
1388 unsigned int c = check_delta_limit(child, n + 1);
1389 if (m < c)
1390 m = c;
1391 child = child->delta_sibling;
1392 }
1393 return m;
1394 }
1396 static void find_deltas(struct object_entry **list, int window, int depth)
1397 {
1398 uint32_t i = nr_objects, idx = 0, processed = 0;
1399 unsigned int array_size = window * sizeof(struct unpacked);
1400 struct unpacked *array;
1401 unsigned last_percent = 999;
1402 int max_depth;
1404 if (!nr_objects)
1405 return;
1406 array = xmalloc(array_size);
1407 memset(array, 0, array_size);
1408 if (progress)
1409 fprintf(stderr, "Deltifying %u objects.\n", nr_result);
1411 do {
1412 struct object_entry *entry = list[--i];
1413 struct unpacked *n = array + idx;
1414 int j;
1416 if (!entry->preferred_base)
1417 processed++;
1419 if (progress) {
1420 unsigned percent = processed * 100 / nr_result;
1421 if (percent != last_percent || progress_update) {
1422 fprintf(stderr, "%4u%% (%u/%u) done\r",
1423 percent, processed, nr_result);
1424 progress_update = 0;
1425 last_percent = percent;
1426 }
1427 }
1429 if (entry->delta)
1430 /* This happens if we decided to reuse existing
1431 * delta from a pack. "!no_reuse_delta &&" is implied.
1432 */
1433 continue;
1435 if (entry->size < 50)
1436 continue;
1437 free_delta_index(n->index);
1438 n->index = NULL;
1439 free(n->data);
1440 n->data = NULL;
1441 n->entry = entry;
1443 /*
1444 * If the current object is at pack edge, take the depth the
1445 * objects that depend on the current object into account
1446 * otherwise they would become too deep.
1447 */
1448 max_depth = depth;
1449 if (entry->delta_child) {
1450 max_depth -= check_delta_limit(entry, 0);
1451 if (max_depth <= 0)
1452 goto next;
1453 }
1455 j = window;
1456 while (--j > 0) {
1457 uint32_t other_idx = idx + j;
1458 struct unpacked *m;
1459 if (other_idx >= window)
1460 other_idx -= window;
1461 m = array + other_idx;
1462 if (!m->entry)
1463 break;
1464 if (try_delta(n, m, max_depth) < 0)
1465 break;
1466 }
1468 /* if we made n a delta, and if n is already at max
1469 * depth, leaving it in the window is pointless. we
1470 * should evict it first.
1471 */
1472 if (entry->delta && depth <= entry->depth)
1473 continue;
1475 next:
1476 idx++;
1477 if (idx >= window)
1478 idx = 0;
1479 } while (i > 0);
1481 if (progress)
1482 fputc('\n', stderr);
1484 for (i = 0; i < window; ++i) {
1485 free_delta_index(array[i].index);
1486 free(array[i].data);
1487 }
1488 free(array);
1489 }
1491 static void prepare_pack(int window, int depth)
1492 {
1493 struct object_entry **delta_list;
1494 uint32_t i;
1496 get_object_details();
1498 if (!window || !depth)
1499 return;
1501 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1502 for (i = 0; i < nr_objects; i++)
1503 delta_list[i] = objects + i;
1504 qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1505 find_deltas(delta_list, window+1, depth);
1506 free(delta_list);
1507 }
1509 static void progress_interval(int signum)
1510 {
1511 progress_update = 1;
1512 }
1514 static void setup_progress_signal(void)
1515 {
1516 struct sigaction sa;
1517 struct itimerval v;
1519 memset(&sa, 0, sizeof(sa));
1520 sa.sa_handler = progress_interval;
1521 sigemptyset(&sa.sa_mask);
1522 sa.sa_flags = SA_RESTART;
1523 sigaction(SIGALRM, &sa, NULL);
1525 v.it_interval.tv_sec = 1;
1526 v.it_interval.tv_usec = 0;
1527 v.it_value = v.it_interval;
1528 setitimer(ITIMER_REAL, &v, NULL);
1529 }
1531 static int git_pack_config(const char *k, const char *v)
1532 {
1533 if(!strcmp(k, "pack.window")) {
1534 window = git_config_int(k, v);
1535 return 0;
1536 }
1537 return git_default_config(k, v);
1538 }
1540 static void read_object_list_from_stdin(void)
1541 {
1542 char line[40 + 1 + PATH_MAX + 2];
1543 unsigned char sha1[20];
1544 unsigned hash;
1546 for (;;) {
1547 if (!fgets(line, sizeof(line), stdin)) {
1548 if (feof(stdin))
1549 break;
1550 if (!ferror(stdin))
1551 die("fgets returned NULL, not EOF, not error!");
1552 if (errno != EINTR)
1553 die("fgets: %s", strerror(errno));
1554 clearerr(stdin);
1555 continue;
1556 }
1557 if (line[0] == '-') {
1558 if (get_sha1_hex(line+1, sha1))
1559 die("expected edge sha1, got garbage:\n %s",
1560 line);
1561 add_preferred_base(sha1);
1562 continue;
1563 }
1564 if (get_sha1_hex(line, sha1))
1565 die("expected sha1, got garbage:\n %s", line);
1567 hash = name_hash(line+41);
1568 add_preferred_base_object(line+41, hash);
1569 add_object_entry(sha1, 0, hash, 0);
1570 }
1571 }
1573 static void show_commit(struct commit *commit)
1574 {
1575 add_object_entry(commit->object.sha1, OBJ_COMMIT, 0, 0);
1576 }
1578 static void show_object(struct object_array_entry *p)
1579 {
1580 unsigned hash = name_hash(p->name);
1581 add_preferred_base_object(p->name, hash);
1582 add_object_entry(p->item->sha1, p->item->type, hash, 0);
1583 }
1585 static void show_edge(struct commit *commit)
1586 {
1587 add_preferred_base(commit->object.sha1);
1588 }
1590 static void get_object_list(int ac, const char **av)
1591 {
1592 struct rev_info revs;
1593 char line[1000];
1594 int flags = 0;
1596 init_revisions(&revs, NULL);
1597 save_commit_buffer = 0;
1598 track_object_refs = 0;
1599 setup_revisions(ac, av, &revs, NULL);
1601 while (fgets(line, sizeof(line), stdin) != NULL) {
1602 int len = strlen(line);
1603 if (line[len - 1] == '\n')
1604 line[--len] = 0;
1605 if (!len)
1606 break;
1607 if (*line == '-') {
1608 if (!strcmp(line, "--not")) {
1609 flags ^= UNINTERESTING;
1610 continue;
1611 }
1612 die("not a rev '%s'", line);
1613 }
1614 if (handle_revision_arg(line, &revs, flags, 1))
1615 die("bad revision '%s'", line);
1616 }
1618 prepare_revision_walk(&revs);
1619 mark_edges_uninteresting(revs.commits, &revs, show_edge);
1620 traverse_commit_list(&revs, show_commit, show_object);
1621 }
1623 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1624 {
1625 int depth = 10;
1626 int use_internal_rev_list = 0;
1627 int thin = 0;
1628 uint32_t i;
1629 off_t last_obj_offset;
1630 const char *base_name = NULL;
1631 const char **rp_av;
1632 int rp_ac_alloc = 64;
1633 int rp_ac;
1635 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1637 rp_av[0] = "pack-objects";
1638 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1639 rp_ac = 2;
1641 git_config(git_pack_config);
1643 progress = isatty(2);
1644 for (i = 1; i < argc; i++) {
1645 const char *arg = argv[i];
1647 if (*arg != '-')
1648 break;
1650 if (!strcmp("--non-empty", arg)) {
1651 non_empty = 1;
1652 continue;
1653 }
1654 if (!strcmp("--local", arg)) {
1655 local = 1;
1656 continue;
1657 }
1658 if (!strcmp("--incremental", arg)) {
1659 incremental = 1;
1660 continue;
1661 }
1662 if (!prefixcmp(arg, "--window=")) {
1663 char *end;
1664 window = strtoul(arg+9, &end, 0);
1665 if (!arg[9] || *end)
1666 usage(pack_usage);
1667 continue;
1668 }
1669 if (!prefixcmp(arg, "--depth=")) {
1670 char *end;
1671 depth = strtoul(arg+8, &end, 0);
1672 if (!arg[8] || *end)
1673 usage(pack_usage);
1674 continue;
1675 }
1676 if (!strcmp("--progress", arg)) {
1677 progress = 1;
1678 continue;
1679 }
1680 if (!strcmp("--all-progress", arg)) {
1681 progress = 2;
1682 continue;
1683 }
1684 if (!strcmp("-q", arg)) {
1685 progress = 0;
1686 continue;
1687 }
1688 if (!strcmp("--no-reuse-delta", arg)) {
1689 no_reuse_delta = 1;
1690 continue;
1691 }
1692 if (!strcmp("--delta-base-offset", arg)) {
1693 allow_ofs_delta = 1;
1694 continue;
1695 }
1696 if (!strcmp("--stdout", arg)) {
1697 pack_to_stdout = 1;
1698 continue;
1699 }
1700 if (!strcmp("--revs", arg)) {
1701 use_internal_rev_list = 1;
1702 continue;
1703 }
1704 if (!strcmp("--unpacked", arg) ||
1705 !prefixcmp(arg, "--unpacked=") ||
1706 !strcmp("--reflog", arg) ||
1707 !strcmp("--all", arg)) {
1708 use_internal_rev_list = 1;
1709 if (rp_ac >= rp_ac_alloc - 1) {
1710 rp_ac_alloc = alloc_nr(rp_ac_alloc);
1711 rp_av = xrealloc(rp_av,
1712 rp_ac_alloc * sizeof(*rp_av));
1713 }
1714 rp_av[rp_ac++] = arg;
1715 continue;
1716 }
1717 if (!strcmp("--thin", arg)) {
1718 use_internal_rev_list = 1;
1719 thin = 1;
1720 rp_av[1] = "--objects-edge";
1721 continue;
1722 }
1723 if (!prefixcmp(arg, "--index-version=")) {
1724 char *c;
1725 index_default_version = strtoul(arg + 16, &c, 10);
1726 if (index_default_version > 2)
1727 die("bad %s", arg);
1728 if (*c == ',')
1729 index_off32_limit = strtoul(c+1, &c, 0);
1730 if (*c || index_off32_limit & 0x80000000)
1731 die("bad %s", arg);
1732 continue;
1733 }
1734 usage(pack_usage);
1735 }
1737 /* Traditionally "pack-objects [options] base extra" failed;
1738 * we would however want to take refs parameter that would
1739 * have been given to upstream rev-list ourselves, which means
1740 * we somehow want to say what the base name is. So the
1741 * syntax would be:
1742 *
1743 * pack-objects [options] base <refs...>
1744 *
1745 * in other words, we would treat the first non-option as the
1746 * base_name and send everything else to the internal revision
1747 * walker.
1748 */
1750 if (!pack_to_stdout)
1751 base_name = argv[i++];
1753 if (pack_to_stdout != !base_name)
1754 usage(pack_usage);
1756 if (!pack_to_stdout && thin)
1757 die("--thin cannot be used to build an indexable pack.");
1759 prepare_packed_git();
1761 if (progress) {
1762 fprintf(stderr, "Generating pack...\n");
1763 setup_progress_signal();
1764 }
1766 if (!use_internal_rev_list)
1767 read_object_list_from_stdin();
1768 else {
1769 rp_av[rp_ac] = NULL;
1770 get_object_list(rp_ac, rp_av);
1771 }
1773 if (progress)
1774 fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1775 if (non_empty && !nr_result)
1776 return 0;
1777 if (progress && (nr_objects != nr_result))
1778 fprintf(stderr, "Result has %u objects.\n", nr_result);
1779 if (nr_result)
1780 prepare_pack(window, depth);
1781 if (progress == 1 && pack_to_stdout) {
1782 /* the other end usually displays progress itself */
1783 struct itimerval v = {{0,},};
1784 setitimer(ITIMER_REAL, &v, NULL);
1785 signal(SIGALRM, SIG_IGN );
1786 progress_update = 0;
1787 }
1788 last_obj_offset = write_pack_file();
1789 if (!pack_to_stdout) {
1790 unsigned char object_list_sha1[20];
1791 write_index_file(last_obj_offset, object_list_sha1);
1792 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
1793 base_name, sha1_to_hex(object_list_sha1));
1794 if (rename(pack_tmp_name, tmpname))
1795 die("unable to rename temporary pack file: %s",
1796 strerror(errno));
1797 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
1798 base_name, sha1_to_hex(object_list_sha1));
1799 if (rename(idx_tmp_name, tmpname))
1800 die("unable to rename temporary index file: %s",
1801 strerror(errno));
1802 puts(sha1_to_hex(object_list_sha1));
1803 }
1804 if (progress)
1805 fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1806 written, written_delta, reused, reused_delta);
1807 return 0;
1808 }