![[tokkee]](http://tokkee.org/images/avatar.png){kind=avatar}
1 #include "builtin.h"
2 #include "cache.h"
3 #include "attr.h"
4 #include "object.h"
5 #include "blob.h"
6 #include "commit.h"
7 #include "tag.h"
8 #include "tree.h"
9 #include "delta.h"
10 #include "pack.h"
11 #include "pack-revindex.h"
12 #include "csum-file.h"
13 #include "tree-walk.h"
14 #include "diff.h"
15 #include "revision.h"
16 #include "list-objects.h"
17 #include "progress.h"
18 #include "refs.h"
20 #ifndef NO_PTHREADS
21 #include "thread-utils.h"
22 #include <pthread.h>
23 #endif
25 static const char pack_usage[] =
26 "git pack-objects [{ -q | --progress | --all-progress }]\n"
27 " [--all-progress-implied]\n"
28 " [--max-pack-size=N] [--local] [--incremental]\n"
29 " [--window=N] [--window-memory=N] [--depth=N]\n"
30 " [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset]\n"
31 " [--threads=N] [--non-empty] [--revs [--unpacked | --all]*]\n"
32 " [--reflog] [--stdout | base-name] [--include-tag]\n"
33 " [--keep-unreachable | --unpack-unreachable \n"
34 " [<ref-list | <object-list]";
36 struct object_entry {
37 struct pack_idx_entry idx;
38 unsigned long size; /* uncompressed size */
39 struct packed_git *in_pack; /* already in pack */
40 off_t in_pack_offset;
41 struct object_entry *delta; /* delta base object */
42 struct object_entry *delta_child; /* deltified objects who bases me */
43 struct object_entry *delta_sibling; /* other deltified objects who
44 * uses the same base as me
45 */
46 void *delta_data; /* cached delta (uncompressed) */
47 unsigned long delta_size; /* delta data size (uncompressed) */
48 unsigned long z_delta_size; /* delta data size (compressed) */
49 unsigned int hash; /* name hint hash */
50 enum object_type type;
51 enum object_type in_pack_type; /* could be delta */
52 unsigned char in_pack_header_size;
53 unsigned char preferred_base; /* we do not pack this, but is available
54 * to be used as the base object to delta
55 * objects against.
56 */
57 unsigned char no_try_delta;
58 };
60 /*
61 * Objects we are going to pack are collected in objects array (dynamically
62 * expanded). nr_objects & nr_alloc controls this array. They are stored
63 * in the order we see -- typically rev-list --objects order that gives us
64 * nice "minimum seek" order.
65 */
66 static struct object_entry *objects;
67 static struct pack_idx_entry **written_list;
68 static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
70 static int non_empty;
71 static int reuse_delta = 1, reuse_object = 1;
72 static int keep_unreachable, unpack_unreachable, include_tag;
73 static int local;
74 static int incremental;
75 static int ignore_packed_keep;
76 static int allow_ofs_delta;
77 static const char *base_name;
78 static int progress = 1;
79 static int window = 10;
80 static uint32_t pack_size_limit, pack_size_limit_cfg;
81 static int depth = 50;
82 static int delta_search_threads;
83 static int pack_to_stdout;
84 static int num_preferred_base;
85 static struct progress *progress_state;
86 static int pack_compression_level = Z_DEFAULT_COMPRESSION;
87 static int pack_compression_seen;
89 static unsigned long delta_cache_size = 0;
90 static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
91 static unsigned long cache_max_small_delta_size = 1000;
93 static unsigned long window_memory_limit = 0;
95 /*
96 * The object names in objects array are hashed with this hashtable,
97 * to help looking up the entry by object name.
98 * This hashtable is built after all the objects are seen.
99 */
100 static int *object_ix;
101 static int object_ix_hashsz;
103 /*
104 * stats
105 */
106 static uint32_t written, written_delta;
107 static uint32_t reused, reused_delta;
110 static void *get_delta(struct object_entry *entry)
111 {
112 unsigned long size, base_size, delta_size;
113 void *buf, *base_buf, *delta_buf;
114 enum object_type type;
116 buf = read_sha1_file(entry->idx.sha1, &type, &size);
117 if (!buf)
118 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
119 base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
120 if (!base_buf)
121 die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
122 delta_buf = diff_delta(base_buf, base_size,
123 buf, size, &delta_size, 0);
124 if (!delta_buf || delta_size != entry->delta_size)
125 die("delta size changed");
126 free(buf);
127 free(base_buf);
128 return delta_buf;
129 }
131 static unsigned long do_compress(void **pptr, unsigned long size)
132 {
133 z_stream stream;
134 void *in, *out;
135 unsigned long maxsize;
137 memset(&stream, 0, sizeof(stream));
138 deflateInit(&stream, pack_compression_level);
139 maxsize = deflateBound(&stream, size);
141 in = *pptr;
142 out = xmalloc(maxsize);
143 *pptr = out;
145 stream.next_in = in;
146 stream.avail_in = size;
147 stream.next_out = out;
148 stream.avail_out = maxsize;
149 while (deflate(&stream, Z_FINISH) == Z_OK)
150 ; /* nothing */
151 deflateEnd(&stream);
153 free(in);
154 return stream.total_out;
155 }
157 /*
158 * The per-object header is a pretty dense thing, which is
159 * - first byte: low four bits are "size", then three bits of "type",
160 * and the high bit is "size continues".
161 * - each byte afterwards: low seven bits are size continuation,
162 * with the high bit being "size continues"
163 */
164 static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
165 {
166 int n = 1;
167 unsigned char c;
169 if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
170 die("bad type %d", type);
172 c = (type << 4) | (size & 15);
173 size >>= 4;
174 while (size) {
175 *hdr++ = c | 0x80;
176 c = size & 0x7f;
177 size >>= 7;
178 n++;
179 }
180 *hdr = c;
181 return n;
182 }
184 /*
185 * we are going to reuse the existing object data as is. make
186 * sure it is not corrupt.
187 */
188 static int check_pack_inflate(struct packed_git *p,
189 struct pack_window **w_curs,
190 off_t offset,
191 off_t len,
192 unsigned long expect)
193 {
194 z_stream stream;
195 unsigned char fakebuf[4096], *in;
196 int st;
198 memset(&stream, 0, sizeof(stream));
199 git_inflate_init(&stream);
200 do {
201 in = use_pack(p, w_curs, offset, &stream.avail_in);
202 stream.next_in = in;
203 stream.next_out = fakebuf;
204 stream.avail_out = sizeof(fakebuf);
205 st = git_inflate(&stream, Z_FINISH);
206 offset += stream.next_in - in;
207 } while (st == Z_OK || st == Z_BUF_ERROR);
208 git_inflate_end(&stream);
209 return (st == Z_STREAM_END &&
210 stream.total_out == expect &&
211 stream.total_in == len) ? 0 : -1;
212 }
214 static void copy_pack_data(struct sha1file *f,
215 struct packed_git *p,
216 struct pack_window **w_curs,
217 off_t offset,
218 off_t len)
219 {
220 unsigned char *in;
221 unsigned int avail;
223 while (len) {
224 in = use_pack(p, w_curs, offset, &avail);
225 if (avail > len)
226 avail = (unsigned int)len;
227 sha1write(f, in, avail);
228 offset += avail;
229 len -= avail;
230 }
231 }
233 static unsigned long write_object(struct sha1file *f,
234 struct object_entry *entry,
235 off_t write_offset)
236 {
237 unsigned long size, limit, datalen;
238 void *buf;
239 unsigned char header[10], dheader[10];
240 unsigned hdrlen;
241 enum object_type type;
242 int usable_delta, to_reuse;
244 if (!pack_to_stdout)
245 crc32_begin(f);
247 type = entry->type;
249 /* apply size limit if limited packsize and not first object */
250 if (!pack_size_limit || !nr_written)
251 limit = 0;
252 else if (pack_size_limit <= write_offset)
253 /*
254 * the earlier object did not fit the limit; avoid
255 * mistaking this with unlimited (i.e. limit = 0).
256 */
257 limit = 1;
258 else
259 limit = pack_size_limit - write_offset;
261 if (!entry->delta)
262 usable_delta = 0; /* no delta */
263 else if (!pack_size_limit)
264 usable_delta = 1; /* unlimited packfile */
265 else if (entry->delta->idx.offset == (off_t)-1)
266 usable_delta = 0; /* base was written to another pack */
267 else if (entry->delta->idx.offset)
268 usable_delta = 1; /* base already exists in this pack */
269 else
270 usable_delta = 0; /* base could end up in another pack */
272 if (!reuse_object)
273 to_reuse = 0; /* explicit */
274 else if (!entry->in_pack)
275 to_reuse = 0; /* can't reuse what we don't have */
276 else if (type == OBJ_REF_DELTA || type == OBJ_OFS_DELTA)
277 /* check_object() decided it for us ... */
278 to_reuse = usable_delta;
279 /* ... but pack split may override that */
280 else if (type != entry->in_pack_type)
281 to_reuse = 0; /* pack has delta which is unusable */
282 else if (entry->delta)
283 to_reuse = 0; /* we want to pack afresh */
284 else
285 to_reuse = 1; /* we have it in-pack undeltified,
286 * and we do not need to deltify it.
287 */
289 if (!to_reuse) {
290 no_reuse:
291 if (!usable_delta) {
292 buf = read_sha1_file(entry->idx.sha1, &type, &size);
293 if (!buf)
294 die("unable to read %s", sha1_to_hex(entry->idx.sha1));
295 /*
296 * make sure no cached delta data remains from a
297 * previous attempt before a pack split occurred.
298 */
299 free(entry->delta_data);
300 entry->delta_data = NULL;
301 entry->z_delta_size = 0;
302 } else if (entry->delta_data) {
303 size = entry->delta_size;
304 buf = entry->delta_data;
305 entry->delta_data = NULL;
306 type = (allow_ofs_delta && entry->delta->idx.offset) ?
307 OBJ_OFS_DELTA : OBJ_REF_DELTA;
308 } else {
309 buf = get_delta(entry);
310 size = entry->delta_size;
311 type = (allow_ofs_delta && entry->delta->idx.offset) ?
312 OBJ_OFS_DELTA : OBJ_REF_DELTA;
313 }
315 if (entry->z_delta_size)
316 datalen = entry->z_delta_size;
317 else
318 datalen = do_compress(&buf, size);
320 /*
321 * The object header is a byte of 'type' followed by zero or
322 * more bytes of length.
323 */
324 hdrlen = encode_header(type, size, header);
326 if (type == OBJ_OFS_DELTA) {
327 /*
328 * Deltas with relative base contain an additional
329 * encoding of the relative offset for the delta
330 * base from this object's position in the pack.
331 */
332 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
333 unsigned pos = sizeof(dheader) - 1;
334 dheader[pos] = ofs & 127;
335 while (ofs >>= 7)
336 dheader[--pos] = 128 | (--ofs & 127);
337 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
338 free(buf);
339 return 0;
340 }
341 sha1write(f, header, hdrlen);
342 sha1write(f, dheader + pos, sizeof(dheader) - pos);
343 hdrlen += sizeof(dheader) - pos;
344 } else if (type == OBJ_REF_DELTA) {
345 /*
346 * Deltas with a base reference contain
347 * an additional 20 bytes for the base sha1.
348 */
349 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
350 free(buf);
351 return 0;
352 }
353 sha1write(f, header, hdrlen);
354 sha1write(f, entry->delta->idx.sha1, 20);
355 hdrlen += 20;
356 } else {
357 if (limit && hdrlen + datalen + 20 >= limit) {
358 free(buf);
359 return 0;
360 }
361 sha1write(f, header, hdrlen);
362 }
363 sha1write(f, buf, datalen);
364 free(buf);
365 }
366 else {
367 struct packed_git *p = entry->in_pack;
368 struct pack_window *w_curs = NULL;
369 struct revindex_entry *revidx;
370 off_t offset;
372 if (entry->delta)
373 type = (allow_ofs_delta && entry->delta->idx.offset) ?
374 OBJ_OFS_DELTA : OBJ_REF_DELTA;
375 hdrlen = encode_header(type, entry->size, header);
377 offset = entry->in_pack_offset;
378 revidx = find_pack_revindex(p, offset);
379 datalen = revidx[1].offset - offset;
380 if (!pack_to_stdout && p->index_version > 1 &&
381 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
382 error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
383 unuse_pack(&w_curs);
384 goto no_reuse;
385 }
387 offset += entry->in_pack_header_size;
388 datalen -= entry->in_pack_header_size;
389 if (!pack_to_stdout && p->index_version == 1 &&
390 check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
391 error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
392 unuse_pack(&w_curs);
393 goto no_reuse;
394 }
396 if (type == OBJ_OFS_DELTA) {
397 off_t ofs = entry->idx.offset - entry->delta->idx.offset;
398 unsigned pos = sizeof(dheader) - 1;
399 dheader[pos] = ofs & 127;
400 while (ofs >>= 7)
401 dheader[--pos] = 128 | (--ofs & 127);
402 if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
403 unuse_pack(&w_curs);
404 return 0;
405 }
406 sha1write(f, header, hdrlen);
407 sha1write(f, dheader + pos, sizeof(dheader) - pos);
408 hdrlen += sizeof(dheader) - pos;
409 reused_delta++;
410 } else if (type == OBJ_REF_DELTA) {
411 if (limit && hdrlen + 20 + datalen + 20 >= limit) {
412 unuse_pack(&w_curs);
413 return 0;
414 }
415 sha1write(f, header, hdrlen);
416 sha1write(f, entry->delta->idx.sha1, 20);
417 hdrlen += 20;
418 reused_delta++;
419 } else {
420 if (limit && hdrlen + datalen + 20 >= limit) {
421 unuse_pack(&w_curs);
422 return 0;
423 }
424 sha1write(f, header, hdrlen);
425 }
426 copy_pack_data(f, p, &w_curs, offset, datalen);
427 unuse_pack(&w_curs);
428 reused++;
429 }
430 if (usable_delta)
431 written_delta++;
432 written++;
433 if (!pack_to_stdout)
434 entry->idx.crc32 = crc32_end(f);
435 return hdrlen + datalen;
436 }
438 static int write_one(struct sha1file *f,
439 struct object_entry *e,
440 off_t *offset)
441 {
442 unsigned long size;
444 /* offset is non zero if object is written already. */
445 if (e->idx.offset || e->preferred_base)
446 return -1;
448 /*
449 * If we are deltified, attempt to write out base object first.
450 * If that fails due to the pack size limit then the current
451 * object might still possibly fit undeltified within that limit.
452 */
453 if (e->delta)
454 write_one(f, e->delta, offset);
456 e->idx.offset = *offset;
457 size = write_object(f, e, *offset);
458 if (!size) {
459 e->idx.offset = 0;
460 return 0;
461 }
462 written_list[nr_written++] = &e->idx;
464 /* make sure off_t is sufficiently large not to wrap */
465 if (*offset > *offset + size)
466 die("pack too large for current definition of off_t");
467 *offset += size;
468 return 1;
469 }
471 /* forward declaration for write_pack_file */
472 static int adjust_perm(const char *path, mode_t mode);
474 static void write_pack_file(void)
475 {
476 uint32_t i = 0, j;
477 struct sha1file *f;
478 off_t offset;
479 struct pack_header hdr;
480 uint32_t nr_remaining = nr_result;
481 time_t last_mtime = 0;
483 if (progress > pack_to_stdout)
484 progress_state = start_progress("Writing objects", nr_result);
485 written_list = xmalloc(nr_objects * sizeof(*written_list));
487 do {
488 unsigned char sha1[20];
489 char *pack_tmp_name = NULL;
491 if (pack_to_stdout) {
492 f = sha1fd_throughput(1, "<stdout>", progress_state);
493 } else {
494 char tmpname[PATH_MAX];
495 int fd;
496 fd = odb_mkstemp(tmpname, sizeof(tmpname),
497 "pack/tmp_pack_XXXXXX");
498 pack_tmp_name = xstrdup(tmpname);
499 f = sha1fd(fd, pack_tmp_name);
500 }
502 hdr.hdr_signature = htonl(PACK_SIGNATURE);
503 hdr.hdr_version = htonl(PACK_VERSION);
504 hdr.hdr_entries = htonl(nr_remaining);
505 sha1write(f, &hdr, sizeof(hdr));
506 offset = sizeof(hdr);
507 nr_written = 0;
508 for (i = 0; i < nr_objects; i++)
509 if (write_one(f, objects + i, &offset) == 1)
510 display_progress(progress_state, written);
512 /*
513 * Did we write the wrong # entries in the header?
514 * If so, rewrite it like in fast-import
515 */
516 if (pack_to_stdout) {
517 sha1close(f, sha1, CSUM_CLOSE);
518 } else if (nr_written == nr_remaining) {
519 sha1close(f, sha1, CSUM_FSYNC);
520 } else {
521 int fd = sha1close(f, sha1, 0);
522 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
523 nr_written, sha1, offset);
524 close(fd);
525 }
527 if (!pack_to_stdout) {
528 mode_t mode = umask(0);
529 struct stat st;
530 const char *idx_tmp_name;
531 char tmpname[PATH_MAX];
533 umask(mode);
534 mode = 0444 & ~mode;
536 idx_tmp_name = write_idx_file(NULL, written_list,
537 nr_written, sha1);
539 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
540 base_name, sha1_to_hex(sha1));
541 free_pack_by_name(tmpname);
542 if (adjust_perm(pack_tmp_name, mode))
543 die_errno("unable to make temporary pack file readable");
544 if (rename(pack_tmp_name, tmpname))
545 die_errno("unable to rename temporary pack file");
547 /*
548 * Packs are runtime accessed in their mtime
549 * order since newer packs are more likely to contain
550 * younger objects. So if we are creating multiple
551 * packs then we should modify the mtime of later ones
552 * to preserve this property.
553 */
554 if (stat(tmpname, &st) < 0) {
555 warning("failed to stat %s: %s",
556 tmpname, strerror(errno));
557 } else if (!last_mtime) {
558 last_mtime = st.st_mtime;
559 } else {
560 struct utimbuf utb;
561 utb.actime = st.st_atime;
562 utb.modtime = --last_mtime;
563 if (utime(tmpname, &utb) < 0)
564 warning("failed utime() on %s: %s",
565 tmpname, strerror(errno));
566 }
568 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
569 base_name, sha1_to_hex(sha1));
570 if (adjust_perm(idx_tmp_name, mode))
571 die_errno("unable to make temporary index file readable");
572 if (rename(idx_tmp_name, tmpname))
573 die_errno("unable to rename temporary index file");
575 free((void *) idx_tmp_name);
576 free(pack_tmp_name);
577 puts(sha1_to_hex(sha1));
578 }
580 /* mark written objects as written to previous pack */
581 for (j = 0; j < nr_written; j++) {
582 written_list[j]->offset = (off_t)-1;
583 }
584 nr_remaining -= nr_written;
585 } while (nr_remaining);
587 free(written_list);
588 stop_progress(&progress_state);
589 if (written != nr_result)
590 die("wrote %"PRIu32" objects while expecting %"PRIu32,
591 written, nr_result);
592 }
594 static int locate_object_entry_hash(const unsigned char *sha1)
595 {
596 int i;
597 unsigned int ui;
598 memcpy(&ui, sha1, sizeof(unsigned int));
599 i = ui % object_ix_hashsz;
600 while (0 < object_ix[i]) {
601 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
602 return i;
603 if (++i == object_ix_hashsz)
604 i = 0;
605 }
606 return -1 - i;
607 }
609 static struct object_entry *locate_object_entry(const unsigned char *sha1)
610 {
611 int i;
613 if (!object_ix_hashsz)
614 return NULL;
616 i = locate_object_entry_hash(sha1);
617 if (0 <= i)
618 return &objects[object_ix[i]-1];
619 return NULL;
620 }
622 static void rehash_objects(void)
623 {
624 uint32_t i;
625 struct object_entry *oe;
627 object_ix_hashsz = nr_objects * 3;
628 if (object_ix_hashsz < 1024)
629 object_ix_hashsz = 1024;
630 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
631 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
632 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
633 int ix = locate_object_entry_hash(oe->idx.sha1);
634 if (0 <= ix)
635 continue;
636 ix = -1 - ix;
637 object_ix[ix] = i + 1;
638 }
639 }
641 static unsigned name_hash(const char *name)
642 {
643 unsigned c, hash = 0;
645 if (!name)
646 return 0;
648 /*
649 * This effectively just creates a sortable number from the
650 * last sixteen non-whitespace characters. Last characters
651 * count "most", so things that end in ".c" sort together.
652 */
653 while ((c = *name++) != 0) {
654 if (isspace(c))
655 continue;
656 hash = (hash >> 2) + (c << 24);
657 }
658 return hash;
659 }
661 static void setup_delta_attr_check(struct git_attr_check *check)
662 {
663 static struct git_attr *attr_delta;
665 if (!attr_delta)
666 attr_delta = git_attr("delta");
668 check[0].attr = attr_delta;
669 }
671 static int no_try_delta(const char *path)
672 {
673 struct git_attr_check check[1];
675 setup_delta_attr_check(check);
676 if (git_checkattr(path, ARRAY_SIZE(check), check))
677 return 0;
678 if (ATTR_FALSE(check->value))
679 return 1;
680 return 0;
681 }
683 static int add_object_entry(const unsigned char *sha1, enum object_type type,
684 const char *name, int exclude)
685 {
686 struct object_entry *entry;
687 struct packed_git *p, *found_pack = NULL;
688 off_t found_offset = 0;
689 int ix;
690 unsigned hash = name_hash(name);
692 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
693 if (ix >= 0) {
694 if (exclude) {
695 entry = objects + object_ix[ix] - 1;
696 if (!entry->preferred_base)
697 nr_result--;
698 entry->preferred_base = 1;
699 }
700 return 0;
701 }
703 if (!exclude && local && has_loose_object_nonlocal(sha1))
704 return 0;
706 for (p = packed_git; p; p = p->next) {
707 off_t offset = find_pack_entry_one(sha1, p);
708 if (offset) {
709 if (!found_pack) {
710 found_offset = offset;
711 found_pack = p;
712 }
713 if (exclude)
714 break;
715 if (incremental)
716 return 0;
717 if (local && !p->pack_local)
718 return 0;
719 if (ignore_packed_keep && p->pack_local && p->pack_keep)
720 return 0;
721 }
722 }
724 if (nr_objects >= nr_alloc) {
725 nr_alloc = (nr_alloc + 1024) * 3 / 2;
726 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
727 }
729 entry = objects + nr_objects++;
730 memset(entry, 0, sizeof(*entry));
731 hashcpy(entry->idx.sha1, sha1);
732 entry->hash = hash;
733 if (type)
734 entry->type = type;
735 if (exclude)
736 entry->preferred_base = 1;
737 else
738 nr_result++;
739 if (found_pack) {
740 entry->in_pack = found_pack;
741 entry->in_pack_offset = found_offset;
742 }
744 if (object_ix_hashsz * 3 <= nr_objects * 4)
745 rehash_objects();
746 else
747 object_ix[-1 - ix] = nr_objects;
749 display_progress(progress_state, nr_objects);
751 if (name && no_try_delta(name))
752 entry->no_try_delta = 1;
754 return 1;
755 }
757 struct pbase_tree_cache {
758 unsigned char sha1[20];
759 int ref;
760 int temporary;
761 void *tree_data;
762 unsigned long tree_size;
763 };
765 static struct pbase_tree_cache *(pbase_tree_cache[256]);
766 static int pbase_tree_cache_ix(const unsigned char *sha1)
767 {
768 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
769 }
770 static int pbase_tree_cache_ix_incr(int ix)
771 {
772 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
773 }
775 static struct pbase_tree {
776 struct pbase_tree *next;
777 /* This is a phony "cache" entry; we are not
778 * going to evict it nor find it through _get()
779 * mechanism -- this is for the toplevel node that
780 * would almost always change with any commit.
781 */
782 struct pbase_tree_cache pcache;
783 } *pbase_tree;
785 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
786 {
787 struct pbase_tree_cache *ent, *nent;
788 void *data;
789 unsigned long size;
790 enum object_type type;
791 int neigh;
792 int my_ix = pbase_tree_cache_ix(sha1);
793 int available_ix = -1;
795 /* pbase-tree-cache acts as a limited hashtable.
796 * your object will be found at your index or within a few
797 * slots after that slot if it is cached.
798 */
799 for (neigh = 0; neigh < 8; neigh++) {
800 ent = pbase_tree_cache[my_ix];
801 if (ent && !hashcmp(ent->sha1, sha1)) {
802 ent->ref++;
803 return ent;
804 }
805 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
806 ((0 <= available_ix) &&
807 (!ent && pbase_tree_cache[available_ix])))
808 available_ix = my_ix;
809 if (!ent)
810 break;
811 my_ix = pbase_tree_cache_ix_incr(my_ix);
812 }
814 /* Did not find one. Either we got a bogus request or
815 * we need to read and perhaps cache.
816 */
817 data = read_sha1_file(sha1, &type, &size);
818 if (!data)
819 return NULL;
820 if (type != OBJ_TREE) {
821 free(data);
822 return NULL;
823 }
825 /* We need to either cache or return a throwaway copy */
827 if (available_ix < 0)
828 ent = NULL;
829 else {
830 ent = pbase_tree_cache[available_ix];
831 my_ix = available_ix;
832 }
834 if (!ent) {
835 nent = xmalloc(sizeof(*nent));
836 nent->temporary = (available_ix < 0);
837 }
838 else {
839 /* evict and reuse */
840 free(ent->tree_data);
841 nent = ent;
842 }
843 hashcpy(nent->sha1, sha1);
844 nent->tree_data = data;
845 nent->tree_size = size;
846 nent->ref = 1;
847 if (!nent->temporary)
848 pbase_tree_cache[my_ix] = nent;
849 return nent;
850 }
852 static void pbase_tree_put(struct pbase_tree_cache *cache)
853 {
854 if (!cache->temporary) {
855 cache->ref--;
856 return;
857 }
858 free(cache->tree_data);
859 free(cache);
860 }
862 static int name_cmp_len(const char *name)
863 {
864 int i;
865 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
866 ;
867 return i;
868 }
870 static void add_pbase_object(struct tree_desc *tree,
871 const char *name,
872 int cmplen,
873 const char *fullname)
874 {
875 struct name_entry entry;
876 int cmp;
878 while (tree_entry(tree,&entry)) {
879 if (S_ISGITLINK(entry.mode))
880 continue;
881 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
882 memcmp(name, entry.path, cmplen);
883 if (cmp > 0)
884 continue;
885 if (cmp < 0)
886 return;
887 if (name[cmplen] != '/') {
888 add_object_entry(entry.sha1,
889 object_type(entry.mode),
890 fullname, 1);
891 return;
892 }
893 if (S_ISDIR(entry.mode)) {
894 struct tree_desc sub;
895 struct pbase_tree_cache *tree;
896 const char *down = name+cmplen+1;
897 int downlen = name_cmp_len(down);
899 tree = pbase_tree_get(entry.sha1);
900 if (!tree)
901 return;
902 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
904 add_pbase_object(&sub, down, downlen, fullname);
905 pbase_tree_put(tree);
906 }
907 }
908 }
910 static unsigned *done_pbase_paths;
911 static int done_pbase_paths_num;
912 static int done_pbase_paths_alloc;
913 static int done_pbase_path_pos(unsigned hash)
914 {
915 int lo = 0;
916 int hi = done_pbase_paths_num;
917 while (lo < hi) {
918 int mi = (hi + lo) / 2;
919 if (done_pbase_paths[mi] == hash)
920 return mi;
921 if (done_pbase_paths[mi] < hash)
922 hi = mi;
923 else
924 lo = mi + 1;
925 }
926 return -lo-1;
927 }
929 static int check_pbase_path(unsigned hash)
930 {
931 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
932 if (0 <= pos)
933 return 1;
934 pos = -pos - 1;
935 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
936 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
937 done_pbase_paths = xrealloc(done_pbase_paths,
938 done_pbase_paths_alloc *
939 sizeof(unsigned));
940 }
941 done_pbase_paths_num++;
942 if (pos < done_pbase_paths_num)
943 memmove(done_pbase_paths + pos + 1,
944 done_pbase_paths + pos,
945 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
946 done_pbase_paths[pos] = hash;
947 return 0;
948 }
950 static void add_preferred_base_object(const char *name)
951 {
952 struct pbase_tree *it;
953 int cmplen;
954 unsigned hash = name_hash(name);
956 if (!num_preferred_base || check_pbase_path(hash))
957 return;
959 cmplen = name_cmp_len(name);
960 for (it = pbase_tree; it; it = it->next) {
961 if (cmplen == 0) {
962 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
963 }
964 else {
965 struct tree_desc tree;
966 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
967 add_pbase_object(&tree, name, cmplen, name);
968 }
969 }
970 }
972 static void add_preferred_base(unsigned char *sha1)
973 {
974 struct pbase_tree *it;
975 void *data;
976 unsigned long size;
977 unsigned char tree_sha1[20];
979 if (window <= num_preferred_base++)
980 return;
982 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
983 if (!data)
984 return;
986 for (it = pbase_tree; it; it = it->next) {
987 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
988 free(data);
989 return;
990 }
991 }
993 it = xcalloc(1, sizeof(*it));
994 it->next = pbase_tree;
995 pbase_tree = it;
997 hashcpy(it->pcache.sha1, tree_sha1);
998 it->pcache.tree_data = data;
999 it->pcache.tree_size = size;
1000 }
1002 static void cleanup_preferred_base(void)
1003 {
1004 struct pbase_tree *it;
1005 unsigned i;
1007 it = pbase_tree;
1008 pbase_tree = NULL;
1009 while (it) {
1010 struct pbase_tree *this = it;
1011 it = this->next;
1012 free(this->pcache.tree_data);
1013 free(this);
1014 }
1016 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1017 if (!pbase_tree_cache[i])
1018 continue;
1019 free(pbase_tree_cache[i]->tree_data);
1020 free(pbase_tree_cache[i]);
1021 pbase_tree_cache[i] = NULL;
1022 }
1024 free(done_pbase_paths);
1025 done_pbase_paths = NULL;
1026 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1027 }
1029 static void check_object(struct object_entry *entry)
1030 {
1031 if (entry->in_pack) {
1032 struct packed_git *p = entry->in_pack;
1033 struct pack_window *w_curs = NULL;
1034 const unsigned char *base_ref = NULL;
1035 struct object_entry *base_entry;
1036 unsigned long used, used_0;
1037 unsigned int avail;
1038 off_t ofs;
1039 unsigned char *buf, c;
1041 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1043 /*
1044 * We want in_pack_type even if we do not reuse delta
1045 * since non-delta representations could still be reused.
1046 */
1047 used = unpack_object_header_buffer(buf, avail,
1048 &entry->in_pack_type,
1049 &entry->size);
1050 if (used == 0)
1051 goto give_up;
1053 /*
1054 * Determine if this is a delta and if so whether we can
1055 * reuse it or not. Otherwise let's find out as cheaply as
1056 * possible what the actual type and size for this object is.
1057 */
1058 switch (entry->in_pack_type) {
1059 default:
1060 /* Not a delta hence we've already got all we need. */
1061 entry->type = entry->in_pack_type;
1062 entry->in_pack_header_size = used;
1063 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1064 goto give_up;
1065 unuse_pack(&w_curs);
1066 return;
1067 case OBJ_REF_DELTA:
1068 if (reuse_delta && !entry->preferred_base)
1069 base_ref = use_pack(p, &w_curs,
1070 entry->in_pack_offset + used, NULL);
1071 entry->in_pack_header_size = used + 20;
1072 break;
1073 case OBJ_OFS_DELTA:
1074 buf = use_pack(p, &w_curs,
1075 entry->in_pack_offset + used, NULL);
1076 used_0 = 0;
1077 c = buf[used_0++];
1078 ofs = c & 127;
1079 while (c & 128) {
1080 ofs += 1;
1081 if (!ofs || MSB(ofs, 7)) {
1082 error("delta base offset overflow in pack for %s",
1083 sha1_to_hex(entry->idx.sha1));
1084 goto give_up;
1085 }
1086 c = buf[used_0++];
1087 ofs = (ofs << 7) + (c & 127);
1088 }
1089 ofs = entry->in_pack_offset - ofs;
1090 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1091 error("delta base offset out of bound for %s",
1092 sha1_to_hex(entry->idx.sha1));
1093 goto give_up;
1094 }
1095 if (reuse_delta && !entry->preferred_base) {
1096 struct revindex_entry *revidx;
1097 revidx = find_pack_revindex(p, ofs);
1098 if (!revidx)
1099 goto give_up;
1100 base_ref = nth_packed_object_sha1(p, revidx->nr);
1101 }
1102 entry->in_pack_header_size = used + used_0;
1103 break;
1104 }
1106 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1107 /*
1108 * If base_ref was set above that means we wish to
1109 * reuse delta data, and we even found that base
1110 * in the list of objects we want to pack. Goodie!
1111 *
1112 * Depth value does not matter - find_deltas() will
1113 * never consider reused delta as the base object to
1114 * deltify other objects against, in order to avoid
1115 * circular deltas.
1116 */
1117 entry->type = entry->in_pack_type;
1118 entry->delta = base_entry;
1119 entry->delta_size = entry->size;
1120 entry->delta_sibling = base_entry->delta_child;
1121 base_entry->delta_child = entry;
1122 unuse_pack(&w_curs);
1123 return;
1124 }
1126 if (entry->type) {
1127 /*
1128 * This must be a delta and we already know what the
1129 * final object type is. Let's extract the actual
1130 * object size from the delta header.
1131 */
1132 entry->size = get_size_from_delta(p, &w_curs,
1133 entry->in_pack_offset + entry->in_pack_header_size);
1134 if (entry->size == 0)
1135 goto give_up;
1136 unuse_pack(&w_curs);
1137 return;
1138 }
1140 /*
1141 * No choice but to fall back to the recursive delta walk
1142 * with sha1_object_info() to find about the object type
1143 * at this point...
1144 */
1145 give_up:
1146 unuse_pack(&w_curs);
1147 }
1149 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1150 /*
1151 * The error condition is checked in prepare_pack(). This is
1152 * to permit a missing preferred base object to be ignored
1153 * as a preferred base. Doing so can result in a larger
1154 * pack file, but the transfer will still take place.
1155 */
1156 }
1158 static int pack_offset_sort(const void *_a, const void *_b)
1159 {
1160 const struct object_entry *a = *(struct object_entry **)_a;
1161 const struct object_entry *b = *(struct object_entry **)_b;
1163 /* avoid filesystem trashing with loose objects */
1164 if (!a->in_pack && !b->in_pack)
1165 return hashcmp(a->idx.sha1, b->idx.sha1);
1167 if (a->in_pack < b->in_pack)
1168 return -1;
1169 if (a->in_pack > b->in_pack)
1170 return 1;
1171 return a->in_pack_offset < b->in_pack_offset ? -1 :
1172 (a->in_pack_offset > b->in_pack_offset);
1173 }
1175 static void get_object_details(void)
1176 {
1177 uint32_t i;
1178 struct object_entry **sorted_by_offset;
1180 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1181 for (i = 0; i < nr_objects; i++)
1182 sorted_by_offset[i] = objects + i;
1183 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1185 for (i = 0; i < nr_objects; i++)
1186 check_object(sorted_by_offset[i]);
1188 free(sorted_by_offset);
1189 }
1191 /*
1192 * We search for deltas in a list sorted by type, by filename hash, and then
1193 * by size, so that we see progressively smaller and smaller files.
1194 * That's because we prefer deltas to be from the bigger file
1195 * to the smaller -- deletes are potentially cheaper, but perhaps
1196 * more importantly, the bigger file is likely the more recent
1197 * one. The deepest deltas are therefore the oldest objects which are
1198 * less susceptible to be accessed often.
1199 */
1200 static int type_size_sort(const void *_a, const void *_b)
1201 {
1202 const struct object_entry *a = *(struct object_entry **)_a;
1203 const struct object_entry *b = *(struct object_entry **)_b;
1205 if (a->type > b->type)
1206 return -1;
1207 if (a->type < b->type)
1208 return 1;
1209 if (a->hash > b->hash)
1210 return -1;
1211 if (a->hash < b->hash)
1212 return 1;
1213 if (a->preferred_base > b->preferred_base)
1214 return -1;
1215 if (a->preferred_base < b->preferred_base)
1216 return 1;
1217 if (a->size > b->size)
1218 return -1;
1219 if (a->size < b->size)
1220 return 1;
1221 return a < b ? -1 : (a > b); /* newest first */
1222 }
1224 struct unpacked {
1225 struct object_entry *entry;
1226 void *data;
1227 struct delta_index *index;
1228 unsigned depth;
1229 };
1231 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1232 unsigned long delta_size)
1233 {
1234 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1235 return 0;
1237 if (delta_size < cache_max_small_delta_size)
1238 return 1;
1240 /* cache delta, if objects are large enough compared to delta size */
1241 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1242 return 1;
1244 return 0;
1245 }
1247 #ifndef NO_PTHREADS
1249 static pthread_mutex_t read_mutex;
1250 #define read_lock() pthread_mutex_lock(&read_mutex)
1251 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1253 static pthread_mutex_t cache_mutex;
1254 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1255 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1257 static pthread_mutex_t progress_mutex;
1258 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1259 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1261 #else
1263 #define read_lock() (void)0
1264 #define read_unlock() (void)0
1265 #define cache_lock() (void)0
1266 #define cache_unlock() (void)0
1267 #define progress_lock() (void)0
1268 #define progress_unlock() (void)0
1270 #endif
1272 static int try_delta(struct unpacked *trg, struct unpacked *src,
1273 unsigned max_depth, unsigned long *mem_usage)
1274 {
1275 struct object_entry *trg_entry = trg->entry;
1276 struct object_entry *src_entry = src->entry;
1277 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1278 unsigned ref_depth;
1279 enum object_type type;
1280 void *delta_buf;
1282 /* Don't bother doing diffs between different types */
1283 if (trg_entry->type != src_entry->type)
1284 return -1;
1286 /*
1287 * We do not bother to try a delta that we discarded
1288 * on an earlier try, but only when reusing delta data.
1289 */
1290 if (reuse_delta && trg_entry->in_pack &&
1291 trg_entry->in_pack == src_entry->in_pack &&
1292 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1293 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1294 return 0;
1296 /* Let's not bust the allowed depth. */
1297 if (src->depth >= max_depth)
1298 return 0;
1300 /* Now some size filtering heuristics. */
1301 trg_size = trg_entry->size;
1302 if (!trg_entry->delta) {
1303 max_size = trg_size/2 - 20;
1304 ref_depth = 1;
1305 } else {
1306 max_size = trg_entry->delta_size;
1307 ref_depth = trg->depth;
1308 }
1309 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1310 (max_depth - ref_depth + 1);
1311 if (max_size == 0)
1312 return 0;
1313 src_size = src_entry->size;
1314 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1315 if (sizediff >= max_size)
1316 return 0;
1317 if (trg_size < src_size / 32)
1318 return 0;
1320 /* Load data if not already done */
1321 if (!trg->data) {
1322 read_lock();
1323 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1324 read_unlock();
1325 if (!trg->data)
1326 die("object %s cannot be read",
1327 sha1_to_hex(trg_entry->idx.sha1));
1328 if (sz != trg_size)
1329 die("object %s inconsistent object length (%lu vs %lu)",
1330 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1331 *mem_usage += sz;
1332 }
1333 if (!src->data) {
1334 read_lock();
1335 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1336 read_unlock();
1337 if (!src->data)
1338 die("object %s cannot be read",
1339 sha1_to_hex(src_entry->idx.sha1));
1340 if (sz != src_size)
1341 die("object %s inconsistent object length (%lu vs %lu)",
1342 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1343 *mem_usage += sz;
1344 }
1345 if (!src->index) {
1346 src->index = create_delta_index(src->data, src_size);
1347 if (!src->index) {
1348 static int warned = 0;
1349 if (!warned++)
1350 warning("suboptimal pack - out of memory");
1351 return 0;
1352 }
1353 *mem_usage += sizeof_delta_index(src->index);
1354 }
1356 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1357 if (!delta_buf)
1358 return 0;
1360 if (trg_entry->delta) {
1361 /* Prefer only shallower same-sized deltas. */
1362 if (delta_size == trg_entry->delta_size &&
1363 src->depth + 1 >= trg->depth) {
1364 free(delta_buf);
1365 return 0;
1366 }
1367 }
1369 /*
1370 * Handle memory allocation outside of the cache
1371 * accounting lock. Compiler will optimize the strangeness
1372 * away when NO_PTHREADS is defined.
1373 */
1374 free(trg_entry->delta_data);
1375 cache_lock();
1376 if (trg_entry->delta_data) {
1377 delta_cache_size -= trg_entry->delta_size;
1378 trg_entry->delta_data = NULL;
1379 }
1380 if (delta_cacheable(src_size, trg_size, delta_size)) {
1381 delta_cache_size += delta_size;
1382 cache_unlock();
1383 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1384 } else {
1385 cache_unlock();
1386 free(delta_buf);
1387 }
1389 trg_entry->delta = src_entry;
1390 trg_entry->delta_size = delta_size;
1391 trg->depth = src->depth + 1;
1393 return 1;
1394 }
1396 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1397 {
1398 struct object_entry *child = me->delta_child;
1399 unsigned int m = n;
1400 while (child) {
1401 unsigned int c = check_delta_limit(child, n + 1);
1402 if (m < c)
1403 m = c;
1404 child = child->delta_sibling;
1405 }
1406 return m;
1407 }
1409 static unsigned long free_unpacked(struct unpacked *n)
1410 {
1411 unsigned long freed_mem = sizeof_delta_index(n->index);
1412 free_delta_index(n->index);
1413 n->index = NULL;
1414 if (n->data) {
1415 freed_mem += n->entry->size;
1416 free(n->data);
1417 n->data = NULL;
1418 }
1419 n->entry = NULL;
1420 n->depth = 0;
1421 return freed_mem;
1422 }
1424 static void find_deltas(struct object_entry **list, unsigned *list_size,
1425 int window, int depth, unsigned *processed)
1426 {
1427 uint32_t i, idx = 0, count = 0;
1428 struct unpacked *array;
1429 unsigned long mem_usage = 0;
1431 array = xcalloc(window, sizeof(struct unpacked));
1433 for (;;) {
1434 struct object_entry *entry;
1435 struct unpacked *n = array + idx;
1436 int j, max_depth, best_base = -1;
1438 progress_lock();
1439 if (!*list_size) {
1440 progress_unlock();
1441 break;
1442 }
1443 entry = *list++;
1444 (*list_size)--;
1445 if (!entry->preferred_base) {
1446 (*processed)++;
1447 display_progress(progress_state, *processed);
1448 }
1449 progress_unlock();
1451 mem_usage -= free_unpacked(n);
1452 n->entry = entry;
1454 while (window_memory_limit &&
1455 mem_usage > window_memory_limit &&
1456 count > 1) {
1457 uint32_t tail = (idx + window - count) % window;
1458 mem_usage -= free_unpacked(array + tail);
1459 count--;
1460 }
1462 /* We do not compute delta to *create* objects we are not
1463 * going to pack.
1464 */
1465 if (entry->preferred_base)
1466 goto next;
1468 /*
1469 * If the current object is at pack edge, take the depth the
1470 * objects that depend on the current object into account
1471 * otherwise they would become too deep.
1472 */
1473 max_depth = depth;
1474 if (entry->delta_child) {
1475 max_depth -= check_delta_limit(entry, 0);
1476 if (max_depth <= 0)
1477 goto next;
1478 }
1480 j = window;
1481 while (--j > 0) {
1482 int ret;
1483 uint32_t other_idx = idx + j;
1484 struct unpacked *m;
1485 if (other_idx >= window)
1486 other_idx -= window;
1487 m = array + other_idx;
1488 if (!m->entry)
1489 break;
1490 ret = try_delta(n, m, max_depth, &mem_usage);
1491 if (ret < 0)
1492 break;
1493 else if (ret > 0)
1494 best_base = other_idx;
1495 }
1497 /*
1498 * If we decided to cache the delta data, then it is best
1499 * to compress it right away. First because we have to do
1500 * it anyway, and doing it here while we're threaded will
1501 * save a lot of time in the non threaded write phase,
1502 * as well as allow for caching more deltas within
1503 * the same cache size limit.
1504 * ...
1505 * But only if not writing to stdout, since in that case
1506 * the network is most likely throttling writes anyway,
1507 * and therefore it is best to go to the write phase ASAP
1508 * instead, as we can afford spending more time compressing
1509 * between writes at that moment.
1510 */
1511 if (entry->delta_data && !pack_to_stdout) {
1512 entry->z_delta_size = do_compress(&entry->delta_data,
1513 entry->delta_size);
1514 cache_lock();
1515 delta_cache_size -= entry->delta_size;
1516 delta_cache_size += entry->z_delta_size;
1517 cache_unlock();
1518 }
1520 /* if we made n a delta, and if n is already at max
1521 * depth, leaving it in the window is pointless. we
1522 * should evict it first.
1523 */
1524 if (entry->delta && max_depth <= n->depth)
1525 continue;
1527 /*
1528 * Move the best delta base up in the window, after the
1529 * currently deltified object, to keep it longer. It will
1530 * be the first base object to be attempted next.
1531 */
1532 if (entry->delta) {
1533 struct unpacked swap = array[best_base];
1534 int dist = (window + idx - best_base) % window;
1535 int dst = best_base;
1536 while (dist--) {
1537 int src = (dst + 1) % window;
1538 array[dst] = array[src];
1539 dst = src;
1540 }
1541 array[dst] = swap;
1542 }
1544 next:
1545 idx++;
1546 if (count + 1 < window)
1547 count++;
1548 if (idx >= window)
1549 idx = 0;
1550 }
1552 for (i = 0; i < window; ++i) {
1553 free_delta_index(array[i].index);
1554 free(array[i].data);
1555 }
1556 free(array);
1557 }
1559 #ifndef NO_PTHREADS
1561 /*
1562 * The main thread waits on the condition that (at least) one of the workers
1563 * has stopped working (which is indicated in the .working member of
1564 * struct thread_params).
1565 * When a work thread has completed its work, it sets .working to 0 and
1566 * signals the main thread and waits on the condition that .data_ready
1567 * becomes 1.
1568 */
1570 struct thread_params {
1571 pthread_t thread;
1572 struct object_entry **list;
1573 unsigned list_size;
1574 unsigned remaining;
1575 int window;
1576 int depth;
1577 int working;
1578 int data_ready;
1579 pthread_mutex_t mutex;
1580 pthread_cond_t cond;
1581 unsigned *processed;
1582 };
1584 static pthread_cond_t progress_cond;
1586 /*
1587 * Mutex and conditional variable can't be statically-initialized on Windows.
1588 */
1589 static void init_threaded_search(void)
1590 {
1591 pthread_mutex_init(&read_mutex, NULL);
1592 pthread_mutex_init(&cache_mutex, NULL);
1593 pthread_mutex_init(&progress_mutex, NULL);
1594 pthread_cond_init(&progress_cond, NULL);
1595 }
1597 static void cleanup_threaded_search(void)
1598 {
1599 pthread_cond_destroy(&progress_cond);
1600 pthread_mutex_destroy(&read_mutex);
1601 pthread_mutex_destroy(&cache_mutex);
1602 pthread_mutex_destroy(&progress_mutex);
1603 }
1605 static void *threaded_find_deltas(void *arg)
1606 {
1607 struct thread_params *me = arg;
1609 while (me->remaining) {
1610 find_deltas(me->list, &me->remaining,
1611 me->window, me->depth, me->processed);
1613 progress_lock();
1614 me->working = 0;
1615 pthread_cond_signal(&progress_cond);
1616 progress_unlock();
1618 /*
1619 * We must not set ->data_ready before we wait on the
1620 * condition because the main thread may have set it to 1
1621 * before we get here. In order to be sure that new
1622 * work is available if we see 1 in ->data_ready, it
1623 * was initialized to 0 before this thread was spawned
1624 * and we reset it to 0 right away.
1625 */
1626 pthread_mutex_lock(&me->mutex);
1627 while (!me->data_ready)
1628 pthread_cond_wait(&me->cond, &me->mutex);
1629 me->data_ready = 0;
1630 pthread_mutex_unlock(&me->mutex);
1631 }
1632 /* leave ->working 1 so that this doesn't get more work assigned */
1633 return NULL;
1634 }
1636 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1637 int window, int depth, unsigned *processed)
1638 {
1639 struct thread_params *p;
1640 int i, ret, active_threads = 0;
1642 init_threaded_search();
1644 if (!delta_search_threads) /* --threads=0 means autodetect */
1645 delta_search_threads = online_cpus();
1646 if (delta_search_threads <= 1) {
1647 find_deltas(list, &list_size, window, depth, processed);
1648 cleanup_threaded_search();
1649 return;
1650 }
1651 if (progress > pack_to_stdout)
1652 fprintf(stderr, "Delta compression using up to %d threads.\n",
1653 delta_search_threads);
1654 p = xcalloc(delta_search_threads, sizeof(*p));
1656 /* Partition the work amongst work threads. */
1657 for (i = 0; i < delta_search_threads; i++) {
1658 unsigned sub_size = list_size / (delta_search_threads - i);
1660 /* don't use too small segments or no deltas will be found */
1661 if (sub_size < 2*window && i+1 < delta_search_threads)
1662 sub_size = 0;
1664 p[i].window = window;
1665 p[i].depth = depth;
1666 p[i].processed = processed;
1667 p[i].working = 1;
1668 p[i].data_ready = 0;
1670 /* try to split chunks on "path" boundaries */
1671 while (sub_size && sub_size < list_size &&
1672 list[sub_size]->hash &&
1673 list[sub_size]->hash == list[sub_size-1]->hash)
1674 sub_size++;
1676 p[i].list = list;
1677 p[i].list_size = sub_size;
1678 p[i].remaining = sub_size;
1680 list += sub_size;
1681 list_size -= sub_size;
1682 }
1684 /* Start work threads. */
1685 for (i = 0; i < delta_search_threads; i++) {
1686 if (!p[i].list_size)
1687 continue;
1688 pthread_mutex_init(&p[i].mutex, NULL);
1689 pthread_cond_init(&p[i].cond, NULL);
1690 ret = pthread_create(&p[i].thread, NULL,
1691 threaded_find_deltas, &p[i]);
1692 if (ret)
1693 die("unable to create thread: %s", strerror(ret));
1694 active_threads++;
1695 }
1697 /*
1698 * Now let's wait for work completion. Each time a thread is done
1699 * with its work, we steal half of the remaining work from the
1700 * thread with the largest number of unprocessed objects and give
1701 * it to that newly idle thread. This ensure good load balancing
1702 * until the remaining object list segments are simply too short
1703 * to be worth splitting anymore.
1704 */
1705 while (active_threads) {
1706 struct thread_params *target = NULL;
1707 struct thread_params *victim = NULL;
1708 unsigned sub_size = 0;
1710 progress_lock();
1711 for (;;) {
1712 for (i = 0; !target && i < delta_search_threads; i++)
1713 if (!p[i].working)
1714 target = &p[i];
1715 if (target)
1716 break;
1717 pthread_cond_wait(&progress_cond, &progress_mutex);
1718 }
1720 for (i = 0; i < delta_search_threads; i++)
1721 if (p[i].remaining > 2*window &&
1722 (!victim || victim->remaining < p[i].remaining))
1723 victim = &p[i];
1724 if (victim) {
1725 sub_size = victim->remaining / 2;
1726 list = victim->list + victim->list_size - sub_size;
1727 while (sub_size && list[0]->hash &&
1728 list[0]->hash == list[-1]->hash) {
1729 list++;
1730 sub_size--;
1731 }
1732 if (!sub_size) {
1733 /*
1734 * It is possible for some "paths" to have
1735 * so many objects that no hash boundary
1736 * might be found. Let's just steal the
1737 * exact half in that case.
1738 */
1739 sub_size = victim->remaining / 2;
1740 list -= sub_size;
1741 }
1742 target->list = list;
1743 victim->list_size -= sub_size;
1744 victim->remaining -= sub_size;
1745 }
1746 target->list_size = sub_size;
1747 target->remaining = sub_size;
1748 target->working = 1;
1749 progress_unlock();
1751 pthread_mutex_lock(&target->mutex);
1752 target->data_ready = 1;
1753 pthread_cond_signal(&target->cond);
1754 pthread_mutex_unlock(&target->mutex);
1756 if (!sub_size) {
1757 pthread_join(target->thread, NULL);
1758 pthread_cond_destroy(&target->cond);
1759 pthread_mutex_destroy(&target->mutex);
1760 active_threads--;
1761 }
1762 }
1763 cleanup_threaded_search();
1764 free(p);
1765 }
1767 #else
1768 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1769 #endif
1771 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1772 {
1773 unsigned char peeled[20];
1775 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1776 !peel_ref(path, peeled) && /* peelable? */
1777 !is_null_sha1(peeled) && /* annotated tag? */
1778 locate_object_entry(peeled)) /* object packed? */
1779 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1780 return 0;
1781 }
1783 static void prepare_pack(int window, int depth)
1784 {
1785 struct object_entry **delta_list;
1786 uint32_t i, nr_deltas;
1787 unsigned n;
1789 get_object_details();
1791 /*
1792 * If we're locally repacking then we need to be doubly careful
1793 * from now on in order to make sure no stealth corruption gets
1794 * propagated to the new pack. Clients receiving streamed packs
1795 * should validate everything they get anyway so no need to incur
1796 * the additional cost here in that case.
1797 */
1798 if (!pack_to_stdout)
1799 do_check_packed_object_crc = 1;
1801 if (!nr_objects || !window || !depth)
1802 return;
1804 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1805 nr_deltas = n = 0;
1807 for (i = 0; i < nr_objects; i++) {
1808 struct object_entry *entry = objects + i;
1810 if (entry->delta)
1811 /* This happens if we decided to reuse existing
1812 * delta from a pack. "reuse_delta &&" is implied.
1813 */
1814 continue;
1816 if (entry->size < 50)
1817 continue;
1819 if (entry->no_try_delta)
1820 continue;
1822 if (!entry->preferred_base) {
1823 nr_deltas++;
1824 if (entry->type < 0)
1825 die("unable to get type of object %s",
1826 sha1_to_hex(entry->idx.sha1));
1827 } else {
1828 if (entry->type < 0) {
1829 /*
1830 * This object is not found, but we
1831 * don't have to include it anyway.
1832 */
1833 continue;
1834 }
1835 }
1837 delta_list[n++] = entry;
1838 }
1840 if (nr_deltas && n > 1) {
1841 unsigned nr_done = 0;
1842 if (progress)
1843 progress_state = start_progress("Compressing objects",
1844 nr_deltas);
1845 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1846 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1847 stop_progress(&progress_state);
1848 if (nr_done != nr_deltas)
1849 die("inconsistency with delta count");
1850 }
1851 free(delta_list);
1852 }
1854 static int git_pack_config(const char *k, const char *v, void *cb)
1855 {
1856 if (!strcmp(k, "pack.window")) {
1857 window = git_config_int(k, v);
1858 return 0;
1859 }
1860 if (!strcmp(k, "pack.windowmemory")) {
1861 window_memory_limit = git_config_ulong(k, v);
1862 return 0;
1863 }
1864 if (!strcmp(k, "pack.depth")) {
1865 depth = git_config_int(k, v);
1866 return 0;
1867 }
1868 if (!strcmp(k, "pack.compression")) {
1869 int level = git_config_int(k, v);
1870 if (level == -1)
1871 level = Z_DEFAULT_COMPRESSION;
1872 else if (level < 0 || level > Z_BEST_COMPRESSION)
1873 die("bad pack compression level %d", level);
1874 pack_compression_level = level;
1875 pack_compression_seen = 1;
1876 return 0;
1877 }
1878 if (!strcmp(k, "pack.deltacachesize")) {
1879 max_delta_cache_size = git_config_int(k, v);
1880 return 0;
1881 }
1882 if (!strcmp(k, "pack.deltacachelimit")) {
1883 cache_max_small_delta_size = git_config_int(k, v);
1884 return 0;
1885 }
1886 if (!strcmp(k, "pack.threads")) {
1887 delta_search_threads = git_config_int(k, v);
1888 if (delta_search_threads < 0)
1889 die("invalid number of threads specified (%d)",
1890 delta_search_threads);
1891 #ifdef NO_PTHREADS
1892 if (delta_search_threads != 1)
1893 warning("no threads support, ignoring %s", k);
1894 #endif
1895 return 0;
1896 }
1897 if (!strcmp(k, "pack.indexversion")) {
1898 pack_idx_default_version = git_config_int(k, v);
1899 if (pack_idx_default_version > 2)
1900 die("bad pack.indexversion=%"PRIu32,
1901 pack_idx_default_version);
1902 return 0;
1903 }
1904 if (!strcmp(k, "pack.packsizelimit")) {
1905 pack_size_limit_cfg = git_config_ulong(k, v);
1906 return 0;
1907 }
1908 return git_default_config(k, v, cb);
1909 }
1911 static void read_object_list_from_stdin(void)
1912 {
1913 char line[40 + 1 + PATH_MAX + 2];
1914 unsigned char sha1[20];
1916 for (;;) {
1917 if (!fgets(line, sizeof(line), stdin)) {
1918 if (feof(stdin))
1919 break;
1920 if (!ferror(stdin))
1921 die("fgets returned NULL, not EOF, not error!");
1922 if (errno != EINTR)
1923 die_errno("fgets");
1924 clearerr(stdin);
1925 continue;
1926 }
1927 if (line[0] == '-') {
1928 if (get_sha1_hex(line+1, sha1))
1929 die("expected edge sha1, got garbage:\n %s",
1930 line);
1931 add_preferred_base(sha1);
1932 continue;
1933 }
1934 if (get_sha1_hex(line, sha1))
1935 die("expected sha1, got garbage:\n %s", line);
1937 add_preferred_base_object(line+41);
1938 add_object_entry(sha1, 0, line+41, 0);
1939 }
1940 }
1942 #define OBJECT_ADDED (1u<<20)
1944 static void show_commit(struct commit *commit, void *data)
1945 {
1946 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1947 commit->object.flags |= OBJECT_ADDED;
1948 }
1950 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1951 {
1952 char *name = path_name(path, last);
1954 add_preferred_base_object(name);
1955 add_object_entry(obj->sha1, obj->type, name, 0);
1956 obj->flags |= OBJECT_ADDED;
1958 /*
1959 * We will have generated the hash from the name,
1960 * but not saved a pointer to it - we can free it
1961 */
1962 free((char *)name);
1963 }
1965 static void show_edge(struct commit *commit)
1966 {
1967 add_preferred_base(commit->object.sha1);
1968 }
1970 struct in_pack_object {
1971 off_t offset;
1972 struct object *object;
1973 };
1975 struct in_pack {
1976 int alloc;
1977 int nr;
1978 struct in_pack_object *array;
1979 };
1981 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1982 {
1983 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1984 in_pack->array[in_pack->nr].object = object;
1985 in_pack->nr++;
1986 }
1988 /*
1989 * Compare the objects in the offset order, in order to emulate the
1990 * "git rev-list --objects" output that produced the pack originally.
1991 */
1992 static int ofscmp(const void *a_, const void *b_)
1993 {
1994 struct in_pack_object *a = (struct in_pack_object *)a_;
1995 struct in_pack_object *b = (struct in_pack_object *)b_;
1997 if (a->offset < b->offset)
1998 return -1;
1999 else if (a->offset > b->offset)
2000 return 1;
2001 else
2002 return hashcmp(a->object->sha1, b->object->sha1);
2003 }
2005 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2006 {
2007 struct packed_git *p;
2008 struct in_pack in_pack;
2009 uint32_t i;
2011 memset(&in_pack, 0, sizeof(in_pack));
2013 for (p = packed_git; p; p = p->next) {
2014 const unsigned char *sha1;
2015 struct object *o;
2017 if (!p->pack_local || p->pack_keep)
2018 continue;
2019 if (open_pack_index(p))
2020 die("cannot open pack index");
2022 ALLOC_GROW(in_pack.array,
2023 in_pack.nr + p->num_objects,
2024 in_pack.alloc);
2026 for (i = 0; i < p->num_objects; i++) {
2027 sha1 = nth_packed_object_sha1(p, i);
2028 o = lookup_unknown_object(sha1);
2029 if (!(o->flags & OBJECT_ADDED))
2030 mark_in_pack_object(o, p, &in_pack);
2031 o->flags |= OBJECT_ADDED;
2032 }
2033 }
2035 if (in_pack.nr) {
2036 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2037 ofscmp);
2038 for (i = 0; i < in_pack.nr; i++) {
2039 struct object *o = in_pack.array[i].object;
2040 add_object_entry(o->sha1, o->type, "", 0);
2041 }
2042 }
2043 free(in_pack.array);
2044 }
2046 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2047 {
2048 static struct packed_git *last_found = (void *)1;
2049 struct packed_git *p;
2051 p = (last_found != (void *)1) ? last_found : packed_git;
2053 while (p) {
2054 if ((!p->pack_local || p->pack_keep) &&
2055 find_pack_entry_one(sha1, p)) {
2056 last_found = p;
2057 return 1;
2058 }
2059 if (p == last_found)
2060 p = packed_git;
2061 else
2062 p = p->next;
2063 if (p == last_found)
2064 p = p->next;
2065 }
2066 return 0;
2067 }
2069 static void loosen_unused_packed_objects(struct rev_info *revs)
2070 {
2071 struct packed_git *p;
2072 uint32_t i;
2073 const unsigned char *sha1;
2075 for (p = packed_git; p; p = p->next) {
2076 if (!p->pack_local || p->pack_keep)
2077 continue;
2079 if (open_pack_index(p))
2080 die("cannot open pack index");
2082 for (i = 0; i < p->num_objects; i++) {
2083 sha1 = nth_packed_object_sha1(p, i);
2084 if (!locate_object_entry(sha1) &&
2085 !has_sha1_pack_kept_or_nonlocal(sha1))
2086 if (force_object_loose(sha1, p->mtime))
2087 die("unable to force loose object");
2088 }
2089 }
2090 }
2092 static void get_object_list(int ac, const char **av)
2093 {
2094 struct rev_info revs;
2095 char line[1000];
2096 int flags = 0;
2098 init_revisions(&revs, NULL);
2099 save_commit_buffer = 0;
2100 setup_revisions(ac, av, &revs, NULL);
2102 while (fgets(line, sizeof(line), stdin) != NULL) {
2103 int len = strlen(line);
2104 if (len && line[len - 1] == '\n')
2105 line[--len] = 0;
2106 if (!len)
2107 break;
2108 if (*line == '-') {
2109 if (!strcmp(line, "--not")) {
2110 flags ^= UNINTERESTING;
2111 continue;
2112 }
2113 die("not a rev '%s'", line);
2114 }
2115 if (handle_revision_arg(line, &revs, flags, 1))
2116 die("bad revision '%s'", line);
2117 }
2119 if (prepare_revision_walk(&revs))
2120 die("revision walk setup failed");
2121 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2122 traverse_commit_list(&revs, show_commit, show_object, NULL);
2124 if (keep_unreachable)
2125 add_objects_in_unpacked_packs(&revs);
2126 if (unpack_unreachable)
2127 loosen_unused_packed_objects(&revs);
2128 }
2130 static int adjust_perm(const char *path, mode_t mode)
2131 {
2132 if (chmod(path, mode))
2133 return -1;
2134 return adjust_shared_perm(path);
2135 }
2137 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2138 {
2139 int use_internal_rev_list = 0;
2140 int thin = 0;
2141 int all_progress_implied = 0;
2142 uint32_t i;
2143 const char **rp_av;
2144 int rp_ac_alloc = 64;
2145 int rp_ac;
2147 read_replace_refs = 0;
2149 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2151 rp_av[0] = "pack-objects";
2152 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2153 rp_ac = 2;
2155 git_config(git_pack_config, NULL);
2156 if (!pack_compression_seen && core_compression_seen)
2157 pack_compression_level = core_compression_level;
2159 progress = isatty(2);
2160 for (i = 1; i < argc; i++) {
2161 const char *arg = argv[i];
2163 if (*arg != '-')
2164 break;
2166 if (!strcmp("--non-empty", arg)) {
2167 non_empty = 1;
2168 continue;
2169 }
2170 if (!strcmp("--local", arg)) {
2171 local = 1;
2172 continue;
2173 }
2174 if (!strcmp("--incremental", arg)) {
2175 incremental = 1;
2176 continue;
2177 }
2178 if (!strcmp("--honor-pack-keep", arg)) {
2179 ignore_packed_keep = 1;
2180 continue;
2181 }
2182 if (!prefixcmp(arg, "--compression=")) {
2183 char *end;
2184 int level = strtoul(arg+14, &end, 0);
2185 if (!arg[14] || *end)
2186 usage(pack_usage);
2187 if (level == -1)
2188 level = Z_DEFAULT_COMPRESSION;
2189 else if (level < 0 || level > Z_BEST_COMPRESSION)
2190 die("bad pack compression level %d", level);
2191 pack_compression_level = level;
2192 continue;
2193 }
2194 if (!prefixcmp(arg, "--max-pack-size=")) {
2195 char *end;
2196 pack_size_limit_cfg = 0;
2197 pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
2198 if (!arg[16] || *end)
2199 usage(pack_usage);
2200 continue;
2201 }
2202 if (!prefixcmp(arg, "--window=")) {
2203 char *end;
2204 window = strtoul(arg+9, &end, 0);
2205 if (!arg[9] || *end)
2206 usage(pack_usage);
2207 continue;
2208 }
2209 if (!prefixcmp(arg, "--window-memory=")) {
2210 if (!git_parse_ulong(arg+16, &window_memory_limit))
2211 usage(pack_usage);
2212 continue;
2213 }
2214 if (!prefixcmp(arg, "--threads=")) {
2215 char *end;
2216 delta_search_threads = strtoul(arg+10, &end, 0);
2217 if (!arg[10] || *end || delta_search_threads < 0)
2218 usage(pack_usage);
2219 #ifdef NO_PTHREADS
2220 if (delta_search_threads != 1)
2221 warning("no threads support, "
2222 "ignoring %s", arg);
2223 #endif
2224 continue;
2225 }
2226 if (!prefixcmp(arg, "--depth=")) {
2227 char *end;
2228 depth = strtoul(arg+8, &end, 0);
2229 if (!arg[8] || *end)
2230 usage(pack_usage);
2231 continue;
2232 }
2233 if (!strcmp("--progress", arg)) {
2234 progress = 1;
2235 continue;
2236 }
2237 if (!strcmp("--all-progress", arg)) {
2238 progress = 2;
2239 continue;
2240 }
2241 if (!strcmp("--all-progress-implied", arg)) {
2242 all_progress_implied = 1;
2243 continue;
2244 }
2245 if (!strcmp("-q", arg)) {
2246 progress = 0;
2247 continue;
2248 }
2249 if (!strcmp("--no-reuse-delta", arg)) {
2250 reuse_delta = 0;
2251 continue;
2252 }
2253 if (!strcmp("--no-reuse-object", arg)) {
2254 reuse_object = reuse_delta = 0;
2255 continue;
2256 }
2257 if (!strcmp("--delta-base-offset", arg)) {
2258 allow_ofs_delta = 1;
2259 continue;
2260 }
2261 if (!strcmp("--stdout", arg)) {
2262 pack_to_stdout = 1;
2263 continue;
2264 }
2265 if (!strcmp("--revs", arg)) {
2266 use_internal_rev_list = 1;
2267 continue;
2268 }
2269 if (!strcmp("--keep-unreachable", arg)) {
2270 keep_unreachable = 1;
2271 continue;
2272 }
2273 if (!strcmp("--unpack-unreachable", arg)) {
2274 unpack_unreachable = 1;
2275 continue;
2276 }
2277 if (!strcmp("--include-tag", arg)) {
2278 include_tag = 1;
2279 continue;
2280 }
2281 if (!strcmp("--unpacked", arg) ||
2282 !strcmp("--reflog", arg) ||
2283 !strcmp("--all", arg)) {
2284 use_internal_rev_list = 1;
2285 if (rp_ac >= rp_ac_alloc - 1) {
2286 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2287 rp_av = xrealloc(rp_av,
2288 rp_ac_alloc * sizeof(*rp_av));
2289 }
2290 rp_av[rp_ac++] = arg;
2291 continue;
2292 }
2293 if (!strcmp("--thin", arg)) {
2294 use_internal_rev_list = 1;
2295 thin = 1;
2296 rp_av[1] = "--objects-edge";
2297 continue;
2298 }
2299 if (!prefixcmp(arg, "--index-version=")) {
2300 char *c;
2301 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2302 if (pack_idx_default_version > 2)
2303 die("bad %s", arg);
2304 if (*c == ',')
2305 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2306 if (*c || pack_idx_off32_limit & 0x80000000)
2307 die("bad %s", arg);
2308 continue;
2309 }
2310 if (!strcmp(arg, "--keep-true-parents")) {
2311 grafts_replace_parents = 0;
2312 continue;
2313 }
2314 usage(pack_usage);
2315 }
2317 /* Traditionally "pack-objects [options] base extra" failed;
2318 * we would however want to take refs parameter that would
2319 * have been given to upstream rev-list ourselves, which means
2320 * we somehow want to say what the base name is. So the
2321 * syntax would be:
2322 *
2323 * pack-objects [options] base <refs...>
2324 *
2325 * in other words, we would treat the first non-option as the
2326 * base_name and send everything else to the internal revision
2327 * walker.
2328 */
2330 if (!pack_to_stdout)
2331 base_name = argv[i++];
2333 if (pack_to_stdout != !base_name)
2334 usage(pack_usage);
2336 if (!pack_to_stdout && !pack_size_limit)
2337 pack_size_limit = pack_size_limit_cfg;
2339 if (pack_to_stdout && pack_size_limit)
2340 die("--max-pack-size cannot be used to build a pack for transfer.");
2342 if (!pack_to_stdout && thin)
2343 die("--thin cannot be used to build an indexable pack.");
2345 if (keep_unreachable && unpack_unreachable)
2346 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2348 if (progress && all_progress_implied)
2349 progress = 2;
2351 prepare_packed_git();
2353 if (progress)
2354 progress_state = start_progress("Counting objects", 0);
2355 if (!use_internal_rev_list)
2356 read_object_list_from_stdin();
2357 else {
2358 rp_av[rp_ac] = NULL;
2359 get_object_list(rp_ac, rp_av);
2360 }
2361 cleanup_preferred_base();
2362 if (include_tag && nr_result)
2363 for_each_ref(add_ref_tag, NULL);
2364 stop_progress(&progress_state);
2366 if (non_empty && !nr_result)
2367 return 0;
2368 if (nr_result)
2369 prepare_pack(window, depth);
2370 write_pack_file();
2371 if (progress)
2372 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2373 " reused %"PRIu32" (delta %"PRIu32")\n",
2374 written, written_delta, reused, reused_delta);
2375 return 0;
2376 }