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 unsigned long 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 /* if we are deltified, write out base object first. */
449 if (e->delta && !write_one(f, e->delta, offset))
450 return 0;
452 e->idx.offset = *offset;
453 size = write_object(f, e, *offset);
454 if (!size) {
455 e->idx.offset = 0;
456 return 0;
457 }
458 written_list[nr_written++] = &e->idx;
460 /* make sure off_t is sufficiently large not to wrap */
461 if (*offset > *offset + size)
462 die("pack too large for current definition of off_t");
463 *offset += size;
464 return 1;
465 }
467 /* forward declaration for write_pack_file */
468 static int adjust_perm(const char *path, mode_t mode);
470 static void write_pack_file(void)
471 {
472 uint32_t i = 0, j;
473 struct sha1file *f;
474 off_t offset;
475 struct pack_header hdr;
476 uint32_t nr_remaining = nr_result;
477 time_t last_mtime = 0;
479 if (progress > pack_to_stdout)
480 progress_state = start_progress("Writing objects", nr_result);
481 written_list = xmalloc(nr_objects * sizeof(*written_list));
483 do {
484 unsigned char sha1[20];
485 char *pack_tmp_name = NULL;
487 if (pack_to_stdout) {
488 f = sha1fd_throughput(1, "<stdout>", progress_state);
489 } else {
490 char tmpname[PATH_MAX];
491 int fd;
492 fd = odb_mkstemp(tmpname, sizeof(tmpname),
493 "pack/tmp_pack_XXXXXX");
494 pack_tmp_name = xstrdup(tmpname);
495 f = sha1fd(fd, pack_tmp_name);
496 }
498 hdr.hdr_signature = htonl(PACK_SIGNATURE);
499 hdr.hdr_version = htonl(PACK_VERSION);
500 hdr.hdr_entries = htonl(nr_remaining);
501 sha1write(f, &hdr, sizeof(hdr));
502 offset = sizeof(hdr);
503 nr_written = 0;
504 for (; i < nr_objects; i++) {
505 if (!write_one(f, objects + i, &offset))
506 break;
507 display_progress(progress_state, written);
508 }
510 /*
511 * Did we write the wrong # entries in the header?
512 * If so, rewrite it like in fast-import
513 */
514 if (pack_to_stdout) {
515 sha1close(f, sha1, CSUM_CLOSE);
516 } else if (nr_written == nr_remaining) {
517 sha1close(f, sha1, CSUM_FSYNC);
518 } else {
519 int fd = sha1close(f, sha1, 0);
520 fixup_pack_header_footer(fd, sha1, pack_tmp_name,
521 nr_written, sha1, offset);
522 close(fd);
523 }
525 if (!pack_to_stdout) {
526 mode_t mode = umask(0);
527 struct stat st;
528 const char *idx_tmp_name;
529 char tmpname[PATH_MAX];
531 umask(mode);
532 mode = 0444 & ~mode;
534 idx_tmp_name = write_idx_file(NULL, written_list,
535 nr_written, sha1);
537 snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
538 base_name, sha1_to_hex(sha1));
539 free_pack_by_name(tmpname);
540 if (adjust_perm(pack_tmp_name, mode))
541 die_errno("unable to make temporary pack file readable");
542 if (rename(pack_tmp_name, tmpname))
543 die_errno("unable to rename temporary pack file");
545 /*
546 * Packs are runtime accessed in their mtime
547 * order since newer packs are more likely to contain
548 * younger objects. So if we are creating multiple
549 * packs then we should modify the mtime of later ones
550 * to preserve this property.
551 */
552 if (stat(tmpname, &st) < 0) {
553 warning("failed to stat %s: %s",
554 tmpname, strerror(errno));
555 } else if (!last_mtime) {
556 last_mtime = st.st_mtime;
557 } else {
558 struct utimbuf utb;
559 utb.actime = st.st_atime;
560 utb.modtime = --last_mtime;
561 if (utime(tmpname, &utb) < 0)
562 warning("failed utime() on %s: %s",
563 tmpname, strerror(errno));
564 }
566 snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
567 base_name, sha1_to_hex(sha1));
568 if (adjust_perm(idx_tmp_name, mode))
569 die_errno("unable to make temporary index file readable");
570 if (rename(idx_tmp_name, tmpname))
571 die_errno("unable to rename temporary index file");
573 free((void *) idx_tmp_name);
574 free(pack_tmp_name);
575 puts(sha1_to_hex(sha1));
576 }
578 /* mark written objects as written to previous pack */
579 for (j = 0; j < nr_written; j++) {
580 written_list[j]->offset = (off_t)-1;
581 }
582 nr_remaining -= nr_written;
583 } while (nr_remaining && i < nr_objects);
585 free(written_list);
586 stop_progress(&progress_state);
587 if (written != nr_result)
588 die("wrote %"PRIu32" objects while expecting %"PRIu32,
589 written, nr_result);
590 }
592 static int locate_object_entry_hash(const unsigned char *sha1)
593 {
594 int i;
595 unsigned int ui;
596 memcpy(&ui, sha1, sizeof(unsigned int));
597 i = ui % object_ix_hashsz;
598 while (0 < object_ix[i]) {
599 if (!hashcmp(sha1, objects[object_ix[i] - 1].idx.sha1))
600 return i;
601 if (++i == object_ix_hashsz)
602 i = 0;
603 }
604 return -1 - i;
605 }
607 static struct object_entry *locate_object_entry(const unsigned char *sha1)
608 {
609 int i;
611 if (!object_ix_hashsz)
612 return NULL;
614 i = locate_object_entry_hash(sha1);
615 if (0 <= i)
616 return &objects[object_ix[i]-1];
617 return NULL;
618 }
620 static void rehash_objects(void)
621 {
622 uint32_t i;
623 struct object_entry *oe;
625 object_ix_hashsz = nr_objects * 3;
626 if (object_ix_hashsz < 1024)
627 object_ix_hashsz = 1024;
628 object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
629 memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
630 for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
631 int ix = locate_object_entry_hash(oe->idx.sha1);
632 if (0 <= ix)
633 continue;
634 ix = -1 - ix;
635 object_ix[ix] = i + 1;
636 }
637 }
639 static unsigned name_hash(const char *name)
640 {
641 unsigned c, hash = 0;
643 if (!name)
644 return 0;
646 /*
647 * This effectively just creates a sortable number from the
648 * last sixteen non-whitespace characters. Last characters
649 * count "most", so things that end in ".c" sort together.
650 */
651 while ((c = *name++) != 0) {
652 if (isspace(c))
653 continue;
654 hash = (hash >> 2) + (c << 24);
655 }
656 return hash;
657 }
659 static void setup_delta_attr_check(struct git_attr_check *check)
660 {
661 static struct git_attr *attr_delta;
663 if (!attr_delta)
664 attr_delta = git_attr("delta");
666 check[0].attr = attr_delta;
667 }
669 static int no_try_delta(const char *path)
670 {
671 struct git_attr_check check[1];
673 setup_delta_attr_check(check);
674 if (git_checkattr(path, ARRAY_SIZE(check), check))
675 return 0;
676 if (ATTR_FALSE(check->value))
677 return 1;
678 return 0;
679 }
681 static int add_object_entry(const unsigned char *sha1, enum object_type type,
682 const char *name, int exclude)
683 {
684 struct object_entry *entry;
685 struct packed_git *p, *found_pack = NULL;
686 off_t found_offset = 0;
687 int ix;
688 unsigned hash = name_hash(name);
690 ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
691 if (ix >= 0) {
692 if (exclude) {
693 entry = objects + object_ix[ix] - 1;
694 if (!entry->preferred_base)
695 nr_result--;
696 entry->preferred_base = 1;
697 }
698 return 0;
699 }
701 if (!exclude && local && has_loose_object_nonlocal(sha1))
702 return 0;
704 for (p = packed_git; p; p = p->next) {
705 off_t offset = find_pack_entry_one(sha1, p);
706 if (offset) {
707 if (!found_pack) {
708 found_offset = offset;
709 found_pack = p;
710 }
711 if (exclude)
712 break;
713 if (incremental)
714 return 0;
715 if (local && !p->pack_local)
716 return 0;
717 if (ignore_packed_keep && p->pack_local && p->pack_keep)
718 return 0;
719 }
720 }
722 if (nr_objects >= nr_alloc) {
723 nr_alloc = (nr_alloc + 1024) * 3 / 2;
724 objects = xrealloc(objects, nr_alloc * sizeof(*entry));
725 }
727 entry = objects + nr_objects++;
728 memset(entry, 0, sizeof(*entry));
729 hashcpy(entry->idx.sha1, sha1);
730 entry->hash = hash;
731 if (type)
732 entry->type = type;
733 if (exclude)
734 entry->preferred_base = 1;
735 else
736 nr_result++;
737 if (found_pack) {
738 entry->in_pack = found_pack;
739 entry->in_pack_offset = found_offset;
740 }
742 if (object_ix_hashsz * 3 <= nr_objects * 4)
743 rehash_objects();
744 else
745 object_ix[-1 - ix] = nr_objects;
747 display_progress(progress_state, nr_objects);
749 if (name && no_try_delta(name))
750 entry->no_try_delta = 1;
752 return 1;
753 }
755 struct pbase_tree_cache {
756 unsigned char sha1[20];
757 int ref;
758 int temporary;
759 void *tree_data;
760 unsigned long tree_size;
761 };
763 static struct pbase_tree_cache *(pbase_tree_cache[256]);
764 static int pbase_tree_cache_ix(const unsigned char *sha1)
765 {
766 return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
767 }
768 static int pbase_tree_cache_ix_incr(int ix)
769 {
770 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
771 }
773 static struct pbase_tree {
774 struct pbase_tree *next;
775 /* This is a phony "cache" entry; we are not
776 * going to evict it nor find it through _get()
777 * mechanism -- this is for the toplevel node that
778 * would almost always change with any commit.
779 */
780 struct pbase_tree_cache pcache;
781 } *pbase_tree;
783 static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
784 {
785 struct pbase_tree_cache *ent, *nent;
786 void *data;
787 unsigned long size;
788 enum object_type type;
789 int neigh;
790 int my_ix = pbase_tree_cache_ix(sha1);
791 int available_ix = -1;
793 /* pbase-tree-cache acts as a limited hashtable.
794 * your object will be found at your index or within a few
795 * slots after that slot if it is cached.
796 */
797 for (neigh = 0; neigh < 8; neigh++) {
798 ent = pbase_tree_cache[my_ix];
799 if (ent && !hashcmp(ent->sha1, sha1)) {
800 ent->ref++;
801 return ent;
802 }
803 else if (((available_ix < 0) && (!ent || !ent->ref)) ||
804 ((0 <= available_ix) &&
805 (!ent && pbase_tree_cache[available_ix])))
806 available_ix = my_ix;
807 if (!ent)
808 break;
809 my_ix = pbase_tree_cache_ix_incr(my_ix);
810 }
812 /* Did not find one. Either we got a bogus request or
813 * we need to read and perhaps cache.
814 */
815 data = read_sha1_file(sha1, &type, &size);
816 if (!data)
817 return NULL;
818 if (type != OBJ_TREE) {
819 free(data);
820 return NULL;
821 }
823 /* We need to either cache or return a throwaway copy */
825 if (available_ix < 0)
826 ent = NULL;
827 else {
828 ent = pbase_tree_cache[available_ix];
829 my_ix = available_ix;
830 }
832 if (!ent) {
833 nent = xmalloc(sizeof(*nent));
834 nent->temporary = (available_ix < 0);
835 }
836 else {
837 /* evict and reuse */
838 free(ent->tree_data);
839 nent = ent;
840 }
841 hashcpy(nent->sha1, sha1);
842 nent->tree_data = data;
843 nent->tree_size = size;
844 nent->ref = 1;
845 if (!nent->temporary)
846 pbase_tree_cache[my_ix] = nent;
847 return nent;
848 }
850 static void pbase_tree_put(struct pbase_tree_cache *cache)
851 {
852 if (!cache->temporary) {
853 cache->ref--;
854 return;
855 }
856 free(cache->tree_data);
857 free(cache);
858 }
860 static int name_cmp_len(const char *name)
861 {
862 int i;
863 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
864 ;
865 return i;
866 }
868 static void add_pbase_object(struct tree_desc *tree,
869 const char *name,
870 int cmplen,
871 const char *fullname)
872 {
873 struct name_entry entry;
874 int cmp;
876 while (tree_entry(tree,&entry)) {
877 if (S_ISGITLINK(entry.mode))
878 continue;
879 cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
880 memcmp(name, entry.path, cmplen);
881 if (cmp > 0)
882 continue;
883 if (cmp < 0)
884 return;
885 if (name[cmplen] != '/') {
886 add_object_entry(entry.sha1,
887 object_type(entry.mode),
888 fullname, 1);
889 return;
890 }
891 if (S_ISDIR(entry.mode)) {
892 struct tree_desc sub;
893 struct pbase_tree_cache *tree;
894 const char *down = name+cmplen+1;
895 int downlen = name_cmp_len(down);
897 tree = pbase_tree_get(entry.sha1);
898 if (!tree)
899 return;
900 init_tree_desc(&sub, tree->tree_data, tree->tree_size);
902 add_pbase_object(&sub, down, downlen, fullname);
903 pbase_tree_put(tree);
904 }
905 }
906 }
908 static unsigned *done_pbase_paths;
909 static int done_pbase_paths_num;
910 static int done_pbase_paths_alloc;
911 static int done_pbase_path_pos(unsigned hash)
912 {
913 int lo = 0;
914 int hi = done_pbase_paths_num;
915 while (lo < hi) {
916 int mi = (hi + lo) / 2;
917 if (done_pbase_paths[mi] == hash)
918 return mi;
919 if (done_pbase_paths[mi] < hash)
920 hi = mi;
921 else
922 lo = mi + 1;
923 }
924 return -lo-1;
925 }
927 static int check_pbase_path(unsigned hash)
928 {
929 int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
930 if (0 <= pos)
931 return 1;
932 pos = -pos - 1;
933 if (done_pbase_paths_alloc <= done_pbase_paths_num) {
934 done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
935 done_pbase_paths = xrealloc(done_pbase_paths,
936 done_pbase_paths_alloc *
937 sizeof(unsigned));
938 }
939 done_pbase_paths_num++;
940 if (pos < done_pbase_paths_num)
941 memmove(done_pbase_paths + pos + 1,
942 done_pbase_paths + pos,
943 (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
944 done_pbase_paths[pos] = hash;
945 return 0;
946 }
948 static void add_preferred_base_object(const char *name)
949 {
950 struct pbase_tree *it;
951 int cmplen;
952 unsigned hash = name_hash(name);
954 if (!num_preferred_base || check_pbase_path(hash))
955 return;
957 cmplen = name_cmp_len(name);
958 for (it = pbase_tree; it; it = it->next) {
959 if (cmplen == 0) {
960 add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
961 }
962 else {
963 struct tree_desc tree;
964 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
965 add_pbase_object(&tree, name, cmplen, name);
966 }
967 }
968 }
970 static void add_preferred_base(unsigned char *sha1)
971 {
972 struct pbase_tree *it;
973 void *data;
974 unsigned long size;
975 unsigned char tree_sha1[20];
977 if (window <= num_preferred_base++)
978 return;
980 data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
981 if (!data)
982 return;
984 for (it = pbase_tree; it; it = it->next) {
985 if (!hashcmp(it->pcache.sha1, tree_sha1)) {
986 free(data);
987 return;
988 }
989 }
991 it = xcalloc(1, sizeof(*it));
992 it->next = pbase_tree;
993 pbase_tree = it;
995 hashcpy(it->pcache.sha1, tree_sha1);
996 it->pcache.tree_data = data;
997 it->pcache.tree_size = size;
998 }
1000 static void cleanup_preferred_base(void)
1001 {
1002 struct pbase_tree *it;
1003 unsigned i;
1005 it = pbase_tree;
1006 pbase_tree = NULL;
1007 while (it) {
1008 struct pbase_tree *this = it;
1009 it = this->next;
1010 free(this->pcache.tree_data);
1011 free(this);
1012 }
1014 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1015 if (!pbase_tree_cache[i])
1016 continue;
1017 free(pbase_tree_cache[i]->tree_data);
1018 free(pbase_tree_cache[i]);
1019 pbase_tree_cache[i] = NULL;
1020 }
1022 free(done_pbase_paths);
1023 done_pbase_paths = NULL;
1024 done_pbase_paths_num = done_pbase_paths_alloc = 0;
1025 }
1027 static void check_object(struct object_entry *entry)
1028 {
1029 if (entry->in_pack) {
1030 struct packed_git *p = entry->in_pack;
1031 struct pack_window *w_curs = NULL;
1032 const unsigned char *base_ref = NULL;
1033 struct object_entry *base_entry;
1034 unsigned long used, used_0;
1035 unsigned int avail;
1036 off_t ofs;
1037 unsigned char *buf, c;
1039 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1041 /*
1042 * We want in_pack_type even if we do not reuse delta
1043 * since non-delta representations could still be reused.
1044 */
1045 used = unpack_object_header_buffer(buf, avail,
1046 &entry->in_pack_type,
1047 &entry->size);
1048 if (used == 0)
1049 goto give_up;
1051 /*
1052 * Determine if this is a delta and if so whether we can
1053 * reuse it or not. Otherwise let's find out as cheaply as
1054 * possible what the actual type and size for this object is.
1055 */
1056 switch (entry->in_pack_type) {
1057 default:
1058 /* Not a delta hence we've already got all we need. */
1059 entry->type = entry->in_pack_type;
1060 entry->in_pack_header_size = used;
1061 if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1062 goto give_up;
1063 unuse_pack(&w_curs);
1064 return;
1065 case OBJ_REF_DELTA:
1066 if (reuse_delta && !entry->preferred_base)
1067 base_ref = use_pack(p, &w_curs,
1068 entry->in_pack_offset + used, NULL);
1069 entry->in_pack_header_size = used + 20;
1070 break;
1071 case OBJ_OFS_DELTA:
1072 buf = use_pack(p, &w_curs,
1073 entry->in_pack_offset + used, NULL);
1074 used_0 = 0;
1075 c = buf[used_0++];
1076 ofs = c & 127;
1077 while (c & 128) {
1078 ofs += 1;
1079 if (!ofs || MSB(ofs, 7)) {
1080 error("delta base offset overflow in pack for %s",
1081 sha1_to_hex(entry->idx.sha1));
1082 goto give_up;
1083 }
1084 c = buf[used_0++];
1085 ofs = (ofs << 7) + (c & 127);
1086 }
1087 ofs = entry->in_pack_offset - ofs;
1088 if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1089 error("delta base offset out of bound for %s",
1090 sha1_to_hex(entry->idx.sha1));
1091 goto give_up;
1092 }
1093 if (reuse_delta && !entry->preferred_base) {
1094 struct revindex_entry *revidx;
1095 revidx = find_pack_revindex(p, ofs);
1096 if (!revidx)
1097 goto give_up;
1098 base_ref = nth_packed_object_sha1(p, revidx->nr);
1099 }
1100 entry->in_pack_header_size = used + used_0;
1101 break;
1102 }
1104 if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1105 /*
1106 * If base_ref was set above that means we wish to
1107 * reuse delta data, and we even found that base
1108 * in the list of objects we want to pack. Goodie!
1109 *
1110 * Depth value does not matter - find_deltas() will
1111 * never consider reused delta as the base object to
1112 * deltify other objects against, in order to avoid
1113 * circular deltas.
1114 */
1115 entry->type = entry->in_pack_type;
1116 entry->delta = base_entry;
1117 entry->delta_size = entry->size;
1118 entry->delta_sibling = base_entry->delta_child;
1119 base_entry->delta_child = entry;
1120 unuse_pack(&w_curs);
1121 return;
1122 }
1124 if (entry->type) {
1125 /*
1126 * This must be a delta and we already know what the
1127 * final object type is. Let's extract the actual
1128 * object size from the delta header.
1129 */
1130 entry->size = get_size_from_delta(p, &w_curs,
1131 entry->in_pack_offset + entry->in_pack_header_size);
1132 if (entry->size == 0)
1133 goto give_up;
1134 unuse_pack(&w_curs);
1135 return;
1136 }
1138 /*
1139 * No choice but to fall back to the recursive delta walk
1140 * with sha1_object_info() to find about the object type
1141 * at this point...
1142 */
1143 give_up:
1144 unuse_pack(&w_curs);
1145 }
1147 entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1148 /*
1149 * The error condition is checked in prepare_pack(). This is
1150 * to permit a missing preferred base object to be ignored
1151 * as a preferred base. Doing so can result in a larger
1152 * pack file, but the transfer will still take place.
1153 */
1154 }
1156 static int pack_offset_sort(const void *_a, const void *_b)
1157 {
1158 const struct object_entry *a = *(struct object_entry **)_a;
1159 const struct object_entry *b = *(struct object_entry **)_b;
1161 /* avoid filesystem trashing with loose objects */
1162 if (!a->in_pack && !b->in_pack)
1163 return hashcmp(a->idx.sha1, b->idx.sha1);
1165 if (a->in_pack < b->in_pack)
1166 return -1;
1167 if (a->in_pack > b->in_pack)
1168 return 1;
1169 return a->in_pack_offset < b->in_pack_offset ? -1 :
1170 (a->in_pack_offset > b->in_pack_offset);
1171 }
1173 static void get_object_details(void)
1174 {
1175 uint32_t i;
1176 struct object_entry **sorted_by_offset;
1178 sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1179 for (i = 0; i < nr_objects; i++)
1180 sorted_by_offset[i] = objects + i;
1181 qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1183 for (i = 0; i < nr_objects; i++)
1184 check_object(sorted_by_offset[i]);
1186 free(sorted_by_offset);
1187 }
1189 /*
1190 * We search for deltas in a list sorted by type, by filename hash, and then
1191 * by size, so that we see progressively smaller and smaller files.
1192 * That's because we prefer deltas to be from the bigger file
1193 * to the smaller -- deletes are potentially cheaper, but perhaps
1194 * more importantly, the bigger file is likely the more recent
1195 * one. The deepest deltas are therefore the oldest objects which are
1196 * less susceptible to be accessed often.
1197 */
1198 static int type_size_sort(const void *_a, const void *_b)
1199 {
1200 const struct object_entry *a = *(struct object_entry **)_a;
1201 const struct object_entry *b = *(struct object_entry **)_b;
1203 if (a->type > b->type)
1204 return -1;
1205 if (a->type < b->type)
1206 return 1;
1207 if (a->hash > b->hash)
1208 return -1;
1209 if (a->hash < b->hash)
1210 return 1;
1211 if (a->preferred_base > b->preferred_base)
1212 return -1;
1213 if (a->preferred_base < b->preferred_base)
1214 return 1;
1215 if (a->size > b->size)
1216 return -1;
1217 if (a->size < b->size)
1218 return 1;
1219 return a < b ? -1 : (a > b); /* newest first */
1220 }
1222 struct unpacked {
1223 struct object_entry *entry;
1224 void *data;
1225 struct delta_index *index;
1226 unsigned depth;
1227 };
1229 static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1230 unsigned long delta_size)
1231 {
1232 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1233 return 0;
1235 if (delta_size < cache_max_small_delta_size)
1236 return 1;
1238 /* cache delta, if objects are large enough compared to delta size */
1239 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1240 return 1;
1242 return 0;
1243 }
1245 #ifndef NO_PTHREADS
1247 static pthread_mutex_t read_mutex;
1248 #define read_lock() pthread_mutex_lock(&read_mutex)
1249 #define read_unlock() pthread_mutex_unlock(&read_mutex)
1251 static pthread_mutex_t cache_mutex;
1252 #define cache_lock() pthread_mutex_lock(&cache_mutex)
1253 #define cache_unlock() pthread_mutex_unlock(&cache_mutex)
1255 static pthread_mutex_t progress_mutex;
1256 #define progress_lock() pthread_mutex_lock(&progress_mutex)
1257 #define progress_unlock() pthread_mutex_unlock(&progress_mutex)
1259 #else
1261 #define read_lock() (void)0
1262 #define read_unlock() (void)0
1263 #define cache_lock() (void)0
1264 #define cache_unlock() (void)0
1265 #define progress_lock() (void)0
1266 #define progress_unlock() (void)0
1268 #endif
1270 static int try_delta(struct unpacked *trg, struct unpacked *src,
1271 unsigned max_depth, unsigned long *mem_usage)
1272 {
1273 struct object_entry *trg_entry = trg->entry;
1274 struct object_entry *src_entry = src->entry;
1275 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1276 unsigned ref_depth;
1277 enum object_type type;
1278 void *delta_buf;
1280 /* Don't bother doing diffs between different types */
1281 if (trg_entry->type != src_entry->type)
1282 return -1;
1284 /*
1285 * We do not bother to try a delta that we discarded
1286 * on an earlier try, but only when reusing delta data.
1287 */
1288 if (reuse_delta && trg_entry->in_pack &&
1289 trg_entry->in_pack == src_entry->in_pack &&
1290 trg_entry->in_pack_type != OBJ_REF_DELTA &&
1291 trg_entry->in_pack_type != OBJ_OFS_DELTA)
1292 return 0;
1294 /* Let's not bust the allowed depth. */
1295 if (src->depth >= max_depth)
1296 return 0;
1298 /* Now some size filtering heuristics. */
1299 trg_size = trg_entry->size;
1300 if (!trg_entry->delta) {
1301 max_size = trg_size/2 - 20;
1302 ref_depth = 1;
1303 } else {
1304 max_size = trg_entry->delta_size;
1305 ref_depth = trg->depth;
1306 }
1307 max_size = (uint64_t)max_size * (max_depth - src->depth) /
1308 (max_depth - ref_depth + 1);
1309 if (max_size == 0)
1310 return 0;
1311 src_size = src_entry->size;
1312 sizediff = src_size < trg_size ? trg_size - src_size : 0;
1313 if (sizediff >= max_size)
1314 return 0;
1315 if (trg_size < src_size / 32)
1316 return 0;
1318 /* Load data if not already done */
1319 if (!trg->data) {
1320 read_lock();
1321 trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1322 read_unlock();
1323 if (!trg->data)
1324 die("object %s cannot be read",
1325 sha1_to_hex(trg_entry->idx.sha1));
1326 if (sz != trg_size)
1327 die("object %s inconsistent object length (%lu vs %lu)",
1328 sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1329 *mem_usage += sz;
1330 }
1331 if (!src->data) {
1332 read_lock();
1333 src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1334 read_unlock();
1335 if (!src->data)
1336 die("object %s cannot be read",
1337 sha1_to_hex(src_entry->idx.sha1));
1338 if (sz != src_size)
1339 die("object %s inconsistent object length (%lu vs %lu)",
1340 sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1341 *mem_usage += sz;
1342 }
1343 if (!src->index) {
1344 src->index = create_delta_index(src->data, src_size);
1345 if (!src->index) {
1346 static int warned = 0;
1347 if (!warned++)
1348 warning("suboptimal pack - out of memory");
1349 return 0;
1350 }
1351 *mem_usage += sizeof_delta_index(src->index);
1352 }
1354 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1355 if (!delta_buf)
1356 return 0;
1358 if (trg_entry->delta) {
1359 /* Prefer only shallower same-sized deltas. */
1360 if (delta_size == trg_entry->delta_size &&
1361 src->depth + 1 >= trg->depth) {
1362 free(delta_buf);
1363 return 0;
1364 }
1365 }
1367 /*
1368 * Handle memory allocation outside of the cache
1369 * accounting lock. Compiler will optimize the strangeness
1370 * away when NO_PTHREADS is defined.
1371 */
1372 free(trg_entry->delta_data);
1373 cache_lock();
1374 if (trg_entry->delta_data) {
1375 delta_cache_size -= trg_entry->delta_size;
1376 trg_entry->delta_data = NULL;
1377 }
1378 if (delta_cacheable(src_size, trg_size, delta_size)) {
1379 delta_cache_size += delta_size;
1380 cache_unlock();
1381 trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1382 } else {
1383 cache_unlock();
1384 free(delta_buf);
1385 }
1387 trg_entry->delta = src_entry;
1388 trg_entry->delta_size = delta_size;
1389 trg->depth = src->depth + 1;
1391 return 1;
1392 }
1394 static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1395 {
1396 struct object_entry *child = me->delta_child;
1397 unsigned int m = n;
1398 while (child) {
1399 unsigned int c = check_delta_limit(child, n + 1);
1400 if (m < c)
1401 m = c;
1402 child = child->delta_sibling;
1403 }
1404 return m;
1405 }
1407 static unsigned long free_unpacked(struct unpacked *n)
1408 {
1409 unsigned long freed_mem = sizeof_delta_index(n->index);
1410 free_delta_index(n->index);
1411 n->index = NULL;
1412 if (n->data) {
1413 freed_mem += n->entry->size;
1414 free(n->data);
1415 n->data = NULL;
1416 }
1417 n->entry = NULL;
1418 n->depth = 0;
1419 return freed_mem;
1420 }
1422 static void find_deltas(struct object_entry **list, unsigned *list_size,
1423 int window, int depth, unsigned *processed)
1424 {
1425 uint32_t i, idx = 0, count = 0;
1426 struct unpacked *array;
1427 unsigned long mem_usage = 0;
1429 array = xcalloc(window, sizeof(struct unpacked));
1431 for (;;) {
1432 struct object_entry *entry;
1433 struct unpacked *n = array + idx;
1434 int j, max_depth, best_base = -1;
1436 progress_lock();
1437 if (!*list_size) {
1438 progress_unlock();
1439 break;
1440 }
1441 entry = *list++;
1442 (*list_size)--;
1443 if (!entry->preferred_base) {
1444 (*processed)++;
1445 display_progress(progress_state, *processed);
1446 }
1447 progress_unlock();
1449 mem_usage -= free_unpacked(n);
1450 n->entry = entry;
1452 while (window_memory_limit &&
1453 mem_usage > window_memory_limit &&
1454 count > 1) {
1455 uint32_t tail = (idx + window - count) % window;
1456 mem_usage -= free_unpacked(array + tail);
1457 count--;
1458 }
1460 /* We do not compute delta to *create* objects we are not
1461 * going to pack.
1462 */
1463 if (entry->preferred_base)
1464 goto next;
1466 /*
1467 * If the current object is at pack edge, take the depth the
1468 * objects that depend on the current object into account
1469 * otherwise they would become too deep.
1470 */
1471 max_depth = depth;
1472 if (entry->delta_child) {
1473 max_depth -= check_delta_limit(entry, 0);
1474 if (max_depth <= 0)
1475 goto next;
1476 }
1478 j = window;
1479 while (--j > 0) {
1480 int ret;
1481 uint32_t other_idx = idx + j;
1482 struct unpacked *m;
1483 if (other_idx >= window)
1484 other_idx -= window;
1485 m = array + other_idx;
1486 if (!m->entry)
1487 break;
1488 ret = try_delta(n, m, max_depth, &mem_usage);
1489 if (ret < 0)
1490 break;
1491 else if (ret > 0)
1492 best_base = other_idx;
1493 }
1495 /*
1496 * If we decided to cache the delta data, then it is best
1497 * to compress it right away. First because we have to do
1498 * it anyway, and doing it here while we're threaded will
1499 * save a lot of time in the non threaded write phase,
1500 * as well as allow for caching more deltas within
1501 * the same cache size limit.
1502 * ...
1503 * But only if not writing to stdout, since in that case
1504 * the network is most likely throttling writes anyway,
1505 * and therefore it is best to go to the write phase ASAP
1506 * instead, as we can afford spending more time compressing
1507 * between writes at that moment.
1508 */
1509 if (entry->delta_data && !pack_to_stdout) {
1510 entry->z_delta_size = do_compress(&entry->delta_data,
1511 entry->delta_size);
1512 cache_lock();
1513 delta_cache_size -= entry->delta_size;
1514 delta_cache_size += entry->z_delta_size;
1515 cache_unlock();
1516 }
1518 /* if we made n a delta, and if n is already at max
1519 * depth, leaving it in the window is pointless. we
1520 * should evict it first.
1521 */
1522 if (entry->delta && max_depth <= n->depth)
1523 continue;
1525 /*
1526 * Move the best delta base up in the window, after the
1527 * currently deltified object, to keep it longer. It will
1528 * be the first base object to be attempted next.
1529 */
1530 if (entry->delta) {
1531 struct unpacked swap = array[best_base];
1532 int dist = (window + idx - best_base) % window;
1533 int dst = best_base;
1534 while (dist--) {
1535 int src = (dst + 1) % window;
1536 array[dst] = array[src];
1537 dst = src;
1538 }
1539 array[dst] = swap;
1540 }
1542 next:
1543 idx++;
1544 if (count + 1 < window)
1545 count++;
1546 if (idx >= window)
1547 idx = 0;
1548 }
1550 for (i = 0; i < window; ++i) {
1551 free_delta_index(array[i].index);
1552 free(array[i].data);
1553 }
1554 free(array);
1555 }
1557 #ifndef NO_PTHREADS
1559 /*
1560 * The main thread waits on the condition that (at least) one of the workers
1561 * has stopped working (which is indicated in the .working member of
1562 * struct thread_params).
1563 * When a work thread has completed its work, it sets .working to 0 and
1564 * signals the main thread and waits on the condition that .data_ready
1565 * becomes 1.
1566 */
1568 struct thread_params {
1569 pthread_t thread;
1570 struct object_entry **list;
1571 unsigned list_size;
1572 unsigned remaining;
1573 int window;
1574 int depth;
1575 int working;
1576 int data_ready;
1577 pthread_mutex_t mutex;
1578 pthread_cond_t cond;
1579 unsigned *processed;
1580 };
1582 static pthread_cond_t progress_cond;
1584 /*
1585 * Mutex and conditional variable can't be statically-initialized on Windows.
1586 */
1587 static void init_threaded_search(void)
1588 {
1589 pthread_mutex_init(&read_mutex, NULL);
1590 pthread_mutex_init(&cache_mutex, NULL);
1591 pthread_mutex_init(&progress_mutex, NULL);
1592 pthread_cond_init(&progress_cond, NULL);
1593 }
1595 static void cleanup_threaded_search(void)
1596 {
1597 pthread_cond_destroy(&progress_cond);
1598 pthread_mutex_destroy(&read_mutex);
1599 pthread_mutex_destroy(&cache_mutex);
1600 pthread_mutex_destroy(&progress_mutex);
1601 }
1603 static void *threaded_find_deltas(void *arg)
1604 {
1605 struct thread_params *me = arg;
1607 while (me->remaining) {
1608 find_deltas(me->list, &me->remaining,
1609 me->window, me->depth, me->processed);
1611 progress_lock();
1612 me->working = 0;
1613 pthread_cond_signal(&progress_cond);
1614 progress_unlock();
1616 /*
1617 * We must not set ->data_ready before we wait on the
1618 * condition because the main thread may have set it to 1
1619 * before we get here. In order to be sure that new
1620 * work is available if we see 1 in ->data_ready, it
1621 * was initialized to 0 before this thread was spawned
1622 * and we reset it to 0 right away.
1623 */
1624 pthread_mutex_lock(&me->mutex);
1625 while (!me->data_ready)
1626 pthread_cond_wait(&me->cond, &me->mutex);
1627 me->data_ready = 0;
1628 pthread_mutex_unlock(&me->mutex);
1629 }
1630 /* leave ->working 1 so that this doesn't get more work assigned */
1631 return NULL;
1632 }
1634 static void ll_find_deltas(struct object_entry **list, unsigned list_size,
1635 int window, int depth, unsigned *processed)
1636 {
1637 struct thread_params *p;
1638 int i, ret, active_threads = 0;
1640 init_threaded_search();
1642 if (!delta_search_threads) /* --threads=0 means autodetect */
1643 delta_search_threads = online_cpus();
1644 if (delta_search_threads <= 1) {
1645 find_deltas(list, &list_size, window, depth, processed);
1646 cleanup_threaded_search();
1647 return;
1648 }
1649 if (progress > pack_to_stdout)
1650 fprintf(stderr, "Delta compression using up to %d threads.\n",
1651 delta_search_threads);
1652 p = xcalloc(delta_search_threads, sizeof(*p));
1654 /* Partition the work amongst work threads. */
1655 for (i = 0; i < delta_search_threads; i++) {
1656 unsigned sub_size = list_size / (delta_search_threads - i);
1658 /* don't use too small segments or no deltas will be found */
1659 if (sub_size < 2*window && i+1 < delta_search_threads)
1660 sub_size = 0;
1662 p[i].window = window;
1663 p[i].depth = depth;
1664 p[i].processed = processed;
1665 p[i].working = 1;
1666 p[i].data_ready = 0;
1668 /* try to split chunks on "path" boundaries */
1669 while (sub_size && sub_size < list_size &&
1670 list[sub_size]->hash &&
1671 list[sub_size]->hash == list[sub_size-1]->hash)
1672 sub_size++;
1674 p[i].list = list;
1675 p[i].list_size = sub_size;
1676 p[i].remaining = sub_size;
1678 list += sub_size;
1679 list_size -= sub_size;
1680 }
1682 /* Start work threads. */
1683 for (i = 0; i < delta_search_threads; i++) {
1684 if (!p[i].list_size)
1685 continue;
1686 pthread_mutex_init(&p[i].mutex, NULL);
1687 pthread_cond_init(&p[i].cond, NULL);
1688 ret = pthread_create(&p[i].thread, NULL,
1689 threaded_find_deltas, &p[i]);
1690 if (ret)
1691 die("unable to create thread: %s", strerror(ret));
1692 active_threads++;
1693 }
1695 /*
1696 * Now let's wait for work completion. Each time a thread is done
1697 * with its work, we steal half of the remaining work from the
1698 * thread with the largest number of unprocessed objects and give
1699 * it to that newly idle thread. This ensure good load balancing
1700 * until the remaining object list segments are simply too short
1701 * to be worth splitting anymore.
1702 */
1703 while (active_threads) {
1704 struct thread_params *target = NULL;
1705 struct thread_params *victim = NULL;
1706 unsigned sub_size = 0;
1708 progress_lock();
1709 for (;;) {
1710 for (i = 0; !target && i < delta_search_threads; i++)
1711 if (!p[i].working)
1712 target = &p[i];
1713 if (target)
1714 break;
1715 pthread_cond_wait(&progress_cond, &progress_mutex);
1716 }
1718 for (i = 0; i < delta_search_threads; i++)
1719 if (p[i].remaining > 2*window &&
1720 (!victim || victim->remaining < p[i].remaining))
1721 victim = &p[i];
1722 if (victim) {
1723 sub_size = victim->remaining / 2;
1724 list = victim->list + victim->list_size - sub_size;
1725 while (sub_size && list[0]->hash &&
1726 list[0]->hash == list[-1]->hash) {
1727 list++;
1728 sub_size--;
1729 }
1730 if (!sub_size) {
1731 /*
1732 * It is possible for some "paths" to have
1733 * so many objects that no hash boundary
1734 * might be found. Let's just steal the
1735 * exact half in that case.
1736 */
1737 sub_size = victim->remaining / 2;
1738 list -= sub_size;
1739 }
1740 target->list = list;
1741 victim->list_size -= sub_size;
1742 victim->remaining -= sub_size;
1743 }
1744 target->list_size = sub_size;
1745 target->remaining = sub_size;
1746 target->working = 1;
1747 progress_unlock();
1749 pthread_mutex_lock(&target->mutex);
1750 target->data_ready = 1;
1751 pthread_cond_signal(&target->cond);
1752 pthread_mutex_unlock(&target->mutex);
1754 if (!sub_size) {
1755 pthread_join(target->thread, NULL);
1756 pthread_cond_destroy(&target->cond);
1757 pthread_mutex_destroy(&target->mutex);
1758 active_threads--;
1759 }
1760 }
1761 cleanup_threaded_search();
1762 free(p);
1763 }
1765 #else
1766 #define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)
1767 #endif
1769 static int add_ref_tag(const char *path, const unsigned char *sha1, int flag, void *cb_data)
1770 {
1771 unsigned char peeled[20];
1773 if (!prefixcmp(path, "refs/tags/") && /* is a tag? */
1774 !peel_ref(path, peeled) && /* peelable? */
1775 !is_null_sha1(peeled) && /* annotated tag? */
1776 locate_object_entry(peeled)) /* object packed? */
1777 add_object_entry(sha1, OBJ_TAG, NULL, 0);
1778 return 0;
1779 }
1781 static void prepare_pack(int window, int depth)
1782 {
1783 struct object_entry **delta_list;
1784 uint32_t i, nr_deltas;
1785 unsigned n;
1787 get_object_details();
1789 /*
1790 * If we're locally repacking then we need to be doubly careful
1791 * from now on in order to make sure no stealth corruption gets
1792 * propagated to the new pack. Clients receiving streamed packs
1793 * should validate everything they get anyway so no need to incur
1794 * the additional cost here in that case.
1795 */
1796 if (!pack_to_stdout)
1797 do_check_packed_object_crc = 1;
1799 if (!nr_objects || !window || !depth)
1800 return;
1802 delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1803 nr_deltas = n = 0;
1805 for (i = 0; i < nr_objects; i++) {
1806 struct object_entry *entry = objects + i;
1808 if (entry->delta)
1809 /* This happens if we decided to reuse existing
1810 * delta from a pack. "reuse_delta &&" is implied.
1811 */
1812 continue;
1814 if (entry->size < 50)
1815 continue;
1817 if (entry->no_try_delta)
1818 continue;
1820 if (!entry->preferred_base) {
1821 nr_deltas++;
1822 if (entry->type < 0)
1823 die("unable to get type of object %s",
1824 sha1_to_hex(entry->idx.sha1));
1825 } else {
1826 if (entry->type < 0) {
1827 /*
1828 * This object is not found, but we
1829 * don't have to include it anyway.
1830 */
1831 continue;
1832 }
1833 }
1835 delta_list[n++] = entry;
1836 }
1838 if (nr_deltas && n > 1) {
1839 unsigned nr_done = 0;
1840 if (progress)
1841 progress_state = start_progress("Compressing objects",
1842 nr_deltas);
1843 qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
1844 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
1845 stop_progress(&progress_state);
1846 if (nr_done != nr_deltas)
1847 die("inconsistency with delta count");
1848 }
1849 free(delta_list);
1850 }
1852 static int git_pack_config(const char *k, const char *v, void *cb)
1853 {
1854 if (!strcmp(k, "pack.window")) {
1855 window = git_config_int(k, v);
1856 return 0;
1857 }
1858 if (!strcmp(k, "pack.windowmemory")) {
1859 window_memory_limit = git_config_ulong(k, v);
1860 return 0;
1861 }
1862 if (!strcmp(k, "pack.depth")) {
1863 depth = git_config_int(k, v);
1864 return 0;
1865 }
1866 if (!strcmp(k, "pack.compression")) {
1867 int level = git_config_int(k, v);
1868 if (level == -1)
1869 level = Z_DEFAULT_COMPRESSION;
1870 else if (level < 0 || level > Z_BEST_COMPRESSION)
1871 die("bad pack compression level %d", level);
1872 pack_compression_level = level;
1873 pack_compression_seen = 1;
1874 return 0;
1875 }
1876 if (!strcmp(k, "pack.deltacachesize")) {
1877 max_delta_cache_size = git_config_int(k, v);
1878 return 0;
1879 }
1880 if (!strcmp(k, "pack.deltacachelimit")) {
1881 cache_max_small_delta_size = git_config_int(k, v);
1882 return 0;
1883 }
1884 if (!strcmp(k, "pack.threads")) {
1885 delta_search_threads = git_config_int(k, v);
1886 if (delta_search_threads < 0)
1887 die("invalid number of threads specified (%d)",
1888 delta_search_threads);
1889 #ifdef NO_PTHREADS
1890 if (delta_search_threads != 1)
1891 warning("no threads support, ignoring %s", k);
1892 #endif
1893 return 0;
1894 }
1895 if (!strcmp(k, "pack.indexversion")) {
1896 pack_idx_default_version = git_config_int(k, v);
1897 if (pack_idx_default_version > 2)
1898 die("bad pack.indexversion=%"PRIu32,
1899 pack_idx_default_version);
1900 return 0;
1901 }
1902 if (!strcmp(k, "pack.packsizelimit")) {
1903 pack_size_limit_cfg = git_config_ulong(k, v);
1904 return 0;
1905 }
1906 return git_default_config(k, v, cb);
1907 }
1909 static void read_object_list_from_stdin(void)
1910 {
1911 char line[40 + 1 + PATH_MAX + 2];
1912 unsigned char sha1[20];
1914 for (;;) {
1915 if (!fgets(line, sizeof(line), stdin)) {
1916 if (feof(stdin))
1917 break;
1918 if (!ferror(stdin))
1919 die("fgets returned NULL, not EOF, not error!");
1920 if (errno != EINTR)
1921 die_errno("fgets");
1922 clearerr(stdin);
1923 continue;
1924 }
1925 if (line[0] == '-') {
1926 if (get_sha1_hex(line+1, sha1))
1927 die("expected edge sha1, got garbage:\n %s",
1928 line);
1929 add_preferred_base(sha1);
1930 continue;
1931 }
1932 if (get_sha1_hex(line, sha1))
1933 die("expected sha1, got garbage:\n %s", line);
1935 add_preferred_base_object(line+41);
1936 add_object_entry(sha1, 0, line+41, 0);
1937 }
1938 }
1940 #define OBJECT_ADDED (1u<<20)
1942 static void show_commit(struct commit *commit, void *data)
1943 {
1944 add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1945 commit->object.flags |= OBJECT_ADDED;
1946 }
1948 static void show_object(struct object *obj, const struct name_path *path, const char *last)
1949 {
1950 char *name = path_name(path, last);
1952 add_preferred_base_object(name);
1953 add_object_entry(obj->sha1, obj->type, name, 0);
1954 obj->flags |= OBJECT_ADDED;
1956 /*
1957 * We will have generated the hash from the name,
1958 * but not saved a pointer to it - we can free it
1959 */
1960 free((char *)name);
1961 }
1963 static void show_edge(struct commit *commit)
1964 {
1965 add_preferred_base(commit->object.sha1);
1966 }
1968 struct in_pack_object {
1969 off_t offset;
1970 struct object *object;
1971 };
1973 struct in_pack {
1974 int alloc;
1975 int nr;
1976 struct in_pack_object *array;
1977 };
1979 static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
1980 {
1981 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->sha1, p);
1982 in_pack->array[in_pack->nr].object = object;
1983 in_pack->nr++;
1984 }
1986 /*
1987 * Compare the objects in the offset order, in order to emulate the
1988 * "git rev-list --objects" output that produced the pack originally.
1989 */
1990 static int ofscmp(const void *a_, const void *b_)
1991 {
1992 struct in_pack_object *a = (struct in_pack_object *)a_;
1993 struct in_pack_object *b = (struct in_pack_object *)b_;
1995 if (a->offset < b->offset)
1996 return -1;
1997 else if (a->offset > b->offset)
1998 return 1;
1999 else
2000 return hashcmp(a->object->sha1, b->object->sha1);
2001 }
2003 static void add_objects_in_unpacked_packs(struct rev_info *revs)
2004 {
2005 struct packed_git *p;
2006 struct in_pack in_pack;
2007 uint32_t i;
2009 memset(&in_pack, 0, sizeof(in_pack));
2011 for (p = packed_git; p; p = p->next) {
2012 const unsigned char *sha1;
2013 struct object *o;
2015 if (!p->pack_local || p->pack_keep)
2016 continue;
2017 if (open_pack_index(p))
2018 die("cannot open pack index");
2020 ALLOC_GROW(in_pack.array,
2021 in_pack.nr + p->num_objects,
2022 in_pack.alloc);
2024 for (i = 0; i < p->num_objects; i++) {
2025 sha1 = nth_packed_object_sha1(p, i);
2026 o = lookup_unknown_object(sha1);
2027 if (!(o->flags & OBJECT_ADDED))
2028 mark_in_pack_object(o, p, &in_pack);
2029 o->flags |= OBJECT_ADDED;
2030 }
2031 }
2033 if (in_pack.nr) {
2034 qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2035 ofscmp);
2036 for (i = 0; i < in_pack.nr; i++) {
2037 struct object *o = in_pack.array[i].object;
2038 add_object_entry(o->sha1, o->type, "", 0);
2039 }
2040 }
2041 free(in_pack.array);
2042 }
2044 static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2045 {
2046 static struct packed_git *last_found = (void *)1;
2047 struct packed_git *p;
2049 p = (last_found != (void *)1) ? last_found : packed_git;
2051 while (p) {
2052 if ((!p->pack_local || p->pack_keep) &&
2053 find_pack_entry_one(sha1, p)) {
2054 last_found = p;
2055 return 1;
2056 }
2057 if (p == last_found)
2058 p = packed_git;
2059 else
2060 p = p->next;
2061 if (p == last_found)
2062 p = p->next;
2063 }
2064 return 0;
2065 }
2067 static void loosen_unused_packed_objects(struct rev_info *revs)
2068 {
2069 struct packed_git *p;
2070 uint32_t i;
2071 const unsigned char *sha1;
2073 for (p = packed_git; p; p = p->next) {
2074 if (!p->pack_local || p->pack_keep)
2075 continue;
2077 if (open_pack_index(p))
2078 die("cannot open pack index");
2080 for (i = 0; i < p->num_objects; i++) {
2081 sha1 = nth_packed_object_sha1(p, i);
2082 if (!locate_object_entry(sha1) &&
2083 !has_sha1_pack_kept_or_nonlocal(sha1))
2084 if (force_object_loose(sha1, p->mtime))
2085 die("unable to force loose object");
2086 }
2087 }
2088 }
2090 static void get_object_list(int ac, const char **av)
2091 {
2092 struct rev_info revs;
2093 char line[1000];
2094 int flags = 0;
2096 init_revisions(&revs, NULL);
2097 save_commit_buffer = 0;
2098 setup_revisions(ac, av, &revs, NULL);
2100 while (fgets(line, sizeof(line), stdin) != NULL) {
2101 int len = strlen(line);
2102 if (len && line[len - 1] == '\n')
2103 line[--len] = 0;
2104 if (!len)
2105 break;
2106 if (*line == '-') {
2107 if (!strcmp(line, "--not")) {
2108 flags ^= UNINTERESTING;
2109 continue;
2110 }
2111 die("not a rev '%s'", line);
2112 }
2113 if (handle_revision_arg(line, &revs, flags, 1))
2114 die("bad revision '%s'", line);
2115 }
2117 if (prepare_revision_walk(&revs))
2118 die("revision walk setup failed");
2119 mark_edges_uninteresting(revs.commits, &revs, show_edge);
2120 traverse_commit_list(&revs, show_commit, show_object, NULL);
2122 if (keep_unreachable)
2123 add_objects_in_unpacked_packs(&revs);
2124 if (unpack_unreachable)
2125 loosen_unused_packed_objects(&revs);
2126 }
2128 static int adjust_perm(const char *path, mode_t mode)
2129 {
2130 if (chmod(path, mode))
2131 return -1;
2132 return adjust_shared_perm(path);
2133 }
2135 int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2136 {
2137 int use_internal_rev_list = 0;
2138 int thin = 0;
2139 int all_progress_implied = 0;
2140 uint32_t i;
2141 const char **rp_av;
2142 int rp_ac_alloc = 64;
2143 int rp_ac;
2145 read_replace_refs = 0;
2147 rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
2149 rp_av[0] = "pack-objects";
2150 rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
2151 rp_ac = 2;
2153 git_config(git_pack_config, NULL);
2154 if (!pack_compression_seen && core_compression_seen)
2155 pack_compression_level = core_compression_level;
2157 progress = isatty(2);
2158 for (i = 1; i < argc; i++) {
2159 const char *arg = argv[i];
2161 if (*arg != '-')
2162 break;
2164 if (!strcmp("--non-empty", arg)) {
2165 non_empty = 1;
2166 continue;
2167 }
2168 if (!strcmp("--local", arg)) {
2169 local = 1;
2170 continue;
2171 }
2172 if (!strcmp("--incremental", arg)) {
2173 incremental = 1;
2174 continue;
2175 }
2176 if (!strcmp("--honor-pack-keep", arg)) {
2177 ignore_packed_keep = 1;
2178 continue;
2179 }
2180 if (!prefixcmp(arg, "--compression=")) {
2181 char *end;
2182 int level = strtoul(arg+14, &end, 0);
2183 if (!arg[14] || *end)
2184 usage(pack_usage);
2185 if (level == -1)
2186 level = Z_DEFAULT_COMPRESSION;
2187 else if (level < 0 || level > Z_BEST_COMPRESSION)
2188 die("bad pack compression level %d", level);
2189 pack_compression_level = level;
2190 continue;
2191 }
2192 if (!prefixcmp(arg, "--max-pack-size=")) {
2193 pack_size_limit_cfg = 0;
2194 if (!git_parse_ulong(arg+16, &pack_size_limit))
2195 usage(pack_usage);
2196 continue;
2197 }
2198 if (!prefixcmp(arg, "--window=")) {
2199 char *end;
2200 window = strtoul(arg+9, &end, 0);
2201 if (!arg[9] || *end)
2202 usage(pack_usage);
2203 continue;
2204 }
2205 if (!prefixcmp(arg, "--window-memory=")) {
2206 if (!git_parse_ulong(arg+16, &window_memory_limit))
2207 usage(pack_usage);
2208 continue;
2209 }
2210 if (!prefixcmp(arg, "--threads=")) {
2211 char *end;
2212 delta_search_threads = strtoul(arg+10, &end, 0);
2213 if (!arg[10] || *end || delta_search_threads < 0)
2214 usage(pack_usage);
2215 #ifdef NO_PTHREADS
2216 if (delta_search_threads != 1)
2217 warning("no threads support, "
2218 "ignoring %s", arg);
2219 #endif
2220 continue;
2221 }
2222 if (!prefixcmp(arg, "--depth=")) {
2223 char *end;
2224 depth = strtoul(arg+8, &end, 0);
2225 if (!arg[8] || *end)
2226 usage(pack_usage);
2227 continue;
2228 }
2229 if (!strcmp("--progress", arg)) {
2230 progress = 1;
2231 continue;
2232 }
2233 if (!strcmp("--all-progress", arg)) {
2234 progress = 2;
2235 continue;
2236 }
2237 if (!strcmp("--all-progress-implied", arg)) {
2238 all_progress_implied = 1;
2239 continue;
2240 }
2241 if (!strcmp("-q", arg)) {
2242 progress = 0;
2243 continue;
2244 }
2245 if (!strcmp("--no-reuse-delta", arg)) {
2246 reuse_delta = 0;
2247 continue;
2248 }
2249 if (!strcmp("--no-reuse-object", arg)) {
2250 reuse_object = reuse_delta = 0;
2251 continue;
2252 }
2253 if (!strcmp("--delta-base-offset", arg)) {
2254 allow_ofs_delta = 1;
2255 continue;
2256 }
2257 if (!strcmp("--stdout", arg)) {
2258 pack_to_stdout = 1;
2259 continue;
2260 }
2261 if (!strcmp("--revs", arg)) {
2262 use_internal_rev_list = 1;
2263 continue;
2264 }
2265 if (!strcmp("--keep-unreachable", arg)) {
2266 keep_unreachable = 1;
2267 continue;
2268 }
2269 if (!strcmp("--unpack-unreachable", arg)) {
2270 unpack_unreachable = 1;
2271 continue;
2272 }
2273 if (!strcmp("--include-tag", arg)) {
2274 include_tag = 1;
2275 continue;
2276 }
2277 if (!strcmp("--unpacked", arg) ||
2278 !strcmp("--reflog", arg) ||
2279 !strcmp("--all", arg)) {
2280 use_internal_rev_list = 1;
2281 if (rp_ac >= rp_ac_alloc - 1) {
2282 rp_ac_alloc = alloc_nr(rp_ac_alloc);
2283 rp_av = xrealloc(rp_av,
2284 rp_ac_alloc * sizeof(*rp_av));
2285 }
2286 rp_av[rp_ac++] = arg;
2287 continue;
2288 }
2289 if (!strcmp("--thin", arg)) {
2290 use_internal_rev_list = 1;
2291 thin = 1;
2292 rp_av[1] = "--objects-edge";
2293 continue;
2294 }
2295 if (!prefixcmp(arg, "--index-version=")) {
2296 char *c;
2297 pack_idx_default_version = strtoul(arg + 16, &c, 10);
2298 if (pack_idx_default_version > 2)
2299 die("bad %s", arg);
2300 if (*c == ',')
2301 pack_idx_off32_limit = strtoul(c+1, &c, 0);
2302 if (*c || pack_idx_off32_limit & 0x80000000)
2303 die("bad %s", arg);
2304 continue;
2305 }
2306 if (!strcmp(arg, "--keep-true-parents")) {
2307 grafts_replace_parents = 0;
2308 continue;
2309 }
2310 usage(pack_usage);
2311 }
2313 /* Traditionally "pack-objects [options] base extra" failed;
2314 * we would however want to take refs parameter that would
2315 * have been given to upstream rev-list ourselves, which means
2316 * we somehow want to say what the base name is. So the
2317 * syntax would be:
2318 *
2319 * pack-objects [options] base <refs...>
2320 *
2321 * in other words, we would treat the first non-option as the
2322 * base_name and send everything else to the internal revision
2323 * walker.
2324 */
2326 if (!pack_to_stdout)
2327 base_name = argv[i++];
2329 if (pack_to_stdout != !base_name)
2330 usage(pack_usage);
2332 if (!pack_to_stdout && !pack_size_limit)
2333 pack_size_limit = pack_size_limit_cfg;
2334 if (pack_to_stdout && pack_size_limit)
2335 die("--max-pack-size cannot be used to build a pack for transfer.");
2336 if (pack_size_limit && pack_size_limit < 1024*1024) {
2337 warning("minimum pack size limit is 1 MiB");
2338 pack_size_limit = 1024*1024;
2339 }
2341 if (!pack_to_stdout && thin)
2342 die("--thin cannot be used to build an indexable pack.");
2344 if (keep_unreachable && unpack_unreachable)
2345 die("--keep-unreachable and --unpack-unreachable are incompatible.");
2347 if (progress && all_progress_implied)
2348 progress = 2;
2350 prepare_packed_git();
2352 if (progress)
2353 progress_state = start_progress("Counting objects", 0);
2354 if (!use_internal_rev_list)
2355 read_object_list_from_stdin();
2356 else {
2357 rp_av[rp_ac] = NULL;
2358 get_object_list(rp_ac, rp_av);
2359 }
2360 cleanup_preferred_base();
2361 if (include_tag && nr_result)
2362 for_each_ref(add_ref_tag, NULL);
2363 stop_progress(&progress_state);
2365 if (non_empty && !nr_result)
2366 return 0;
2367 if (nr_result)
2368 prepare_pack(window, depth);
2369 write_pack_file();
2370 if (progress)
2371 fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2372 " reused %"PRIu32" (delta %"PRIu32")\n",
2373 written, written_delta, reused, reused_delta);
2374 return 0;
2375 }